Datasets:

Daniel-P-Gonzalez

/

OCD

like 0

# Introduction Soil-transmitted helminths (STHs) are considered the most prevalent of the Neglected Tropical Diseases (NTDs). STHs are worms transmitted through soil contaminated with faecal matter. These worms cause infections due to lack of sanitation typically resulting from the practice of open defecation, lack of hygiene such as hand washing or walking barefoot on contaminated soil (for hookworm infection). The three main species of STH are roundworm (*Ascaris lumbricoides)*, the whipworm (*Trichuris trichiura)* and hookworm (*Ancylostoma duodenale* and *Necator americanus*). The infections caused by these worms occur largely in impoverished rural areas of sub-Saharan Africa, Latin America, Southeast Asia, and China. For instance, the global prevalence of *Trichuris trichiura* and *hookworm* reaches 790 million and 740 million respectively, and sub-Saharan Africa and China account for over 50% of hookworm prevalence. The global prevalence of *Ascaris lumbricoides* is over 1.2 billion, with China accounting for over 50% of cases. STH infections are known to affect all age groups. However, school-age children, particularly of the low-income communities, are the most vulnerable to infections due to poor nutrition, inadequate sanitation, and other factors that favour the survival of the parasites. The health consequences of STH infections may plunge children further from low-income neighbourhoods into poverty since infected children possibly have worse school performance. Given that infection intensity determines the severity of morbidities associated with STH infections, the treatment approach to STH infection is periodic drug treatment (deworming) to all children living in endemic areas with albendazole (400mg) or mebendazole (500mg). Specifically, drug treatment is recommended once a year if the prevalence of STHs is over 20% and twice a year if the prevalence of STHs is over 50%. Besides, health and hygiene education, as well as sanitation, is recommended as part of the STH control strategy. However, several previous studies have reported rapid rates of re-infection with STH infections soon after treatment, with roundworm and whipworm infections reoccurring in less than a year. Malnutrition by reducing the effectiveness of the immune response may increase susceptibility to SHT infections. Thus, nutritional supplementation has been regarded as a feasible means of controlling the morbidity of STH infections, considering that appropriate consumption of nutritional supplements plays a critical role in building up immune defences against pathogens. A strong immune system may potentially decrease infection intensity and consequently the chances of re-infection. An additional attraction of nutritional supplementation is their assumed safety over an extended period, an important parameter in pediatric treatments. Previous systematic reviews have concluded that improved nutritional status of individuals can be useful in reducing STH infections through nutritional supplementation intervention. However, these reviews did not address several issues. Firstly, the effect of a nutritional supplement on reducing re-infection rates of STH in school-aged children was not explored. Secondly, the effect of nutritional supplements on the re-infection rate of different worm species types was not established. Lastly, the length of the follow-up period after administering nutritional supplementation needed to observe recovery from the infection was not assessed. Our systematic review mitigated some of these shortcomings by focusing on the effects of nutritional supplements on species-specific re-infection rates in school-aged children assessed within different follow-up periods. School-aged children were chosen specifically since the burden of STH is high in this age group. # Aim This systematic review aimed to assess the effects of nutritional supplements on the re-infection rates and infection intensity of different STH species in infected school-age children. Considering an assumed link between malnutrition, education performance, and worm infections, the review also explores whether there is published evidence on the impact of nutritional supplements on nutritional status and education-related outcomes. # Methods This systematic review, based on a pre-defined protocol, was prepared in line with the Preferred Reporting Items for Systematic Reviews and Meta-analysis (PRISMA). ## Eligibility criteria The review included randomised controlled trials (RCTs) and cluster RCTs on school-aged children and adolescents defined by the World Health Organization as 5–17 years old. Since we wanted to investigate the direct cause-effect relationship between the interventions and study outcomes, we limited the review to studies of this design to minimize the risk of confounding factors. Trials investigating nutrition and or supplements, diet and food for the treatment of STH infections, such as fortified vitamins, multi-micronutrients, minerals, sugars, iron, sodium and iodine, were eligible. Included trials were those with a follow-up period of at least three months to enable assessment of long-term effects. Only studies written in the English language were included. ## Study outcome The primary outcomes of this review were infection rate of each STH species and infection intensity at different follow-up periods. The secondary outcomes were indicators of nutritional status (weight (kg), mid-upper arm circumference (MUAC-for-age)), school attendance and school productivity. We included markers of nutritional status as a secondary outcome because it is one of the main risk factors and consequences of STH infection development. ## Search strategy The search strategy was developed together with a qualified librarian. Initially, we carried out a scoping search in the Scopus database to identify relevant keywords for the final search strategy. Scopus database was searched using a combination of “soil-transmitted helminths” and “nutrition”, and the first hundred results were screened. Besides that, a query search was conducted on PubReminer search tool by combining “soil-transmitted helminths”, “nutrition” and “reinfection” to identify MeSH terms, the most used words in titles and abstracts and the most active authors in the subject field. The identified search terms were used to systematically search Medline (via OvidSP), CENTRAL (via Cochrane Library), EMBASE, and EBSCO on 19^th April 2019 and African Index Medicus (AIM) on 12^th June 2019 (See Appendix 1) from inception. Additionally, we searched grey literature using ClinicalTrials.gov and EBSCO to identify ongoing or unpublished studies. The final search strategies for the systematic search were minimally adjusted for the compatibility of each database (Appendix 1). We imported the final search results into EndNote® online via Web of Science to organize references and identify duplicates. Afterwards, we thoroughly screened the reference lists of all included studies. ## Screening The study selection process followed the PRISMA flow diagram presented in the results section. All the titles/abstracts were liberally screened by the first author (including the abstracts unless there was a high confidence that exclusion criteria were not satisfied) with the full texts screened by two authors independently. Any disagreements were solved by consensus. ## Data extraction Following recommendations of the Cochrane Handbook for Systematic Reviews of Intervention, we designed a data extraction form in Microsoft Excel (version 16.27), validated it by the second reviewer, and piloted on a sub-sample of studies. The first author (AI) extracted data into the validated form, enlisting titles of included studies, study characteristics, and outcomes of interest at each of the follow-up periods. The second or last authors verified the extraction accuracy. Continuous outcome data included prevalence rates or weight (kg), the mean, standard deviations (SDs) and any other reported summary statistics such as medians and inter-quartile range. Where SDs were not available, they were derived from absolute p-values and sample sizes reported within the studies as described in the Cochrane handbook. Dichotomous outcomes included the number of participants with events in both treatment and control groups and their total sample sizes. ## Assessment of risk of bias in individual studies We used the Cochrane tool to assess the risk of bias (RoB) in original evidence. Two reviewers (AI and OE) independently assessed the risk of bias of all included studies following the criteria from the Cochrane RoB tool. The domains for the assessment using this tool include random sequence generation, allocation of sequence concealment, blinding of participants and personnel, incomplete outcome data, selective outcome reporting and other potential sources of bias. Each domain was assessed based on the judgment of “low risk of bias”, “high risk of bias” and “unclear risk of bias”, followed by a supporting statement underlying each judgment. As this review included cluster RCTs, we assessed additional sources of bias under methodological heterogeneity. These sources of bias included recruitment bias, baseline imbalance, loss of clusters, incorrect analysis and comparability with RCTs. ## Assessment of heterogeneity We assessed methodological heterogeneity across the included studies by exploring the differences in the underlying factors leading to statistical heterogeneity such as study designs, length of follow-up, risk of bias and reported outcomes between studies. A meta-analysis employing the random-effect model was used to assess clinical heterogeneity across studies to provide a more conservative estimate and minimise bias within studies. Additionally, when clinical and methodological variations were too high across studies to combine the estimates, we used a narrative synthesis to summarize the results of the intervention effects. Where meta-analysis was possible, we considered statistical heterogeneity by inspecting forest plots and I² statistic values. I² values of 0% indicated no heterogeneity, values less than 50%—moderate heterogeneity, and values greater than 50%—substantial heterogeneity in synthesized outcomes. ## Summary measures of treatment effects Where meta-analysis is possible, the protocol included the following summary measures: (a) the mean difference (MD) for continuous outcomes reported on the same scale, standardized mean difference for outcomes reported on different scales, and risk or odds ratios for dichotomous data. Statistical significance was considered for outcome measures with a p-value \<0.05. We used recommendations of the Cochrane Handbook to interpret the size of the effect of interventions: values of 0.2, 0.5 and 0.8 represented small, moderate and large effects respectively. ## Method of analysis We used Review Manager (RevMan) 5.3 to analyze trials’ data, categorized by baseline prevalence of infection (high/moderate/low) and infection intensity (heavy/moderate/light) for each of the three STH species separately. For similar nutritional interventions, the results for each STH species infection were pooled separately. Each trial was presented at the longest reported follow-up periods for each STH species. Considering the limited number of studies retrieved we included in meta-analysis both RCTs and cluster-RCTs, while analyzing an impact of exclusion of cluster RCTs from the quantitative synthesis. We used a narrative synthesis to summarize the results that could not be included in the meta-analysis. Due to the observed variations in the length of follow-up across individual studies, the effect of length of follow-up on the intervention’s effect was explored in a subgroup analysis. For studies that reported multiple treatment groups, we combined the multiple treatment groups into one to create a single- pairwise comparison as recommended by the Cochrane handbook, where pooling of results was required. ## Subgroup analysis Subgroup analysis investigated how the length of follow-up affects the re- infection rates of each STH species across studies. The follow-up periods to be analysed using random-effects meta-analysis included 3–5 months, 6–9 months and 10–12 months. Also, subgroup analyses for weight, MUAC-for-age, school attendance, and school productivity outcomes were planned to be conducted. ## Sensitivity analysis Sensitivity analysis examined the effect of including trials that were cluster- randomized on the pooled effects. Another sensitivity analysis evaluated any changes in intervention effects using a fixed-effects model compared to the random-effects model. # Result ## Search results and study selection The literature search identified 1,816 studies from all electronic databases. Five other studies were identified from screening the reference lists. After deleting the duplicates, 1,350 studies remained. Details of the study selection process are presented in a PRISMA flow diagram. ## Characteristics of the included studies A total of six studies were included in this review. Five of the studies included in this review were RCTs except for one, which was a cluster RCT. There was variability in the follow-up data used within trials. While the most common follow-up period was three and six months, the longest follow-up period across all studies was up to twelve months. The trials were published between 2000 and 2016, each conducted in five different countries: Kenya, Zambia, Sri Lanka, Malaysia, and Cambodia. The six trials with sufficient reporting included participants with 4,272 children at baseline. All trials also included male and female participants, with a majority of them being female. The participants’ ages ranged between 7–18 years. However, one of the trials did not report any age range of its participants. All trials included children screened for at least one of the worm infestations, regardless of their intestinal worm load. Only three trials reported data for all three STH species. ## Interventions All trials assessed different nutritional supplementation interventions, and none of the trials used the same dosage. The most commonly used intervention, analysed in three studies, was iron supplements. One study used vitamin A supplement which was supplemented with micronutrient fortified food. In four out of six included trials, participants in both arms received either single-dose albendazole or mebendazole before and/or after nutritional supplements were administered. ## Comparators Each trial had a placebo group as a comparator. One study gave additional fortified food to its placebo group, and in two studies, additional doses of mebendazole were given to the placebo groups. ## Outcome measures All studies reported data separately for each STH species. Three studies measured the actual re-infection rates as percentages with corresponding 95% confidence intervals (CIs), prevalence rates and infection intensity. Two studies reported only the prevalence rates and infection intensities to define the strength of re-infection for each STH species. Secondary outcomes were not reported in the majority of the included studies. No study reported weight changes, MUAC-for-age, or school productivity (standard test performance), and only one study reported data on school attendance. ## Risk of bias of individual studies The risk of bias in included trials is presented in. The overall risk of bias in five included studies was low, though one study was considered of unclear risk of bias because of poor reporting. Attrition bias due to missing data and other biases were the main threats for the validity of the outcomes. ## Effect of nutritional supplements All the studies except one (Ebenezer et al., 2013) were included in the meta- analysis to summarise the re-infection rate among children. The exclusion of this study was to avoid the result being confounded by the effect of deworming since the study added deworming in the intervention arm but not in the control arm. Infection intensity was summarised using narrative synthesis for all included studies due to the presence of high statistical heterogeneity in the reported outcomes and missing relevant data (See for further details). A narrative synthesis was used to report the results of single trials on the use of vitamin A. ### 1. Effect of nutritional supplements on re-infection rates of Ascaris lumbricoides *Iron*. Two studies reported sufficient data to assess a pooled effect of using iron supplements to decrease *Ascaris lumbricoides* re-infection rate. The meta- analysis of these studies did not report statistically significant effectiveness of the intervention with the average effect size of 1.00 (-13.61, 15.62). *Vitamin A*. Only one study reported the use of vitamin A supplements to decrease the re-infection rate of *Ascaris lumbricoides* infection, hence a meta-analysis was not possible. While the authors reported a slight decrease in the re-infection rates of *Ascaris lumbricoides* at three months of follow-up in the treatment group compared to the control group, this difference was not significant at six months (p-value = 0.453). Thus, the author suggested that the observed effects were a result of the antihelminthic drugs administered after baseline data were collected. Furthermore, the study reported that rates of re- infection had fallen back to the baseline rates towards the end of the intervention. *Multi-micronutrients*. Two studies reported using multimicronutrients as a single intervention to reduce re-infection rate of *Ascaris lumbricoides* infections. These studies were not able to prove the effectiveness of the intervention with the average effect size of -0.30(-5.23, 4.63). ### 2. Effect of nutritional supplements on re-infection rates of Trichuris trichiura *Iron*. Two studies reported using iron supplements as an intervention to decrease re-infection rates of *Trichuris trichiura* infection. However, a statistically significant effect was reported only by cluster RCT of Ebenezer et al (2013), in which an iron supplement was assessed in combination with mebendazole. *Vitamin A*. Only one study analysed effects of vitamin A supplement, concluding on no significant difference in the prevalence rate of *Trichuris trichiura* between the two treatment groups at six months follow-up (p\>0.05). The study, however, reported a slight decrease in the re-infection rate with *Trichuris trichiura* at three but not the six months of the follow-up. The comparisons of the re-infection rates with baseline values and interpretation of the findings by the authors for this infection were identical to those on *Ascaris lumbricoides*. The infection intensities between the two intervention groups were also similar at follow-up and were described as heavy (p = 0.847). *Multi-micronutrients*. Only a study with unclear risk of bias reported the use of multi-micronutrients as single interventions to decrease the re-infection rate of *Trichuris trichiura* infections, reporting no statistically significant difference in reinfection rates and reinfection intensities between the intervention and placebo arms. ### 3. Effect of nutritional supplements on re-infection rates of hookworm *Iron*. An RCT and a cluster-RCT reported the use of iron supplements as an intervention to decrease the re-infection rate of hookworm infection. Both studies failed to show a decrease in re-infection rates of *hookworm* infections in the treatment group compared to the control group. *Vitamin A*. One study reported the use of vitamin A supplements on re-infection rates with *hookworm* infections. Hence not enough data was available to perform a meta-analysis. In the study, no significant differences were reported in the re-infection rates between the two treatment groups at six months follow up (p-value = 0.411). Overall, the limited quantity of available evidence suggests that vitamin A does not decrease the re-infection rate of hookworm. Furthermore, the length of follow-up in this study was not sufficient to determine if the intervention has a lasting effect on the re-infection rates. *Multi-micronutrients*. Two trials reported the use of multi-micronutrient (fortified rice and multi-micronutrient tablets) as a treatment approach to decrease the re-infection rate of hookworm infections. As shown in, both studies failed to show a decrease in re-infection rates of *hookworm* infections in the treatment group compared to the control group reporting an average effect size value of 0.18 (0.06,0.30). ### 4. Effect of nutritional supplements on weight and MUAC-for-age None of the included studies assessed weight and MUAC-for-age of the study participants as an outcome measure. ### 5. Effect of nutrition on school attendance and school productivity Only one study assessed and reported school attendance of its participants. The results reported no difference in school attendance at follow-up between the two intervention groups. No study included in this review assessed school productivity of its participants. ## Adverse outcomes None of the included studies reported data on any adverse outcomes. ## Supplementary analysis For all three STH species, the effect of the nutritional supplements at the different follow-up periods, i.e. 3 to 12 months were summarised as inconclusive (Figs –). The effect of the nutritional supplements failed to reach statistical significance across the three different follow-up periods. It was not feasible to carry out the planned subgroup analysis by weight, MUAC-for-age, school attendance and school productivity outcomes due to insufficient data. ## Sensitivity analysis We could not compare the results of the RCTs and the cluster-RCTs study because of an insufficient number of studies identified; expectedly, the cluster-RCT tend to report a higher impact of an intervention (iron) on STH re-infection rate. Using fixed-effect versus to random-effect model did not change the conclusions derived in this review (Figs i–k). # Discussion This review shows that nutritional supplements did not significantly reduce the re-infection rate of the different STH species. This effect is apparent in the observed wide confidence interval from the meta-analysis, which suggests that the effect of nutritional supplementation interventions is too small to be clinically relevant. It is also likely that the limited number of studies used in this review contributed to the inconclusive effect of nutritional supplementation interventions. While findings of our review do not encourage nutritional supplements as a deworming intervention among children, they are especially important considering the previous contradictory evidence. With children being the most vulnerable and infected group with STH infections, little evidence exists on the effect of nutritional supplements on the re-infection rate of the different STH species. The World Health Organization, alongside the Food and Agriculture Organization of the United Nations and the United Nations Environment Programme, has long advocated for the need to create safe and urgent interventions to control the incidence of worm infections, particularly among children. Previous studies have reported that the impact of infections is dependent on the worm load and nutrition status of the hosts and that some multi-micronutrients have important effects on the developing immune system. Hence, receiving the right nutrition in early lives may support building children’s immune systems against any form of diseases. Our review though demonstrates that promotions of the nutritional supplements as a safe and effective intervention among children for the treatment of soil-transmitted helminthic infections require stronger supporting evidence. The results of our synthesis are partially contradictory to a previous systematic review on this topic. Yap et al (2014) reported a positive odds ratio effect of 0.75 (0.54,1.05) of iron supplementation interventions in decreasing re-infection rates with *Ascaris lumbricoides* infection. This difference in results and conclusion could be related to the differences in methodologies of both reviews. Yap et al (2014) applied broader eligibility criteria, enabled them to retrieve and include up to 15 studies (three times more than the number of studies included in this review). The population of our review was limited to only school-aged children, while the previous review considered participants of all age groups, including both pre-school and school-aged children, adolescents, and adults. Thus, the observed effects in the previous review may be as a result of the combined intervention effects in other population groups. Children groups are more likely to be prone to re-infection with STH species compared to adults, possibly because of their poor sanitation and hygiene management. Besides, all the included trials in this review were conducted in rural areas, where proper sanitation practices may not exist and thus contributed to the weak effect of the nutritional supplements. Hence, reviews involving adult participants may show better effects of nutritional supplements compared to reviews using only children participants. Secondly, Yap et al (2014) considered both RCTs and prospective cohort studies in their synthesis, while this review only used RCTs and cluster-RCTs in the data analysis. Prospective cohort trials are prone to various types of bias, including selection bias, information bias, and confounding. Hence, including only RCTs in this review increases the certainty in its findings. Thirdly, both reviews performed data analysis at the longest reported follow-up periods. However, in this review, additional data analysis was done at baseline and different follow-up periods. The study authors reported slightly lower re-infection rates at the first follow-up periods, but this effect diminished towards the end of the intervention follow-up periods. Besides, the previous study did not consider the infection status of its participants in its data analysis but rather their nutrition statuses. Thus, this consideration in the analysis of the previous review may have influenced the observed effect of iron supplementation on the different STH infections. Finally, the differences in the tools used in assessing the quality of the included studies may have affected the reporting of biases of the original evidence. ## Limitations There are several limitations in both the included individual studies and the review that may have influenced the findings of this review. Despite the developed comprehensive search, we acknowledge that certain articles may be missed by this systematic review. Evidence gathered in this review did not fully explore the effect of nutritional supplements on the re-infection rate of each STH species due to the presence of methodological heterogeneity across studies. As a result of methodological heterogeneity (varying reporting scales, follow-up periods, infection types) across studies, only four studies were included in the meta-analysis. It must be noted that applying the random-effects model to the few studies in this review can result in poor performance, leading to the observed wider CIs with compromised coverage probability. Furthermore, the use of a few studies could result in poor estimation of heterogeneity between studies. Secondly, none of the studies administered nutritional supplements according to the study participant’s worm loads. Participants with heavier worm loads are likely to require a larger dosage of intervention compared to those with lighter worm loads. Also, participants with heavier worm loads may absorb nutrients less efficiently than those with lighter ones, since the interactions of heavy infections affect the intestinal walls. With regards to the effects of a nutritional supplement on STH species, each participant was given the same type of nutrients regardless of the species infecting each individual. Thus, this may have affected the effectiveness of the interventions. Limiting this meta-analysis to school-based interventions to avoid reporting, selection, and loss to follow up bias, could also result in including a healthier population in the synthesis. More studies are needed to provide sufficient evidence for the recommendation of nutritional supplements as a deworming strategy in school-aged children since the small number of studies included in this review did not show that supplements decrease re-infection rate in this population group. Also, each study included in this review used different types of nutritional supplementation with varying dosages. Hence, the methodological heterogeneity of the included study could have possibly contributed to the observed negative result. The majority of the studies included in this review were RCTs reporting small sample sizes. Trials using large sample sizes have been reported to present more significant weight on intervention effects, by producing narrower CIs and more precise effect estimates compared to trials using small sample sizes. Therefore, larger RCTs are needed to confirm the effect of nutritional supplements on reducing re-infection rate of STHs. This meta-analysis targeted to synthesize RCTs as the most robust evidence; considering the limited evidence retrieved, a supplementary review of quasi-experimental studies would complement the reported analysis, though in this analysis we observed the difference in outcomes reported by randomization approach (truly-randomized versus cluster-randomized). ## Implications The finding of this review has implication for practice. Even though there is little evidence on the long-term effects of iron supplements to decrease re- infection rates of each STH species in infected children, the overall effects of nutritional supplements remain inconclusive. Also, the strength of evidence generated from this systematic review is too low to provide a base for policymakers to make recommendations at both national and international levels. Thus, a review of high-quality prospective cohort studies with a long follow-up would contribute to strengthening the conclusion on the impact of nutritional supplements on the re-infection rate of soil-transmitted helminths in children. More studies with larger samples are needed to confirm the potential long-term benefits of nutritional supplement interventions in children infected with STH infections. Further research should also consider the specific nutrition dosage required for each type of STH infection, as the different species could require different dosages for more effective results. Even though the included trials generally had low risks of bias, they were still prone to some design flaws such as inadequate randomisation, selective reporting and unblinded outcome assessments that may lead to biases and decrease the validity of results. More rigorous methods of randomisation methods can be applied to future studies to increase the reliability and statistical significance of the intervention effects on the outcomes of interest. For example, infected individuals have been shown to have significant differences in the intensity of infections. Hence, participants can be stratified according to their egg loads before random allocations are carried out. This may provide researchers with the bases for treatment dosage guidelines. The included trials were mainly RCTs randomized at the individual level and with small sample sizes. However, there is lack of consensus regarding the most effective type of trial design, considering the difficulty in reaching large sample sizes in RCTs. Hence, future studies could benefit by comparing the outcomes reported in truly RCT, cluster RCTs and factorial designs, as they are feasible to use on larger sample sizes. While cluster RCTs have been reported to have more potential to detect treatment effects in the most affected groups within the clusters, the risk of bias in such studies should be detailed, exploring the differences in outcomes reported. Furthermore, each type of STH species is likely to react to nutritional supplements in distinctive ways. Hence future research could also consider the implementation of nutritional supplement treatments according to the type of infection. Finally, no RCT reported adverse events related to the use of nutritional supplements in the deworming of children. Studies powered to assess negative impacts of nutritional supplements (for instance multimicronutrients) would be valued to highlight their possible negative impact on infection rate and health in general. # Conclusions This systematic review is the first to investigate the effects of nutritional supplements on the strength of STH species-specific re-infection rates in school-aged children. The current evidence gathered in this review is weak to conclude that nutritional interventions had an impact on the prevalence rates and infection intensities of each STH species. Thus, nutritional supplements for treatment of STH in children should not be encouraged unless better evidence emerges. Conclusion of earlier reviews on general populations may not necessarily apply to children since children possibly have a higher re-infection rate due to hygiene. # Supporting information 10.1371/journal.pone.0237112.r001 Decision Letter 0 Wieringa Frank Academic Editor 2020 Frank Wieringa This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 28 May 2020 PONE-D-20-08643 Effects of nutritional supplements on the re-infection rate of soil-transmitted helminths in school-age children: A systematic review and meta-analysis PLOS ONE Dear Dr. Ekwunife, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. Please pay due attention to the remarks of reviewer 1 on the discrepancy between the \# of children reported in the text and Figure 3 for the Ebenezer study.  Please ensure that your decision is justified on PLOS ONE’s [publication criteria](https://journals.plos.org/plosone/s/criteria-for-publication) and not, for example, on novelty or perceived impact. Please submit your revised manuscript by Jul 12 2020 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <plosone@plos.org>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript: A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols> We look forward to receiving your revised manuscript. Kind regards, Frank Wieringa, M.D., Ph.D. Academic Editor PLOS ONE Journal Requirements: When submitting your revision, we need you to address these additional requirements. 1\. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at <https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main _body.pdf> and <https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_titl e_authors_affiliations.pdf> 2\. Please ensure that you refer to Figures i, j and k in your text as, if accepted, production will need this reference to link the reader to each figure. \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Partly Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 2\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: No Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 3\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 4\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 5\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: -systematic review is well written, follows accepted guidelines for reporting (PRISMA) and is based on a pre-defined protocol. -what is the “liberal accelerated approach” to screening- this needs a reference and some detail -is the protocol referenced? -risk of bias is well-done according to accepted Cochrane standards -it is unclear why nutritional supplements would be expected to decrease reinfection rate of children- I see the authors have cited prior systematic reviews in this area, but I do not understand the biological/physiological rationale for why nutritional supplements would have this effect. I suggest the authors provide 2-3 sentences about why nutritional supplements are expected to have this effect -with respect to the statistical analysis and whether data support conclusions, the authors need to consider the independent effect of deworming in these studies. It appears that 3 of the studies had deworming in the intervention arm, and two did not. Also, two studies reported additional doses of mebendazole for the placebo group (page 14, line 302). I think that studies with deworming in the intervention arms should be analyzed separately from the studies which have no deworming. Without doing this, the results may be confounded by the effect of deworming on reinfection, and the conclusions may be misleading also For example, in the figure 3 of iron supplementation, Ebenezer 2013 is a study of iron+deworming vs no deworming, thus it is not surprising that it has a beneficial effect on reinfection rates. Figure 3- it is unclear why the Ebenezer study has less than 300 children in this analysis when the table of included studies indicates this study has over 1000 participants. Similarly, figure 4 shows only 79 participants from the Ebenzer study. These seem to indicate that the Ebenezer study sampled only a few children to conduct STH analysis-this should be described in the description of studies somehow. -Analysis and results- there is no assessment of publication bias. -the main findings do not consider the strength of evidence for the interpretation of results, which is a PRISMA reporting standard. ie standard 24: Summarize the main findings including the strength of evidence for each main outcome; consider their relevance to key groups (e.g., healthcare providers, users, and policy makers). -Results section, some statements are written in a way that presumes effectiveness, such as “The three studies were not 347 able to prove the effectiveness of the intervention with the average effect size of -0.15(-0.48, 348 0.17)”. I suggest authors review the manuscript to write results in a more neutral way. -Discussion section “Hence, including only RCTs in this review may have contributed to its high quality.” I disagree with this sentence. I think including only RCTs may increase the certainty of findings, but does not necessarily increase the quality of the systematic review. I suggest the authors reword. -Discussion- the discussion needs to consider that there may be other reasons to recommend nutritional supplements-ie to promote nutritional status. Thus I think the wording of sentences such as this one: “More studies are needed to provide sufficient evidence for the recommendation of nutritional supplements in school- aged children since the small number of studies included in this review did not show that nutritional supplements reduce the re-infection rate of the different STH species in school-aged children.” This suggests there are no benefits of nutritional supplementation, which seems to go beyond the review since this review focused only on reinfection rates Reviewer \#2: This is a an important manuscript, reporting important findings on the effects of nutritional supplements on re-infection rate of soil transmitted helminths. I do have some comments, please see attachment for my comments. \*\*\*\*\*\*\*\*\*\* 6\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No Reviewer \#2: No \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <figures@plos.org>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0237112.r002 Author response to Decision Letter 0 12 Jun 2020 Reviewer \#1: -systematic review is well written, follows accepted guidelines for reporting (PRISMA) and is based on a pre-defined protocol. -what is the “liberal accelerated approach” to screening- this needs a reference and some detail Reply: We corrected the text. -is the protocol referenced? Reply: The protocol was registered by the student ID 180250633 at the blackboard system of the University of Sheffield. Though since the access to the system is internal, we did not include this information as not useful for the reader. -risk of bias is well-done according to accepted Cochrane standards Reply: Thank you -it is unclear why nutritional supplements would be expected to decrease reinfection rate of children- I see the authors have cited prior systematic reviews in this area, but I do not understand the biological/physiological rationale for why nutritional supplements would have this effect. I suggest the authors provide 2-3 sentences about why nutritional supplements are expected to have this effect Reply: The rationale for this review is provided on lines 96 – 103, just before citing the previous reviews and their limitations. We included an additional sentence on relation of malnutrition and SHT infections (line 96). -with respect to the statistical analysis and whether data support conclusions, the authors need to consider the independent effect of deworming in these studies. It appears that 3 of the studies had deworming in the intervention arm, and two did not. Also, two studies reported additional doses of mebendazole for the placebo group (page 14, line 302). I think that studies with deworming in the intervention arms should be analyzed separately from the studies which have no deworming. Without doing this, the results may be confounded by the effect of deworming on reinfection, and the conclusions may be misleading also For example, in the figure 3 of iron supplementation, Ebenezer 2013 is a study of iron+deworming vs no deworming, thus it is not surprising that it has a beneficial effect on reinfection rates. Figure 3- it is unclear why the Ebenezer study has less than 300 children in this analysis when the table of included studies indicates this study has over 1000 participants. Similarly, figure 4 shows only 79 participants from the Ebenzer study. These seem to indicate that the Ebenezer study sampled only a few children to conduct STH analysis-this should be described in the description of studies somehow. Reply: We agree with the reviewer that Ebenzer study confounds the effect of deworming. We excluded this trial from meta-analysis and added the relevant statements into the Results section (lines 325-328). The second trial by Nchito et al 2009 does not bias the results because both intervention and control group did not have supplementary treatment while the intervention effect is measured in comparison to the control. The results of this trial are homogenous to other trials. Additionally, we corrected the number of participants in Ebenezer study although it was excluded in the meta-analysis. The initial mistake was because the authors presented the observed outcome as N which we mistakenly considered as number of participants. -Analysis and results- there is no assessment of publication bias. Reply: Because of the limited number of studies retrieved we were not able to assess publication bias, as it is instructed by the Cochrane Handbook 5.1.: “As a rule of thumb, tests for funnel plot asymmetry should be used only when there are at least 10 studies included in the meta-analysis, because when there are fewer studies the power of the tests is too low to distinguish chance from real asymmetry.” -the main findings do not consider the strength of evidence for the interpretation of results, which is a PRISMA reporting standard. ie standard 24: Summarize the main findings including the strength of evidence for each main outcome; consider their relevance to key groups (e.g., healthcare providers, users, and policy makers). Reply: The hierarchy of evidence places RCTs on the top of the pyramid with observational studies having much less strengths in comparison to quazi- experimental and even more experimental designs. This review included only RCTs with low risk of bias. Thus, comparison of the outcomes “by the strength of evidence” will not be applicable for this review, which is already based on strong evidence itself. Since almost no secondary outcomes were identified, we can neither detail nor conclude on impact of nutritional supplements on the secondary outcomes. We did though considered the uncertainty in results what is reflected in the Discussion section. For instance, the lines 446-449, mentioning the wide confidence intervals of the main outcomes. We also discussed the limitations of the included studies and the review itself (lines 515-532). The relevance of the outcomes to key groups are reported on lines 534-538. -Results section, some statements are written in a way that presumes effectiveness, such as “The three studies were not 347 able to prove the effectiveness of the intervention with the average effect size of -0.15(-0.48, 348 0.17)”. I suggest authors review the manuscript to write results in a more neutral way. Reply: This sentence refers to the objectives and the design of the included trials. Since we included trials of superiority (and not non-inferiority) design, the presumed effectiveness refers to the null hypothesis of the trials, attempting to measure a positive effect of interventions. -Discussion section “Hence, including only RCTs in this review may have contributed to its high quality.” I disagree with this sentence. I think including only RCTs may increase the certainty of findings, but does not necessarily increase the quality of the systematic review. I suggest the authors reword. Reply: We agreed with the reviewer and re-phrased this sentence. -Discussion- the discussion needs to consider that there may be other reasons to recommend nutritional supplements-ie to promote nutritional status. Thus I think the wording of sentences such as this one: “More studies are needed to provide sufficient evidence for the recommendation of nutritional supplements in school- aged children since the small number of studies included in this review did not show that nutritional supplements reduce the re-infection rate of the different STH species in school-aged children.” This suggests there are no benefits of nutritional supplementation, which seems to go beyond the review since this review focused only on reinfection rates Reply: We agree with the reviewer, the sentence referred to recommendations on deworming strategies. We re-phrased this sentence. Page 4: line 72-73; For the last part of the sentence: ‘and walking barefoot on contaminated soil’ is only the case for hookworm not forthe other helminths Reply: We added “for hookworm infection” to the text. Page 4: line 84: ‘The health consequences of STH infections may plunge further the children from low-income neighborhoods into poverty since infected children possibly have worse school performance’ I think it would be better to write: The health consequences of STH infections maytheplunge childrenfurther from low-income neighborhoods into poverty since infected children possibly have worse school performance Reply: We corrected the text. Page 6: line 136: the secondary outcome: interesting that the authors also want to investigate this. However how does this relateto theresearch question? I would suggest to either add this as a sub question in the introduction or explain better how this relatedto the RQ. Also this secondary outcome measure is not mentioned in the rest of the methods section. Reply: We added a secondary aim to the Introduction (lines 117-119), explaining the reason behind including the secondary outcomes. The methods section reports the review’s methodology independently on the outcomes (eligibility criteria, search strategy, data extraction, etc.). The lines 265-267 state that we planned to conduct the sub-group analysis for the secondary outcomes as well, though it was not possible because of insufficient number of studies retrieved. Page 7: it would be good to add the exact search terms that were used and also the dates that the searches were performed. Reply: We report the search dates on lines 152-154 and the detailed search strategy in the Appendix 1, referenced on lines 152-154. Page 9, figure 1: \- ‘additional records” ; i find this is a bit vague. elsewhere in the manuscript (page 13) the authors state that these record were identified from the reference list from the selected papers. It would be better tobe consistent. \- Full text excluded with reasons; this is also too vague. Please reformulate. Reply: The Figure 1 was corrected accordingly Page 12: line 276; the selection process is described to superficially. It would be good to elaborate on this process. Reply: We apologize for the mistake in this sentence. This sentence referred to a trial, included in abstract screening deviating from the protocol, though excluded afterwards because of a short follow up (\<3 months). The minimum age in all included studies was \>7 years (Table 1). We deleted the mentioned statement. Page 23: line 347: ‘the three studies were not able to prove’. Think a better formulation would be were not able to report. Prove is too strong in my opinion. Reply: The sentence was re-formulated. Page 27; line 415 multi-micronutrients; this is the only meta-analysis which is statistically significant, however this is not mentioned. I wonder why? Also this meta-analysis (even though based on 2 studies) shows that MM’s might have adverse effects. I think it would be good to give important this finding more attention. Please also elaborate on this important finding in the discussion. Reply: We revised the statistical analysis using mean difference instead of the standardized mean difference. An average effect size was 0.18 (0.06,0.30) for hookworm infections in the treatment group compared to the control group. While there is a small favorable difference towards placebo, we consider that this finding is dangerous to interpret directly as an evidence on the harmful effect of the intervention because of the superiority design of the trials (and not non-inferiority), especially considering that only two studies were identified. To address this comment of the reviewer, we added a sentence to the discussion on a value of additional evidence on this regard(lines 598-601). Page 28: line 428: effect on school productivity. This outcome is not mentioned before. Please also elaborate on this in the methods. Reply: School productivity was included in secondary outcomes (line 300) and was reported in Table 2. We clarified that it means standard test performance (line 301). For the discussion I would recommend to; \- Include the value of observationational (longitudinal) studies, as eventhough they may be more prone to confounding, they do often have longer followup time and may thusprovide more evidence. \- Give more attention to the possibility of unfavourable effect of MM supplementation. Reply: The requested statements were added to the Discussion section (lines 527-529). We added a sentence to the discussion on a value of additional evidence on this regard, considering that superiority design of the trials may not allow to retrieve direct conclusions on unfavorable effect of MM supplementation (lines 598-601). 10.1371/journal.pone.0237112.r003 Decision Letter 1 Wieringa Frank Academic Editor 2020 Frank Wieringa This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 21 Jul 2020 Effects of nutritional supplements on the re-infection rate of soil-transmitted helminths in school-age children: A systematic review and meta-analysis PONE-D-20-08643R1 Dear Dr. Ekwunife, We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements. Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication. An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at <http://www.editorialmanager.com/pone/>, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to- date. If you have any billing related questions, please contact our Author Billing department directly at <authorbilling@plos.org>. If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact <onepress@plos.org>. Kind regards, Frank Wieringa, M.D., Ph.D. Academic Editor PLOS ONE Additional Editor Comments (optional): Reviewers' comments: 10.1371/journal.pone.0237112.r004 Acceptance letter Wieringa Frank Academic Editor 2020 Frank Wieringa This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 24 Jul 2020 PONE-D-20-08643R1 Effects of nutritional supplements on the re-infection rate of soil-transmitted helminths in school-age children: A systematic review and meta-analysis Dear Dr. Ekwunife: I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department. If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact <onepress@plos.org>. If we can help with anything else, please email us at <plosone@plos.org>. Thank you for submitting your work to PLOS ONE and supporting open access. Kind regards, PLOS ONE Editorial Office Staff on behalf of Dr. Frank Wieringa Academic Editor PLOS ONE [^1]: The authors have declared that no competing interests exist.

# Introduction During mammalian preimplantation development, two sequential cell fate decisions occur that result in three cell populations (reviewed in). Upon the first decision, cells become either trophectoderm (TE) or inner cell mass (ICM) cells. Descendants of TE cells form the foetal portion of the placenta. The ICM cells make a further decision: they differentiate either into Epiblast (Epi) or into Primitive Endoderm (PrE). Epi cells predominantly give rise to the embryo proper while PrE cell descendants mainly generate the endodermal part of the yolk sac. In mice, three major processes have been proposed for ICM cell differentiation into Epi or PrE. In early blastocysts, ICM cells co-express Epi and PrE markers such as NANOG and GATA6, respectively. As time progresses, Epi and PrE progenitors arise. Epi progenitors express high levels of NANOG, and almost no GATA6, while PrE progenitors express high levels of GATA6, and almost no NANOG. FGF/MAPK signalling reinforces PrE commitment: Epi progenitors secrete FGF4, which binds to FGFR1 on Epi, and FGFR1 and FGFR2 on PrE biased cells. This results in a distribution of Epi and PrE progenitors in the ICM without an obvious spatial pattern. As development progresses, PrE progenitors migrate to occupy their position before the embryo implants. This results in the spatial segregation of the two lineages. PrE progenitors are polarized and positioned at the surface of the ICM, where they form an epithelium in contact with the blastocyst cavity or blastocoele. The Epi progenitors occupy a central position between TE and PrE. Epi versus PrE differentiation has been extensively studied in the context of marker expression dynamics and the involved signalling pathways (reviewed in). Technical developments have made it possible to study cell fate decisions during preimplantation mouse development at single-cell resolution (reviewed in). Invasive studies based on single cell transcriptomics have been used. However, transcriptomic techniques disrupt the cell positional information within the ICM. A complementary approach is single cell resolution imaging based on immunofluorescence stainings or fluorescent reporters. Combined with quantitative image analysis, the immunofluorescence approach provides protein expression levels together with cell positional information. Applying this technique in our recent study in mouse embryos and ICM organoids has revealed a local cell fate clustering during PrE differentiation. Here, we broaden the three-dimensional analysis of the spatial distribution of NANOG and GATA6 expressing cells in the ICMs of mouse embryos at different stages. We combine the positional information of a cell and its expression levels of NANOG and GATA6 with information of its neighbouring cells to obtain the local cell neighbourhood features and a global positional feature (see Terminology Box). We find a complex pattern in the three-dimensional cell neighbourhood in the ICM of early blastocysts, characterized by local positional features and expression level clusters (see Terminology Box). Highest NANOG expression levels are found in cells with a specific number of neighbours. The GATA6 level of a cell correlates with the levels of GATA6 in its neighbours, resulting in GATA6 expression level clusters. We apply a rule-based computer simulation to show that these two local cell neighbourhood features are sufficient to describe the complex population distribution found in early embryos. We further demonstrate that the simulations are also applicable in a *Nanog* mutant background. As potential regulators of the expression level clustering, we identify NANOG and FGF/MAPK signalling. Patterns in the global positional features start in mid blastocyst and are obvious in late blastocysts with *Nanog* regulating this feature in GATA6 expressing cells. In summary, we present three-dimensional cell neighbourhood analyses that allow a novel approach in the study of Epi versus PrE differentiation in relation to nearby cells and fate marker expression levels. Our results point at NANOG and FGF/MAPK-dependent mechanisms as responsible for the spatial arrangement of NANOG and GATA6 expressing cells in the ICM. These mechanisms become obvious in local cell neighbourhood features. Importantly, we present for the first time a signature that correlates with Epi precursor specification. # Materials and methods ## Ethics statement Mouse work was approved by the University of Bath Animal Welfare and Ethical Review Body (AWERB) and undertaken under UK Home Office license PPL 30/3219 in accordance with the Animals (Scientific Procedures) Act incorporating EU Directive 2010/63/EU. Additional mouse work was approved by the Consejería de Agricultura, Ganadería, Pesca y Aguas of the Gobierno de Canarias (CEEA-ULPGC 08/2018). ## Mice, embryos and immunohistochemistry Wild-type CD1 and *Nanog*^*+/+*, *Nanog*^*+/-* and *Nanog*^*-/-* embryos were generated by in-house breeding and natural mating. Detection of copulation plug confirmed successful mating; the resulting embryos were then considered embryonic day (E) 0.5. Embryos were isolated in M2 medium (Millipore). Embryos were prepared for immunofluorescence as previously described (89). Primary antibodies used were: anti-NANOG (eBiosciences 14–5761, 1:100), anti-GATA6 (R&D, AF1700, 1:200). Nuclei were stained using DAPI or Hoechst (1:1000, Invitrogen). Embryos were mounted on microscopy slides with Vaseline bridges to prevent their crushing. Three independent immunofluorescence stainings, each with E3.5 and E4.5 embryos from 7 litters, were performed for the first wild-type data set. *Nanog* mutant embryos were obtained as previously described and genotyped by NANOG antibody staining. ## Imaging and automated image analysis For the first wild-type data set ( and Tables, data I) a total of 45 embryos was imaged in four batches of 19, 15, 2 and 9 embryos. Images were acquired using a Zeiss LSM 510-META and a Plan-Apochromat 63x/1.4 Oil Ph3 objective, with optical section thickness of 1 μm. *Nanog* wild-type, heterozygous and mutant embryos ( and Tables, data III and IV) were similarly imaged in 5 confocal sessions using a Zeiss LSM700 and a Plan-Apochromat 40x/1.3 Oil DIC (UV) VIS-IR M27 objective. All images in each imaging session were obtained using the sequential scanning mode, with the same conditions of laser intensity, gain, and pinhole, and were processed in exactly the same way. The range indicator palette option (Zeiss AIM/ZEN software) was used to ensure that no oversaturated images were taken. For a schematic representation of the image and data pre-processing and further analysis, see also. The three-dimensional image stacks were segmented using MINS, cells were automatically assigned to ICM or TE, the features of the cell nuclei were extracted including the nuclear centroid and volume, together with the mean intensity of NANOG and GATA6 for each nucleus. The automatically assigned TE or ICM fate was manually checked ( Step1). Given the extension of the analysed data sets (over 27.000 cells) a manual correction of the segmentation results was not performed. Extreme errors (over-segmentation and pyknotic nuclei) in the segmentation were removed manually when correcting the classification of TE versus ICM. ## Data analysis The calculations were performed with *Mathematica* 11.1 (Wolfram Research). Details on the total number of embryos and cells in each population type analysed are shown in and Tables. For further details, see and. ## Pre-processing and staging of data from (data II) We used the embryos labelled as “littermate”, available from GitHub. This resulted in 147 additional data sets ( and Tables, data II). Compared to data I, the experimental setup was slightly different. Specifically, a different NANOG antibody was used and the embryos were imaged without being mounted. Given the extra thickness of the samples, the correction of fluorescent decay along the z-axis was required. Furthermore, while the same algorithm was used for the segmentation, a different thresholding method was applied to obtain the four populations (k-means clustering). We used the same cell number-based staging method as the one used for our data set, which resulted in 64 early, 34 mid and 49 late blastocysts. We excluded all NANOG and GATA6 levels from the distribution that were two standard deviations away from the respective mean as we noticed that there were some oversaturated nuclei images. The calculations were performed with *Mathematica* 11.1 (Wolfram Research). For further details, see. ## Cell graph generation and neighbourhood analyses We derived a cell graph representation to characterize the spatial distribution of the cells in each embryo in our wild-type data set (data I), 73 *Nanog*^*+/+* or *Nanog*^*+/-* embryos (data III) and 19 *Nanog*^*-/-* (data IV), and the data set from (data II) (Step 3(ii)). The calculations were performed with *Mathematica* 11.1 (Wolfram Research). For further details, see. ## Correlation analyses The Spearman’s correlation analysis and bootstrapping were performed in Matlab R2012b (MathWorks). The simulations of the null model were performed with *Mathematica* 11.1 (Wolfram Research). To classify the strength of the correlations we used the criteria by Evans: 1. 0.00–0.19: ‘very weak’ 2. 0.20–0.39: ‘weak’ 3. 0.40–0.59: ‘moderate’ 4. 0.60–0.79: ‘strong’ 5. 0.80–1.0: ‘very strong’. For further details, see. ## Analyses of global positional feature We aimed to investigate global patterns within the ICM at all three blastocysts stages. To this end, a reference point is needed. The one that immediately springs to mind is the embryo centroid. However, distributions of cells with respect to the embryo centroid would mainly highlight effects from the sorting process in mid and late embryos. Therefore, we decided to use the ICM centroid as reference point and analysed the expression levels of NANOG and GATA6 with respect to a cell’s distance to the ICM centroid. The ICM centroid was determined as the mean of the positions of all ICM cells. For the graphs, the distance in μm was binned into 5 μm intervals, which is the typical radius of an ICM cell. ## Rule-based simulations of population composition in ICM of early blastocysts The calculations were performed with *Mathematica* 11.1 (Wolfram Research). For further details, see. For the simulations shown in (*Nanog* mutant embryos), we use the cell positions, cells proportions from early *Nanog*^*+/+* or *Nanog*^*+/-* embryos (44 embryos; *p*_*GATA*6 = 94% and *p_NANOG* = 78%) and *startNumNeigh* = 9 to simulate the wild-type situation. For the *Nanog* mutants, we set the proportion of N+ cells (*p_NANOG*) to 0. ## Statistics For the comparison of expression levels, Mann-Whitney tests were applied as the distribution do not follow a Gaussian (performed in Matlab). To compare populations proportions, z-tests with Bonferroni corrections were applied. The used statistical test is indicated in the figure legends. ## Data accessibility For the segmentation and the Delaunay cell graph calculations, we used previously published tools, which can be obtained from the respective references. The code for the neighbourhood analyses is available as part of the electronic supplementary material. This further includes the data for data I, III and IV obtained from MINS as.csv files, as well as the processes data as.json files. Data II as well as V-VIII from MINS has previously been published and is available from. The processed data as.json files, codes and can be found at: <https://github.com/scfischer/fischer-et-al-2020>. # Results ## Pipeline for quantitative three-dimensional neighbourhood analyses of mouse preimplantation development In this study, we quantitatively analyse the three-dimensional spatial distribution of cell fate markers, taking into account the single cell levels as well as the levels of the neighbouring cells (Figs). The quantitative immunofluorescence (QIF) analysis of NANOG and GATA6 at the single cell level in mouse preimplantation embryos at different stages of development using MINS (Modular Interactive Nuclear Segmentation) provides the cell position within the embryo, the mean expression level per nucleus and the distinction between ICM and TE ((I-II) and, Step1). Our data set consists of 45 embryos from three independent experimental replicas, imaged in four confocal sessions, and staged based on the total cell number (early: 32–64, mid: 65–90, late: \>90 cells; and Tables, data I). Due to variability in the experimental and imaging setup, we observe quantitative differences between replicas ( Step 2). To correct this, we align the data according to NANOG and GATA6 threshold values for each experiment. Based on the common thresholds, we identify four discrete cell populations: double positive (DP: N+ and G6+), double negative (DN: N- and G6-), NANOG+/GATA6- (Epi progenitor) and NANOG-/GATA6+ (PrE progenitor). The proportions of the populations in the ICM at the different developmental stages show similar trends as previously published data. In particular, the proportion of DP cells decreases from early to mid to late blastocysts, and the proportion of PrE progenitor cells increases more than the proportion of Epi progenitor cells. To investigate the three-dimensional distribution of cells, we use the Delaunay Cell Graph (DCG) to approximate the nearest neighbours of each cell ( (III), S1 Step 3, see also and. The population analyses of TE cells neighbouring ICM cells and all TE cells shows that in early and mid embryos, the TE contains a large proportion of cells that is not DN and hence can provide FGF signalling to the ICM cells. Therefore, for our analysis, we consider all ICM cells and those TE cells that are neighbouring at least one ICM cell ( Step 4). Consistent with the presence of the blastocoele, we observe a higher number of neighbours, i.e. a higher cell density, close to the ICM centroid compared to the edge of the ICM irrespective of developmental stage and cell population type, except for DN cells in early and mid blastocysts (Figs). We also find that for all developmental stages, the cells of the different population types have comparable numbers of neighbours. Altogether, our data processing allows us to obtain for each ICM cell and its neighbours their position, the expression levels for NANOG and GATA6, the population type and the number of neighbours, resulting in a description of the local cell neighbourhood (see Terminology Box) (. Similar to previous work, the population analyses further show the presence of DN cells in the ICM already in early blastocysts with proportions increasing as development progresses. It has been proposed that these cells might correspond to Epi cells that have downregulated NANOG expression. In order to assess if this cell population has any identifiable pattern of neighbour number or location within the ICM, we analysed their distribution and neighbour number. While the other three cell population types have more neighbours closer to the ICM centroid, consistent with a higher cell density, we only find this pattern in the DN cell population in late embryos. This indicates that at the late stage, the DN cells could indeed be Epi cells that have downregulated NANOG. In early embryos, there is no clear pattern and the cell density around DN is comparable at any position within the ICM at this stage. Further studies will be needed to characterize the nature, origin and fate of these DN cells. In summary, we integrate QIF measurements with an approximation for the nearest neighbours of a cell to obtain a data set that enables studying three- dimensional local cell neighbourhoods in the ICM of early, mid and late mouse blastocysts. ## Clustering of population types is observed in the ICM of early embryos We have recently shown that ICM organoids and mid/late embryos show local clustering of cells with the same population type. Here, we extend this analysis to include early embryos (Figs and for statistical analyses). In the early blastocysts, around 40% of the neighbouring cells of an ICM cell are TE. Surprisingly, we already observe a pattern: the ICM neighbours of DP cells are mostly DP cells and the ICM neighbours of Epi progenitor are either TE or mostly Epi progenitor cells. In mid embryos, the clustering of DP cells and Epi progenitor cells remains, as previously observed. For late blastocysts, we observe the expected pattern. Epi cells have a large proportion of Epi neighbours and the lowest proportion of TE neighbours, as they occupy the internal positions. PrE cells have a larger proportion of TE neighbours and a large proportion of PrE neighbours. This is consistent with PrE cells forming an epithelium between the blastocoele and the Epi cells at this late stage with some of them already migrating (see late embryo in, and). Finally, the proportion of PrE neighbours increases for all four populations from early to mid blastocysts, consistent with the increase of PrE cells in the population distribution of the ICM. In summary, our results indicate population clustering of DP and Epi progenitor cells, which is already present in early embryos. ## DP cells in early blastocysts exhibit GATA6 expression level clusters Our analyses show a novel population clustering of DP and Epi progenitor cells in early blastocysts. To investigate this further, we step back from the discrete categorisation of the cells and their neighbours into population types based on high or low expression of NANOG and GATA6. Instead, we consider NANOG and GATA6 expression levels as continuous parameters. This approach allows us to measure the correlation strength of the levels of NANOG or GATA6 in a cell with the NANOG or GATA6 levels of its neighbours, respectively. We chose to use Spearman’s correlation coefficient as it requires less assumptions, i.e. it does not require bivariate normal data and it measures monotonic, not just linear relationships like the Pearson´s correlation coefficient does ( and, see for the classification of correlation strengths, and also for further details on the analysis). Given the dependence of the validity of a correlation analysis on cell number and the typically low number of cells present in mouse preimplantation embryos, we perform a correlation sensitivity analysis. This sensitivity analysis shows that more than 108 cells per cell population are required to obtain robust results with less than 3% variability. For completeness, we included all results in the following plots. However, those that are obtained from less than 108 cells are marked with stripes and should not be relied upon. The correlation analysis of the GATA6 level in a cell and the median GATA6 level of its neighbours indicates strong positive correlations in DP cells in early and moderate in PrE progenitor cells in late blastocysts. The neighbourhood correlation analyses of NANOG result in no correlation, very weak or weak correlations for all populations in early and late blastocysts. To ensure that the positive correlation results do not arise randomly or are affected by specific constraints on NANOG/GATA6 distributions, we also investigated whether the correlation values are significantly different from those of a null model (see for further details about the null models tested and used). We find that the neighbourhood correlations in the null model are significantly lower than those found for DP cells in early blastocysts and Epi progenitor as well as PrE progenitor cells in late blastocysts. Hence, the correlation values observed for GATA6 do not arise randomly. This is the first time that local correlations of GATA6 levels have been documented for early blastocysts. It has been proposed that Epi fate reinforces PrE fate in neighbouring cells via FGF4. This hypothesis would predict a positive correlation of NANOG levels of a cell with GATA6 levels in its neighbouring cells and GATA6 levels of a cell and NANOG levels in its neighbours. However, this hypothesis cannot be tested in this data set as all cell populations in mid blastocysts contain less than 108 cells (; see below and for more on this issue). These results suggest that DP cells in the ICM form GATA6 expression level clusters. This local distribution of DP is first present in the early blastocysts. Furthermore, NANOG expressing cells do not cluster according to their expression levels. ## NANOG expression levels correlate with the number of neighbours in early blastocysts We next test whether Epi progenitor clustering is related to positional information within the ICM. To do this, we tested whether there are local or global positional features related to NANOG and GATA6 expression levels. To investigate the local positional features, we analysed the relation between expression levels and total number of neighbours. We observe, particularly in early blastocysts, a peak in NANOG expression in cells with 7 to 14 neighbours, with the maximal level in cells with 9 neighbours. Three-dimensional illustrations of the number of neighbours and the NANOG levels in the ICMs of the individual early blastocysts support this finding. Performing the same analyses for GATA6 expression levels, we detected no clear relation at any stage. The number of neighbours favouring higher NANOG expression levels in a cell might be an artefact of our DCG approach. To assess this, we performed a sensitivity analysis of the DCG (see for further details). It demonstrates that the DCG does not favour a 9 neighbour topology for the ICMs. In early embryos, the DCG favours 2–4 neighbours, which is lower than what we observe in our results for the NANOG analysis. To investigate further the robustness of our results, we analysed whether there was a specific effect from the particular geometry of the embryos. Therefore, we compared the results for expression level versus number of neighbours with those for the null model, introduced for the correlation analysis (and see for further details). We find that in the null model, the expression levels of NANOG or GATA6 do not correlate with the number of neighbouring cells at any stage. For the global positional features, we investigate the relation between expression levels and the position of a cell relative to the centroid of the ICM (for the null model results, S7C for the statistical analysis of the experimental data, and Materials and Methods). Only in late blastocysts, we observe statistically significant higher NANOG levels closer to the centroid and lower levels away from it. This is consistent with Epi cells being located at the centre. The equivalent analysis for GATA6 expressing cells also shows a global positional effect: the highest GATA6 expressing cells are located distally from the ICM centroid in late blastocysts, consistent with their final position facing the blastocoele. In summary, these results show that there is a clear interrelation between the number of neighbouring cells and NANOG expression levels in early blastocysts. Conversely, the number of neighbours does not correlate with GATA6 expression levels. A global pattern is only present in the ICM of late embryos, likely coincident with the resolution of the sorting process. ## GATA6 level clusters arise in DP and PrE progenitor cells in early and mid blastocysts In the next step, we extend our analysis to a larger data set (, and Tables, data II). This allows us to ensure the robustness of our observation. Furthermore, we are able to extend the correlation analysis to the populations for which our initial data (data I) did not include enough cells. The larger data set was generated in a different laboratory with a slightly different experimental set up (see for details). Factors like the variability of the data or the presence of outliers affect the value of a correlation coefficient (, see also Sup. Info for further details). Therefore, we expect qualitative but not quantitative agreement between the results for data I and data II. We find that the main conclusions from our previous observations also hold in this second data set: in early blastocysts, we obtain GATA6 expression level clusters in DP cells and NANOG levels are highest in cells with nine neighbours. NANOG expression level clusters and a correlation between number of neighbours and GATA6 levels was not observed. Interestingly, in this second data set, we further find moderate positive correlations for GATA6 in DP cells in mid blastocysts as well as PrE progenitor cells in early and mid blastocysts. These correlations do not arise randomly (comparison with null model) nor are they an artefact of inter-embryo variability. In addition, we could test the hypothesis of Epi fate reinforcing PrE fate in neighbouring cells. The correlation between GATA6 levels in a cell and NANOG levels in its neighbours is very weak (or weak in the DP cells) and between NANOG levels in a cell and GATA6 in its neighbours is weak or moderate. Hence Epi fate reinforcing PrE fate in neighbouring cells is still an outstanding question. These results indicate that even if Epi cells can promote PrE fate in the neighbours, the mechanism does not rely on a direct translation of the levels of NANOG in a cell to the GATA6 levels expressed in its neighbours. Furthermore, we observe a global pattern for NANOG expression in mid and late embryos. In mid embryos, cells at the edge of the ICM have the highest expression of NANOG. In late embryos, we get a pattern with NANOG expression highest in the centre of the ICM and GATA6 expression highest in cells at the edge of the ICM ( and for statistical analysis). Taken together with the results from data I, we conclude that a clear global pattern starts to arise in mid blastocysts. Altogether, our results reveal a novel three-dimensional pattern in the distribution of the population types in ICM cells in early blastocysts. Local positional features and local expression level features characterize this pattern (Figs – and – Figs). ## Two simple rules can generate the population composition observed in early blastocysts Our results indicate that the complex three-dimensional distribution of the population types can be broken down into GATA6 level clustering and NANOG level dependence on number of neighbours. To test this, we implemented a computer simulation based on these two rules and compared the results to the population composition of the experimental data. Different to traditional rule-based models, we do not aim at modelling cellular mechanisms. Our approach aims at validating the simple rules that we identified to describe the population type pattern. As geometrical basis for the simulations, we used the measured cell centroid positions. To obtain the four populations, we assigned each cell a G6+ or G6- and N+ or N- cell state (see Terminology Box), respectively, according to the following two rules with three parameters : 1. G6+ cells are clustered according to their GATA6 levels; in the model, this is achieved by randomly assigning the cell state G6+ to a cell with a probability of 85% (*p*_*GATA*6 = 85%), otherwise the cell is G6-; 2. N+ cell state correlates with the number of neighbours; in the model, this is achieved by setting N+ cells as those with nine or close to nine neighbours (*startNumNeigh* = 9) up to 82% (*p_NANOG* = 82%), otherwise the cell is N-. Hence, we input expression levels for GATA6 and NANOG separately and as output we obtain the four populations as combinations of the two expression levels. The values for all three parameters are obtained from the experimental data II. The parameters *p*_*GATA*6 and *p_NANOG* are the proportions of ICM cells positive for GATA6 or NANOG expression, respectively. Hence, *p*_*GATA*6 is the proportion of DP and PrE progenitor cells and *p_NANOG* is the proportion of DP and Epi progenitor cells. Combining this information and rules for each cell, we determine its simulated population type. The results of the simulations for these parameter values are comparable to the experimental data, indicating that implementing these two rules allows the generation of the embryos with the observed population composition in early blastocysts. To assess the robustness of the model, we perform a parameter sensitivity analysis. This analysis shows that the simulation results are very robust with respect to the starting number of neighbours. Hence, within the observed range, cell density is not determinant for the proportion of cell fate allocation. The sensitivity analysis further shows that the model is sensitive to changes in the proportion of G6+ and N+ cells. Altogether, this indicates that the main parameter affecting the cell population composition in early embryos is the proportion of cells in a particular cell state. In summary, these results show that two simple rules for assigning cell state are capable of representing, to a very good approximation, the population composition observed in the ICM of early blastocysts. ## NANOG promotes GATA6 expression level clusters Previous studies indicate that NANOG represses GATA6 during Epi versus PrE differentiation. To analyse if NANOG also regulates GATA6 neighbourhood features, we performed our neighbourhood analyses in *Nanog* mutant embryos. We analyse and compare *Nanog*^*+/+* or *Nanog*^*+/-* (73 embryos;, data III) with *Nanog*^*-/-* (19 embryos;, data IV) results. In this case, we pool together *Nanog*^*+/+* and *Nanog*^*+/-* as there is no dosage effect. As previously shown, early and mid *Nanog* mutant blastocysts do not show any phenotypic defects until late stages. The detailed single cell quantitative analysis of GATA6 expression in the absence of *Nanog* shows decreased values in mid and late embryos, consistent with previous reports indicating PrE specification defects in the mutants. We start by analysing the clustering of cells according to their GATA6 expression levels. In the absence of *Nanog*, we still observe clusters of GATA6 expressing cells, reflected in the strong correlation found in GATA6 levels in a cell and median levels in the neighbours in early embryos. However, in the absence of *Nanog*, the correlation decreases from moderate to weak in mid blastocysts. Finally, in the late mutant blastocysts, there are fewer than 108 N-G6+ cells, hence the observed weak anti-correlation cannot be relied upon. This low cell number probably results from the apoptosis of ICM cells at this stage in *Nanog* mutants. We next analyse the global positional feature (see Terminology Box;). This analysis shows that the distribution of GATA6 expressing cells is altered in the absence of *Nanog* ( and for statistical analysis): the clear distribution of highest GATA6 expressing cells located away from the ICM centroid due to cell sorting disappears and cells express similar GATA6 levels independently of their position in late blastocysts. Finally, we were interested in testing whether our simulations can generate the population composition in *Nanog* mutants. To simulate the wild-type situation, we use data from early *Nanog*^*+/+* or *Nanog*^*+/-* embryos (44 embryos; see). In the *Nanog* mutant simulations, we set the proportion of N+ cells to 0. We did not detect any statistically significant differences between simulations and experimental results both in *Nanog*^*+/+* or *Nanog*^*+/-* and *Nanog*^*-/-* embryos. Hence, our model is also capable of reproducing the mutant phenotype. Note: we also had access to quantitative single cell data from a previously published data set composed of 19 *Gata6*^*+/+*, 28 *Gata6*^*+/-*, and 15 *Gata6*^*-/-* embryos (5). However, we decided not to perform an analysis, since in most cases, the number of cells per population type was below 108 cells. Altogether, these results suggest NANOG is involved in the neighbourhood regulation of GATA6 expressing cells, coordinating GATA6 expression levels to form the observed clusters and their global position in late embryos. ## FGF/ERK signalling promotes GATA6 expression level clusters and inhibits NANOG expression level clusters As NANOG and GATA6 expression is affected by FGF/ERK signalling (reviewed in), we next investigated if this signalling pathway is involved in the regulation of the local three-dimensional cell neighbourhood features (see Terminology Box). We used the available data sets (;; and Tables, data V-VIII). We focused our three-dimensional analyses on mid blastocysts treated for 24 h or 20 h with PD03, an inhibitor of FGF/ERK signalling. This regime promotes NANOG upregulation, without completely abolishing GATA6 expression. The data set also includes the use of an FGFRi, however in most cases, the number of cells per population type was below 108 cells. Hence, an analysis of this data would not yield reliable results. The data set further includes FGF4 treatments, which result in almost entirely PrE progenitor cells and therefore do not allow investigating the effect on NANOG or GATA6 neighbourhood features. We first analyse the effect of the decreased FGF/ERK signalling on GATA6 expression level clustering. This shows decreased GATA6 correlation between PrE progenitor cells and their neighbours. These results indicate that active FGF/ERK signalling is required to coordinate GATA6 expression levels between neighbouring cells to form the clusters. Concomitantly, we also observe an increased NANOG correlation between Epi progenitor cells and their neighbours, which reflects a NANOG expression level clustering for this population. The analysis of the local positional features (see Terminology Box) gives inconclusive results, as the control-cultured embryos did not show the clear pattern observed in the freshly flushed embryos. The same applies for the global positional features related to GATA6 expression levels and cell position within the ICM. Regarding global positional features related to the NANOG expression levels, there are no differences between control and treated embryos, albeit the absolute levels: highest NANOG expressing cells are closest to the centroid, consistent with these embryos being in late stages. These results, together with previously published results, are consistent with a scenario in which active FGF/ERK signalling is required for regulating NANOG expression in neighbouring cells, and for generating GATA6 expression level clusters. # Discussion In this study, we present a single cell quantification study, which includes three-dimensional neighbourhood analyses to evaluate how NANOG and GATA6 expressing cells are positioned within the ICM with respect to local and global features during cell fate decisions in mouse embryos. The cell neighbourhood is defined by the levels of fate markers expressed by the cell and its neighbours, the number of neighbouring cells and the population type of the cell and its neighbours. We also study a global positional feature by calculating the position of the cell relative to the ICM centroid. These novel three-dimensional analyses allow us to propose a model of how Epi and PrE fates arise from the early blastocyst based on cell neighbourhood descriptors and relative cell position dependant on FGF/MAPK signalling. ## Three-dimensional cell graphs provide local cell neighbourhood Cell fate decisions rely on groups (communities) of cells showing a coordinated and collective behaviour to achieve the determined fate. Hence, the features of an individual cell have to be put into context of the cell neighbourhood. This need for investigating small groups of cells has also been identified in other developmental contexts. Despite advancements in three-dimensional imaging of developing tissues, investigating the local cell neighbourhood features remains challenging. Even high-end imaging and image analysis protocols focus on the nucleus and rarely include the cell membrane, because the number of markers is restricted and the segmentation methods of three-dimensional membrane structures are only slowly evolving. We have shown that the Delaunay Cell Graph (DCG) allows the approximation of the cell neighbourhood from image data for nuclei. Our analysis combines local cell features, providing a good description of the individual cells and the structure of the tissue, with correlation analyses, enabling the identification of relationships between two variables. Extending the correlation analysis by the rule-based computer simulation provides a means to describe quantitatively complex three-dimensional population distributions during cell fate decision that is readily applicable to other systems. To perform a reliable analysis, sufficient amounts of data are required. For the mouse embryos, a robust correlation analysis requires at least 108 cells per measurement. Therefore, a good strategy for future analyses of different signalling pathways might be to first test their features in *in vitro* cultures such as ICM organoids or blastoids and only after a thorough analysis, transfer the results back into mouse embryos. Here, we obtain the spatial distribution of NANOG and GATA6 expressing cells. To our knowledge, these patterns have not been quantified before. However, we believe they are key to advancing in our understanding of cell differentiation in the preimplantation embryos. There are currently two main models for PrE differentiation. The main conceptual difference between them is that the model by Schroter et al. suggests that a bistable system for the NANOG-GATA6 interaction is sufficient, while the second work claims that a tristable system is required. While the model by Schroter et al. focuses on the ratio between Epi and PrE cells, the model by Tosenberger et al. has been fitted to the prevailing notion of the expression pattern, i.e. PrE progenitor cells are mainly surrounded by Epi progenitor cells (see Fig 3D in). Our results differ from this proposed pattern. The interesting question is whether a parameter regime exists, for which either of the two models can reproduce the patterns of the three- dimensional local cell neighbourhood. ## Early blastocysts exhibit patterns in local positional features and expression level clustering Our neighbourhood analyses reveal a pattern in the ICM cells of early blastocysts based on NANOG and GATA6 expression levels. Although most of the ICM cells co-express both markers, the levels of each vary among the different cells. This is reflected in patterns of the local positional features and expression level clustering. NANOG expression levels in the ICM cells correlate with the cellular arrangement. In early embryos, cells with 8 to 10 neighbouring cells display highest NANOG levels. Several mechanisms could link the positional information to NANOG expression levels. Our results are consistent with previous results of mechanical cues inducing high expression of NANOG in the central cells and their differentiation into Epi cells. Furthermore, it has been shown that the spatial confinement of cells in a three-dimensional microenvironment results in the maintenance of pluripotency even in the absence of LIF. In the early mouse embryo, we might observe a similar effect. The mechanical cues might be sensed via Hippo signalling which has been involved in interpreting positional information (reviewed in). Hippo signalling is clearly determining the first fate choice (TE versus ICM) in the mouse embryos and the second fate choice (Epi versus PrE) is linked to the first one. What we observe here might be a reflection of this: first and second cell fate decisions being entwined and Hippo signalling being involved in Epi formation as recently shown. In this study, the authors show attenuation of Hippo signalling promoting nuclear accumulation of YAP in the forming epiblast. In the light of our results, it is plausible that the attenuation of the Hippo signalling might start in those cells having high NANOG levels and around 9 neighbours, and hence contribute to epiblast differentiation. For GATA6, expression level clustering is observed. The expression levels are independent of the cell localization within the ICM. The functional relevance of the clustering effect might be to ensure an early coordinated PrE cell behaviour during their migration in later stages to occupy their final position at the blastocoele. Recent modelling results for cell population development in ICM organoids show that clonal expansion can play a role in clustering. In addition, the substantial cellular rearrangements taking place during preimplantation development might have a positive effect on cell fate clustering. Our results using *Nanog* mutant embryos further indicate a direct or indirect regulation of GATA6 expression level clustering by NANOG. Since PrE fate is regulated through FGF/MAPK signalling (reviewed in), this pathway might also be involved in regulating the spatial distribution of GATA6 and NANOG expressing cells (see below). The three-dimensional analyses of two independent data sets show a qualitative agreement between the results. The quantitative disagreement between the correlation values obtained are due to the mathematical properties of the correlation coefficient and not uncommon in the literature. We have previously used correlations between NANOG and OCT4 levels in individual cells as a pluripotency measurement of mouse embryonic stem cell (mESC) populations. The correlation values decrease as cells differentiate. In this scenario, there are quantitative differences between different wild-type cell lines (Tg2A vs *Tcf3*^*+/+*) cultured under the same conditions, which coincided with them having different pluripotency potential. Given the embryonic origin of the mESCs, one can envisage a similar situation in the mouse early embryos. The quantitative differences found in the correlation values between the data sets might be due to differences in the variability of the measurements related to the experimental setups or embryos being in slightly different developmental stage. All data sets were staged according to total cell number, the usual method to stage preimplantation embryos. However, this method might not be a reliable timing mechanism and variations can have a quantitative effect on accuracy similar to what we observe here. Indeed alternative ways of measuring developmental timing have been proposed, like the number of DP cells we proposed in ICM organoids, (continuous staging;), or morphogenetic events in rabbit embryos. The theoretical model allows us to break down the complex three-dimensional population pattern into two simple rules with three parameters. Eliminating one of the rules reproduces the *Nanog* mutant situation and the experimental results agree with the simulation. Hence, the population composition in ICMs of early embryos can be derived from the local neighbourhood features. Altogether, our results are consistent with positional information impinging on cell fate decision in early blastocysts. This, together with previous results, suggests that very early in development, when ICM cells are co-expressing NANOG and GATA6, the two transcription factors as well as FGF/MAPK signalling impact on their expression levels and that the cells are already deciding about their fate. ## A global pattern of NANOG and GATA6 expression in the ICM starts arising in mid blastocysts Our results show that the global positional features (see Terminology Box) of NANOG and GATA6 in early blastocysts do not show a pattern. This lack of a pattern might allow for the previously observed plasticity during the cell fate decision process. Starting in mid and fully established in late blastocysts, once the decision has been made and fate reversal does no longer occur, we see the expected distribution. Higher NANOG level expressing cells are located at the centre of the ICM and higher GATA6 level expressing cells are at the edge. Hence, our results indicate that the cell fate specification does not correlate with the global position of a cell in the ICM. Only once the cell fate is specified, the cells arrange in a global pattern. Our results show that *Nanog* is involved in the evolvement of the global pattern in late blastocysts as in its absence the GATA6 pattern disappears. A previous study has shown that differential adhesion between ICM cells and directional movement, together with differential adhesion between ICM and TE cells or forces pushing from the blastocoele are responsible for the final distribution of Epi and PrE cells in late blastocysts. According to the Krupinski study, Epi cells would have stronger adhesion between them than with other cell types, while PrE cells would show a directed movement towards the blastocoele. In this context, the absence of *Nanog* would be interpreted as absence of differential adhesion. This scenario would result in a lack of global pattern, which is in agreement with our data. ## FGF/MAPK signalling affects NANOG and GATA6 expression level clustering It has been shown that FGF/MAPK signalling is the main signalling pathway involved in Epi versus PrE differentiation (reviewed in). Our three-dimensional analyses of FGF/MAPK signalling inhibitor treated embryos also implicates this pathway in the regulation of the three-dimensional local clustering of NANOG and GATA6 expressing cells. We did not obtain a clear effect of this pathway on the local or global positional features. There are several explanations for this: FGF/ERK signalling is not involved in establishing the global pattern, the long- term culture of embryos affects their global pattern, or the embryos analysed here are in a different stage from the freshly flushed ones (more than 150 cells *versus* less, respectively). We favour the explanation of an issue with the long-term culture of embryos since cultured embryos until E4.5 stage clearly do not have the same shape (spherical) as freshly flushed E4.5 embryos (prolate; compare embryos shown in stages 120–150 and \>150 to those shown in Fig 2C in). Furthermore, it has been shown that embryo culture delays their development. The change in the overall shape of the cultured embryos together with their delay are key differences that should not be overlooked as they will affect the global pattern and we believe both differences are at the core of the results obtained. Our results, together with previous work, allow us to suggest the following series of events during cell fate decision making in early embryos. *Fgf4* expression is directly regulated by the OCT4-SOX2 dimer (as is *Nanog*,), and it is expressed in a subpopulation of ICM cells. We hypothesize that the secretion of FGF4 starts, or is higher, in the subpopulation of NANOG positive cells with 8 to 10 neighbours. Binding to (mainly) FGFR1 presented in the cells activates the signalling pathway. Activation of the pathway has opposite effects: it promotes autocrine NANOG degradation via ERK1 phosphorylation and paracrine GATA6 upregulation via ERK1/2 phosphorylation. Autocrine NANOG inhibition might result in the low correlation found in NANOG levels between neighbouring cells. Upregulation of GATA6 will result in the upregulation of FGFR2 expression in those cells as suggested by ChIP-seq experiments. GATA6 expression level clusters might be due to neighbouring cells receiving similar levels of FGF4, hence activating the downstream effectors to a similar extend. The clusters might also be related to directed active movement of the cells at this stage towards the cavity. The effect of FGF/MAPK signalling on the GATA6 clustering could be related to both, as signal inhibition results in decreased correlations and reduced cell movement of ICM cells. As *Fgf4* expression depends on NANOG, the decrease in GATA6 expression level clustering observed in the absence of NANOG reinforces the idea that the three- dimensional cell neighbourhood features are regulated by FGF/MAPK signalling. However, this poses the question of how FGF4 is propagated extracellularly once secreted and how its activity is inhibited in the direct neighbours. One possibility is that it is via differential expression of heparan sulfate (HS) chains, which has been associated with heterogeneous di-phosphorylated Erk at this embryonic stage. Another alternative is changes in the internalization and spreading related to endocytosis rates as shown for FGF8 in zebrafish embryos. In summary, we propose that the coordinating mechanism behind the three- dimensional distribution of NANOG and GATA6 expressing cells in early blastocysts is FGF/MAPK signalling. However, we cannot rule out that other major signalling pathways involved in patterning fields or groups of cells, such as Notch, Wnt, BMP, Hippo or EGF might also have an input. In support of this, there are reports of Notch signalling involved in early mouse development, as well as Wnt signalling, BMP signalling, p38/MAPK signalling and EGF signalling. In the light of our results, it will be important to revisit how these signalling pathways might be involved in cell fate decisions in early blastocysts, investigating how they affect the local cell neighbourhood features and the global positional feature within the ICM. # Supporting information We thank Christian Schröter, Jennifer Nichols, Joaquín de Navascués and Alfonso Martínez-Arias for helpful comments and Carl-Magnus Svensson for critical reading of the manuscript. We also want to thank Néstor Sáiz and Kat Hadjantonakis for their comments, sharing their quantitative data through GitHub as well as through personal communication. We also want to thank Claire Chazaud for kindly sharing *Nanog* heterozygous mice. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. [^1]: The authors have declared that no competing interests exist. [^2]: Current address: Center for Computational and Theoretical Biology, Department of Biology, Universität Würzburg, Würzburg, Germany [^3]: Current address: Wellcome-MRC Cambridge Stem Cell Institute, Jeffrey Cheah Biomedical Centre Cambridge Biomedical Campus, University of Cambridge, Cambridge, England, United Kingdom [^4]: ‡ These authors are joint senior authors on this work.

# Introduction Numerous studies have demonstrated the utility of motion-tracking technologies for elucidating the biomechanics and gait patterns associated with various types of mobility impairments across much of the lifespan. However, these technologies are typically available only in specially equipped laboratories at major medical centers or research institutions, not in community hospitals or clinics. Many older adults, particularly those who are physically frail or have cognitive impairments, are unwilling or unable to undergo such assessments due to the required travel, as well as the testing burden itself. Repeated testing to obtain longitudinal data is also impractical. These testing limitations have led to critical gaps in our knowledge regarding gait biomechanics across the full health spectrum of older adults and their relation to mobility impairments. Recent advances in technology now offer new, innovative means of quantifying multiple dimensions of mobility in the community setting. For example, portable gait mats and wearable devices incorporating electronic accelerometers and gyroscopes now allow objective assessment of spatiotemporal features of gait and other mobility performances in large numbers of community-dwelling older adults. Typically, however, joint motion of older adults is not examined as part of gait assessment in the community setting. Newly available depth-sensing cameras, shown in laboratory studies to be accurate in capturing certain aspects of gait are portable, inexpensive, and user-friendly and can therefore potentially be used to record three-dimensional body motion of older adults during instrumented mobility testing in the community, from which joint motion metrics can be extracted. The purpose of this study was to assess the utility of a portable, depth-sensing camera system for three-dimensional motion tracking of older adults in the community setting to obtain clinically relevant metrics of lower extremity joint motion during gait. We first verified that the depth-sensing camera captured joint motion similar to that obtained in a state-of-the-art motion analysis laboratory. We then deployed the depth-sensing camera in the community setting, where we recorded three-dimensional gait motion data from older adults during a uniform, structured mobility testing session. After extracting metrics of hip and knee joint motion, we assessed their potential clinical relevance by examining their correlations with quantitative mobility metrics derived from recordings of a wearable sensor during the same testing session, and also with a diverse range of other mobility-related physical function and health measures. # Materials and methods ## Participants Community-dwelling, ambulatory older adults without dementia were enrolled from the Rush Memory and Aging Project (MAP), a longitudinal clinical-pathologic study of chronic conditions of aging that recruits from retirement communities in northeastern Illinois. To participate in MAP, individuals must be free of known dementia at the time of enrollment, agree to annual structured examinations, and sign an Anatomical Gift Act for autopsy at the time of death. MAP participants provided written consent at enrollment, and as part of this pilot study, we obtained additional written consent for video-based motion capture and questionnaires concerning hip and knee pain and dysfunction. The Institutional Review Board of Rush University Medical Center approved this study. In addition, the individual in the figures gave written informed consent (as outlined in the PLOS consent form) to publish the series of video frames. ## Assessment of joint motion with a depth-sensing camera ### Hardware, data collection and processing, and testing A portable, depth-sensing video camera originally developed for video gaming (Kinect for Windows, Microsoft Corporation, Redmond, WA) was employed to capture participants’ biomechanics during gait. Using infrared light, this device records “depth” video, which quantifies an object’s distance from the camera at 30 frames per second. The camera was positioned as shown in. Its recording was controlled via software (iPi Recorder version 3.1.1.34, iPi Soft, Moscow, Russia) running on a notebook computer (Pavilion 13 x360 Convertible PC, HP, Palo Alto, CA) as participants traversed a walking path. We transferred this data to a workstation (Z620, HP) to run frame-by-frame body pose estimation software (iPi Mocap Studio version 3.1.2.177, iPi Soft). In order to compare hip and knee motion captured using the depth-sensing camera with that obtained using a state-of-the-art optoelectronic motion capture system (Qualysis, Gothenborg, Sweden), we first enrolled 10 participants who were able to travel to and undergo testing in the Motion Analysis Laboratory at Rush University Medical Center. These two systems reconstructed similar patterns of hip and knee flexion and extension throughout the gait cycle; more of the variation in the tracked joint angles was attributable to between-person sources than stemmed from trial- to-trial differences (within person) or differences between the two systems (depth-sensing camera vs. lab-based system). Additional details of motion data collection, processing, and testing are presented in. ## Motion capture in the community setting We then proceeded with testing in the community setting. At their places of residence, participants underwent the annual MAP structured mobility testing protocol, which included a 32-foot walking performance consisting of four traversals of an 8-foot path separated by 180-degree turns, as described in earlier work. Using the techniques outlined above, we recorded the 32-foot walk with the depth-sensing camera. For the two traversals during which the participant was moving toward the camera, we extracted the left and right hip and knee ranges of motion (ROM) computed as the difference between their maximum flexion and extension angles within a gait cycle, as further detailed in We averaged the four measurements (left and right legs from each of two traversals), yielding a mean hip ROM and a mean knee ROM for each participant. ## Assessment of mobility with a wearable sensor We also instrumented the structured mobility testing protocol in MAP with a sensor affixed to the lower back. This sensor records acceleration along and rotation rate around each of three orthogonal axes, from which we extracted objective metrics of mobility from the 32-foot walk, the Timed Up and Go (TUG), and the 20-second standing with eyes closed performances. We combined correlated metrics into composite summary scores capturing participants’ gait and balance abilities, as in previous work. ## Other mobility-related physical function and health measures We collected other mobility-related health measures during the same annual MAP testing cycle, as previously described. Age at the time of mobility testing was computed based on the participant’s self-reported date of birth recorded at MAP study entry along with sex and years of education. Objective measures of physical activity were derived from a wrist-worn activity monitor. A summary motor measure was computed based on 10 conventional tests of gait, dexterity, and strength. We assessed parkinsonian gait, a relatively common clinical phenotype that is likely to affect objective measures of lower limb kinematics, using a modified version of the United Parkinson’s Disease Rating Scale. BMI was calculated from height and weight. Hip and knee joint pain and dysfunction were assessed via the Hip disability and Osteoarthritis Outcome Survey (HOOS) and the Knee injury and Osteoarthritis Outcome Survey (KOOS). Other data gleaned via participants’ self-report included mobility disability based on the Rosow- Breslau scale, fall history, vascular disease and exposure to vascular disease risk factors, and frequency or duration of participants’ engagement in specific physical, social, and cognitively stimulating activities. Expanded descriptions of the methods by which these data were collected and computed are included in. ## Statistical analysis We examined the correlations between hip and knee ROMs and age, sex, education, and the sensor-derived mobility metrics obtained during the same mobility testing session. In further analyses we examined the relationship of joint motion metrics with other mobility-related physical function and health measures that were obtained during the same annual testing cycle and are described in. We computed the Pearson correlation coefficient between each pair of measures examined and reported the two measures as being correlated if their correlation coefficient met the threshold for statistical significance, set at p\<0.05. Statistical analyses were programmed in SAS 9.3 for Linux. # Results ## Descriptive data All 52 individuals who we approached consented to the study, and all but 2 completed gait testing while recorded by the depth-sensing camera in the community setting. Of those 50, we failed to extract reliable hip and knee ROMs for one participant due to their unusual gait pattern in which one leg obscured the other from view of the camera during key portions of the gait cycle. The remaining 49-person sample included individuals with a range of self-selected gait speeds (mean = 84 cm/s, SD = 25, min = 35, max = 122). Their joint motion measures, mobility metrics, and other characteristics are summarized in. ## Association of hip and knee ROMs with demographics Hip and knee ROMs were correlated with each other (r = 0.67) and were both negatively correlated with age (hip/knee r = -0.55/-0.40). There were no associations of ROMs with education or sex. The correlations between hip and knee ROMs and demographics and sensor-derived mobility metrics are presented in. ## Association of hip and knee ROMs with sensor-derived mobility metrics Several mobility performance metrics derived from a sensor worn during the structured mobility testing session were associated with both hip and knee ROM, while others exhibited a more joint-specific association. For example, there was stronger association of stride regularity with hip ROM than with knee ROM. ## Association of hip and knee ROMs with other mobility-related physical function and health measures Total daily physical activity derived from a wrist-worn monitor was associated with hip ROM (r = 0.37) but not knee ROM. A summary motor measure derived from 10 conventional motor tests was correlated with both hip and knee ROM (hip/knee r = 0.59/0.52). Parkinsonian gait had a negative correlation with hip and knee ROM (r = -0.57/-0.70), as did self-reported indicators of mobility disability (r = -0.51/-0.47) and fall history (r = -0.42/-0.32). BMI was not related to either ROM. Knee ROM was correlated with all 5 domains of self-reported hip pain and dysfunction assessed by the HOOS (r = 0.31 to 0.47) and with 3 domains of self- reported knee pain and dysfunction from the KOOS (pain, quality of life, activities of daily living, r = 0.40 to 0.47). Hip ROM exhibited similar trends as knee ROM in its correlations with HOOS and KOOS domains, but these were not significant at the p\<0.05 level. Of the different types of activity for which participants reported their level of engagement, it was the frequency of social activities rather than physical or cognitive activities that was most highly correlated with hip and knee ROMs (r = 0.29). Claudication but no other individual vascular diseases or risk factors were correlated with hip and knee ROM (hip/knee r = -0.45/-0.31), although the number of reported vascular diseases was negatively correlated with hip ROM (r = -0.44). # Discussion We investigated the use of a portable, depth-sensing camera to quantify joint motion of older adults during mobility testing in the community. Using this camera, we successfully acquired three-dimensional motion data from nearly all 52 enrollees. Hip and knee ROMs extracted from the motion recordings were associated with mobility metrics acquired during the same testing session via a wearable sensor. In addition, joint motion metrics were related to a wide range of other mobility-related physical function and health measures. These results support the notion that a portable, depth-sensing camera can be used to expand instrumented gait testing conducted outside the laboratory setting. Further studies using this approach in larger numbers of older adults are needed to elucidate the independent contributions of joint motion and spatiotemporal mobility metrics to late life mobility impairments. This work brings together two separate but related lines of research. Our group and others have successfully employed an unobtrusive wearable sensor for instrumented gait testing of older adults in the community setting. Although a variety of mobility performance metrics can be extracted from recordings of this device, the single, belt-mounted sensor does not provide information about joint motion. Newer systems that incorporate multiple such devices worn on the legs and trunk are capable of tracking lower extremity joint motion, but these may not be well tolerated by older adults due to the extra time required for setup and calibration and the need to wear multiple tight-fitting elastic sleeves. In another line of research, investigators have explored the use of depth-sensing camera technology for gait analysis. However, with few exceptions, these studies have taken place in controlled laboratory environments. Conversely, older adults participating in the current study were assessed while wearing everyday attire at their own residences by personnel who were trained in the use of the depth- sensing camera but otherwise inexperienced in motion capture technology and techniques. The high rate of successful data acquisition under these conditions suggests that the camera was not an overly burdensome addition to the single wearable sensor during the instrumented mobility testing session for either the study participants or the staff who administered the testing. We also gleaned valuable information from the 3 cases in which we were not successful in obtaining hip and knee ROMs. In particular, two individuals withdrew from the study at the time of gait testing due to difficulties encountered by staff in setting up the depth-sensing camera system. Further refinements to reduce the required physical space may help decrease the refusal rate. One other participant’s atypical gait pattern (one foot placed directly in front of the other, obscuring it from view of the camera) precluded extraction of reliable hip and knee ROM values, indicating that the depth-sensing camera may fail to accurately track the biomechanics of a small subset of individuals due to certain aspects of their gait, body habitus, or other unforeseen factors. We explored the potential clinical relevance of hip and knee ROM values obtained using the depth-sensing camera. Earlier studies concluded that these metrics were not accurate enough to serve as reliable gait metrics. A recent review article affirmed these studies’ findings that angular measures of hip and knee motion were less accurate than spatiotemporal measures such as stride length. However, the review also suggested that in some instances, the benefits of a low-cost, portable, and user-friendly depth-sensing camera might outweigh the accompanying penalty in accuracy. It is therefore noteworthy that in this work, in which data was collected in the community setting, hip and knee angular ROMs were correlated with several previously validated mobility metrics obtained using a belt-worn sensor during the same mobility testing session. These metrics, which probe a diverse range of mobility dimensions but do not provide direct information on lower limb kinematics, exhibited statistically significant correlation coefficients ranging from 0.31 to 0.58 with hip ROM and from 0.29 to 0.51 with knee ROM. One interpretation of these low- to moderate-strength correlations is that the hip and knee ROMs derived from the depth-sensing camera capture an additional facet of mobility that is related to, but not duplicative of, sensor-derived mobility metrics. Furthermore, although this was a small pilot study with limited power, there were some indications of differential associations of hip and knee motion measures with the sensor-derived mobility metrics. Specifically, stride regularity was correlated with hip ROM but not knee ROM, while sway during quiet standing with eyes closed was more strongly correlated with knee ROM. This preliminary work hints at the possibility of harnessing emerging technologies to uncover specific biologic substrates of mobility impairments. Further work in larger number of older adults is warranted to clarify the potentially independent associations of joint motion and sensor-derived mobility metrics with impaired mobility. We found additional support for the clinical relevance of joint motion metrics in their associations with a wide range of other mobility-related phenotypes and measures. These included objective metrics of physical activity, several other metrics of motor function, self-report indicators of joint pain and dysfunction, mobility disability, and falls. These findings suggest that the portable depth- sensing camera system could help to uncover unique patterns of gait alteration that portend or cause specific phenotypes and mobility impairments that are common in later life. This work diverges from the majority of earlier studies that utilized depth- sensing cameras for gait analysis in that we employed a third party developer’s software to carry out post hoc frame-by-frame body pose estimation rather than using the camera manufacturer’s software development kit (SDK) for real-time pose estimation. This post hoc pose estimation software requires extra time and resources to process the data, but in the current work, this drawback was outweighed by an improvement in the consistency and accuracy of three- dimensional body tracking over the SDK’s real-time pose estimates. Thus, the use of post hoc pose estimation software might partially account for the current study’s findings of widespread association between hip and knee joint motion derived from depth-sensing camera recordings and mobility-related measures and phenotypes, many of which have not been previously reported. This technical detail may therefore have an important impact on the future clinical utility of depth-sensing camera technology. Work is underway to further automate the pose estimation processing as part of a high throughput system that can accommodate large data sets. Overall, results of this study show that the depth-sensing camera is promising but has limitations. It cannot, for example, match the accuracy, temporal resolution, or functionality of a dedicated, laboratory-based gait analysis system. Such systems have very high temporal and spatial resolution and incorporate force plates to capture ground reaction forces, enabling determination of inverse kinematics. Bringing a force plate into the community setting or estimating ground reaction forces with wearable accelerometers are technically challenging endeavors that will require additional study in order to successfully integrate this data with the depth-sensing camera. Thus, we are currently limited to deriving kinematic measures from the depth-sensing camera in the community setting. This work may have important implications for further aging research and clinical studies of mobility in both the short and long term. In particular, depth-sensing camera technology adds to a growing portfolio of cost-efficient, portable devices that are now available for more comprehensive and objective quantification of multiple dimensions of mobility and other diverse behaviors in large, community-based cohort studies of aging. By collecting three-dimensional kinematic data longitudinally and combining it with objective measures of mobility and habitual physical activity from other wearable sensors, investigators and clinicians can obtain a more comprehensive characterization of healthy mobility and impairments. This practice may help to identify certain alterations or combinations of alterations (beyond simple hip and knee ROMs) that specifically portend different types of mobility impairment and facilitate risk stratification of older adults, allowing early, targeted interventions. In addition to its utility in research, depth-sensing camera technology could eventually provide patients and physicians with wider access to quantitative mobility testing, either in a primary care clinic or via telemedicine in the comfort of the patient’s home. This conceptual paradigm for easier, lower cost, and more frequent gait testing using portable devices has the potential to impact clinical practice by facilitating the detection of early signs of mobility impairment, adding easily collected functional data on joint motion to treatment decision trees (e.g. regarding joint replacements), and monitoring the progression of gait alterations and their response to clinical interventions. # Supporting information We thank the participants of the Rush Memory and Aging Project and the faculty and staff of the Rush Alzheimer’s Disease Center. More information regarding obtaining MAP data for research use can be found at the RADC Research Resource Sharing Hub ([www.radc.rush.edu](http://www.radc.rush.edu/)). [^1]: The authors have declared that no competing interests exist.

# Introduction RAF protein kinases were originally identified as viral oncogenes, found in murine and avian retroviruses. *RAF* genes encode protein serine/threonine kinases, that mediate transduction of extracellular mitogenic signals from activated Ras GTPases at the plasma membrane to a MAP kinase module (RAF-MEK- ERK), the mitogenic cascade (reviewed). As a result, complex physiological responses to growth factor stimulation take place at multiple cellular levels. Insect genomes contain only a single *RAF* gene whereas vertebrates have refined RAF signaling and employ three isoforms that target ERK signalling to different subcellular compartments. Specialized functions are reflected in differential regulation of RAF kinase activity and varying phenotypes of RAF knockout mice. The three RAF isoforms and their splice variants share common structural features comprising three conserved regions, CR1, CR2, CR3. The N-terminal CR1 encompasses the Ras binding domain (RBD) and the cysteine-rich domain (CRD), CR2 contains a conserved 14-3-3 binding motif and the C-terminally located CR3 encodes the kinase domain. Whereas B- and C-RAF were studied extensively, little is known about A-RAF function. A-RAF is marked by a low basal kinase activity, which has been attributed to substitutions in its N-region, where tyrosine 296 plays a central role. In contrast to animals with a genomic deletion of B- or C-RAF, A-RAF ^−/− mice are viable, but die perinatally, depending on the genetic background. Budding yeast *Saccharomyces cerevisiae* is an established eukaryotic model organism that has played a key role in the elucidation of the MAP kinase signaling pathway. Despite the presence of two redundant RAS genes and at least 6 MAPK cascades, no RAF kinases are present in *S. cerevisiae*. Nevertheless, yeast was important for defining RAF function as an activator of a prototypical MAP kinase cascade in experiments that involved ectopic expression of C-RAF. RAF isoforms are known to function as homo and heterodimers in mammalian cells, complicating assignment of individual function. Therefore yeast is an attractive system for investigation of a single RAF isoform. Endocytosis is a process essential for many aspects of cellular life, including receptor internalization and recycling. ARF6 GTPase was shown to regulate endocytosis at several levels. The activation state of ARF6 is determined by the bound nucleotide, GTP or GDP, which also affects intracellular localization. Nucleotide loading is regulated by specialized guanine nucleotide exchange factors (GEFs) and GTPase activating proteins (GAPs) that catalyze hydrolysis of bound GTP. Endocytosis and signal transduction are known to be functionally linked and regulating each other (Von Zastrow and Sorkin, 2007, Polo and Di Fiore, 2006). ARF6 as a central regulator of endocytic trafficking was shown to activate ERK. Based on changes in endocytosis upon inhibition of ERK signaling Robertson et al. (2006) suggested a role of ERK signaling in the regulation of clathrin-independent endocytosis. Here we describe the role of A-RAF in membrane trafficking and identify its function at a specific step of endocytosis. This work led to the discovery of a C-terminally truncated version of A-RAF, AR149 that strongly interfered with cell growth and polarization in yeast and with endocytosis and actin polymerization in mammalian cells. As this work was in progress two splicing isoforms of A-RAF, termed DA-RAF1,2 were described that act as natural inhibitors of RAS-ERK signaling during myogenic differentiation. DA-RAF2 contains the first 153 aar of A-RAF and thus is nearly identical with AR149. AR149 localized specifically to the recycling endosomal compartments as confirmed by colocalization and coprecipitation with ARF6. Expression of AR149 interferes with recycling of endocytosed transferrin (Tfn) and with actin polymerization. siRNA-mediated depletion of endogenous A-RAF or inhibition of MEK by U0126 mimic AR149 function, supporting a role of A-RAF regulated ERK signaling at endosomes that is controlled by AR149/DA-RAF2 and targets ARF6. # Materials and Methods If not otherwise stated reagents were of p.a. purity (Sigma, USA), restriction enzymes were from New England Biolabs (USA) and Fermentas (Lithuania). ## Plasmids used in this study Deletion mutants of A-RAF were generated by insertion of PCR products into the pUG36 plasmid. In the case of N-terminal mutants of A-RAF, BamHI and HindIII recognition sequences have been attached to the sequence of forward and reverse primers respectively. Primers for C-terminal mutants of A-RAF contained SmaI and HindIII respectively. Doubly truncated mutants contained BamHI and XhoI restriction sites. PCR products were generated using primers listed in. For expression of the GFP fusion proteins in mammalian cells we used pEGFP-C-1 and pDS-Red2 (Invitrogen, USA). ARF6 constructs tagged with hemagglutinin (HA) in pLNCX, FLAG-EFA6 and GST-GGA3 were generous gift from Margaret Chou, University of Pennsylvania. GFP-tagged ARF6 was kindly provided by Antoine Galmiche. GFP-ARF6(Q67L) and GFP-ARF6(T27N) were generated by site directed mutagenesis (QuikChange, Stratagene) with the primers listed in. GFP-A-RAF in pEGFP-C1 was kindly donated by Angela Baljuls. ## Cell culture and cell fractionation Media and reagents were purchased from Invitrogen (USA). HeLa, NIH 3T3 and COS7 cells were maintained in DMEM supplemented with 10% fetal bovine serum (FBS), penicillin and streptomycin. For starvation, DMEM was supplemented with 0,03% FBS. Cell transfections utilized jetPEI (Biomol). Subcellular fractionation of HeLa cell lysates employed “ProteoExtract Subcellular Proteome Extraction Kit “(Calbiochem) according to manufacturer's instructions. ## Fluorescence Microscopy Fluorescence microscopy was done with an Openlab software (Improvision, UK) controlled inverted DMIRBE microscope (Leica, Germany) with Leica oil immersion objective. All images were captured and stored as Openlab LIF files. Images were subsequently processed using Photoshop software. ### Yeast live cell imaging Yeast cells were transferred into a self-made chamber slide for imaging. ### Fixed yeast cells imaging Cells were fixed in 3.7% paraformaldehyde in PBS, washed and subsequently digested for 1 hour with lyticase (Sigma). After washing and mounting, samples were either stored at 4°C or processed for imaging. ### Mammalian cells imaging Cells were grown on cover slips, treated with growth factors or serum as indicated and subsequently fixed in 3.7% paraformaldehyde, permeabilized with 0.1% Triton X-100. Stainings were performed with specific antibodies and fluorescently labeled secondary antibodies. ## Indirect immunofluorescence after cytosol depletion HeLa cells were grown on coverslips overnight. After two washes with PBS cells were treated with 0,05% digitonin in isotonic sucrose buffer for 4 min on ice. After digitonin treatment, cells were fixed with 3.7% paraformaldehyde in microtubule-stabilizing buffer (MSB; 0.1 M PIPES, pH 6.9, 2 mM MgCl₂, 2 mM EGTA), washed and subsequently permeabilized with 0,1% w/v Saponin in MSB with 0.5% BSA for 10 min. To stain non-cytosolic A-RAF, cells were incubated with anti-A-RAF antibodies (Santa Cruz, USA) in combination with anti β-Tubulin antibodies (Chemicon International) at concentration of 20 µg/ml in MSB buffer with 0,5% BSA and 0,1% Saponin at room temperature for 2 h. Unbound antibodies were removed by 3 washes with the same buffer. The coverslips were incubated with appropriate secondary antibody (conjugated to TRITC or CY5) diluted 1∶200 for 1 h. After three washes with MSB and brief wash with deionized water the coverslips were mounted using MOWIOL (Calbiochem, USA). ## siRNA-mediated depletion of human A-RAF For generation of A-RAF specific siRNA we used “X-tremeGENE siRNA Dicer Kit” (Roche). Prepared siRNA mix containied about 15 different siRNAs. The target sequence for human A-RAF located at the 3′end of A-RAF coding region was from “Human esiRNA resource” (German Resource Center for Genome). Scrambled siRNA was from QIAGEN. Transfection was carried out using 2 µg of siRNA mixture and 10 µl of “X-tremeGENE siRNA Transfection Reagent” (Roche) for 6-well culture plates, according to the instructions provided by the manufacturer. ## Transferrin internalization HeLa cells were grown on coverslips overnight, transfected with either GFP or GFP-fused AR149, ARF6(Q67L), ARF6(T27N) respectively for 48 h. The cells were pre-incubated in serum-free medium for 1 h at 37°C. For continuous Tfn uptake, the cells were incubated in internalization medium (HBSS medium plus 1% BSA) containing 5 µg/ml Alexa Fluor 546-conjugated human Tfn (Invitrogen) at 37°C for indicated time. After Tfn internalization the cells were extensively washed three times with ice-cold PBS and fixed with 3.7% PFA. ## Yeast strains and techniques Protease deficient strains cI3 ABYS 86, BJ 5459 (Yeast Genetic Stock Center, University of California, Berkeley) were used in order to prevent degradation of expressed proteins. Standard protocols for yeast growth, transformation and manipulations were employed. Yeast transformation was performed by a modified litihum acetate method. The following plasmids were used in this study: pUG36 and pUG36 for detection of GFP-fusion proteins; pEG-KT for galactose-inducible expression of GST-fusion proteins. Membrane fractionation on 20–35% sucrose gradients and indirect immunofluorescence microscopy were done as described previously. ## Immunoprecipitation of GFP-AR149 and ARF6 COS7 cells were transfected with GFP-AR149 and HA-ARF6wt, HA-ARF6(Q67L) or HA- ARF6(T27N) respectively with jetPEI (Biomol). After 24 h, cells were lysed in ARF6 lysis buffer (50 mM Tris-HCl pH 7.0, 2 mM MgCl₂, 100 mM NaCl, 10% Glycerol, 0.75% NP-40) containing protease inhibitors. To avoid high signal from heavy and light chains of antibodies on the Western blot we used Mouse IgG TruhBlot Set (NatuTec) including beads and secondary HRP-conjugated antibody. The clarified lysates were divided into two equilibrated parts, each of them was incubated for 2 h at 4°C with anti-GFP or anti-HA antibody respectively (Santa Cruz Biotechnology, USA). Next, the probes were precipitated with mouse-TruhBlot agarose (NatuTec) for 1 h at 4°C. Beads were washed three times in the lysis buffer with 0.2% NP-40 and protease inhibitors. ## GGA3 pulldown assay Activated ARF6•GTP was monitored by binding to its effector GGA3 as described previously. Briefly, COS7 cells were co-transfected with HA-ARF6 plus indicated plasmids using jetPEI and grown for 24 h. EGF stimulation was performed after 24 h starvation in 0.03% serum with 100 ng/ml EGF (Cell System Biotechnologie Vertrieb) for 10 min at 37°C. Clarified lysates were incubated with 25 µg of GST-GGA3 immobilized on glutathione-Sepharose beads for 2 h at 4°C. The beads were washed three times with PBS, resuspended in SDS PAGE loading buffer and boiled. Bound proteins were size-fractionated by SDS-PAGE and detected by immunoblotting. # Results ## Isotype specific distribution of RAF proteins in yeast All three human *RAF* genes were fused with the C-terminus of green fluorescent protein (GFP) and expressed under the control of a moderately inducible *MET25* promoter in *S. cerevisiae*. GFP-B-RAF and GFP-C-RAF fluorescence was evenly distributed in the cytosol of induced cells. In contrast, A-RAF fluorescence localized to distinct punctate cortical structures, which were polarized toward the tips of small buds and bud necks of larger buds. Treatment with α-factor, concentrated GFP-A-RAF at the tips of mating projections termed shmoos. Sucrose density gradient centrifugation was employed to test the observed differences in localization of GFP-RAF proteins. Consistent with the flourescence data, both, C- and -B-RAF segregated with cytosolic proteins in a sucrose density gradient, whereas A-RAF was predominantly found in the heavy membrane fraction. Taken together, of the three RAF isoforms only A-RAF interacts with yeast membrane components or proteins that are polarized during cell division and mating. ## Multiple lipid binding motifs of A-RAF are required for its unique localization To test, which A-RAF domain was responsibe for its subcellular distribution in yeast we generated a set of amino- (N-) and carboxy-terminal (C-) A-RAF deletion mutants fused to GFP. The localization of each construct was controlled by fluorescence microscopy. A schematic overview of their subcellular distribution is presented in. Representative micrographs are shown in. Expression and correct size of the deletion mutants was ascertained by Western blot analysis. Three types of staining patterns were observed: (i) homogeneous cytosolic as seen with full length B- and C-RAF (see 142–606 construct in), (ii) a punctate pattern (see 88–606 construct), and (iii) an intermediate pattern, with intensive homogeneous labeling of plasma membranes (see, 1–149 construct as an example). Two different lipid-binding domains have previously been identified in the structure of C-RAF. A phosphatidic acid (PA) binding domain, located in the N-terminal part of CR3, and a phosphatidyl serine (PS) binding domain located in the cysteine-rich region of CR1. Moreover, deletion of CR1 of A-RAF (which encompasses RBD+CRD) led to loss of A-RAF-specific binding to PtdIns(4,5)P₂ suggesting a third lipid binding site. As these sites are well conserved among RAF isoforms, we mutagenized their key residues, which have previously been shown to affect the interaction of C-RAF with PA and PS. Substitution of basic residues in the PS site led to localization of mutant protein in the cytosol. This pattern strongly resembled that of N-terminally truncated deletion mutants. In contrast, mutant protein with substitution of basic residues in the PA binding site led to homogeneous distribution in the plasma membrane, the typical pattern for C-terminally truncated mutants. We conclude that the A-RAF specific punctate pattern requires both, the PS/PIP₂ and PA lipid binding sites. ## Expression of AR149 in yeast causes growth inhibition, defects in cell polarization, in nuclear morphology and in membrane trafficking In the course of deletion analysis, we noted that cells expressing several C-terminally truncated mutants showed homogenous membrane localization, different from that of full-length protein. In addition, a significant fraction of cells expressing the C-terminal deletion mutants was larger and apolar (round in shape instead of oval) as compared to GFP or GFP-A-RAF expressing cells. Polarization during cell growth by directed membrane trafficking is essential for cell division, therefore we hypothesized that GFP-AR149 expressing cells may exhibit retarded cell cycle progression and as a consequence loss of viability. To test this, we grew GFP, GFP-A-RAF and GFP-AR149 in liquid culture, under promoter de-repressing conditions. Cells expressing GFP-AR149 showed sustained growth inhibition. The difference of 28 minutes in duplication time confirms that expression of GFP-AR149, but not that of GFP or full-length A-RAF exhibits an inhibitory effect on expressing cells. We stained actin and DNA in GFP-AR149 and GFP-A-RAF expressing cells. In agreement with the apolar phenotype, no polymerized actin could be detected in the large round cells expressing higher levels of GFP-AR149. Intriguingly, in the same cells (large, apolar) no intact nuclei could be detected by DAPI staining. Viability tests showed about 20% decrease in survival rate of GFP-A-RAF149 expressing cells, which is in good agreement with the percentage of apolar cells found in the culture. In a FM4-64 uptake assay we found that yeasts expressing the AR-149 were affected in endocytosis conversely endocytosis was unaffected when GFP-A-RAF was expressed. To test the effect of the fusion partner and of high level expression of A-RAF and of AR149 on yeasts, these proteins were expressed as GST-fusions under the strongly inducible GAL10 promoter. Cells were transformed with GST-A-RAF or GST- AR149, streaked on glucose- (“promoter off”) or galactose (“promoter on”) media and grown at 30°C. shows that whereas all three strains grew well in the repressive medium, in the inductive medium the GST-AR149 expressing strain did not grow. Taken together, AR149, but not full-length A-RAF, localizes homogeneously to the plasma membrane, and blocks yeast cell polarization, actin polymerization, and endocytosis, resulting in cell growth inhibition. Moreover cells overexpressing AR149 from a strong GAL10 promoter were not viable. ## Localization of A-RAF and AR149 in human cell lines The dramatic differences in localization between B- and C-RAF versus A-RAF and its N-terminal fragment, AR149, observed in yeast, prompted us to analyze the latter proteins in mammalian cell lines. AR149 expression had a strong negative effect on actin polymerization in yeast. To test this effect in mammalian cells, we chose NIH 3T3 cell line known for extended fibers of polymerized actin called stress fibers. Similar to yeast, stress fibers in NIH 3T3 cells were diminished by GFP-AR149 expression. This effect was also described for DA-RAF1, 2 by Yokoyama et al. (2007). We next examined the subcellular localization of A-RAF and AR149 in HeLa cells. Cells were transfected with GFP-tagged A-RAF and AR149 respectively. For control, an N-terminal fragment of C-RAF (aar 1–256) described originally as a dominant inhibitory RAF mutant that corresponds to AR149 (named C-RAF-C4) was included in the analysis. As observed in yeast, AR149 was unique in that it labeled the plasma membrane. Additionally in HeLa cells “beads on a string” structures were seen, demonstrating isotype specificity of the lipid binding site in CR1 as B- and C-RAF were reported to localize instead to ER/Golgi complex or mitochondria respectively. Disruption of microtubules by Nocodazole treatment separated beads from the strings. The stained vesicular tubular structures most likely represent recycling endosomes. To test this hypothesis, we studied the localization of RFP-AR149 and that of an established endosomal marker, ARF6. As documented in, there is a high degree of colocalization between RFP-AR149 and GFP-ARF6. ARF6 was reported to regulate central steps of endocytosis and recycling of endocytosed material by early and recycling endosomes (see for review). Like nearly all GTPases, ARF6 functions as a molecular switch alternating between GTP-bound (active) and GDP bound (inactive) states. Consequently, overexpressed GTP-, GDP-locked or nucleotide-free mutants of ARF6 (Q67L, T27N and N121I respectively) have distinct dominant effects on endosomal morphology and endocytosis. In co-transfected cells we found colocalization of RFP-AR149 or wild type A-RAF with GFP-ARF6 in tubular vesicular structures. In summary, AR149 and wild type A-RAF, but not C-RAF-C4 localizes to tubular endosomes, as proven by specific colocalization with ARF6. ## A fraction of endogenous A-RAF localizes to vesicles located along microtubuli Considering the dominant negative effect of AR149 on endocytosis in yeast, we set out to determine whether endogenous A-RAF in HeLa cells was localized on endocytic vesicles. After cytosol depletion, fixation and immunostaining with specific antibodies the remaining endogenous full-length A-RAF was found associated with vesicular structures. A-RAF positive vesicles were localized in the periplasmic area and at nearly each plus end of microtubules. A significant fraction of A-RAF- positive punctate structures is lining the microtubular network that extends to the perinuclear area. A-RAF siRNA removed the punctate labeling, but not the diffuse staining over the nucleus that was also seen with normal rabbit IgG, and is therefore considered non-specific background. Consistently in cell fractionation experiments a significant part of A-RAF fractionates with cytoskeletal, but not with the nuclear fraction. ## Overexpressed AR149 exhibits a dominant inhibitory effect on transport and traps internalized transferrin in ARF6 and RAB11 positive endosomes We asked whether overexpression of AR149 affects endocytic trafficking in mammalian cells. First, uptake and recycling of Tfn was investigated. Tfn is internalized upon binding to its receptor via a clathrin-dependent pathway and, after pH-regulated release, recycles back from the recycling/late endosomes. These endosomes are characterized by their accumulation in the pericentriolar space. GFP or GFP-AR149 transfected HeLa cells were incubated with Alexa-546 labelled Tfn for the indicated time. In control cells, after 30 to 40 minutes, endocytosed Tfn accumulated in the pericentriolar compartment. In GFP-AR149 expressing cells, relocation of endocytosed Tfn to pericentriolar compartments was significantly inhibited. The difference is clearly visible at the 40 min timepoint, where transfected and non-transfected cells are next to each other. Therefore expression of AR149 does not influence the endocytic uptake of Tfn, but strongly inhibits its relocation to recycling endosomes. To pinpoint the cellular endosome compartment, which Tfn is trapped in, EEA1, RAB11 and ARF6 were used as markers for early and recycling endosomes. The data clearly show significant overlap between AR149, internalized Tfn and ARF6- as well as between AR149, Tfn and RAB11-positive vesicles. Consistently, EEA1 positive vesicles did not accumulate Tfn. Thus we conclude that the block in endocytic trafficking is at the level of tubular recycling endosomes. ## siRNA silencing of A-RAF and MEK inhibition mimicks the AR149 overexpression phenotype The splice variant DA-RAF was shown to function as an effective inhibitor of RAF-ERK signaling. As AR149 is expected to share this function, AR149 overexpression or A-RAF depletion should have the same effect on Tfn trafficking. For depletion we prepared a set of A-RAF specific siRNAs and transfected them into HeLa cells. Degree of depletion and specificity of A-RAF siRNAs were determined by immunoblotting. As documented in significant depletion of A-RAF but not of the other two RAF isoforms was achieved. Semi-quantitative RT PCR showed that siRNA down-regulates A-RAF mRNA selectively without affecting the mRNA levels of DA-RAF, which was expressed at a lower level than full length A-RAF in this cell line. To test the effect of A-RAF depletion on Tfn trafficking, A-RAF siRNA- transfected, and control (scrambled siRNA-transfected) cells were subjected to a Tfn uptake assay. Similar to GFP-AR149 expressing cells, defective or significantly delayed accumulation of endocytosed Tfn in the pericentriolar space was observed indicating that A-RAF kinase activity and thus the activation of the mitogenic cascade is required for normal Tfn trafficking. Consistent with a requirement of localized ERK activation, chemical inhibition of the mitogenic cascade by U0126, a specific inhibitor of the RAF effector MEK, prevented the aggregation of Tfn positive endosomes in a similar way as A-RAF depletion. As the phenotypes of A-RAF depletion and mitogenic cascade inhibition overlap, we conclude that localized ERK signalling is required for endosomal maturation and AR149/DA-RAF functions as a dominant negative inhibitor of A-RAF on endosomes. ## Expression of GDP-locked ARF6(T27N) also blocks accumulation of Tfn in the pericentriolar endosomal compartment Dominant negative ARF6 has been shown previously to interfere with internalisation of Tfn. To test whether the pattern of Tfn accumulation by ARF6(T27N) and AR149 are comparable transfected HeLa cells were examined for co- localization of internalized Tfn and ARF6(T27N). Inspection of and reveals similar re-distribution of Tfn in HeLa cells in which A-RAF has been knocked down or inhibited (AR149 or U0126) and HeLa cells that express ARF6(T27N). This data suggests that ARF6 may be a target for regulation by A-RAF on endosomes. ## Tfn accumulates in ARF6 positive and RAB11 positive vesicles after A-RAF knock down To characterize the endosome compartment in which Tfn accumulates after A-RAF knock down ARF6 and RAB11 were used as markers. Additionally we included EEA1 as marker of early endosomes. Co-localisation experiments in HeLa cells identify vesicles with accumulated Tfn as RAB11- and ARF6- positive. No EEA1 Tfn double positive vesicles were found. We conclude that the block in Tfn trafficking in the absence of A-RAF lies between tubular- and TGN-assotiated recycling endosome compartments. ## AR149 Interacts with ARF6 independent of nucleotide status We have shown that AR149 and both (active and inactive) ARF6 mutants partially colocalize on endocytic vesicles and the plasma membrane. We were interested whether interaction of these two proteins could be shown at the biochemical level. HeLa cells were co-transfected with GFP, GFP-AR149 and HA-tagged ARF6 wild type, GTP-locked (Q67L) or GDP-locked (T27N) mutants. After immunoprecipitation with anti-GFP antibodies, the precipitates were separated by SDS PAGE and tested for bound ARF6 by immunoblotting against HA. In a reverse experiment ARF6 mutants were first precipitated and the presence of GFP-A-RAF in the precipitate was demonstrated by immunoblotting with anti-GFP antibodies. documents that similar amounts of ARF6, ARF6(Q67L) and ARF6(T27N) co- precipitated with GFP-AR149, irrespective of which protein was precipitated first. Thus interaction of A-RAF with ARF6 is independent of the ARF6 nucleotide status. ## A-RAF and AR149 regulate ARF6 activation Next we addressed whether overexpression of dominant negative AR149 has an effect on activation of ARF6. ARF6 interacts specifically with its effector GGA3 in a nucleotide dependent manner. Therefore binding of GGA3 can be used as readout of the activation status of ARF6. In a control experiment, cell lysates from COS7 cells transfected with wild type, dominantly activated (Q67L) and dominant negative (T27N) mutants of HA-ARF6 were incubated with immobilized GST- GGA3 and washed. As shown in, GGA3 pulls effectively ARF6(Q67L), and less effectively ARF6wt. In contrast, no binding of ARF6(T27N) to GGA3 could be detected. In subsequent experiments, COS7 cells were starved and induced with EGF, a strong elicitor of mitogenic signalling. Activated ARF6 was pulled-down by immobilized GGA3 as described above. Large amounts of ARF6 were pulled-down from EGF stimulated cells compared to non-stimulated cells. Thus, ARF6 is activated by the mitogenic cascade. This activation is significantly diminished by the coexpression of AR149 or A-RAF knock down. Of note pulled-down ARF6 contains coexpressed AR149, which precludes competition of AR149 and GGA3 for binding to activated ARF6. To pinpoint the position of AR149 action in a signalling pathway, COS7 cells were co-transfected with HA-ARF6 and either AR149, EFA6 or both. It has been previously described that overexpression of the ARF6 exchange factor, EFA6, increases the amount of ARF6 bound to GST-GGA3 beads. As depicted in, less ARF6 was pulled down in the presence of AR149. Similarly, A-RAF knock down reduced the amount of ARF6 that could be pulled-down with GST-GGA3. EFA6 partially rescued this effect. We conclude that AR149 co-expression or A-RAF knock down negatively affects the activation of ARF6 by EFA6, i.e. A-RAF functions upstream of ARF6. # Discussion ## A-RAF is a distinct member of the RAF kinase family In this work we unveil an unexpected link between A-RAF and regulation of ARF6 activity. A-RAF possesses several features that set it apart from the other RAF kinases. *A-RAF* maps to the X chromosome and is the only steroid hormone-regulated *RAF* isoform. A-RAF protein has substitutions in a negatively charged region immediately upstream of the kinase domain (N-region), which is at least partially responsible for its low basal activity. In contrast to C-RAF, feedback phosphorylation of A-RAF by its downstream effector, ERK, has an activating effect on its kinase activity. Interestingly, out of 590 kinases tested, the three RAF isoforms were among 208 kinases affecting clathrin- or caveolae- dependent endocytosis. si-RNA mediated silencing of *A-RAF* inhibits, whereas that of *C*- and *B-RAF* activate SV40 uptake. ## Role of lipid binding domains, identification and properties of inhibitory A-RAF, AR149 Our initial studies on RAF distribution in a heterologous yeast system showed that of the three RAF isoforms, only A-RAF located to the cell cortex of yeast cells. This pattern resembles that of components of the endocytic machinery, such as Sla1p, Sla2p, clathrin light chain protein Clc1p. Differences in localization of A-RAF and AR149 can be explained by the existence of two lipid- binding domains in the structure of A-RAF, each domain binding to specific sets of lipids. The unique distribution pattern of A-RAF in yeast resembles the distribution of a PtdIns(4,5)P₂ sensor, AP180 N-terminal homology (ANTH) domain. Consistently, A-RAF is the only RAF isoform, which binds to immobilized PtdIns(4,5)P₂. Using a set of point mutations, Johnson et al. (2005) located PtdIns(4,5)P₂ binding site in the RAS binding domain of A-RAF. In contrast, we mapped PtdIns(4,5)P₂ binding site to the cysteine rich domain of A-RAF because a RAS binding domain deleted mutant (88–606) still shows wild type distribution. In agreement with Johnson et al (2005), replacement of arginine 52 in the RBD with leucine, which prevents binding of C-RAF RBD to RAS, disturbs the membrane localization of both full-length A-RAF and AR149 in yeast. We interpret this discrepancy by an interaction between RBD and CRD. It is known that RBD and CRD bind to RAS cooperatively. The R52L mutation in the RBD may cause structural rearrangements disabling the interaction of the PtdIns(4,5)P₂ binding site in the CRD with membranes. CRDs are conserved among RAF isoforms, but only the A-RAF CRD possesses the unique property of PtdIns(4,5)P₂ binding. However as we show here the PtdIns(4,5)P₂ binding site *per se* is not sufficient to specify the A-RAF wild type localization pattern. The wild type pattern of membrane binding requires in addition the PA binding site in the catalytic domain. In regard to the colocalization of A-RAF with ARF6 in mammalian cells, it is noteworthy that ARF6 is known to regulate the PtdIns(4,5)P₂ content of membrane microdomains to which it binds. In yeast Arf3p is the functional homolog of ARF6. Arf3p regulates PtdIns(4,5)P₂ levels, endocytosis and actin polymerization,. Therefore it is likely that the dominant lethal phenotype of AR149 in yeast is mediated by interference with Arf3p. ## Localization of A-RAF and AR149 in mammalian cells The differences between RAF isoforms were most pronounced in yeast, where evidence for membrane attachment was restricted to A-RAF. The punctate cortical structures that are sites of A-RAF accumulation in yeast are known to contain proteins associated with early steps of endocytosis such as AP180 as well as Pan1p, that is associated with cortical actin patches. It is tempting to speculate that in mammalian cells the localization of A-RAF not only brings A-RAF into the neighbourhood of ARF6 on tubular endosomes, but additionally involves interaction with mammalian homologs of AP180/Pan1p that function at the interface of clathrin coated vesicles and an actin cytoskeleton regulatory complex. Such complexes are essential for endocytosis and also linked with the microtubule network. Consistently GFP-AR149 decorated predominantly tubulo- vesicular endosomes, as confirmed by co-localization with ARF6. These endosomes lost their beads on the strings appearance upon Nocodazole treatment. ## Role of A-RAF and AR149 in endocytosis Definitions of endocytic compartments are poorly standardized. Here we used definitions given in, where endocytosed Tfn is first enclosed in small vesicles. These vesicles fuse to form early endosomes. Further maturation of early endosomes through tubular recycling endosomes leads to the pericentriolar TGN- associated recycling compartment. AR149 is targeted to tubular endosomes where it colocalizes with ARF6. Exogenous expression of AR149 or dominant negative ARF6(T27N) mutant causes trapping of endocytosed Tfn in tubular endosomes, which prevents transfer to pericentriolar recycling compartment and subsequent return to the plasma membrane. AR149 and ARF6 occur in the same complex as demonstrated both by fluorescence microscopy and coprecipitation experiments. A-RAF kinase was previously reported to participate in regulation of caveolae/raft-mediated endocytosis by stabilizing the caveolar coat. Our discovery of a requirement for A-RAF activity at a step subsequent to endocytosis, in the transfer to the recycling compartment, point to a broader role of A-RAF in membrane trafficking that additionally involves feedback regulation by the inhibitory splice variants DA-RAF 1,2. Our data suggest that DA-RAF does not necessarily work as a general inhibitor of mitogenic signalling as initially described. More likely, due to its unique intracellular localization, AR149/DA-RAF primarily inhibits just a specific endosome- associated branch of mitogenic signalling, which is responsible for regulation of receptor recycling and/or restoration of signalling molecules. The classical mechanism of cytoskeleton dependent endocytosis is subdivided in two parts: first, a short-distance step in the cortical area depends on actin and second long-distance microtubule dependent vesicular transport. Our micrographs do not provide time resolution, but considering our functional analyses, it is conceivable that A-RAF functionally associates with the endosomes shuttling along microtubules between plasma membrane and the perinuclear recycling compartment. From the functional analysis of Tfn endocytosis, we conclude that not only localization to endosomes but also activity of A-RAF kinase in the mitogenic cascade on endosomes are a prerequisite for the translocation of Tfn-positive endosomes to the pericentriolar region. ## ARF6 regulation by A-RAF and AR149 Our data suggest that A-RAF functions upstream of ARF6. There has been an earlier report by Robertson, who suggested, based on epistasis experiments with ERK and ARF6 that ERK functions upstream and downstream of ARF6. Several other reports also suggested dependence of ERK activation on ARF6, which is in direct contrast with the results of our epistasis analysis. This contradiction may be explained by a pleiotropic effect of active ARF6, which is known to block internalization of activated receptors, leading to sustained signalling through the mitogenic cascade. Our data additionally suggest EFA6 as a cooperation partner of A-RAF in the activation of ARF6. Of note inspection of the primary structure of EFA6 revealed multiple potential ERK/MAP kinase phosphorylation sites that suggest that EFA6 is a substrate of ERK downstream of A-RAF. There are several examples for crosstalk between endocytosis and signal transduction. In addition, there is a growing body of evidence that at least part of mitogenic signalling takes place on endosomes: for example, the MEK partner MP1 was isolated as a component of endosomal vesicles and associates with MEK and participates in signal transduction. Another scaffold of the mitogenic cascade, KSR1, was also shown to localize and mediate signalling to endosomes. ## Model of A-RAF function in regulation of endocytosis The novel findings on the A-RAF localization and the interaction with ARF6 have led to a new model of A-RAF function shown in. Stimulation of growth factor receptors is followed by RAS activation. RAS•GTP recruits RAFs and leads to assembly of the RAF-MEK-ERK module at membranes. The three RAF isozymes become activated consecutively and mediate diversification of the signal to different subcellular compartments. A-RAF activation is delayed because it has a corequirement for ERK phosphorylation, which uncouples A-RAF from B- and C-RAF that in turn are inhibited by feedback phosphorylation. Due to the unique localization of A-RAF to endosomes, this delay is optimal for regulation of later endocytic events such as recycling of receptors. A-RAF function at endosomes also involves the mitogenic cascade and triggers ARF6 activation possibly via EFA6. A-RAF involvement in ARF6 dependent endocytic recycling provides a new perspective for explaining the phenotype of A-RAF knock out mice, which exhibit severe neurological defects such as ataxia, rigidity of the musculature and continuous tremor. Intriguingly, A-RAF, ARF6 and EFA6 are strongly expressed in Purkinje cells of mouse cerebellum. Along this line, DA-RAF expression is particularly strong in brain. Endocytosis and rapid recycling of synaptic vesicles is critically important for the physiological function of neurons, which may further stress the role of A-RAF in the nervous system. # Supporting Information We thank Margaret M. Chou and Xiang Dong Gao (University of Pennsylvania) for plasmids and helpful comments on this work, Angela Baljuls and Antoine Galmiche (this laboratory) for donation of plasmids and sharing protocols and results prior to publication, Eugen Kerkhoff (University of Augsburg) for kindly provided pEGFP-Rab11 plasmid. [^1]: Conceived and designed the experiments: EN SA MB URR. Performed the experiments: EN SA MB. Analyzed the data: EN SA MB URR. Contributed reagents/materials/analysis tools: URR. Wrote the paper: EN SA MB URR. [^2]: The authors have declared that no competing interests exist.

# Introduction Ceacam1 is a member of a large family of carcinoembryonic antigen proteins. It is primarily a type I transmembrane protein with multiple splice variants, though soluble forms also exist. Ceacam1 is widely expressed on a variety of tissues including endothelium, epithelium, hematopoietic cells and both hematologic and solid tumors, and interacts in a homophilic and heterophilic fashion with physiological and pathogen-associated ligands, including carcinoembryonic antigen and the *Neisseria spp.* proteins. Some Ceacam1 isoforms contain intracellular ITIM motifs, and activation of Ceacam1 results in the recruitment of the SHP-1 and SHP-2 phosphatases, which dephosphorylate substrates across a range of signaling pathways. Ceacam1 thus inhibits T cell receptor (TCR) signaling and suppresses multiple aspects of T cell function. Ceacam1 agonists attenuate cytokine secretion, T cell polarization and cytolytic function. *In vivo*, ligation of Ceacam1 with soluble ligands or over-expression of ITIM-containing Ceacam1 isoforms on T cells attenuates experimental colitis. Additionally, Ceacam1 is also expressed on intestinal T cells in patients with Celiac disease and ulcerative colitis, and may represent an attempt by the immune system to negatively regulate these inflammatory processes. In addition to immune regulation, Ceacam1 exerts a wide variety of other biological functions. It is a cell-cell adhesion molecule, and a receptor for a variety of commensal and pathogenic microbes in mouse and man. Ceacam1 also regulates angiogenesis, energy homeostasis, and tumor biology. Ceacam1 regulates the tumorigenesis of colon cancers, and is a prognostic factor in lung adenocarcinoma. Tumor expression of Ceacam1 may regulate tumor angiogenesis and invasion, and the expression of both Ceacam1 and CEA by tumors may inhibit the functions of tumor infiltrating lymphocytes. Allo-BMT is an established therapy with curative intent for a variety of hematologic malignancies and non-malignant conditions. Alloreactive T cells of donor origin play a criticial role in both GVHD, a major complication of allo- BMT, and graft-versus-tumor activity, a major contributor to the efficacy of allo-BMT as a cancer therapy. Donor-recipient antigenic disparity, donor T cells, and tissue injury resulting in inflammation due to the conditioning regimen all contribute to GVHD, which primarily affects intestines, liver, skin and thymus. Ceacam1 is expressed both on leukocytes (especially T cells), as well as on epithelial and endothelial cells, which are prominent components of the parenchyma of the above-mentioned GVHD target organs. In addition, Ceacam1 is upregulated on many tumors. In this report, we assess the impact of Ceacam1 on alloreactive T cells in the donor allograft, as well as the effects of Ceacam1 deficiency on recipients of allo-BMT with respect to GVHD and GVT activity. # Results ## Ceacam1 on donor T cells and recipient tissues can regulate GVHD mortality We assessed Ceacam1 regulation of GHVD on donor T cells or recipients in two well-described major histocompatibility complex (MHC) class I/II-disparate models C57BL/6 (B6, H-2^b)→BALB/c (H-2^d) and BALB/c→B10.BR (H-2^k). We used Ceacam1^−/− B6 mice, Ceacam1-transgenic (Tg) B6 mice (described), and Ceacam1^−/− BALB/c mice as the source of donor T cells or recipients. In all experiments, recipients received split-dose lethal irradiation (BALB/c: 8.5 Gy, B10.BR: 11 Gy) and a graft of 5×10⁶ allogeneic T cell depleted bone marrow (TCD-BM) of wildtype (WT) origin, with or without splenic T cells. We first transplanted irradiated BALB/c mice with B6 TCD-BM with WT or Ceacam1^−/− T cells, and observed that recipients of Ceacam1^−/− T cells had significantly increased mortality compared to recipients of WT T cells (left). We confirmed this in a second MHC-disparate allo-BMT model, BALB/c→B10.BR (right). We next asked whether T cells overexpressing Ceacam1 would cause less disease, and transplanted BALB/c recipients with 0.5×10⁶ or 1×10⁶ donor WT or Ceacam 1-Tg T cells. At both doses, recipients of Ceacam1-Tg T cells showed attenuated mortality. Finally, we assessed the role of Ceacam1 on tissues of allo-BMT recipients, and transferred TCD-BM+T cells into WT vs. Ceacam1^−/− BALB/c recipients. This revealed that Ceacam1^−/− recipients had increased early (but not overall) mortality, with nearly 50% of mice succumbing within the first week. ## Ceacam1 is an important regulator of GVHD target organ damage We next asked whether Ceacam1 regulated GVHD target organ damage, and again assessed effects of Ceacam1 deficiency or overexpression on donor T cells, and Ceacam1^−/− allo-BMT recipients. We observed that recipients of Ceacam1^−/− T cells had more severe large intestinal GVHD. Surprisingly however, these mice exhibited less thymic GVHD, as determined by thymic cellularity and numbers of CD4⁺CD8⁺ double-positive (DP) thymocytes. Thymic cellularity from age/sex-matched non-transplanted animals are shown in. We also observed a modest trend towards less skin GVHD in recipients of Ceacam1^−/− T cells, suggesting that Ceacam1^−/− T cells caused preferential damage to the (large) intestines. In experiments comparing recipients of WT and Ceacam1-Tg T cells on day 21 post- transplant, we found that recipients of Ceacam1-Tg T cells demonstrated significantly less GVHD of the liver, intestines, and thymus compared to recipients of WT T cells, but similar skin GVHD. This appears to suggest that Ceacam1-Tg T cells caused less GVHD overall, with relatively little organ specificity. Finally, we assessed Ceacam1^−/− allo-BMT recipients on day 14 post- transplant. In correspondence with increased early GVHD mortality, Ceacam1^−/− allo-BMT recipients showed increased large bowel damage and thymic GVHD. ## Ceacam1 regulates donor T cell numbers in lymphoid tissues and target organs during GVHD We next assessed the numbers of donor CD4 and CD8 effector T cells after transfer of Ceacam1^−/− or Ceacam1-Tg T cell-containing allografts, or in Ceacam1^−/− allo-BMT recipients. Comparing recipients of WT T cells with those receiving Ceacam1^−/− T cells, we observed increased numbers of Ceacam1^−/− donor alloactivated effector T cells in the spleen, MLN, and IEL of allo-BMT recipients, which was associated with a concomitant decrease in the number of Ceacam1^−/− alloactivated CD4 and CD8 T cells in the PLN and liver. When we analyzed organs of recipients of allografts containing WT vs. Ceacam1-Tg T cells, we noted decreased numbers of donor effector T cells in the MLN, PLN, and liver. Via histopathological analysis, we also observed decreased numbers of total lymphocytic infiltrates into the liver, small and large bowels by histopathology, as well as decreased neutrophilic infiltrates in these organs (data not shown). Finally, we assessed infiltrating T cells in WT and Ceacam1^−/− allo-BMT recipients, and observed increased numbers of donor alloactivated effector T cells in the MLN and IEL of Ceacam1^−/− allo-BMT recipients, but decreased numbers of these cells in the PLN and liver. ## Ceacam1 regulates the sensitivity of the small intestine to radiation injury The accelerated early mortality of Ceacam1^−/− allo-BMT recipients, together with increased accumulation of donor T cells in GI tract and mesenteric lymph nodes, but decreased numbers peripheral lymph nodes, led us to ask whether Ceacam1 had differential effects in regulating GVHD target organ damage for various target organs and tissues. In the context of Ceacam1^−/− recipients, we therefore tested the radiation sensitivity of Ceacam1^−/− mice used as hosts, by irradiating WT and Ceacam1^−/− BALB/c mice and assessing survival. Ceacam1^−/− animals showed increased kinetics of mortality, and in some cases, overall mortality after radiation injury. Similar results were obtained on the B6 background. We then enumerated regenerating and surviving crypts in the small intestine (terminal ileum) at 84 hours after irradiation to assess intestinal radiation damage, and observed that Ceacam1^−/− mice had fewer regenerating and surviving crypts as compared with WT counterparts, indicating greater damage to the small intestine across a wide range of radiation doses. It is thus quite likely that our radiation-containing conditioning regimen for transplant recipients also contributes in part to their survival kinetics, selective GVHD target organ damage, and the selective accumulation of donor T cells in lymphoid tissues and target organs in these recipients. ## Donor Ceacam1^−/− CD8 T cells express higher levels of integrin α₄β₇ post-transplant Next, we studied a variety of possible mechanisms by which Ceacam1 may regulate donor T cell function. We analyzed donor WT and Ceacam1^−/− alloactivated splenic T cells on day 14 after allo-BMT for trafficking molecules, and found that Ceacam1^−/− CD8⁺ CD44⁺CD62L⁻ effector T cells expressed higher levels of integrin β₇ subunit and the gut homing integrin α₄β₇, which is important for intestinal GVHD. However, WT vs. Ceacam1−/− CD4 effector T cells had similar integrin β₇ subunit expression, yet also accumulated in greater numbers in the gut, suggesting that regulation of target organ damage by Ceacam1 is very likely to involve multiple additional mechanisms beyond trafficking molecule expression. Additionally, levels of the α_E subunit, which forms integrin α_Eβ₇, were similar, as were levels of CCR9, CD31, PSGL1, CCR7, CXCR3, and LFA1 (data not shown). When we assessed the expression of trafficking molecules in recipients of WT vs. Ceacam1-Tg T cell allografts, we found no significant differences in levels of β₇ subunit, integrin α₄β₇, or any other molecules. This is consistent with the systemic reduction in GVHD in recipients of Ceacam1-Tg T cells. Finally, we assessed trafficking molecules in irradiated WT vs. Ceacam1^−/− recipients of identical donor allografts, and observed again that donor CD8, but not CD4 splenic T cells in Ceacam1^−/− recipients had a trend towards increased expression of the β₇ subunit, although this was not directly reflected in increased expression of integrin α₄β₇.. Taken together, these observations suggest that regulation of trafficking molecule expression by Ceacam1 is only one component of how it regulates GVHD target organ damage. ## Ceacam1 is expressed on T cells during alloactivation Ceacam1 can be found on activated T cells, and, we thus performed a kinetic analysis of Ceacam1 expression on T cells during alloactivation. We adoptively transferred CFSE-labeled B6 T cells into irradiated BALB/c recipients, and observed transient expression only on day 2 after alloactivation ( and data not shown). Furthermore, only CFSE^lo alloactivated T cells, which have divided ≥4 times in 48 hours, expressed low but consistently detectable levels of Ceacam1. These kinetics are consistent with a role for Ceacam1 in regulating early events in T cell alloactivation. ## Ceacam1 regulates the alloactivation and proliferation of T cells Because the expression of Ceacam1 on alloreactive T cells after adoptive transfer occurred *in vivo* with similar kinetics as T cell alloactivation, we asked whether Ceacam1 on either donor alloreactive T cells or radio-resistant cells in allo-BMT recipients could regulate this process. We transferred CFSE- labeled purified B6 WT or Ceacam1^−/− splenic T cells into irradiated BALB/c recipients and analyzed donor T cells in spleens on day 3. We observed that relative to isotype control staining, an increased percentage of alloactivated CFSE^lo CD4 Ceacam1^−/− T cells were positive for the alloactivation marker CD25, and that a greater percentage of these cells downregulated CD62L than WT T cells , suggesting that more of them became activated. Additionally, an increased percentage of donor Ceacam1^−/− CD4 T cells had divided to a CFSE^lo alloactivated state, suggesting enhanced proliferation in the absence of Ceacam1. We repeated these experiments with alloreactive Ceacam1-Tg T cells and as expected, observed a decrease in numbers of CFSE^lo T cells as assessed by CFSE dilution. This is consistent with an inhibitory role for Ceacam1 in the proliferation of alloreactive T cells. However, we did not observe significant differences in alloactivation between Ceacam1-Tg vs. WT donor T cells (data not shown). Lastly, we assessed the role of Ceacam1 expression on radio-resistant cells in allo-BMT recipients for donor T cell alloactivation. We transferred CFSE-labeled B6 T cells into irradiated WT vs. Ceacam1^−/− BALB/c mice, and analyzed donor T cells in spleens on day 3. Here, we did not observe differences in proliferation (data not shown), but donor CD4 T cells in Ceacam1^−/− allogeneic recipients did exhibit an increase in alloactivation as measured by CD25. ## Ceacam1 does not significantly influence T cell polarization, cytolysis or dendritic cell function in GVHD We measured serum cytokines in recipients of WT, Ceacam1-Tg and Ceacam1^−/− T cells on days 7 and 14 post-transplant, and observed that levels of IFNγ, TNF, IL-2, IL-4, IL-6, IL-10, and IL-12p70 were similar (data not shown). Percentages of FoxP3⁺ donor regulatory T cells and expression of T-bet were also similar between recipients of WT, Ceacam1-Tg and Ceacam1^−/− T cells (data not shown), and stimulation of splenocytes harvested on day 14 after BMT post-transplant from these three groups revealed essentially no IL-17⁺ donor T cells (not shown), and similar percentages of donor IFNγ⁺ T cells (data not shown). As Ceacam1 can regulate the cytolytic responses of lymphocytes, we assessed the cytolytic function of WT vs. Ceacam1^−/− alloactivated CD8 T cells from the spleens of allo-BMT recipients on day 14. Ceacam1^−/− CD8 T cells and WT CD8 T cells demonstrated similar cytolysis against ⁵¹Cr- radiolabeled allogeneic A20 B cell lymphoma cells and EL4 controls. Lastly, we found no differences in DC numbers, activation state (CD80, CD86, MHC class II) from the infusion of Ceacam1^−/− or Ceacam1-Tg T cells, or in Ceacam1^−/− allo-BMT recipients. ## Ceacam1^−/− donor T cells have enhanced graft-versus-tumor activity towards A20 lymphoma but not renal cell carcinom Finally, we assessed the GVT activity of Ceacam1^−/− donor alloreactive T cells against A20 lymphoma and RENCA renal cell carcinoma. Recipients of Ceacam1^−/− donor T cells had improved survival in the A20 lymphoma model, but both T cell replete groups showed comparable survival in the RENCA solid tumor model. When we analyzed these two tumor lines for Ceacam1 expression, we noted that all A20 lymphoma cells uniformly expressed high levels, while only a subset of RENCA cells expressed some Ceacam1. # Discussion In this report, we show that Ceacam1, which is found on both donor alloreactive T cells as well as non-hematopoietic tissues such as gastrointestinal and hepatic epithelium, can regulate both donor T cell function and the sensitivity of allo-BMT recipients to radiation-containing preparative regimens. In addition, Ceacam1 on donor T cells and tumors may modulate GVT activity. Ceacam1 on both the donor allograft and recipient tissues thus appears to represent an important regulator of GVHD and GVT morbidity and mortality via both T cell dependent and independent mechanisms, suggesting that therapeutic approaches which modulate Ceacam1 may need to assess and balance GVHD vs. GVT. Ceacam1 on T cells has previously been shown to restrain CD4 T cell polarization, cytokine secretion and cytotoxicity. In our GVHD model systems however, we found similar T cell polarization and cytokine secretion when we analyzed donor alloreactive T cells *ex vivo*. We ascribe this to the strongly proinflammatory cytokine milieu found in recipients following myeloablative radiation treatment, as well as the ubiquitous presence of alloantigen, which together promote strong Th1 differentiation regardless of Ceacam1 expression. However, in our model systems Ceacam1 regulated T cell activation, and numbers of donor alloactivated T cells in both lymphoid tissues and GVHD target tissues, in patterns that generally correlated with levels target organ damage ( **and** ). We also assessed the role of Ceacam1 in allo-BMT recipients. In our model systems, WT T cells in a Ceacam1-deficient environment showed a phenotype similar to that of Ceacam1^−/−alloactivated T cells: both showed increased activation, selective damage to the large intestines, and preferential accumulation in the MLN and intestinal parenchyma of mice with GVHD, and correspondingly decreased infiltration of the liver and PLN, ultimately leading to exacerbation of disease, with accelerated mortality in the first two weeks post-transplant. This suggests that Ceacam1 on donor T cells interacts with recipient tissues, and that Ceacam1 “fraternal” interactions between cells of the donor graft, were not sufficient to restrain GVHD in Ceacam1^−/− recipients. However, the increased early mortality of Ceacam1^−/− allo-BMT recipients with GVHD also led us to ask whether Ceacam1^−/− mice were sensitive to radiation injury. While Ceacam1^−/− mice were not significantly defective for hematopoiesis after sublethal irradiation at 3.5 and 4.5 Gy (data not shown), they did exhibit significantly increased damage to the small intestines after lethal irradiation. Ceacam1 also directly regulates intestinal epithelia. Due to enhanced Wnt/β-catenin signaling, Ceacam1^−/− jejunal and ileal enterocytes exhibit higher levels of the positive cell cycle regulators c-Myc and cyclin D1. Dysregulated c-Myc may sensitize cells to apoptosis, and higher levels of these proteins may render Ceacam1^−/− enterocytes more sensitive to radiation injury. Finally, Ceacam1 also regulates cell-cell adhesion, under normal and pathological conditions; it may therefore also be possible that loss of Ceacam1 regulates radiation-induced sloughing of intestinal epithelium. It is difficult to directly assess the relative importance of gastrointestinal radiation sensitivity versus increased GVHD in Ceacam1^−/− allo-BMT recipients, as radiation-induced gut damage may both be directly manifested in intestinal pathology, yet transmural migration of bacterial superantigens is an important first step for the initiation of GVHD, and increased damage to the intestines of Ceacam1^−/− mice may thus amplify the development of GVHD in these mice, and also explain in part the specifically increased large intestinal GVHD we observed. In experiments with Ceacam1^−/− donor T cells, we also observed a trend for splenic donor CD8 alloactivated T cells to express higher levels of α₄β₇. Although integrin α₄β₇ is important for GVHD pathogenesis, and we have previously shown that β₇^−/− T cells cause a sustained decrease in acute systemic and intestinal GVHD, differential expression of integrin α₄β₇ by Ceacam1^−/− T cells is almost certainly only one part of how Ceacam1 regulates target organ GVHD. Indeed, donor alloactivated CD4 T cells expressed comparable levels of integrin α₄β₇ as wildtype cells, yet were also found in increased numbers in the gut. This suggests that other mechanisms, such as Ceacam1 regulation of donor T cell activation may also contribute to its regulation of GVHD target organ damage. Moreover, recipients of Ceacam1-Tg T cells also had reduced intestinal infiltrates despite similar integrin α₄β₇ expression, suggesting that Ceacam1 regulates the accumulation of donor T cells in target tissues via multiple mechanisms. Thus, our results on donor lymphocyte infiltrates into GVHD target tissues and secondary lymphoid tissues must be interpreted cautiously, as they must be influenced by T cell proliferation, retention and apoptosis, in addition to trafficking. Although Ceacam1^−/− and Ceacam1-Tg T cells displayed overall symmetric and opposite phenotypes, we also noted differences. Ceacam1-Tg T cells primarily showed decreased proliferation, whereas Ceacam1^−/− T cells showed changes in proliferation, but also trafficking and activation. Some of these differences may be due to our models: on WT T cells, Ceacam1 is only briefly and transiently upregulated during activation. Consequently, Ceacam1^−/− T cells are “missing” Ceacam1 only transiently, while Ceacam1-Tg T cells constitutively over-express Ceacam1. Furthermore, while Ceacam1^−/− T cells are effectively insensitive to all Ceacam1 ligands and interactions, Ceacam1-Tg T cells which over-express the protein may have increased fraternal Ceacam1 interactions with other donor T cells, but may not necessarily experience increased Ceacam1 interactions with donor BM or host hematopoietic and non-hematopoietic components. These differences may explain why their activation and trafficking phenotypes are not directly opposed. We were interested to note that in our GVT experiments, recipients of Ceacam1^−/− T cells had significantly improved survival when challenged with A20 lymphoma but not renal cell carcinoma. Although both A20 lymphoma and renal cell carcinoma express Ceacam1, A20 cells uniformly expressed Ceacam1 at high levels, while only a subset of RENCA cells showed (somewhat lower) expression. Indeed, a number of hematologic tumors, including EL4 leukemia, P815 mastocytoma, and C1498 myeloid leukemia all express substantial levels of Ceacam1 (data not shown), whereas some solid tumors, such as mouse 4T1 breast epithelial cancer and CT51 colon tumor normally express only lower or even minimal levels of Ceacam1, similar to the lower level of expression we found with RENCA (not shown). Therefore, one possibility is that the GVT activity of T cells can be negatively regulated by tumors expressing high levels of Ceacam1, but is less important for tumors that express low levels or only on a subset of cells in the first place. However, RENCA in our GVT model systems is found primarily in the liver, and to a lesser extent, the lungs. Since donor allografts with Ceacam1^−/− T cells showed decreased numbers of donor alloreactive T cells in the liver as compared with wildype in GVHD experiments, interpretation of GVT activity against RENCA with respect to Ceacam1 on T cells must also consider this aspect of its biology. In conclusion, our results show that Ceacam1 on both donor T cells and allo-BMT recipients controls the proliferation, activation, and trafficking of donor alloreactive T cells, and the sensitivity of gastrointestinal tissues to irradiation. Consequently, Ceacam1 may represent a viable target for reducing radiation-associated gastrointestinal toxicity, for the control of GVHD and GVT activity after allo-BMT. # Materials and Methods ## Ethics Statement All animal protocols were approved by the Memorial Sloan-Kettering Cancer Center (MSKCC) Institutional Animal Care and Use Committee (protocol \#99-07-025). ## Mice C57BL/6 (B6, H-2^b), BALB/c (H-2^d), and B10.BR (H-2^k) mice were obtained from The Jackson Laboratory (Bar Harbor, ME). B6 and BALB/c Ceacam1^−/− mice, and B6 Ceacam1-Tg mice were generated at McGill University (B6 and BALB/c from Harlan (Montreal, Quebec, Canada)), and maintained at Memorial Sloan-Kettering Cancer Center. Mice used were between 8 and 12 weeks old. ## Bone Marrow Transplantation BM cells removed from femurs and tibias were T cell-depleted (TCD) with anti- Thy-1.2 and low-TOX-M rabbit complement (Cedarlane Laboratories, Hornby, ON, Canada). Enriched splenic T cells were obtained by nylon wool column passage. Cells were resuspended in DMEM and injected into lethally irradiated recipients on day 0 after total body irradiation (¹³⁷Cs source) as a split dose 3 hours apart. ## T cell carboxyfluorescein diacetate succinimidyl ester (CFSE) labeling and transfer Purified splenic T cells were incubated with CFSE (Invitrogen, Carlsbad, CA) at a concentration of 2.5–5 µM in PBS (5×10⁷ cells/mL) at 37°C for 20 minutes, washed twice with PBS, resuspended in DMEM and infused intravenously into lethally irradiated allogeneic recipients. Splenocytes from recipients were harvested at varying time points and analyzed by FACS as described. ## Assessment of GVHD Survival was monitored daily, and mice were scored weekly for 5 clinical parameters (weight, posture, activity level, fur ruffling, and skin lesions) on a scale from 0 to 2. A clinical GVHD score was generated by summation of the 5 criteria scores; mice scoring 5 or greater were considered moribund and euthanized. ## Histopathologic analysis Small and large bowel, liver, and skin were assessed by experts in a blinded fashion. Organs were preserved in formalin, transferred to 70% ethanol, and then embedded in paraffin, sectioned, stained with hematoxylin and eosin, and scored with a semi-quantitative scoring system. Bowel and liver were scored for 19 to 22 different parameters associated with GVHD (detailed); skin was evaluated for number of apoptotic cells/mm² of epidermis via terminal deoxynucleotide transferase dUTP nick end labeling (TUNEL). ## Statistical analysis Histopathologic scores, median fluorescence intensities and cell counts were compared between groups using the nonparametric unpaired Mann-Whitney U test; the Mantel-Cox log-rank test was used for survival data. ## Additional methods Additional methods are described in. # Supporting Information The authors thank the staff of the Memorial Sloan-Kettering Cancer Center Research Animal Resources Center for excellent animal care. [^1]: Conceived and designed the experiments: SXL LWK A-MC-A RA AMH CT VMH JAR CL GM OA MRMvdB. Performed the experiments: SXL LWK A-MC-A RA AMH CT VMH JAR CL GM MS DS CK UKR NY JLB RRJ OP I-KN. Analyzed the data: SXL JAR CL GM OA MRMvdB. Contributed reagents/materials/analysis tools: RSB FM KVH NB. Wrote the paper: SXL OA MRMvdB. [^2]: ¶ These authors also contributed equally to this work. [^3]: The authors have declared that no competing interests exist.

# Introduction Spoken language, like other communication systems, evolved as a means for influencing the attitudes and behaviours of communication partners –. That spoken language is particularly powerful in this influence likely has two reasons. First, language is the only biological communication system that is truly generative. Unlike nonverbal messages, which are limited in number and scope, language comprises a set of arbitrary symbols whose combination allows for an infinite number of potentially complex and abstract messages. A second and equally important fact is that language uses as its vehicle the voice–a communication system already present in our pre-linguistic ancestors. Emotion induced bodily changes affect the functioning of the voice thereby modulating the rate, intensity, and spectral quality of vocalizations. These modulations, also referred to as prosody, add emotional significance to a verbal message thereby increasing its persuasive power. Past research investigated whether and how prosody augments the influence of spoken language on listeners. Of particular interest has been the question whether emotional prosody captures attention more readily than neutral prosody. Behavioral evidence to this effect comes from an investigation of spatial attention. Spatial locations are more effectively cued by emotional as compared to neutral vocalizations. Additionally, neuroimaging research provides evidence. For example, fMRI studies found larger activity in the superior temporal sulcus (STS) for emotional as compared to neutral prosody regardless of whether prosody was task-relevant. Given the role of the STS in higher order auditory processing, this observation suggests that emotional prosody recruits more processing resources and is thus more likely to be noticed. A similar conclusion was derived from auditory odd-ball studies using event-related potentials (ERPs). In such studies, participants typically perform a foreground task while a task-irrelevant auditory sequence is presented in the background. Rare auditory deviants elicit a mismatch negativity (MMN) indicative of pre-attentive change detection (for a review see). Importantly, this negativity is larger for vocal emotional as compared to neutral deviants, again suggesting that listeners are more likely to notice the former kind of utterance. A second focus of interest in the study of prosody has been the integration of prosodic and verbal information. This has been investigated using both explicit emotion judgments and implicit priming paradigms. Explicit emotion judgment studies typically presented semantically neutral, negative, or positive valence words spoken with neutral, negative, or positive prosody. Thus, word valence and prosody were emotionally congruous or incongruous. Participants performed word valence judgments faster and more accurately when emotional prosody was congruous as compared to incongruous. Similar results emerged from implicit priming studies. Here participants performed lexical decisions on emotion words whose valence was congruous or incongruous to that of a preceding prosodic prime. Faster lexical decisions were observed for the earlier as compared to the latter condition. Functional neuroimaging evidence suggests that these effects reflect the retrieval of word information from semantic memory. Accordingly, such retrieval appears to be facilitated for congruous relative to incongruous prosodic and verbal emotions allowing congruous messages to be more easily understood and acted on. While these immediate effects of prosody on language processing are relatively well established, little is known about potentially sustained effects on listener attitudes and behavior. In particular, one may ask whether prosody influences the representation of a verbal message in long-term memory as that representation will determine whether and how people act on the message. One way such an influence may occur is by enhancing memory for verbal messages that are delivered with an emotional as compared to neutral prosody. Indirect support for this proposition comes from published memory research (for a review see –). Words, like other stimuli, were found to be better remembered when they convey an emotional as compared to neutral meaning. More importantly, memory for neutral words can be improved when they are embedded in an emotional sentence relative to when they are embedded in a neutral sentence. Thus, one may conclude that verbal context modulates memory for individual words and speculate that prosody, another form of context, may have a similar effect. This speculation is partially supported by a study on incidental speech processing. In this study, participants engaged in a numeric short-term memory task while passively listening to sentences pronounced with positive, neutral, or negative prosody. Incidental memory for negatively spoken sentences was higher than that for neutral or positively spoken sentences suggesting that negative prosody facilitated the storage of verbal information. However, as this finding was specific to an incidental encoding condition with a high short-term memory load, it is unclear how pervasive the influence of emotional prosody really is and whether it extends to a situation in which speech processing is intentional. A second way in which prosody could influence the representation of a verbal message in memory is by adjusting its emotional significance or valence. After all, words are just arbitrary combinations of phonemes that derive their valence from what they symbolize, which in turn derives its valence from experience. This experience can be direct through interactions with a word's referent or indirect through communications that relay such interactions. For example, after being bitten by a dog or learning from another individual that dogs bite, the word that represents dogs may come to symbolize threat and acquire a negative valence. Evidence for this comes from classical conditioning research demonstrating that individuals fear symbols that have been paired with an electric shock or for which they have been told that such a shock may occur. In both cases, they respond with increased physiological arousal relative to a symbol for which neither a direct nor an indirect negative experience is available. Given that words are symbols, one may infer that their emotional significance is equally malleable. Moreover, one may speculate that a word's context, such as speaker prosody, continuously modulates word valence. The present study probed this speculation and investigated whether and how prosody influences the storage of intentionally processed speech. Participants were asked to memorize a series of neutral words spoken with neutral or sad prosody. Subsequently, these words were presented together with new words in a visual word recognition test. In this test, participants indicated whether a word was old or new. Both old and new decisions were followed by a word valence rating for which participants judged each word on a 5 point scale ranging from −2 (very negative) to +2 (very positive). If emotional prosody influences word processing in the ways outlined above, we should observe better word recognition of old words that were studied with sad as compared to neutral prosody. Additionally, old words studied with sad prosody should be rated as more negative than old words studied with neutral prosody. # Methods ## Experiment 1 ### Participants Thirty-two undergraduate students participated in the experiment. Half the participants were female with an average age of 21.8 years (SD 2.4). Male participants were on average 22.7 years (SD 1.5). Participants were enrolled in an introductory level psychology module and received course credit for participating. All participants reported normal or corrected to normal vision as well as normal hearing. They signed informed consent prior to the experiment. ### Materials A set of 500 words was rated by a group of 30 independent raters (15 female) on two 5-point scales, one ranging from −2 (very negative) to +2 (very positive) for word valence and one ranging from 0 (non-arousing) to 4 (highly arousing) for arousal. Based on these ratings, 240 neutral valence (mean 0.16, SD 0.20), weakly arousing (mean 0.58, SD 0.24) words were selected. Frequency measures (Kucera-Francis Written Frequency: mean 57.2, SD 76.5) were obtained from the MRC Psycholinguistic Database. The speaker for this and the experiments reported below was selected based on a rating study. For this study, we invited four individuals with drama experience. These individuals were asked to portray the selected 240 words with anger, sadness, happiness and neutrality. All words were recorded and digitized at a 16 bit/44.1 KHz sampling rate. Word amplitude was normalized at the root-mean- square value using Adobe Audition 2.0. A subset of the same 15 words was selected for each prosodic condition and each speaker. These words were presented in random order to a group of 30 listeners (15 female) who were asked to indicate whether the speaker pronouncing a given word was in an angry, sad, happy, or neutral emotional state or in an emotional state not listed (e.g., disgust). They then had to rate each vocalization on a five-point scale from −2 (very negative) to +2 (very positive) with respect to emotional valence and on a five-point scale ranging from 0 (not aroused) to 4 (very aroused) with respect to arousal. For Experiment 1, we selected the speaker who portrayed sadness and neutrality better than all the other speakers. Her rating results are presented in. The average duration of words produced by this speaker was 1132.4 ms (SD 245.5) for sad prosody and 777.6 ms (SD 149) for neutral prosody. ### Procedure Experiment 1 employed a verbal memory paradigm consisting of two blocks with a study phase and a test phase each. A study phase comprised 60 trials. Each trial started with a fixation cross. After 500 ms, a word was presented over headphones while the fixation cross remained on the screen. The fixation cross disappeared at word offset. On half the trials, words were spoken with a sad prosody whereas the remaining trials used neutrally spoken words. The order of trials was randomized and the inter-trial interval (ITI) was 1000 ms. Each study phase was followed by a test phase comprising 120 trials. Again, each trial started with a 500 ms fixation cross. The cross was replaced by a word in the center of the computer screen. On half the trials, the word was from the preceding study phase, whereas on the remaining trials the word was new. Upon reading a word, participants were asked to press one of two buttons indicating whether the word was “old” or “new”. Once participants pressed the appropriate button, the word disappeared from the screen and they were now prompted to rate the valence of the word on a 5-point scale ranging from −2 (very negative) to +2 (very positive). The screen turned black after participants completed the rating and the next trial started after 1000 ms. The stimulus set of 240 words was split into four sets of 60 words each. These sets were presented as (1) old words with sad prosody, (2) old words with neutral prosody, and (3/4) new words. A given word was presented only once to a given participant. However, across participants, they appeared equally often as old and new words and equally often as words with sad and neutral prosody. Words were rotated in this way to avoid any stimulus confound on the effects of interest. To avoid a dexterity related response confound, we also counterbalanced the assignment of old/new judgments to left and right response buttons. Prior to the experiment, participants were instructed to listen to the words in each study phase and were informed that their memory for these words would be assessed in a subsequent word recognition test. In order to clear any doubts about the general procedure, participants performed a practice run composed of six study trials followed by 12 test trials using the dummy words from the stimulus recording. Test trials in this practice run comprised old/new decisions only. Participants were informed about the word valence rating only when commencing the word recognition test in the actual experiment. ### Results The results of Experiment 1 are illustrated in. The uncorrected probability of recognizing an old word as old was 0.77 (SD 0.15) for sad prosody and 0.79 (SD 0.14) for neutral prosody. A signal detection framework was applied to the analysis of the word recognition data. To this end, the probability of false alarms was calculated by dividing the number of new words incorrectly classified as old by the actual number of new words. Please note that this value did not differ as a function of prosody as all new words appeared in written form only. The probability of hits was calculated by dividing the number of correctly recognized old words by the actual number of old words in each prosody condition. Thus, hits differed as a function of prosody. A d' score was calculated by subtracting the normalized probability of false alarms from the normalized probability of hits for each prosody condition. The obtained d' scores were subjected to an ANOVA with *Prosody* as a repeated measures factor and *Sex* as a between subjects factor. This analysis revealed no significant effects (ps\>.2). A second ANOVA with reaction times to correctly recognized old words as the dependent variable was performed to assess the speed of memory access as a function of *Prosody* and *Sex*. This analysis was also non- significant (ps\>.18). The effect of speaker prosody on word valence was assessed by subjecting the valence ratings of correctly recognized old words to an ANOVA with *Prosody* as a repeated measures factor and *Sex* as a between subjects factor. This analysis revealed a main effect of *Prosody* (F(1,30) = 8.09, p\<.01) indicating that participants evaluated words as more negative, when these words were spoken with sad (mean 0.23, SD 0.74) as compared to neutral prosody during study (mean 0.43, SD 0.62). ### Discussion The results of Experiment 1 support the claim that speaker prosody has a sustained influence on listener attitudes and behaviour. Participants rated words as more negative if they had studied these words with sad as compared to neutral prosody. Contrary to expectation, however, Experiment 1 failed to reveal an influence of prosody on the accuracy or speed of word recognition. There are at least two possible reasons for this. First, prior work establishing a relationship between emotion and memory has relied on threat-related and/or highly arousing stimuli, (for a review see). Moreover, emotional memory effects have been linked to activation of the sympathetic nervous system and feedback from this system to brain structures implicated in memory consolidation (for a review see). As such, stimuli that are emotional but minimally arousing may not effectively enhance memory. Given that some consider sadness to be a low-arousal emotion, the sad prosody used here may not have been appropriate to study emotional memory. Alternatively, however, prosody may be irrelevant for the intentional storage of verbal information. Previous emotional context effects on intentional speech processing were based on a within-stimulus manipulation. Written words were presented together with other words of emotional or neutral meaning. In the present study, the context was of a different quality than the content. While the former was non-linguistic, the latter was linguistic in nature. Under these conditions transfer of emotional significance may not readily occur. A second experiment was conducted to probe these possibilities. While this experiment was comparable to the previous one in most respects, it differed in that study prosody was either happy or neutral. Happy prosody was selected because it reflects a high-arousal emotion and thus should induce arousal dependent memory facilitation if such facilitation exists for spoken words. Additionally, happy prosody allowed us to determine whether the observed prosodic effect on word valence could be replicated for a positive emotion. If true, words with happy study prosody should induce more positive subsequent ratings than words with neutral study prosody. ## Experiment 2 ### Participants Thirty-five undergraduate students participated in the experiment. Three participants were excluded from the data analysis. Two had a false alarm probability greater than 0.88 suggesting non-compliance with the task. One participant rated all word meanings with 0, suggesting exceptionally low emotion sensitivity or non-compliance with the task. Half of the remaining participants were female with an average age of 21 years (SD 0.8). Male participants were on average 21.44 years old (SD 1.3). Participants were enrolled in an introductory level psychology module and received course credit for participation. All participants reported normal or corrected to normal vision as well as normal hearing. They signed informed consent prior to the experiment. ### Materials The set of words selected for Experiment 1 was also used for Experiment 2. However, the words were spoken by a different female speaker. As for the first experiment, this speaker was selected based on her being best at conveying happiness and neutrality. The rating results for this speaker are presented in. The average duration of words spoken by her with a happy prosody was 680 ms (SD 98.7 ms) and that of words with a neutral prosody was 840.8 ms (SD 157). ### Procedures The procedures were identical to Experiment 1. ### Results The results of Experiment 2 are illustrated in. The uncorrected probability of recognizing an old word as old was 0.74 (SD 0.15) for happy prosody and 0.73 (SD 0.15) for neutral prosody. Discrimination sensitivity as a function of study prosody was again assessed by computing d' scores and subjecting these scores to an ANOVA with *Prosody* as a repeated measures factor and *Sex* as a between subjects factor. With all other effects being non-significant (ps\>.2), a marginal main effect of *Sex* suggested better word recognition in female as compared to male participants (F(1,30) = 4.13, p = .051). Again an ANOVA for reaction times was non-significant (ps\>.12). The effect of speaker prosody on word valence was assessed by subjecting the valence ratings of correctly recognized old words to an ANOVA with *Prosody* as a repeated measures factor and *Sex* as a between subjects factor. This analysis revealed a main effect of *Prosody* (F(1,30) = 4.89, p\<.05) indicating that participants evaluated words as more positive, if these words had been spoken with happy (mean 0.52, SD 0.45) as compared to neutral prosody during study (mean 0.35, SD 0.47). ### Discussion The results of Experiment 2 largely replicated those of Experiment 1. Prosody again failed to influence verbal memory, but significantly modulated word valence. Happy prosody resulted in more positive word valence ratings than neutral prosody. Together with the results from Experiment 1, this suggests that prosodic context modulates a word's affective representation in semantic memory. Positive and negative prosody increase and decrease the pleasantness associated with a given word, respectively. This effect may arise at three different processing stages. First, it may be a reflection of stimulus encoding. Specifically, a perceived mismatch between word valence and speaker prosody during study may lead to an immediate adjustment of word valence. Second, it may be a reflection of memory consolidation. Here, the adjustment would not be immediate but result from consolidation processes that bind prosodic context and word information (for a review see). As in the first case, however, the adjustment would be complete upon word retrieval and possibly independent from the listeners' ability to recollect study prosody. Finally, one may speculate that the influence of prosody on word valence arises during memory retrieval. Participants may remember prior prosodic context during word recognition and base their valence ratings on this memory. This could occur implicitly, without the participants being aware of it, or explicitly with participants consciously adjusting the valence ratings to accord with the remembered prosody. In either case, however, the word valence effect would depend on and therefore correlate with the participants' memory for prosody. Experiment 3 investigated this issue. As in Experiments 1 and 2, participants were presented with emotionally and neutrally spoken words during study and asked to memorize these words for a later recognition test. During test, they again performed an old/new judgment for each word. However, following this judgment they were now asked to either rate word valence or to indicate whether a word's prosody during study was neutral or emotional. The secondary judgments were performed in separate blocks and recorded as a within-participant variable to allow for a correlation analysis. If prosody modulates word valence during memory encoding or consolidation, memory for prosody should be irrelevant and hence may not correlate with the word valence effect. If, however, prosody modulates word valence during memory retrieval, memory for prosody should positively predict this modulation. ## Experiment 3 ### Participants Forty-eight undergraduate students participated in the experiment. Half the participants were female and on average of 21 years old (SD 1.9). Male participants were on average 22.2 years old (SD 1.5). Participants were enrolled in an introductory level psychology module and received course credits for participating. All participants reported normal or corrected to normal vision as well as normal hearing. They signed informed consent prior to the experiment. ### Materials The materials were identical to Experiment 1. ### Procedure Each participant completed two study phases each followed by one test phase. The instructions for both study phases were identical to Experiment 1 and 2. Participants were again asked to focus on the words and to remember the words for a later recognition test. Moreover, as in the preceding experiments, participants were instructed to make old/new judgments in both test phases. However, only in one test phase was this judgment followed by a word valence rating. In the other test phase, participants were asked to indicate for any word that was judged as “old” whether its study prosody was sad or neutral. These latter judgments were made by pressing one of two buttons on the response box. As for the preceding experiments, word lists were created, which were rotated across conditions and participants such that across participants each word appeared equally often as old or new word, equally often with sad or neutral prosody, and equally often in the word valence and the prosody memory tasks. We also counterbalanced the order of tasks and the assignment of left and right response buttons to the old/new and sad/neutral judgments. Prior to the experiment, participants were instructed to listen to the words in each study phase and informed that their memory for these words would be assessed in a subsequent word recognition test. In order to clear any doubts about the general procedure, participants performed a practice run composed of six study trials followed by 12 test trials using the dummy words from the stimulus recording. Test trials in this practice ran comprised old/new decisions only. Participants were informed about the word valence rating and the prosody memory task only when commencing the respective test block in the actual experiment. ### Results The results from Experiment 3 are presented in. The uncorrected probability of recognizing an old word as old was 0.68 (SD 0.18) for the emotional condition and 0.67 (SD 0.21) for the neutral condition. d' scores and reaction times were subjected to separate ANOVAs with *Prosody* as a repeated measures factor and *Sex* as a between subjects factor. Both analyses failed to reveal significant effects (ps\>.16). The influence of study prosody on a word's affective representation in semantic memory was assessed by subjecting the valence ratings of correctly recognized old words to an ANOVA with *Prosody* as a repeated measures factor and *Sex* as a between subjects factor. This analysis revealed a main effect of *Prosody* (F(1,44) = 6.56, p\<.05) with the other main effect and interaction being non- significant (ps\>.2). Thus, as in the two previous experiments, participants rated the valence of a word as more emotional if that word was presented with emotional (mean 0.24, SD 0.47) as compared to neutral prosody during study (mean 0.38, SD 0.41). Participant's ability to accurately remember a word's study prosody was assessed by calculating a d' score. False alarms were identified as correctly recognized old words for which study prosody was incorrectly specified as sad. Hits were identified as correctly recognized old words for which study prosody was correctly specified as sad. The normalized probability of false alarms (i.e., number of false alarms divided by the number of correctly recognized old words with neutral study prosody) was subtracted from the normalized probability of hits (i.e., number of hits divided by the number of correctly recognized old words with sad study prosody). The obtained d' scores were relatively small (Mean 0.54, SD 0.74) but differed significantly from zero (t(47) = 5.02, p\<.0001). Therefore, one can conclude that participants were better than chance in remembering study prosody. Finally, we assessed whether conscious recollection of study prosody accounts for the observed word valence effect in two separate analyses. First, we subtracted mean valence ratings of correctly recognized words with sad study prosody from those with neutral study prosody. Across participants, this score was positive as words with sad study prosody tended to have a more negative rating than words with neutral study prosody. A one-tailed Pearson correlation analysis was used to test for a positive relationship between this score and the prosody memory d'. This analysis was non-significant (r = .09, p = .27) suggesting that participants' ability to recollect prosody does not predict whether and how prosody affects their affective representation of words in semantic memory. A second analysis was aimed at verifying that the word valence effect reported above would still be significant if inter-subject variation in prosody memory was entered into the model. To this end, an analysis with *Prosody* as a repeated measures factor, *Sex* as a between subjects factor, and *Prosody Memory d'* as a co-variate was performed. The *Prosody* main effect was again significant (F(1,44) = 6.47, p\<.05). ### Discussion Experiment 3 replicates and extends the results of Experiments 1 and 2. Consistent with prior observations, the prosody effect on the speed and accuracy of verbal memory was non-significant reinforcing the idea that words are remembered equally well regardless of whether they are spoken with a neutral or an emotional prosody. Moreover, prosody again influenced word valence ratings indicating sustained prosodic effects on listeners. Analysis of prosody memory indicated that although participants were better than chance in remembering study prosody, their performance was nevertheless poor. Compared to the average d' for word recognition (mean 1.7, SD 1), the average d' associated with prosody recognition (mean 0.5, SD 0.7) was low. More importantly, however, the latter value failed to correlate with the word valence effect. Listeners who were good at remembering study prosody were not necessarily showing an influence of study prosody on word valence and vice versa. Thus, memory for prosody and the influence of prosody on word valence appear to be independent. # Discussion The present study investigated the influence of speaker prosody on the representation of verbal information in memory. Compared to neutrally spoken words, emotionally spoken words were expected to attract greater attention and to induce bodily arousal thereby enhancing memory for concurrent verbal information. Contrary to this expectation, however, word memory was comparable for neutrally and emotionally spoken words suggesting that prosody has little impact on memory storage of intentionally processed speech. This may be explained in several ways. First, an effect of prosody on memory formation presupposes that listeners perceive the intended emotion state implicitly. Thus, one may question whether the prosodic manipulation used here was strong enough to enable such perception. While the word recognition results may suggest a lack of emotional strength, the word valence ratings speak to the contrary. Specifically, across three experiments, participants reliably discriminated between emotional and neutral study prosody. Moreover, this discrimination was evident during word recognition when no prosody information was provided and showed regardless of whether prosody was task-relevant. Hence, one can conclude that the emotions conveyed by prosody during study could be processed implicitly and should have been available for memory formation. A second possible explanation for the failure of prosody to modulate verbal memory is that the emotions used here were inappropriate. To date, major evidence for an emotional facilitation of memory comes from studies that used threat related stimuli, raising the possibility that this facilitation is threat specific. However, some researchers identified memory facilitation for positive stimuli providing evidence that such facilitation exists across emotion categories. Moreover, a recent verbal memory study conducted in our lab compared the effect of neutral and angry prosody and obtained similar results. If asked to remember a series of spoken words, participants' subsequent word recognition did not benefit from the prosodic threat context. Interestingly, a benefit emerged when participants were instructed to forget the studied words. Based on this and the present evidence, one can conclude that emotional prosody, regardless of valence and quality, leaves intentional memory storage unaffected but has sustained effects on existing memory representations by modulating their affective connotation. That prosody fails to enhance intentional memory storage may be surprising. Comparable research using images, facial expressions, or words with affective or neutral connotations revealed relatively robust effects of emotion on memory. However, such stimuli also reliably activate one of the key brain structures implicated in emotional processing - the amygdala. In contrast, prosodic stimuli activate the amygdala less reliably. Most neuroimaging studies that compared emotional with neutral prosody in a whole brain analysis failed to identify amygdala contribution. Moreover, when such a contribution was identified it typically involved a regions-of-interest approach, suggesting that the emotion evoked by prosody is not as strong as that evoked by other stimuli. Thus, prosody may fail to evoke sufficient bodily arousal to enable amygdala-dependent memory facilitation. A potential reason for this is that prosodic emotional expression is constrained by language. Emotions can only be conveyed to the extent that they allow speakers to articulate a verbal message. If emotional vocal modulations become too dominant they may interfere with linguistic production and communication may break down. Support for this argument comes from studies investigating non- linguistic vocalizations such as laughing or crying. Their emotional connotation is more accurately identified than that of speech prosody. Moreover, like their facial analogues, these expressions reliably excite the amygdala and elicit bodily arousal – suggesting that vocalizations gain in emotional significance if they are freed from language. Moreover, like facial expressions or words, they may then be powerful enough to modulate memory storage. Although the present study revealed no influence of prosody on verbal memory it nevertheless points to sustained prosodic effects on listener attitudes towards words and, by association, word referents. Words heard with a negative prosody assimilate negativity and words heard with a positive prosody assimilate positivity. Through this process, prosodic context moderates whether an individual will approach or avoid a word's referent in the future. Interestingly, this occurs independently from an individual's ability to remember prosody suggesting that prosodic moderation of word valence precedes word retrieval. Moreover, given that in two of the three experiments prosody was task-irrelevant it appears to be an implicit process. Past research on the processing of prosody may offer insights into the mechanisms that underlie the observed valence effect. Specifically, work by Bach and colleagues identified the amygdala and left STS as being particularly important for implicit prosodic processing. In their study, both structures were more strongly activated when participants categorized prosodic emotion as compared to when they categorized speaker sex. Moreover, these activations emerged when collapsing emotional and neutral prosody suggesting that they represent processes that are emotion-unspecific. In the amygdala these processes likely reflect relevance detection and the modulation of regions associated with stimulus processing. In the STS these processes likely reflect higher order auditory functions such as the mapping of acoustic cues onto stored vocal representations with a particular significance to the individual. Additionally, through connections with other temporal and frontal lobe structures, both the amygdala and the STS communicate with regions involved in language processing. As such they may be critical in mediating the effects observed in the present study. For example, one could envision that vocal information represented in the STS is matched against verbal representations in regions posterior and inferior to the STS. In case of incongruity, the largely biologically determined vocal representations may shape the stored linguistic symbols. Evidence in support for this speculation comes from functional neuroimaging research that identified greater activation for emotional words spoken with incongruous as compared to congruous emotional prosody. Positive and negative words spoken with a negative or positive prosody, respectively, were found to activate the inferior frontal gyrus , –a structure implicated in word retrieval. Additional evidence comes from studies measuring event-related potentials (ERPs). These revealed a larger negativity around 400 ms following words with incongruous as compared to congruous emotional prosody (e.g., happily spoken “success”;). This is comparable to a negativity with frontal and temporal generators that is elicited for words presented in a semantically incongruous as compared to congruous sentence context. Importantly, the observed negativity is not only increased for complete incongruity but also for a partial incongruity as arising from a neutral word meaning and an emotional prosody. Based on this and the present results, one may speculate that in addition to modulating word retrieval, incongruity between prosody and word meaning triggers processes that calibrate linguistic representations to better map onto accompanying vocal context. Future research involving online measures of neural processing will be necessary to validate this hypothesis and contrast it with a potential modulation occurring after stimulus processing. Rather than stimulus encoding, it is possible that prosodic modulation of word valence occurs during memory consolidation where content and context are bound to enable integrative event memories (for a review see). Taken together, the present results extend the existing literature by highlighting sustained changes in verbal representations as a function of speaker prosody. As such they point to a mechanism by which words - in the course of repeated interactions and through the integration with other contextual cues - acquire an emotional significance that may be salient enough to excite automatic appraisal and lead to bodily arousal. The functionality of such a mechanism is easy to conceive. Among others, it would allow individuals to acquire adequate emotional responses, not just to a word's referent, but to the word itself allowing the word to effectively guide behaviour. This notion is in line with observations of language learning in childhood. Such observations revealed that adults use a different mode of speech when interacting with infants and young children as compared to adults. This mode, termed infant- directed (ID) speech, is produced at a higher pitch and with greater prosodic variation than the so called adult-directed (AD) speech. Researchers have proposed that ID speech serves attentional engagement and language learning by allowing infants to identify important units of speech. Additionally, ID speech has been implicated in emotional communication. Research by Trainor and colleagues revealed strong similarities between ID speech and emotionally expressive speech directed at adults. The authors, therefore, proposed that ID speech promotes emotional exchanges and bonding with the infant. The present results extend this idea. ID speech conveys not only relational emotional information but emotional information about communication referents. The child can thus learn which emotions correspond to which objects or events in the environment and link these emotions to the accompanying words. As for the adult participants tested here, these words then acquire a valence that informs subsequent behaviour. While providing intriguing evidence for sustained effects of speaker prosody on listeners, the present results should nevertheless be viewed preliminary. To better understand the modulation of stored word valence by speaker prosody, one may wish to examine the relationship between memory for prosody and word valence within a participant and within a given item. This was not possible here as different items were presented in the different tasks. Participants performed the word valence judgment on a different set of words than the prosody memory task. The rational for this was that if asked to remember prosody and judge word valence for the same item, participants would potentially confound the two. Future research could address this issue by using the same stimuli in a word valence task and a prosody memory task but separating them by several days. Alternatively, one could measure neuronal activity during initial and subsequent encounters with a word. This might allow the identification of encoding processes that predict later changes in word valence. To conclude, the present study found speaker prosody to be irrelevant for subsequent word recognition but important for shaping a word's affective representation in memory. Words produced with an emotional tone assimilate that tone thereby becoming more emotional themselves. Given that this occurs without intention and independently of memory for prosody, one can infer this process to be automatically triggered during speech processing. Through this, speakers can produce attitude changes in their listeners that outlast the moment and that allow their message to have a long-term influence on listener behaviour. The author would like to thank Trevor Penney for earlier comments on the manuscript. [^1]: Conceived and designed the experiments: AS. Performed the experiments: AS. Analyzed the data: AS. Wrote the paper: AS. [^2]: The author has declared that no competing interests exist.

# Introduction Hairdressers are exposed to a variety of chemicals everyday. Chemical waving, a very common hairstyling procedure, uses permanent-waving solutions (PWSs). The active ingredient of PWSs is thioglycolic acid (TGA), which is highly toxic. Hence, concerns on the use of PWSs have arisen given the escalating popularity of hairdressing over the last few decades. Hairdressers are frequently in contact with high amounts of TGA through the skin, inhalation, and even ingestion. Results of animal studies have indicated that contact with PWSs could lead to acute toxicity, such as irritation and burning of the eye, nose, throat, and lungs. Chronic toxicity, which lasts for months, could also occur. Frequent contact with PWSs has been reported to induce allergic reactions, immunological dysfunctions, and mutageneses. The risk of having a small-for-gestational-age infant – in hairdressers also increases. Indeed, intensive exposure to TGA potentially damages the health of both hairdressers and their customers. Percutaneous TGA administration in mice has been found to cause damages to the liver and DNA, signifying the possible mutagenic property of TGA. Other reports indicated that TGA exposure alters the ultrastructures of ovaries and testes, decreases the number of sperm, disturbs the estrous cycle in rats, and increases menoxenia in hairdressers. TGA was found to be harmful to animal offspring as well. In our previous studies, TGA was found to inhibit the *in vitro* maturation of mice and *Xenopus* oocytes. However, information on female *in vivo* reproductive toxicity of TGA is not yet available. The purpose of the present study is to investigate the effect of percutaneous TGA treatments on mouse oocyte maturation *in vivo*, and to observe the related phenomena. # Materials and Methods ## Chemicals Pregnant mare serum gonadotropin (PMSG) and human chorionic gonadotropin (hCG) were from the Ningbo Second Hormone Factory (China). Bovine serum albumin (BSA) was obtained from Promega (USA). Triton X-100 was purchased from Ameresco (USA). Rabbit anti-mouse tubulin antibody was purchased from the Lab Vision Corporation (USA), and fluorescein isothiocyanate (FITC)-conjugated goat anti-rabbit IgG was from the Zhongshan Goldenbridge Biotechnology Co. Ltd. (China). CZB medium was formulated as previously described. Unless noted, all other chemicals were from Sigma. ## Animals Female Kunming mice were provided by the Experimental Animal Center of the Second Hospital of the Harbin Medical University. The mice were maintained under the conditions of a 14L∶10D photoperiod, constant temperature, and 65% relative humidity. Food and water were available *ad libitum*. All animal care and experimental procedures were performed in compliance with the policies on the care and use of animals of the Ethical Committee of the Harbin Medical University. The mice were allowed to adapt for at least 1 week in the experimental conditions, and were then randomly assigned into four groups. A 2 cm×2 cm area on the back each mouse was shaved for TGA treatment. The mice were superovulated with an intraperitoneal injection of 10 IU PMSG followed by 10 IU hCG after 48 h. ## TGA treatment Based on earlier studies, the median lethal dose (LD₅₀) for percutaneous TGA administration into mice was 1210 mg TGA/kg body weight. Different body weight doses were designed for the experiments to investigate whether relatively low TGA doses, which theoretically cause no toxicity, are actually toxic to female reproductive functions. The doses were 37.81 mg/kg (1/32 LD₅₀), 75.62 mg/kg (1/16 LD₅₀), and 151.25 mg/kg (1/8 LD₅₀). TGA solutions of different concentrations were prepared in distilled water not more than 30 min before administration. The solutions were gently smeared (0.05 mL/10 g body weight) percutaneously on the depilated back of each mouse simultaneously with the hCG treatment. After 1 h, the backs of the mice were thoroughly washed with warm water, wiped dry, and coated with Vaseline. The animals were observed all the time throughout the treatment to prevent them from licking and rubbing each other. Using the same method above, the animals in the control group were treated with distilled water. TGA was administered to three mice per group. ## Collection of oocytes About 13–14 h after hCG injection, the mice were sacrificed via cervical dislocation, and the ovulated oocytes were collected. Cumulus cells were released from the oviductal ampullae by a brief exposure to 0.1% hyaluronidase in M2 medium. The cumulus-free oocytes were then washed in M2 medium three times. Only oocytes with first polar bodies were used for the observation of spindle morphology, CG distribution, and parthenogenetic activation. ## Immunofluorescence staining Oocyte fixation and labeling were performed according to the method of Zhu *et al.*. Briefly, oocytes were fixed with 4% paraformaldehyde in phosphate-buffered saline (PBS) for 30 min at room temperature. They were then permeabilized in incubation buffer (0.5% Triton X-100 in 20 mM Hepes, pH 7.4, 3 mM MgCl₂, 50 mM NaCl, 300 mM sucrose) for 15 min at 37°C in an incubator. Blocking in 1% BSA for 1 h and overnight incubation at 4°C with rabbit anti-α-tubulin antibody for spindle staining followed. After three washes in PBS with 0.1% Tween-20 and 0.1% Triton X-100 for 5 min, the oocytes were incubated with FITC-conjugated goat anti-rabbit IgG at 37°C for 1 h. The oocytes of the control group were only incubated with the FITC-conjugated secondary antibody. The total area (in µm²) of each spindle was calculated based on the method of Sanfins and on confocal microscope observations:*a*, *b*, and *c* are the widths of different parts of the spindle, and *h* is the total spindle length. For observing CG distribution, oocytes were first incubated at 37°C in acidified M2 medium for 1–2 min to remove the zona pellucida. After thorough washing in PBS with 0.1% Tween-20 and 0.1% Triton X-100, oocytes were incubated in 100 µg/mL FITC-Lens culinaris agglutinin for 1 h. Following three washes, the cells were incubated in 5 µg/ml propidium iodide (PI) for chromosome staining. After extensive rewashing, the samples were placed in droplets of 1,4-diazobicyclo(2,2,2)octane to avoid photobleaching, and were mounted under coverslips. The mounted samples were kept frozen and protected from light until observation. The oocytes were classified according to CG distribution. Oocytes with CGs arranged in clusters throughout the entire cytoplasms were classified as “immature,” those with CGs only in the periphery were “mature,” and those without CG (CG exocytosis) were “activated.” ## Oocyte activation The activating medium was Ca²⁺-free CZB supplemented with 10 mM SrCl₂. The oocytes were treated with SrCl₂ for 6 h, stained with 5 µg/ml PI in PBS for 5 min, and microscopically examined for evidence of activation. Oocytes were considered activated when each cell contained one or two well-developed pronuclei, or when two cells have pronuclei. ## Data analysis At least three replicates were conducted for each treatment. Data were expressed as means ± SD and were analyzed using ANOVA. *P* values less than 0.05 were considered significant. # Results ## Effect of TGA on the number of ovulated oocytes shows that the average numbers of ovulated oocytes were 32±4, 28±5, and 13±5 after TGA treatment at doses of 37.81, 75.62, and 151.25 mg/kg, respectively. There was no significant difference between the control (38±6) and 37.81 mg/kg TGA (32±4) groups. In contrast, the ovulated oocytes in the 151.25 mg/kg (13±5) TGA group were significantly less than in the control and 37.81 mg/kg TGA groups. The ovulated oocytes in the 75.62 mg/kg TGA group were less than in the 37.81 mg/kg TGA group significantly, but it is not different from the control group. ## Effect of TGA on the spindle morphology of ovulated oocytes shows that the ovulated oocytes displayed bipolar spindles with focused poles in the control and 37.81 mg/kg TGA groups. In contrast, the oocytes displayed a large barrel configuration in the 75.62 and 151.25 mg/kg TGA groups. The spindle areas in the four groups were 241.08±47.48, 271.88±59.66, 502.95±57.03, and 623.02±67.53 µm², respectively. Spindle area significantly increased with increased TGA dose, demonstrating a dose-dependent relationship. The spindle areas in the 75.62 and 151.25 mg/kg TGA groups were both significantly larger than in the control and 37.81 mg/kg TGA groups. The spindle area in the 151.25 mg/kg TGA group was also significantly larger than in the 75.62 mg/kg TGA group. ## Effect of TGA on CG distribution within ovulated oocytes CG distribution within the oocytes was observed under confocal microscopy by staining the CGs with immunocytochemicals. shows that after percutaneous TGA treatment in all groups, the CGs migrated to the cortex and formed a continuous layer under the oolemma. A CG-free domain (CGFD) was found, and there was no significant difference among the control and TGA-treated groups. ## Effect of TGA on *in vitro* ovulated oocyte activation After the oocytes were parthenogenetically activated by 10 mM SrCl₂, the percentage of activated oocytes decreased with increased TGA dose. shows that although treatment with 37.81 mg/kg TGA decreased oocyte activation, this change had no statistical significance. In the 75.62 and 151.25 mg/kg TGA groups, the percentages of activated oocytes were significantly lower than in the control and 37.81 mg/kg TGA groups. The percentage of activation in the 151.25 mg/kg TGA group was also significantly lower than in the 75.62 mg/kg group. # Discussion Exposure to TGA has become a very common and severely hazardous phenomenon given the rising popularity of hairdressing. Numerous studies have shown that TGA exposure could cause acute and subchronic damage to rats and mice, as well as reproductive toxicity. The latter requires further attention. Our previous study revealed that mice oocytes treated with TGA presented abnormal chromosomal arrangements and spindle configurations, as well as inhibited *in vitro* maturation. TGA inhibited the *in vitro* maturation of *Xenopus* oocytes by increasing germinal vesicle breakdown frequency as well as altering protein expression and phosphorylation involved in the maturation promoting factor (MPF) and mitogen-activated protein kinase (MAPK) pathways. To identify further the effect of TGA on reproductive functions as well as determine possible differences between *in vitro* and *in vivo* experiments, we investigated the activities of TGA *in vivo*. TGA doses that corresponded to 1/32, 1/16, and 1/8 LD₅₀ were percutaneously administered to female mice. The purpose was to investigate whether such low doses related to the experimental LD₅₀ could interfere in the *in vivo* development and maturation of mouse oocytes. Our results demonstrated that TGA doses of 1/8 LD₅₀ (151.25 mg/kg) and 1/16 LD₅₀ (75.62 mg/kg) decreased both the number of ovulated oocytes and the percentage of parthenogenetically activated oocytes, as well as caused abnormal and larger spindles. Actually, the doses we used also were low doses relevant to real-life exposure levels. Chinese national standard for permissible concentration of TGA in permanent-waving solutions (PWSs) (GB 7916-87) is 8%–10%. According to our treatment method (smearing percutaneously), the highest dose of TGA (151.25 mg/kg) was only close to 3.03%, which is much lower than the practical touching concentration for hairdressers and their customers. Of course, besides TGA concentration, the volume and the duration of using PWSs also influence the real-life exposure levels for human beings. Several factors are involved in the regulation of ovulation rate, such as intraovarian factors, exogenous hormone administration, and environmental factors. Among these, the hypothalamic-pituitary-ovary axis is believed to be the most important. In the present study, all ovulated oocytes were obtained from superovulated mice, but only after TGA treatment was mice ovulation inhibited. Therefore, we speculate that this inhibition could be attributed to TGA disturbing the hypothalamic-pituitary-ovary axis. However, if such a disturbance did occur, TGA did not completely block gonadotropin-induced superovulation at the doses we used. Moreover, nutrition affects follicle development. Zak et al demonstrated that the size of follicles and the rate of maturation of oocytes obtained from restricted-fed sows could be affected by the nutritonal history. TGA was reported to increase the glucose-6-phosphatase activity and cause protein and carbohydrate metabolism disorders in rat liver. Hence, we propose that these metabolism problems may contribute to the inhibiting effect of TGA on ovulation. Other factors such as ovarian autocrine, paracrine, and endocrine systems may also affect ovulation. Spindle formation and CG distribution are both indicators of oocyte maturation. A meiotic spindle is composed of microtubules and appears from the late G2 phase to the end of the M phase. Meiotic spindles are crucial for normal chromosome alignment as well as for the separation of maternal chromosomes during meiosis I and II. The normal MII spindle has a characteristic barrel shape. Spindles are sensitive to temperature , xenobiotic chemicals, and drugs. Recently, researchers used spindle analysis to assess oocyte quality as well as the effects of toxins and drugs on oocytes. The present study showed that *in vivo* TGA treatment induced abnormal, larger-barrel spindles, which agrees with the results of a previous *in vitro* experiment. Although the *in vivo* mechanism of spindle damage is not yet clear, MAPK and MPF may play essential roles in the regulation of microtubule dynamics. The effects of MAPK and MPF signals on *in vitro* oocyte maturation after TGA treatment was observed. This finding makes it rational to speculate that TGA altered the spindle morphology of mouse oocytes through the MAPK and MPF signaling pathways, but the mechanism warrants further investigation. CGs, which are about 0.2–0.6 µm in diameter, are round and membrane-bound organelles found in many invertebrate and vertebrate oocytes. Oocyte CGs originate from the Golgi apparatus and first appear at the onset of follicular growth. The CGs then migrate to the cortex and form a continuous layer under the oolemma during oocyte growth and maturation. When the oocytes are mature enough, a CG-free domain (CGFD) forms where the plasma membrane lacks microvilli. Previous studies in mice have demonstrated that sperm are less likely to penetrate an egg in the amicrovillar region. Therefore, CGs play a significant role in the blocking of polyspermic penetration in mammalian oocytes. An increasing number of studies have also suggested that CG distribution measures cytoplasmic maturation. In the present study, we detected CG distribution to determine if TGA also affected *in vivo* oocyte cytoplasmic maturation in mice and found that a CGFD still appeared after percutaneous TGA administration. This result was contrary to that observed in *in vitro* matured oocytes, where CG distribution was inhibited and the CGFD disappeared after TGA treatment. This difference may be explained on one hand by the fact that different doses were used in the *in vivo* and *in vitro* experiments, and that a relatively low TGA dose of 1/8 LD₅₀ was used in the present study. On the other hand, the metabolism and detoxification of TGA *in vivo* cannot be neglected. Spindle disturbance may severely affect oocyte development after activation. Therefore, the abnormal spindle configuration that resulted from the TGA treatment in mice suggested that oocyte activation might be compromised to investigate further the following oocoyte fate. When oocytes are activated, there is a series of Ca²⁺ oscillations. In the present paper, a strontium chloride-containing medium was used to induce intracellular Ca²⁺ release that triggers a series of Ca²⁺-dependent biological responses, and results in parthenogenetic activation. Considering that the parthenogenetic activation rate decreased in TGA-treated mice, we propose that TGA may alter Ca²⁺ concentration in oocytes or affect intracellular Ca²⁺ release, although the mechanism remains unclear. Furthermore, the inhibitory effect on the parthenogenetic activation of ovulated oocytes was significant in TGA groups at 75.62 and 151.25 mg/kg, in consistence with the effect on spindle area. Hence, we speculate that the decreased oocyte activation may have resulted from spindle formation perturbation caused by TGA. We also hypothesize that TGA affects the fertilization process, but further studies are needed. In conclusion, we found that *in vivo* TGA treatment decreased the number of ovulated oocytes and induced abnormal spindle configurations. TGA also inhibited the parthenogenetic activation of ovulated oocytes. TGA toxicity to reproductive functions deserves further investigations. The authors would like to thank Doctor Ling Zhang for his technical guidance, as well as our colleagues for the data analysis. [^1]: Conceived and designed the experiments: ZW. Performed the experiments: LX SH. Analyzed the data: LX XR. Contributed reagents/materials/analysis tools: LX SH. Wrote the paper: SH. Participated in the preparation and revision of the manuscript: LX ZW. Contributed to technical support: SH. [^2]: The authors have declared that no competing interests exist.

# Introduction Bark beetles have coexisted with their tree hosts since the early Mesozoic, and while often regarded as pests, bark beetles and their associated fungi also play important roles in nutrient cycling, forest dynamics and biodiversity. But of the more than 5800 described bark beetle species, less than a dozen, mostly in the genera *Dendroctonus* and *Ips* (e.g., *Ips typographus*, *Dendroctonus ponderosae*, *D. frontalis*) are known to colonize and kill even healthy host trees when population densities are high. Aggregation pheromones released by beetles while they are boring into and excavating mating galleries in host trees elicit attraction of conspecifics of both sexes, and the greater the number of beetles attacking, the greater their probability of exhausting host defences and achieving successful oviposition. The beetles also vector presumably mutualistic microorganisms (mostly fungi), some of which contribute to tree mortality. The Pinaceae have evolved defences against bark beetles and their associated fungi : resin-filled ducts can mechanically seal off the entrance holes, a number of compounds (terponoids and phenolics) with inhibitory or toxic effects on the beetles and fungi increases in concentration, and cell structure changes in the surrounding tissue helps contain the infection. The effectiveness of these defences varies over time within and between trees, as it is vulnerable to water stress and other biological factors. While a fallen tree with remaining root contact may still have partially active defences, broken trees are defenceless, and suppressed trees have reduced defences. As modular organisms, the “death” of a tree is neither instantaneous nor necessarily affecting the whole individual. Here, however, we define a “dead” tree as one that has no effective defense against a given species of bark beetle. This often results from loss of root contact, extreme drought stress, mechanical damage or parasites. Failure to find a suitable host is a major source of beetle mortality, and usable breeding habitat is patchy, stochastic and transient, as dead host trees appear randomly in the landscape through wind-felling or logging, after which the phloem decays within months, while forest succession take decades to centuries. On the other hand, partaking in an unsuccessful aggregation is at best a waste of time and at worst fatal, and the risk is likely to be greatest for the ones initiating the attack. Since the maximum reproductive success may be achieved at low to intermediate gallery densities, the beetles seem to face both positive and negative density dependence, on slightly different spatial scales. At very low densities, mate finding may be problematic, while increasing densities both cause crowding and facilitates colonization of living trees. Early colonizers of living trees meet the strongest tree defences and uncertain success, latecomers meet higher competition and, perhaps, increased predation. From an evolutionary perspective, this raises several questions: Why do individual beetles initiate or join with unrelated conspecifics in risky “cooperative” attacks? This is especially puzzling considering that the beetles initiating the attack face the greatest risks, and thus are the least likely to reap the benefits. Conversely, if such attacks represent an adaptive strategy, why are they nevertheless only sporadic, usually local and self-limiting, but at other times forming large-scale outbreaks that kill virtually all host trees over large areas, persisting long after the triggering resource pulse has ceased, ? Here we develop an individual-based model, hereafter called the Sequential Restricted Distribution (SRD), to study the adaptive strategies of host selection (dead vs. living hosts) that lie behind the dynamics approached at the population level by traditional population models. The model explores what behavior is adaptive (i.e., maximizing fitness, here measured as expected reproductive success) under what circumstances. It then uses the results to explore the population dynamics that emerge from the predicted behavior. Thus, we can approach the evolutionary mechanisms behind the conditional strategies that make some species and populations switch between endemic and epidemic states. (The term “epidemic” is here used to denote all populations that kill living trees. This does not imply that all populations entering the “epidemic” state will produce large-scale tree killing, only that large-scale forest mortality results when populations in a large number of patches become epidemic within a relatively short time due to a combination of dispersal and autocorrelated climate and landscape effects. Also, some obligatory saprophagous species are not necessarily described in this model.) To do this, we couple a set of monotonic functions representing the trade-offs facing the individual beetle: the probability that a living tree will become colonisable (Eq. 1), the risk suffered by the first beetles boring into a still living tree (Eq. 2), the decrease in reproductive success caused by high gallery densities (Eq. 3), and migration risk (Eq. 4). We scale the parameters (see section belo) so that model runs are biologically reasonable, investigate the effect of varying parameters representing biological differences, and calculate to what degree our inclusions of relative risk and sequential choice predict divergence from the Ideal Free Distribution (IFD). The IFD (conceptually similar to the game theory term Nash Equilibrium) is a central idea in evolution and behavioral ecology. It predicts that organisms should distribute themselves proportionately to the amount of resources available in each patch. The relationship with evolutionary game theory is that when an IFD is achieved, no individual can do better by changing patch. Imperfect information, unequal competitive abilities, time lags and costs of redistribution can inhibit the formation of an IFD, but it is a very useful null assumption, as systematic deviations from it alerts us to the existence of costs and/or constrains that need to be accounted for. Finally, the results are compared to empirical data and existing population models, and discussed in relation to evolutionary, ecological and management issues. # Analysis Scaled logistic functions are used as approximations of the actual risk and fitness trade-offs, as we can assume monotonic but not linear transitions between the biologically plausible extremes. For instance, a single beetle will never overwhelm a tree but an infinite number of beetles always will, the risks of attacking or migrating are between zero and certain death, and mean number of offspring per adult is between zero and the maximum for the species. Our results are general over a biologically plausible range of parameters (see below), and do not depend on the exact functions or parameter values as long as they are monotonic and scale relatively to each other. The following assumptions are made in the SRD model: 1. **The beetles show an adaptive behavioural reaction norm.** Thus, we assume that they have had time to evolve, and that there are no strong evolutionary trade-offs with processes invisible to this model, or manipulation from other organisms such as parasites. Kin-selected altruism is assumed to play no significant part, since at least some important species are outbreeding and widely dispersing, 2. **The beetles respond to the density of conspecifics, and to the defence level of their host trees.** While uncertain, the beetles' estimates are assumed to be unbiased. This is supported by the observation that beetles respond to host volatiles, conspecific pheromone concentrations and post-landing host inspection. 3. **The beetles act sequentially, within a limited time (flight period).** Individuals must at some point make a choice of one resource over the other. As they can only be aware of conspecifics that have already settled and started releasing volatiles and pheromones, they have no information about the presence of unsettled individuals or about individuals that will arrive later. The model proceeds through several steps. First, beetle no. 1 selects the strategy (settling in a dead tree, settling in a living tree, or migrating away) that gives it the highest expected number of offspring. Then beetle no. 2 does the same, but the outcome may be influenced by the choice already done by beetle no. 1. This is repeated until all N beetles that encounter the patch during a swarming period have settled or migrated. The result of these choices determines the adult (gallery) densities in dead and live trees respectively, and thus the number of realized offspring. The patch size is an abstraction of the maximum distance over which the beetles integrate information about their hosts and conspecifics using olfactory and, at close range, even visual/tactile clues. From pheromone trapping experiments, we consider that about a hundred meters or less in radius seems a realistic scale approximation, and it can reasonably be visualized as a stand of host trees. N is the number of beetles/patch, defined as the number of beetles that will respond to a patch during the flight period. Using the patch (stand) as the spatial unit, N is hence referred to as “population density”. The amount of breeding material in a patch exists either in the form of dead (i.e., undefended) and living (i.e., defended but susceptible to colonization) trees, which are denoted K_D and K_L respectively. This is conceptually related to the “carrying capacity”, and is scaled to the density at which mean reproductive success is halved. The subscripts D, L and M are used throughout, denoting dead trees, live trees and migration respectively. Assuming a probabilistic relationship regulated by c₀, the expected probability of beetle *i* reproducing successfully if it chooses a living host is a function of tree resistance (T), how many conspecifics have already attacked (N_L), and possibly also of how many conspecifics have settled in the dead trees (N_D) in the patch. Since some of these may have first sampled living trees, penetrating resin channels, their average contribution towards successful colonization is reduced to a proportion τ, where τ ∈ (0,1). Thus:This is closely related to the number of trees being killed, although some species, such as *D. ponderosae*, can succeed in so-called “strip- kills” where only a section of the tree dies. In addition, the first *i* beetles settling on a tree may suffer an increased risk C_i of not rearing offspring due to the still-vigorous tree defences. As more beetles settle, defences are exhausted and the risk faced by subsequent settlers decrease:We assume essentially no risk for initial settlers in dead trees (c_D\>3, T_D\<−5/c₀), but a substantial risk for the first settlers in live trees (c_L\<3, T_L\>5). The number of offspring F produced by the i^th beetle can reach its maximum value of R up until the point where competition starts reducing reproduction with increasing gallery density (eq.3). Here α and β regulate the onset and steepness of negative density dependence, and a₁ and a₂ regulate a reduced probability of reproduction due to mate finding failure or other Allee effects. The expected reproductive output from out-of patch migration F_M,i is the better of the expected (mean) values of (1−Ω_i)F_D,i and (1−Ω_i)F_D,i where Ω is the probability of dying before finding a new patch to settle in. This may increase with time as fat reserves are depleted and time runs out so thatF_M assume that K_D and K_L are mean values of a Poisson-distributed resource landscape, and that N, N_L,i and N_D,i are mean values with Gamma- distributed values in the receiving patches, thus including a shape parameter to regulate the standard deviance γ. If the population is perfectly synchronized (γ = 0), a migrant will always meet the same N_L and N_D (though K still varies). As migration range increases relative to the scale of spatial population synchrony (γ increases), the population state experienced by emigrants is increasingly independent of the state they left. Thus, the local model implicitly accounts for the larger-scale process of migration. As each beetle optimizes its expected reproductive success, the probability P that the i^th beetle will choose each strategy iswhere regulates the sensitivity of beetles to differences in fitness between substrates (i.e., imperfect information). If the reproductive outputs of all beetles in a tree react equally to the final density, the realized number of offspring after the settlement process is complete becomes The total number of offspring produced in the patch is thus But if each beetle monopolized the amount of bark it can utilize for breeding when it settled, thus suffering no interference from later arrivals, the total number of offspring becomesA summary of the SRD is given in. For numerical investigation of the model, we scale K_D, K_L, T_L and the other parameters so that when 0≤N≤100 most model runs will be biologically reasonable and non-trivial (i.e., populations non-zero and not increasing to infinity, positive “carrying capacities”, live trees not succumbing to one single beetle etc.). We here scale N to the range of 0–100 individuals to make numerical analysis computationally manageable, but running the model upscaled to more realistic population numbers gives the same biological predictions. For an overview of model predictions see, and for the population dynamics following from the model and. Grégoire et al. describe a site in Southern Belgium with 70 spruce that had been felled by wind in February 1990 and left on site. The number of windfalls and living trees in the same stand that were colonized by *I. typographus* were estimated on 13 occasions from April 10^th to November 12^th. From the asymptotic increase, we assume that the number of trees settled at the end of the period represent all that were available in the patch, and that beetles encountered the patch at an approximately constant rate. With dead and live trees of equal size, the data tells us that K_L/K_D = 44/70, and to run the model from 0 to 100 representing the (unknown) number of real beetles that arrived over the period, we scale this with a factor g₀ = 0.1. We then investigate whether the colonization pattern will be reproduced using plausible values of the within the investigated range for the other, unknown, variables. The proportion of trees observed to be colonized is expected to increase logistically in proportion to the number of individuals settling in them, so that the number of trees observed to be colonized at time t is. One of the few population models incorporating resources and beetle populations is the resource depletion model by Økland and Bjørnstad. Models of this type assume that when N≤T, K = K_D, and when N\>T, K = K_D+K_L, and combine this with a standard population growth model like the Gomperz function. A parameterisation of eq.3, with R = 50, K_D = 6, a₁ = 0,a₂ = 2, β = 0 and A = N = Attack density of *I.cembrae* is consistent with field data The SRD obviously involves a number of assumptions and trade-offs between process clarity and realism, being an exploratory model of general use. One key assumption is that beetles can assess their own population density and host availability on “patch” scales seems supported: highly developed olfactory systems allow them to sense different conspecific pheromones, host tree volatiles, and their relative concentrations. On smaller scales other senses may be used. Tree-killing scolytids continuously produce aggregation pheromones until the moment when the host resistance threshold is reached. Manipulating ratios of the predominant monoterpene compound in Norway spruce to *I. typographus* pheromones shows a strong, positive effect on attraction to increasing monoterpene∶pheromone ratios. This is in contrast to *I. pini*, which rarely attack healthy trees and shows a parabolic attraction effect of host monoterpene∶conspecific pheromone ratios, intuitively explained as a trade-off between avoiding already densely populated trees and too vigorously defending trees not likely to be successfully colonized. # Results The SRD predict that only dead trees will initially be colonized, but as the local population density increases, these grow increasingly crowded until density dependence outweighs the risks of either settling in a living tree or emigrating. The observed dynamics emerge from a single, flexible strategy shared by the whole population, but individuals maximizing their fitness does not necessarily imply maximized population growth: the distributions between dead and living trees mostly result in population growth rates far from maximum. At low and very high population densities, the SRD and IFD converge, but for a range of population densities when living trees are being colonized, they diverge considerably. ## Population dynamics There is a considerable volume of parameter space where aggregative attacks may occur ( summarizes model behaviour for such a case), and in a subset of these the population growth function gives three non-zero population equilibrium points. The population trajectories can thus be rather complex, and have two attractor basins; one lower (“endemic”) and one higher (“epidemic”). The position of the attractor basins are found to depend on the expected payoff from migration (ω,γ), the risks incurred by initiating an attack (c_L), the beetles having a substantial effect on the trees (c₀), the degree to which beetles settling in dead trees first sample random live trees (τ), and of course the abundance of dead (K_D) and living trees (K_L) that are not too vigorously defended (T_L). The population can shift between the endemic and epidemic attractor basins by several mechanisms. An epidemic state can be triggered by increased population density, such as following an increased abundance of dead hosts (K_D), immigration, or increased survival rates. It can also be triggered by drought stress, or increased aggressiveness (in the beetles or the composition of their host-pathogenic fungi), both in effect lowering the colonization threshold (T_L), increasingly explorative search patterns (increased τ), or even spontaneously when the dynamics are unstable. Even increased environmental variability alone can increase the odds that a population will enter the epidemic attractor basin within a given time period. Likewise, epidemics may cease from density-dependent offspring reduction, emigration, poor survival (low R), lack of resources (low K), abundant rainfall (increased T_L) or decreased aggressiveness or exploration (increased T or decreased τ). Another factor shaping dynamic structure is whether early- arriving individuals are able to monopolize resources, thus making them less affected by increasing density than latecomers and stabilizing the dynamics around an equilibrium point. Low-threshold systems where few beetles are needed to kill moderately healthy hosts can appear stable and endemic as long as beetle populations are low (for instance due to winter mortality, predation and low host abundance), but can easily switch to stable epidemic (i.e., potentially causing large outbreaks) when populations increase (for instance due to increased survival or decreased predation). However, when defence thresholds are very high, living trees are colonized only at densities where reproduction is severely depressed by density dependence and migration mortality, and the population tends to decline rapidly and return to the endemic cycle. ## Evolutionary dynamics The Ideal Free Distribution (IFD) is a central theoretical concept in ecology, behavioral ecology and evolutionary biology. The term was first coined by Fretwell and Lucas in 1970 to 1971, and has been of central importance in theory development and studies for a range of ecological and evolutionary systems. It describes the way in which animals distribute themselves among resource patches, stating that individual animals will aggregate proportionately to the amount of resources available in each patch. So for instance, if patch A contains twice as much food as patch B, there will be twice as many individuals foraging in patch A as in patch B. The relationship with evolutionary game theory is that when an IFD is achieved, no individual can do better by changing patch. Simple IFDs are rarely observed in nature, as imperfect information, unequal competitive abilities, time lags and costs of redistribution can inhibit the formation of an IFD. However, it is a very useful null assumption, because systematic deviations from it alerts us to the existence of costs and/or constrains that need to be accounted for, or to search for a cause of maladaptive behavior. The density at which living trees are settled is determined by the beetles' evolved “expectations” of density dependence, but also migration mortality and probability of successfully colonizing living trees. Thus, the patchy and unpredictable distribution of dead trees is a prerequisite for the risky colonization strategy to arise. As long as redistribution is penalized, there is a considerable population density interval over which individuals settling in living trees, despite the risks from tree defences, have higher expected reproductive output than those in dead trees, thus deviating from the IFD. However, selection for aggressiveness is decreased at very high densities, and reversed at very low densities, and as all populations exhausting their local host base quicker than it replenishes sooner or later are likely to return to low population densities, selection is highly unlikely to stay directional. As the selective pressure changes depending on the time scale of population fluctuations, this suggests a fitness premium on rapid trait selection such as maternal or epigenetic effects. Such rapid responses to selection would materialize as heterogeneities in aggressive behaviour between otherwise genetically homogenous populations. (It also suggests that if some species could redistribute so as to follow the IFD, this would be antagonistic to the evolution of aggressive strategies, as the best a beetle could hope for when initiating a risky attack on a healthy host would be to break even with its conspecifics in the same patch.) The “random sampling coefficient” τ represents a little explored effect of beetle behaviour. It denotes the proportion of beetles in a patch that will sample one or more trees by burrowing, and thus contribute to the depletion of tree defences, before they decide to settle in a dead tree. The density threshold at which colonization occurs increases markedly with low values of τ, suggesting that species with finely grained host localisation will have fewer outbreaks. Simulations of the model suggests that there is selection for different levels of “aggressiveness” depending on population density (N) relative to defence threshold (T) and landscape connectivity/patch abundance (Ω). ## Comparing with observations As predicted by the SRD, *Ips typographus'* reproductive success has been found to be consistently higher in trees colonized while still living than in dead trees at the same sites, despite a clear general preference (i.e., when sufficient amounts are available for both substrates) for dead trees. Experiments suggest that host tree volatiles attract beetles on a “patch” scale, and that most trees are visited and assessed by beetles landing randomly on a finer spatial scale. Using close range chemical signals and/or penetrating into the phloem to assess host suitability, they simply abort the attempt if the tree is either too strongly defended or crowded, implying that τ\>0 for some aggressive species. Fat content is negatively affected both by activity (migration) and density, and less fat means less chance of surviving to find another host. The SRD model implies that an adaptive reaction norm to signs of increasing migration risk include a higher propensity for attacking trees. Thus, beetles emerging from crowded, brood trees should be more likely to start an infestation close to their parental tree instead of migrating, both due to their density and to their lower energy reserves. This seems to be supported by multiple observations. The colonization sequence observed by by Grégoire et al. were easily reproduced and predicted by the SRD model. Best fit was found when assuming increasing migration risk as the season progresses (ω₁ = −2, ω₂ = 0.05), density dependence is moderately steep (c₀ = 0.2, α = 2), some random sampling from early settlers occur (τ = 0.5), and a medium-range tree defence (T_L = 15). Field data on the number of offspring per adult *I. cembrae* as a function of gallery density suggest a steep density-dependence for this species. If about 20% of the offspring survive to breed, populations will decrease for all but the lowest recorded attack densities. This indicates that density dependence can be a local demographic force, without which the SRD predicts that no beetle would take the risk of initiating an attack on a living tree, and the shape is consistent with the SRD general function choice. The density per m² of newly emerging *I. typographus* in 68 trees was recorded at six sites over two years in the Vosges mountains of France ( Grégoire unpublished data). The living trees were colonized only when the dead trees nearby were colonized at high density, as measured by emergence. Living trees with decreased defences (low T) (i.e., trees with partial root contact) follow the expected pattern, being colonized at lower levels of crowding in nearby dead trees. The assumption implicit in some population models that beetles colonize trees to produce new resources when possible (i.e., whenever N\>T) creates considerable discrepancies with the SRD, both with respect to densities at which a population can enter an epidemic phase, and to how stable the epidemic can be. The beetle density at which tree mortality will occur responds in a strongly non-linear fashion to the interaction between the colonization threshold and the amount of dead trees present. This may be one of the reasons why classical risk and population models have shown relative rather than absolute predictive capability. # Discussion The SRD model supports and provides mechanistic underpinnings for the proposed bimodal population growth curve of aggressive bark beetles,. The bimodal density dependence suggests that zones of moderate beetle density may attract individuals from surrounding areas, facilitating waves of attack emanating from outbreak patches. A good example may be the *D. ponderosae* on the Chilcotin Plateau, where strong spatial and temporal dependencies at small scales (\<18 km), indicate feedback from local epidemic processes. The beetles may position galleries so as to minimize interference from later arrivals, and the larvae develop quickly. But larvae also have a limited ability to cross consumed phloem, and the net outcome seems to be a strong decrease in reproductive success with increasing gallery densities. Whatever the consequences for the residents, late-arriving beetles will be selected for entering a crowded bole. The exception would be under strong kin selection, which is unlikely in widely dispersing species. We see that simply dividing the number of beetles on the available resources may give wrong estimates of population growth and stability properties, as the population distributions may not follow an IFD or maximize tree colonization. The exact host defence threshold (T) only matters when it is higher than the beetle density at which crowding occur in dead trees, meaning that there is no fixed defence threshold below which trees become at risk (complicating predictive risk models). Thus, the gallery densities found in colonized trees are not just a reflection of how many beetles it takes to kill a tree, but the risk of initiating a colonization attempt in a living tree. This is in accordance with the observation that even healthy hosts can be successfully colonized if pheromone-baited. When the SRD and IFD deviate, the “random landing” on a stand scale observed for several species may be adaptive: beetles sampling living trees may increase their fitness by being more likely to become part of an early-stage attack on a living tree than beetles that steer directly to the available dead trees. The exploratory landing pattern is therefore in a feedback relationship with aggressiveness, as less aggressive species will have less to gain by exploratory sampling of living hosts, which again makes colonization less likely and selects for more precise localisation of dead hosts etc. Landing and search patterns are thus predicted to also shape larger-scale spatial dynamics: if the beetles do not sample living hosts to assess defence levels and quality, and are “invisible” to each other while flying (i.e., not emitting pheromones), no living trees will be colonized without sufficient dead hosts nearby to inform beetles about high conspecific densities. Thus landing patterns determine to what degree epidemics can start at random locations and spread efficiently through healthy forests. The deviation from the IDF may point to a feedback process facilitating the evolution of aggressiveness: as sex pheromones easily lead to beetles aggregating in considerable numbers there is a fitness gain from settling in a living but possibly succumbing tree at the right moment. However, the expected fitness differences converge when dead wood is scarce and beetle densities are high, and the selection pressure is reversed when resource depletion, density dependence or reduced survival (cold winters, predation, etc.) depresses the population below the threshold where successful colonization is likely. Thus, “aggressiveness” in bark beetles and their associated fungi seem likely to exist in a state of evolutionary flux or cycles. As heritability on the traits shaping aggressiveness might be mediated through maternal and epigenetic effects as well as allele differences, this may explain evidence for multiple strategies and strongly heritable differences between individuals and populations of the same species when it comes to aggressiveness. Also, the similarities in selective forces and starting points suggest that aggressive life strategies may easily have evolved independently from non- aggressive ancestors in different genera such as *Ips* and *Dendroctonus*. If beetles continue to be aware of the rest of the patch after starting to burrow into a host, and can relocate without fitness costs, they will distribute proportionally between dead and live trees, and an IFD state will be reached. However, once gallery construction has begun, this is not likely to be the case. Beetles may re-emerge to start a second brood after some time, but for the purposes of this model, any re-emerged individuals are treated as new arrivals and fitness is calculated per brood. The possibility of re-emergence is an incentive to breed as soon as possible. For some individuals, it may still be adaptive to “wait and see” instead of being the first to attack, and the complex pheromone systems of many bark beetles show signs of individual variation. But time is limited, and the beetles with the least fat reserves must commit first, enforcing the sequential individual decisions driving the SRD. While increasing migration range generally may increase the range of spatial synchrony, this may not be the case under chaotic or strongly resource-driven population dynamics, with consequences for adaptive dispersal patterns. Everything else being equal, species with a lower cost of migration are less prone to attack living trees. But when they do, the epidemics may be more prolonged, as individuals can migrate away rather than cause population crashes. Similarly, if early settlers can monopolize gallery space, density dependence will rarely cause populations to crash. Integrating data and processes on different scales is difficult, but these results show the need to examine population models for implicit evolutionary assumptions, and doing so generates hypotheses that may explain otherwise puzzling aspects of the species' ecology. In general, the inherent bistability and strong endogenous feedbacks of bark beetle systems makes them particularly sensitive to perturbations such as climate change, and adaptations to rapidly fluctuating selective pressures may allow species to respond quickly to ecological changes such as decreased host defences following changes in climate or forestry. We thank Derek Johnson, Sandy Liebhold, James Powell, Mark Lewis and two anonymous referees for valuable discussions. [^1]: Designed and implemented the model: KLK. Contributed data: J-CG. Designed research: J-CG OS NE MG NCS BØ. Wrote the manuscript: KLK J-CG OS NE MG NCS BØ. [^2]: The authors have declared that no competing interests exist.

# Introduction Macroautophagy (hereafter referred to as autophagy), a degradation process conserved from plants and yeast to higher eukaryotes, has been described to play an essential role in cellular mechanisms such as lifespan extension, cellular development and differentiation as well as in the development of several diseases including cancer, Huntington's, Alzheimer's and Parkinson's diseases. The yeast autophagy-related protein Atg8 is an ubiquitin-like modifier that localizes to and aids in the formation of autophagosomes. In mammals, the *atg8* family is divided into two subfamilies based on when they intervene in autophagosomal formation, the *MAP-lc3* subfamily (*Microtubule-Associated Protein light chain 3 A*, *B* and *C*) and the *gabarap* subfamily (*gabarap* for *GABA_A receptor-associated protein*; *gabarapl1/gec1* for *GABA_A receptor-associated protein like 1/glandular epithelial cells 1* and *gabarapl2/gate16* for *GABA_A receptor-associated protein like 2/GTPase activating protein of 16 kDa*). The LC3 subfamily is involved in earlier events in the process of autophagosomal formation and is necessary for elongation of the double membrane vesicle, whereas the GABARAP subfamily is implicated in latter stages of autophagosomal formation and is necessary for the closure of the vesicle. The members of these two subfamilies, although similar, differ in their expression patterns, transcriptional regulation, and affinity for different protein partners, suggesting that their contribution to the autophagic pathway and even their particular role may vary depending on the tissue type and condition. To date, two *atg8* knockout mice have been created: for *lc3ß* and *gabarap*. Both knockout mice produce viable offspring and the *lc3ß* knockout mouse displays no difference in its autophagic phenotype, suggesting that in the absence of one family member, there is compensation by another. It is not known if the same is true for the members of the GABARAP subfamily since autophagy levels were not assessed in the *gabarap* knockout. *gabarapl1/gec1*, one of the *gabarap* subfamily members, was originally discovered as an estrogen-regulated gene in guinea-pig glandular epithelial cells and has since been identified as a member of the *gabarap* subfamily due to sequence similarity. It is the most highly expressed transcript of the *atg8* mammalian homologues in the central nervous system, and its mRNA expression has been mapped out in the rat central nervous system in 2008. *gabarapl1* mRNA expression varies throughout the rat brain, with a more abundant staining visible in neurons such as the pyramidal cells of the cerebral cortex and hippocampus, magnocellular neurons in the hypothalamus, purkinje cells of the cerebellum, and motoneurons in the brainstem and spinal cord, and a less expression visible in structures such as the striatum and the various reticular nuclei. In the same article, mRNA expression appeared to be limited to neurons, based on the lack of labelling in the fiber tracts and the morphology of the cells labelled in the gray matter. At the protein level, GABARAPL1 expression has been identified by western blot analysis in different rat brain regions and confirmed in a few regions in which GABARAPL1 immunohistochemical labelling was visible such as the cortex, medial septal nucleus, diagonal band, hippocampus and motoneurons of the ventral horn. In the latter study, authors used an antibody that appeared specific to GABARAPL1 in immunohistochemistry but that was not in immunoblotting techniques. *gabarapl1* mRNA expression varies in different pathologies and following multiple stimuli. In the brain, a decrease in *gabarapl1* transcript levels have been observed in both the cortex of macaques treated with MPTP (1-methyl-4-phenyl-1,2,3,6- tetrahydropyridine), a neurotoxin which mimics the effect of the development of Parkinson's disease in animal models, and in the *substantia nigra* neurons of human Parkinson's patients. We recently set out to study the specificity of several antibodies directed against GABARAPL1 and identified an antibody that is specific to the latter protein in both western blot and immunohistochemical experiments. With this tool in hand, we have mapped out the protein expression of GABARAPL1 in the adult mouse brain and during the development. Our results demonstrate a protein expression throughout the adult mouse brain similar to the *gabarapl1* mRNA distribution previously described in rat. During development and in the adult mouse, GABARAPL1 is distributed throughout the brain, from the olfactory bulb to brainstem, with varying intensity. Although GABARAPL1 is found only in neurons in the mature brain, it is also observed in fiber tracts during neuronal development. At a cellular level, GABARAPL1 displays a weak cytoplasmic and sometimes dendritic staining with stronger intensity punctate structures clustered within the cytoplasm. These structures are increased in the presence of a drug that inhibits the autophagy flux and the majority co-localizes with the specific autophagic substrate p62, suggesting that they are indeed autophagosomes. GABARAPL1 staining was observed in long projection neurons and interneurons, and co-localized with antibodies against choline acetyl transferase (ChAT), the dopaminergic transporter (DAT), calbindin and parvalbumin. This study describes for the first time the expression of an autophagic protein, GABARAPL1, in the mouse brain and during its development. Our data demonstrate that this expression is not ubiquitous and shows high differences throughout the mouse brain. These data suggest that the basal autophagy levels might vary in the different types of neurons at different stages of their life and that this process might be highly regulated. # Materials and Methods ## Animals Cerebral tissues from wild type (WT) embryo, WT adult male mice and GABARAP−/− adult male mice were used for GABARAPL1 protein expression mapping. WT Swiss mice were obtained from Janvier, Le Genest-Saint-Isle, France. GABARAP−/− 129/SvEv-C57BL/6 mice were kindly provided by Pr. Heinrich Betz, Department of Neurochemistry, Max-Planck-Institute for Brain Research, Frankfurt, Germany. The time of conception for embryos was documented by vaginal plug observation (embryonic day 0, E0). Timed pregnant mice and adult male mice were euthanized by cervical dislocation. Mice pups at E11, E12, E13, E14 and E17 were sacrificed by decapitation. ## Tissue fixation and cryoprotection Adult male mice were perfused as previously described with 0.9% NaCl followed by ice- cold 1% paraformaldehyde (PFA, ROTH) fixative in PBS (0.137 M NaCl, 3.3 mM KCl, 10 mM Na₂HPO4, 1.8 mM KH₂PO4). For immunohistochemistry experiments, extracted adult and embryonic brains were post-fixed in 1% PFA for several hours at 4°C and cryoprotected by saturation in a 15% sucrose solution (Sigma, 17 994-9) at 4°C overnight. Fresh tissue contained within an Eppendorf tube (for western-blotting experiments) or cryoprotected tissue (for immunohistochemistry experiments) was frozen by immersion in isopentane using the Snap-Frost**™** system (Excilone, France), sectioned at −20°C into 10 µm sections using a cryostat-microtome (Microm), placed on gelatin-coated slides (ROTH) and stored at −40°C. ## Immunohistochemistry For immunohistochemical staining, slides were prepared by rehydration in PBS-T (PBS with 0.3% Triton X100) for three 5 min washes at room temperature. Tissue was subsequently incubated with the appropriate primary antibody dissolved in milk diluant (PBS-T containing 1% bovine serum albumine, 10% lactoproteins and 0.01% sodium azide) overnight at room temperature. The primary anti-GABARAPL1 specific antibody used was anti-ATG8L (rabbit polyclonal, Proteintech, 11010-1-AP, 1∶500). Slides were then washed three times with PBS-T for 5 min each at room temperature before being incubated with the appropriate secondary antibody diluted in milk solution for 1 h at room temperature. The secondary antibody used was the Alexa Fluor 488 goat anti-rabbit IgG (Invitrogen, 1∶800). Finally, the slides were washed with PBS-T three times 5 min and cover slips were mounted with 70∶30 glycerol∶PBS-T. ## Immunostaining of mouse embryonic cortical primary neurons Dissected cortices from 8 embryos were resuspended in 5 ml of HBSS (Invitrogen) and then incubated with 26 U/ml of Papain (Worthington) for 20 min at 37°C. After addition of 2 ml of NBM+ medium (Neurobasal medium, X B27, 100 µg/ml Penicillin-Streptomycin, 0.5 mM L-Glutamine, Invitrogen), the cortices were grounded and centrifuged for 5 minutes at 1000 g. The pellet was then resuspended in 2 ml of complete NBM+ medium and 500 µl were loaded on 2 ml of a 4% BSA solution before a centrifugation step at 1000 g for 5 min to remove debris. The pellet was then resuspended in 2 ml NBM+ medium and the cell concentration determined before plating on coverslips in 24 well plates at a concentration of 2.10⁵ cells per well. The primary neurons were then cultured in an incubator at 37°C in a 5% CO₂ atmosphere for 7 days with half the medium changed on day 2 and 5. On day 7, half of the medium was replaced with fresh medium with or without Baf A1 (Sigma) at a final concentration of 100 nM and incubated for a further 16 h. The cells were then washed once in PBS and fixed for 45 min in a 4% PFA solution in PBS. Cells were then washed twice in PBS and subsequently permeabilized for 3 min in a 0.5% Triton-X-100 solution in PBS, washed four times with PBS, blocked for 1 h in 3% BSA in PBS and then incubated overnight at 4°C with the anti- ATG8L (rabbit polyclonal, Proteintech, 11010-1-AP, 1∶500), anti-p62 (mouse monoclonal, Abnova, 1∶500), anti-GFAP (rabbit polyclonal, Dako, 1∶500) and the anti-NeuN antibody (mouse monoclonal, Millipore, 1∶500) dissolved in blocking buffer. Coverslips were then washed three times with PBS for 5 min each at room temperature before being incubated with an Alexa Fluor 488 goat anti-rabbit IgG (Invitrogen, 1∶500) and an Alexa Fluor 568 goat anti-mouse IgG (Invitrogen, 1∶500) diluted in PBS for 1 h at room temperature in the dark. Finally, the cells were washed four times 5 min with PBS, incubated with a 2 µg/ml Hoescht solution in PBS for 10 min at room temperature, washed once in PBS and coverslips were mounted on slides in Fluoromount-G (Southern Biotechnology Associates, Inc). ## Microscopy Fluorescent tissue sections were examined with either a fluorescent microscope (BX51, Olympus, France), a fluorescence laser scanning confocal microscope (IX81, Olympus, France, SFR IBCT FED 4234 microscopy core Besançon, France) or a Zeiss LSM 710 confocal microscope (UAB microscopy core, Birmingham, AL, USA). Images were captured with either a DP50 or DP75 numeric cameras (Olympus, France) using the Fluoview FV1000 software, AnalySIS3.1 software (Soft Imaging System) or Zen 2008 software, respectively. All figures were assembled using Adobe Photoshop CS2 and Adobe Illustrator CS software. ## Western blot analysis Brain tissue was ground and lysed using the following lysis buffer: \[10 mM Tris-HCl pH 7.4, 100 mM NaCl, 5 mM EDTA, 10 mM MgCl₂, 0.5% NP-40, 1% Triton X-100, protease inhibitor cocktail (Sigma, P8340)\] then 40 µg of tissue lysates were resolved on a 15% polyacrylamide gel in running buffer (25 mM Tris base, 200 mM glycine and 0.1% SDS) at 20 mA using a Biorad Power Pack 1000. Proteins were transferred onto Immun-Blot PVDF 0.2 µm membranes (Biorad) for 1 h at 4°C at 200 mA in Western Blot transfer buffer (25 mM Tris base, 200 mM glycine and 10% methanol). Membranes were subsequently blocked in TBS-T (199 mM Tris-HCl, pH 7.4, 1.36 mM NaCl, 0.1% Tween 20) with 5% skim milk powder. Membranes were blotted in TBS-T supplemented with 0.5% skim milk and anti-ATG8L (rabbit polyclonal, Proteintech, 1∶1000) and anti-Actin (rabbit polyclonal, Sigma, 1∶3000). The secondary antibody horseradish peroxydase-coupled anti- rabbit IgG was prepared in TBS-T containing 0.5% skim milk powder. Binding of antibodies to membranes was detected by Enhanced Chemiluminescence Plus Reagent (ECL Plus, GE Healthcare Life Sciences), according to the manufacturer's protocol.. Image Lab software (Bio-Rad Laboratories) was used to quantify protein band density. ## Statistical analysis Statistical analyses were carried out using a Student's t test on GraphPad PRISM® software. Data are expressed as the mean ± SEM, \*\*p\<0.05 versus E11, \*p\<0.1 versus E11 (n = 3). ## Ethics Statement The appropriate institutional ethics committee approved all experiments in this study. All animal care and experimental protocols adhere to the institutional guidelines and are in compliance with the University of Alabama at Birmingham and the University of Franche-Comte Institutional Animal Care and Use Committee guidelines. # Results ## Anti-GABARAPL1 antibody specificity and staining Since GABARAPL1 shares a very high sequence identity (86% in the case of the human proteins) with its closest homologue GABARAP, we used the only specific anti-GABARAPL1 antibody, to our knowledge, available on the market for our study. This antibody from Proteintech is specific of GABARAPL1, not only in western blot but also in immunohistochemistry experiments. At a cellular level, GABARAPL1 localization is displayed as weak labelling throughout the cytoplasm and some neuronal extensions with brighter punctate labelling within the cell body ( in adult and 12d in embryo). This pattern is observed in all the labelled neurons throughout the adult brain, but is more prominent at E17 than E11 in the embryonic brain. This localization is similar to the one observed previously in cellular overexpression models, and is consistent with its previously described roles in receptor transport and the formation of autophagosomes. Although our study was performed using the same antibody lot previously demonstrated to be specific to GABARAPL1, we nevertheless took supplementary precautions to confirm the specificity of the staining observed. Firstly, we performed *in-situ* hybridization on the preceding slide in the wild type adult or embryonic mouse series with a probe recognizing a 3′-UTR region of *gabarapl1* mRNA whose specificity has been previously confirmed. These experiments allowed us to correlate the mRNA *gabarapl1* expression to its protein expression in the same region. Secondly, we studied the expression of GABARAPL1 in the GABARAP knockout mouse to confirm its distribution pattern. To do so, we performed immunohistochemical experiments, using the same GABARAPL1 antibody, on a series of coronal sections produced from a GABARAP knockout mouse or GABARAP knockout embryos and verified the features of the staining produced in the different regions examined in the wild type mouse brain. In all the regions examined, the GABARAPL1 staining pattern was found to be the same as the one observed in the wild type animals. ## GABARAPL1 expression in embryonic cortical primary cultures The GABARAPL1 labeling pattern in the whole adult brain revealed that this protein, like its mRNA, is found predominantly in the grey matter and thus likely only within neurons. In order to confirm the neuronal expression of GABARAPL1, we utilized mouse embryonic cortical primary cell cultures to study GABARAPL1 staining in an *ex vivo* model. We performed immunocytofluorescence experiments with our specific anti-GABARAPL1 antibody and a marker of neuronal cell types, anti-NeuN (neuronal nuclei), or a marker of glial cells, anti-GFAP (glial fibrillary acidic protein), on primary cell cultures grown in either neurobasal media (NBM) or Dulbecco's Modified Eagle Medium supplemented with 10% FBS (DMEM), producing primarily neuronal or glial cell cultures, respectively. In both *ex vivo* models, all GABARAPL1 positive cells were also positive for NeuN. Consistent with the small percentage of neuronal cells in these cultures, very little GABARAPL1 staining was visible in the DMEM grown cultures (predominantly glial cells). Together, these results demonstrate that GABARAPL1 protein expression is found primarily in neurons. Since GABARAPL1 has been previously described to be involved in receptor transport in neurons, and to be associated with autophagic vesicles, the punctate staining observed in the neurons of the mouse brain is most likely associated with secretion or autophagic vesicles. Previously published studies have demonstrated that the visualisation of autophagic proteins such as p62 or LC3 in the normal brain is difficult. Indeed, these studies describe a lack of p62 or punctate LC3 staining in normal brain tissue. Like these other studies, we were unable to identify p62 labeling or LC3 punctate labeling in the mouse brain tissue. As such, we performed co-labeling experiments in primary neuronal cultures. To determine whether GABARAPL1 is linked to autophagic vesicles in neurons, we incubated the primary cultures in the presence of Baf A1, a described inhibitor of the autophagic flux, allowing the accumulation of autophagic vesicles in the cells. Like the *in vivo* staining described earlier, GABARAPL1 displayed a weak cytoplasmic and a stronger punctiform staining in primary neuron cultures. When the autophagic flux is blocked with Baf A1, we observed an increase in the number of GABARAPL1-positive vesicles in the cells, suggesting that the punctiform staining observed in these neurons, and therefore in the mouse brain, is most likely linked to autophagic vesicles. We then stained these primary neuron cultures with an anti-p62 antibody. p62 has been described as a protein specifically degraded through the autophagosome/lysosome pathway and shown to accumulate in autophagosomes in presence of BafA1. As seen in, in control cells, GABARAPL1 and p62 staining are diffuse throughout the cytoplasm with some more intense vesicle staining and we observe little co- localization. Conversely, when the cells are incubated with BafA1, we observe an increase in the number of these vesicles and an accumulation of the two proteins in most, but not all, of these structures. We can therefore conclude that the GABARAPL1 staining observed in mouse neurons is primarily linked to autophagosomal structures. ## Distribution of GABARAPL1 protein in the adult brain The protein expression of GABARAPL1 in the brain mapped out in this study describes an expression pattern congruent with that of its mRNA. We confirmed that the GABARAPL1 protein is abundantly expressed in the brain from the olfactory bulb to the brainstem. GABARAPL1 distribution, however, was not uniform throughout the brain and the intensity of labelling varied in labelled cells. A detailed list of the specific brain regions labelled with the anti- GABARAPL1 antibody is displayed in. Cells found to be positive for GABARAPL1 were designated by a relative level of staining from 0 to +++ based on their intensity. Examples of intense, moderate and weakly labelled cells can be seen in. ### Telencephalon GABARAPL1 expression was observed throughout the telencephalon, like its mRNA, with a higher expression found in the cerebral cortex and moderate to weak expression observed in most of the basal nuclei with the exception of the magnocellular preoptic and medial septal nuclei and the nucleus of the diagonal band, in which very intensely labelled cells were observed. GABARAPL1 protein distribution within the telencephalon is depicted in with representative images of staining in the cerebellar cortex, the striatum, and nucleus of the diagonal band. Within the telencephalon, GABARAPL1 expression was identified in cholinergic neurons visible in the cerebral cortex (ChAT-positive cells,) the striatum and the pallidum as well as calbindin and parvalbumin-positive cells in the striatum. ### Diencephalon GABARAPL1 protein distribution within the diencephalon is depicted in. GABARAPL1 expression could be seen throughout the diencephalon with varying intensities. Except for the anterodorsal nucleus of the thalamus (AD), the thalamus was only weakly to moderately stained with anti-GABARAPL1. GABARAPL1 expression was overall higher in the hypothalamus, particularly in the paraventricular nucleus (PVH), the medial preoptic nucleus, the medial mammillary nucleus and the lateral hypothalamus area. Double staining with anti-ChAT and anti-GABARAPL1 revealed that all cholinergic neurons within the diencephalon were GABARAPL1 positive. ### Mesencephalon Weak to moderate staining was observed throughout most regions of the mesencephalon with the exception of some stronger labelling. In particular, the ventral tegmental area (VTA) and *substantia nigra* (SN) expressed a higher level of GABARAPL1. Interestingly, the GABARAPL1 expression observed in the *substantia nigra* was particularly intense in the *pars compacta* region compared to the *pars reticula* region. Double staining in these regions revealed that GABARAPL1 is found in neurons that express the dopamine transporter (DAT). Moderate labelling in the pyramidal layer of the hippocampal formation is depicted in. ### Rhombencephalon GABARAPL1 labelling in the hindbrain was the strongest in the motor nuclei, in particular in the purkinje cells within the cerebellum, in the dorsal motor nucleus of the vagus nerve (DMX) and in the hypoglossal nucleus (XII). ## GABARAPL1 expression throughout development GABARAPL1 protein expression was detectable in whole head embryonic tissue by western blot analysis as early as embryonic day 11 (E11), after which it increased progressively throughout development, reaching a value six fold higher at E17 than at E11. However, expression levels of GABARAPL1 at E17, shortly before birth, were significantly inferior to GABARAPL1 expression in the adult mouse brain. In accordance with western blot results, GABARAPL1 staining was present at E11 to E17 throughout the developing brain from the telencephalon to the brainstem. GABARAPL1 labelling became ubiquitous throughout the brain as the mantle layer differentiated. This labelling was visible throughout the mantle layer in the pallium but never in the neuroepithelial layer of the developing brain. This result is in accordance with the presence of GABARAPL1 in immature neurons. During earlier developmental stages (E11/E12) cells in the mantle layer were uniformly labelled for GABARAPL1. Differential patterns of GABARAPL1 labelling only started to become visible at intermediate stages (E14) and were more obvious at later stages (E17), at which point the adult expression pattern appeared to develop. This differential pattern similar to the adult was particularly evident in the motoneurons, such as surrounding the 4^th ventricle in the brainstem. In the pallium, organized layers of GABARAPL1 labelling were visible, in which the intermediate zone and cortical plate strongly expressed the GABARAPL1 protein ( or). Contrary to the adult mouse, we also observed a labelling of fibers in the embryo as is visible in the pyramidal tract at E11, the posterior commissure at E14 and the optic tract at E17. Fiber labelling with the anti-GABARAPL1 was also demonstrated in the intermediate zone of the cortical plate, which gives rise to the corpus callosum. At a cellular level, GABARAPL1 labelling appeared diffused throughout the cell at earlier stages of development and, as mentioned before, only showed signs of the stronger punctate-like structures visible in the cytoplasm at later stages, around E17. # Discussion Our study is the first to map out the differential expression of the GABARAPL1 protein in the adult and developing embryonic murine brain. In this study, we confirmed the strong presence of this protein throughout the brain, visible in immature neurons and fibers during differentiation and in mature neurons in the adult. Immunohistochemical labelling with a specific anti-GABARAPL1 antibody revealed an overall similar distribution in the adult of the mouse GABARAPL1 protein compared to that of the rat *gabarapl1* mRNA, suggesting that in this case, mRNA levels do depict protein levels. Examples of these coherences include the cerebral cortex, which was strongly labelled in both studies, and the pallidum, which was weakly labelled in both studies. The similar distribution of *gabarapl1* mRNA and protein is particularly evident through the labelling pattern of the *substantia nigra*, in which the *pars compacta* was particularly strongly labelled with both the *gabarapl1* mRNA probe and the GABARAPL1 protein specific antibody, while the *pars reticula* remained less strongly labelled in both cases. Despite the overall similar distribution between the GABARAPL1 protein and its mRNA, some variations were observed. These differences are due either to a difference between mRNA and protein expression levels or between species. For example, in the telencephalon, the relative protein expression of GABARAPL1 was found to be lower than its relative mRNA levels in the main olfactory bulb, the anterior olfactory nucleus, the fundus of the striatum and the bed nucleus of the accessory olfactory tract. A few dissimilarities were visible between the two species: rat and mouse. *gabarapl1* mRNA expression in the induseum griseum of the cortex and certain regions of the diencephalon varied. In the induseum griseum, intense mRNA staining was observed in the rat and weak mRNA staining, associated with weak to moderate protein expression, were observed in the mouse. The mRNA expression and corresponding protein expression of GABARAPL1 in the mouse was found to be weak in several sensory nuclei of the thalamus as well as in the *zona incerta*, and several hypothalamic nuclei compared to a higher expression observed in the rat. On the other hand, *gabarapl1* expression appeared to be higher in the mouse anteroventral periventricular hypothalamic nucleus compared its expression in the rat. Lastly, we also observed a lower expression of *gabarapl1* in the mouse interstitial nucleus of Cajal and the nucleus of Darkschewitsch and a higher expression in the external cuneate nucleus compared to the mRNA expression in the rat. What was particularly interesting in the adult was the strong expression of GABARAPL1 restricted to neuronal cells. Since GABARAPL1 is an important effector of the autophagic process, and autophagy is a globally important cellular process, one would expect to still see some labelling in the surrounding cells. This difference could be explained by the higher importance of housekeeping autophagy in post-mitotic cells, such as neurons, compared to their surrounding supporting cells, such as glia. Not only did GABARAPL1 display a differential expression between different cells types, but its expression also varied in different brain regions. By use of double labelling with different antibodies, we demonstrated that GABARAPL1 is expressed in different types of neurons, including both GABAergic (calbindin and parvalbumin positive neurons in the pallidum) and glutamatergic (pyramidal cells of the cortex) neurons. Double labelling experiments with an anti-DAT also revealed that GABARAPL1 punctate staining was strong in dopaminergic neurons throughout the *substantia nigra* as well as the ventral tegmental area. While GABARAPL1 was present in different neuronal types, the level of its expression varied from one region to another. For example, GABARAPL1 labeling was visible in the GABAergic Purkinje cells but absent from the glutamatergic granule cells of the cerebellum. GABARAPL1 labeling was also visible in the pyramidal and granular layers (containing excitatory neurons) of the hippocampus and dentate gyrus but not in the molecular layers (composed of inhibitory interneurons). More detailed studies of GABARAPL1 expression with respect to the organization of specific brain regions combined with functional approaches will therefore be interesting for future studies. To date, no other mammalian Atg8 protein homologue has been mapped out in the adult brain, although their mRNA expression levels have been observed. As such, protein mapping of the other members of the GABARAP and LC3 families in the brain would be very interesting in order to determine if there is a compensation of the other members in the cells in which GABARAPL1 is less strongly expressed. This map of basal levels of GABARAPL1 in the brain is also a valuable tool that can be used to compare GABARAPL1 protein levels under different conditions such as neurodegenerative models using different chemical molecules, experimentation methods or knockout models, such as the neuronal specific *atg5* and *atg7* mouse knockout models. In the embryo, similar GABARAPL1 distribution patterns compared to the adult became visible at E14 and even more pronounced at E17, at which point neuroblast expression of GABARAPL1 corresponded to the adult expression of this protein. Unlike in the adult, however, GABARAPL1 staining was also observed in both fibers. In particular, a strong signal was detected in fiber tracts at E11, suggesting a role for GABARAPL1 in axonal growth and elongation during development. Indeed, GABARAPL1, like GABARAP and LC3, is a microtubule- associated protein known to bind to tubulin and promote tubulin assembly and microtubule bundling and, therefore, most likely plays an important role in axon and dendritic projection during development. Contrary to the mantle layer, no labelling in the neuroepithellium was visible, suggesting that GABARAPL1 does not play a role in stem cells surrounding the ventricles. *lc3α* and *lc3ß* mRNA expression have also been examined in the developing murine central nervous system. In these studies, the authors demonstrate identification of *lc3 in-situ* staining in the forebrain, midbrain and commissural neurons of the developing dorsal neural tube, similar to the expression we observed for the GABARAPL1 protein. In the adult and later stages of development, GABARAPL1 labelling displays a weak signal in the cytoplasm and some neuronal extensions with brighter punctate structures throughout the cytoplasm and a complete absence of labelling in the nucleus. This staining is consistent with the GABARAPL1 distribution we have previously observed in our *in vitro* models when studying the role of GABARAPL1 in autophagy. The importance of a basal amount of autophagy in neurons has been demonstrated by the two conditional knockouts: *atg5*−/− and *atg7*−/−. In both these mouse models, a deficiency in autophagy in the central nervous system resulted in severe neurodegeneration. As such, one would expect to see a basal level of autophagy visible in the murine brain, suggesting that the punctate structures visible with our GABARAPL1 staining likely, at least partially, represent autophagosomal membranes. Indeed, primary cortical cultures (E17) incubated with or without Baf A1, an inhibitor of the autophagic flux described to inhibit the fusion of autophagosomes with lysosomes, revealed a significant increase in the number of intracellular vesicles positive for GABARAPL1 and/or p62, a described substrate of autophagy. These *ex vivo* data demonstrate that GABARAPL1 is, at least partially, associated with autophagosomes in primary neurons and consequently in the mouse brain. Nonetheless, some vesicles are positive for both proteins while others are only positive for GABARAPL1 or p62. The p62 only positive vesicles are likely to be autophagosomes containing none or little GABARAPL1. Indeed, we are aware of the existence of 6 members of the Atg8 family, most of which are expressed in neurons to different extents, and it is still unknown if they all participate in the formation of each autophagosome or, more likely, if they are specific to different structures. Interestingly, while LC3 expression has been observed in the murine brain, the GFP-LC3 mouse does not display any kind of punctate structure in the brain. This differs from GABARAPL1, which does display a strong punctate labelling, and may suggest that GABARAPL1 plays a more prominent role in the autophagic process under normal conditions in the brain. This question will only be answered with the creation of a knockout mouse model for GABARAPL1. Indeed, the study of two previous knockout models against LC3B and GABARAP did not portray autophagic deregulation or pathologies linked to the disruption of the autophagosome/lysosome pathway, suggesting that a compensatory mechanism by the other members of the family is in place. Since GABARAPL1 is predominant in the mouse brain and it interacts with several adaptor proteins for selective autophagy, which degrades unwanted protein aggregates and damaged organelles, such as p62, NBR1 (neighbor of BRCA1) and NIX (NIP3-like protein X), its extinction might present a more severe phenotype, especially in this organ. GABARAPL1 has also been shown to bind to and is potentially implicated in trafficking of the GABA_AR (gamma-aminobutyric acid type A receptor). Regulation of GABA_AR numbers and clustering is important for modulation of synaptic strength at inhibitory synapses, which is mediated by an accumulation of the GABA_AR at the postsynaptic dendrite to maintain neural network homeostasis. Given that GABARAPL1 interacts with this receptor and is expressed in GABAergic neurons, it may also play a role for in neural network homeostasis. GABARAPL1 expression in GABAergic neurons becomes particularly interesting for studies of depression. Indeed, *gabarapl1* and *GABA_AR* were found to be upregulated in the hippocampus and certain regions of the cortex of human brains affected by major depression regulated. kappa opioid receptor (KOR)-induced signalling has also been implicated in depression-like behaviour. Since *gabarapl1* is upregulated in major depression and GABARAPL1 interacts with and mediates KOR expression, then another of its functions might be to increase the transport of KOR to the plasma membrane to enhance depressive-like behaviour. Conversely, a suppression of GABAergic inhibition has been linked to the etiology of some neurodevelopmental disorders, such as autism and schizophrenia. In fact, *gabarapl1* was found to be deleted in a genome wide experiment using in an European ancestry case-control data set for autism. *gabarapl1* expression is also shown to be deregulated in Parkinson's disease. *gabarapl1* is downregulated in a MPTP model of Parkinson's in macaque monkeys as well as in the dopaminergic neurons of the *SNpc* of Parkinson's patients, a region of the brain in which we have found a high expression of the GABARAPL1 protein. Moreover, GABARAPL1 can interact with α-synuclein, a protein shown to form protein aggregates and deleterious lewy bodies in Parkinson's disease, and displays a higher binding affinity for α-synuclein oligomers compared to monomers. GABARAPL1 may, therefore, play a protective role in this disease by aiding in the degradation of unwanted α-synuclein aggregates until such time as it is downregulated with the progression of the disease. Indeed, *gabarapl1* transcription is upregulated in the presence of estrogens and estrogen has been show to have a protective effect on an MPTP murine model of Parkinson's disease. Overall, our study describes, for the first time, the protein expression map of an autophagy factor in the adult and embryonic murine brain. Its expression and cellular localization is consistent with previous data obtained relating to its mRNA expression in the rat, suggesting a similar expression and role in different mammals. Its expression is also consistent with its involvement in receptor transport and autophagy. Moreover, its high expression in the brain and in particular in disease-linked areas such as the *substantia nigra pars compacta* might set up GABARAPL1 as a potential therapeutic target for diverse neurological disorders. The GABARAP−/− 129/SvEv-C57BL/6 mouse line was kindly provided by Pr. Heinrich Betz, Department of Neurochemistry, Max-Planck-Institute for Brain Research, Frankfurt, Germany. Authors are grateful to Sophie Launay and Shawn Williams for their technical help (microscopy cores, SFR IBCT FED 4234, Besançon and UAB, Birmingham, AL) and to Gabrielle Bernard for her technical help in *in situ* hybridization and animal expertise. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: AF JZ MJ P-YR MB-G RD-M. Performed the experiments: JNLG KB MB-G. Analyzed the data: JNLG KB. Wrote the paper: JNLG KB MB-G.

# Introduction Dynamin is a large multidomain GTPase known for its role in catalyzing membrane fission in clathrin-mediated endocytosis (CME). It consists of five functional domains: an N-terminal GTPase domain (G domain); the middle domain and GTPase effector domains (GEDs), which together form the stalk of dynamin; a pleckstrin homology (PH) domain; and a C-terminal proline- and arginine-rich domain (PRD), which interacts with many SH3 domain-containing proteins. Dynamin assembles at the necks of invaginated clathrin-coated pits and catalyzes scission and release of clathrin-coated vesicles from the plasma membrane. Dynamin is recruited to nascent coated pits in its unassembled state and also plays a regulatory role during the early stages of CME. Most GTPase family members that function as regulatory proteins do so by switching between GTP-bound ‘active’ conformations and GDP-bound ‘inactive’ states. Their intrinsic GTP hydrolysis rates are slow, and rate-limited by the exchange of tightly-bound GDP for GTP. These two steps in the GTP hydrolytic cycle are regulated by GTPase activating proteins (GAPs) and GTP exchange factors (GEFs), respectively. In this regard, dynamin is an atypical GTPase as it has a low affinity for GTP (2–5 μM), a high rate of GDP dissociation (\~ 60–90 s^-1), and a comparatively robust and measurable basal rate of GTP hydrolysis (\~ 1 min^-1 at 37°C). However, upon self-assembly, interactions between G domains can stimulate GTPase activity *in trans*. *In vivo*, dynamin self-assembles into short helical structures that surround the necks of deeply invaginated coated pits. *In vitro*, dynamin assembles into long helical arrays around lipid nanotubes whereby its GTPase activity is stimulated \> 100-fold. Dynamin’s GTPase activity can also be stimulated, albeit to a lesser extent, through interactions with divalent SH3 domain containing partners such as Grb2 or under low salt conditions that favor dynamin self-assembly. Given its importance for clathrin-mediated endocytosis, coupled to the fact that it is one of the few enzymes known to be required for CME, small molecule inhibitors of dynamin’s GTPase activity have been sought as potentially powerful tools for studying CME. Indeed, several chemical inhibitors of dynamin have been reported and are commercially available, including Dynasore and its structural derivative, Dyngo-4a. However, recent findings have brought into question the specificity of these compounds. For example, Dynasore and Dyngo-4a continue to inhibit endocytosis in triple dynamin-1, 2, and 3 knockout cells, thus revealing potential off-target effects. These off-target effects on endocytosis may reflect their reported ability to perturb plasma membrane cholesterol levels and destabilize actin filaments. Recently, Dynasore was shown to impair VEGFR2 signaling in an endocytosis-independent manner. Based on the clear evidence for dynamin-independent, off-target effects of these compounds, there remains a need to develop more specific and robust dynamin inhibitors. Previous screens for small molecule inhibitors of dynamin’s GTPase activity were based on the detection of released phosphate using a malachite green colorimetric assay. However, this assay lacks sufficient sensitivity to detect dynamin’s basal GTPase activity, especially when measured at room temperature and at the low concentrations of dynamin and GTP practically needed for the design of a high-throughput assay. To circumvent this, previous high throughput screens measured dynamin’s stimulated GTPase activity either in the presence of GST-Grb2 or with sonicated phosphatidylserine (PS) liposomes at low salt. Dynasore, and by extension Dyngo-4a, was shown to be a noncompetitive inhibitor of dynamin’s GTPase activity. Hence its mechanism of action, which remains unknown, may reflect indirect effects on dynamin assembly or aggregation. Here we report the optimization of a new, highly sensitive, and robust HTS- compatible assay to detect the basal GTPase activity of dynamin and its validation in a preliminary screen of 8000 compounds. # Materials and methods ## Dynamin expression, purification, and preparation Dynamin-1 (Dyn1) was expressed in Sf9 (*Spodoptera frugiperda*, GIBCO-BRL, Gaithersburg, MD) insect cells and purified by affinity chromatography, as previously described. Nucleotide-free protein aliquots were stored in assay buffer containing 20 mM HEPES, 150 mM KCl, 1 mM EDTA, 1 mM EGTA and 1 mM DTT at pH 7.4. The aliquots were flash frozen in liquid nitrogen and stored in -80°C in 5% v/v glycerol. Prior to running assays, frozen aliquots of Dyn1 were thawed and centrifuged at 100,000 g for 15 minutes to remove any aggregates. Dyn1 concentration was determined by measuring its absorbance at 280 nm with a UV/Vis spectrophotometer (Beckman Coulter Inc.) using a molar absorptivity coefficient of 73,800 M^-1 cm^-1. ## Transcreener GDP fluorescence polarization assay The Transcreener^® GDP FP (BellBrook Labs) assay is an immune- competition assay based on a mouse monoclonal antibody that selectively binds Alexa633-conjugated GDP over GTP. The preformed Alexa633-GDP antibody complex is added at the end of the reaction, and the GDP generated displaces bound Alexa633 fluorescent tracer, resulting in a decrease in its far-red fluorescence polarization (FP). The assay detects GDP production with high sensitivity at low substrate concentrations and has been used to develop an HTS compatible assay for the enzymatic activity of ARFGAP. Stocks of 5 mM GTP, 10x Stop & Detect Buffer B, 400 nM GDP Alexa633 Tracer, and 3100 μg/mL GDP antibody were purchased from BellBrook Labs (Madison, WI). Assays were performed in Corning 384-well plates at room temperature. All assays were conducted in the endpoint format, which measures total GDP production after 60 minutes of incubation at room temperature. The optimized reaction (total volume 15 μL) was initiated by adding 5 μL of GTP (3x stock of 30 μM, final concentration 10 μM) to wells containing 10 μL Dyn1 (1.5x stock, final concentration 50 or 100 nM). Reactions were terminated after 60 minutes with the addition of 5 μL of GDP detection mixture (4x stock of 8 nM GDP Alexa633 Tracer, 40 mM HEPES, 80 mM EDTA, 0.04% Brij and 34.4 μg/mL GDP antibody). The plates were incubated for another 60 minutes, allowing the GDP antibody-GDP binding to reach equilibrium. The plate was then read for fluorescence polarization in millipolarization units (mP) using an EnVision multi-modal microplate reader (PerkinElmer, Inc.). The mP values of the reaction-containing wells were subtracted from those containing no enzyme to obtain ΔmP values. To convert the polarization data to GDP released, a standard curve representing 0 to 100% GDP conversion from 10 μM GTP was generated. Using GraphPad Prism, the ΔmP values were fitted to the standard curve to obtain the total GDP production. ## Mock screen 15 μL reactions with 50 nM Dyn1 were incubated for 60 minutes with 0.3 μL of 100% DMSO in the presence of 10 μM GTP. Reactions that lacked Dyn1 were included as positive controls for complete inhibition. The mock screen was run in accordance with the Transcreener GDP FP assay protocol. ## 8000-compound library screen A screen was performed using the optimized Transcreener GDP FP assay on a diverse library subset of 8000 small molecule compounds, provided by the UT Southwestern HTS Core. For the HTS screen, the final concentration of Dyn1 in the reaction mixture was 50 nM. 0.3 μL of compound in 100% DMSO (final concentration 10 μM compound, 2% DMSO) was added to the Dyn1 and pre-incubated for 30 minutes. Controls for the screen included reactions that lacked enzyme (positive control for inhibition) and uninhibited reactions containing only DMSO (negative control for inhibition), which were dispensed in single columns in each plate. Solutions were dispensed using automated liquid handling devices. ## Data analysis The primary screen data were analyzed using Genedata Screener^® software. The Z’ factors for the mock screen and the 8000-compound pilot screen were calculated using the equation below: $$Z^{\prime}\ factor = 1 - \frac{3\left( {\sigma_{positive\ control} + \sigma_{sample\ or\ negative\ control}} \right)}{\left| {\mu_{positive\ control} - \mu_{sample\ or\ negative\ control}} \right|}$$ where σ_{positive control} is the standard deviation of the positive controls for inhibition, and σ_{sample or negative control} is the standard deviation of the samples or negative controls for inhibition, respectively. μ_{positive control} is the mean of the positive control for inhibition, and μ_{sample or negative control} is the mean of the samples or neutral DMSO controls, respectively. The samples were normalized by a two-point correction method using the equation below: $$Two\ point\ normalized\ values = \frac{{raw\ value}_{sample} - {median}_{total\ samples}}{{median}_{positive\ controls} - {median}_{total\ samples}}\ \times \ 100$$ where median_{total samples} is defined as the median of all library compound-containing reaction wells within the plate. The two-point normalized activity values were adjusted using a correction factor to account for systematic errors within and across assay plates. The correction factor of a well in a given plate is calculated using pattern detection algorithms that are proprietary to the Screener^® software (Genedata, Inc.). The corrected activity values were then used to determine the robust Z (RZ) score with the following equation: $$Robust\ Z\ score = \frac{normalized\ value\ of\ sample - median\ of\ DMSO\ controls}{robust\ STD\ of\ DMSO\ controls}$$ where robust STD is the standard deviation calculated using the median of the DMSO controls (negative control for inhibition). For the confirmation screen and dose response curves, the data were analyzed by normalizing the sample GDP released to the control GDP released using the following equation: $$normalized\ GTPase\ activity = \frac{{sample}_{GDP\ released}}{{control}_{GDP\ released}}\ \times \ 100$$ ## Malachite green assay The lipid nanotube (NT)-stimulated malachite green assays were performed in 96-well plates at 37°C. The final reaction consisted of 100 nM Dyn1, 25 μM GTP, and 300 μM lipid nanotubes. The assay and reagent preparations were performed according to our published protocol. All general chemicals were purchased from Sigma-Aldrich (St. Louis, MO). Compounds tested in the malachite green assay were purchased from Chembridge and ChemDiv (both located in San Diego, CA). # Results ## Optimization of a fluorescence polarization assay to detect basal GTPase activity of dynamin Fluorescence polarization (FP) is a method that allows for rapid and quantitative analysis of diverse molecular interactions and enzyme activities. Polarization measures the change in the molecular movement of the labeled species. It is the ratio of the difference between the vertical and horizontal components of the emitted light over their sum. In recent years, FP has been successfully used in HTS of compound libraries to identify small molecule inhibitors of protein-protein interactions. Bellbrook labs has developed an assay that detects GDP using a competitive FP immunoassay. The GDP antibody binds both GDP and the GDP-Alexa633 tracer with similar affinities. The GDP released upon hydrolysis of GTP by GTPases displaces the fluorescent tracer from the antibody, resulting in a decrease in polarization due to increased rotational mobility. The antibody has 140-fold specificity for GDP versus GTP, which allows sensitive measurement of GDP in the presence of excess GTP. Nonetheless, given that the antibody used has a finite selectivity for GDP over GTP, it was necessary to determine the optimal concentration of the Alexa-GDP antibody conjugate needed for maximum mP measured in the presence of 10, 100 and 500 μM GTP, our initial substrate concentrations. For this purpose, the Alexa-GDP antibody conjugate was titrated into the reaction mixture containing GTP (10, 100 or 500 μM) and assay buffer. The data were fitted to a variable slope sigmoidal dose-response curve using GraphPad Prism. From the titration curves, we determined the optimal GDP antibody concentrations to be 8.6, 81.5 and 405.5 μg/mL for 10, 100 and 500 μM GTP, respectively (highlighted data points). These concentrations were chosen near signal saturation and represent a good compromise between sensitivity and maximal polarization value. To convert ΔmP to μM GDP released, we generated a standard curve by titrating increasing concentrations of GDP in the presence of GTP to mimic reaction conditions. The assay accurately measures GTP hydrolysis in the range of 0.05% to 10% of the substrate converted. To determine the optimal conditions for high throughput screening, we measured ΔmP for increasing concentrations of Dyn1 (0.3 nM to 5000 nM) at three different concentrations of GTP (10, 100, and 500 μM). These titrations established that 50 nM Dyn1, assayed in the presence of 10 μM GTP for 60 minutes, resulted in excellent signal-to-noise with high reproducibility. We further confirmed that, under these conditions, the basal rate of GTP hydrolysis by Dyn1 (\~ 0.04 min^-1 at 10 μM GTP) was linear for 60 min. These results are consistent with assays performed at room temperature and under low substrate concentrations. We chose 60 minutes to ensure that substrate consumption remained below 10%. Importantly, no signal was detected at any concentration of GTP when S45N mutant Dyn1, which is unable to bind GTP, was used as a negative control. We further confirmed that under these conditions, GTPase activity is directly proportional to the concentration of Dyn1. Thus, there is no evidence of cooperativity and the assay measures the basal rate of GTP hydrolysis of unassembled Dyn1. ## Mock screen We measured the robustness of the assay under our optimized HTS conditions to determine whether ‘hits’ could be identified with high confidence. mP values obtained from the DMSO containing reaction wells were compared to the control wells.. The average Z’ factor for the mock screen, which was calculated to be 0.56, indicated that the assay was sufficiently robust for screening purposes. Importantly, the assay was insensitive to DMSO concentrations up to 3%. ## Pilot screen, hit selection and validation A pilot screen using an 8000-compound diversity subset of the chemical library at UT Southwestern was conducted using the optimized Transcreener GDP FP assay. The compounds were tested for their inhibitory effects on the Dyn1 GTPase activity at a concentration of 10 μM. Intrinsically fluorescent compounds interfere with fluorescence polarization measurement; therefore, compound-containing wells with total fluorescence intensities greater than 3 times those of control wells were eliminated. After careful analysis of the data, we identified 42 compounds with a robust Z score greater than 3 as primary hits. These compounds were re-tested in a confirmation screen at three different concentrations (1, 3, and 10 μM) to validate their inhibitory effects. The confirmation screen yielded 4 compounds chosen based on concentration-dependent inhibition and their commercial availability. To confirm the inhibitory effects of the 4 compounds, we conducted 11-point dose response curves, with concentrations ranging from 1 nM to 100 μM. The IC₅₀ values of these compounds ranged from \< 1 μM to \> 50 μM. We focused on compound 24 which had an IC₅₀ of 0.58 μM. ## Secondary assay and comparison with Dynasore and Dyngo-4a Compound 24 was compared with the two commercially available dynamin inhibitors, Dynasore and Dyngo-4a, in a dose-response assay measuring inhibition of basal GTPase activity under high salt (150 mM KCl) conditions. The concentration of inhibitors ranged from 1 nM to 100 μM. As seen in, Dynasore and Dyngo-4a do not appear to inhibit basal GTPase activity even at high concentrations, in contrast to previous findings in which assays were performed under conditions that measure dynamin’s stimulated, assembly-dependent GTPase activity. Therefore, to more closely parallel previous studies we tested both commercial inhibitors in comparison to compound 24 for their effects on dynamin’s stimulated GTPase activity. Assays were performed in the presence of PI(4,5)P₂-containing lipid nanotubes (NT), whose diameter (\~ 20 nm) resembles the neck of an invaginated coated pit, using the malachite green assay. Under these conditions (100 nM Dyn1, 300 μM lipid nanotubes, 25 μM GTP), both Dynasore and Dyngo-4a inhibited the NT-stimulated GTPase activity of Dyn1 with IC₅₀ values of 83.5 and 45.4 μM, respectively, as compared to compound 24, which exhibited an IC₅₀ of 6.4 μM in this assay. # Discussion We have optimized a robust, high-throughput assay to measure the basal GTPase activity of unassembled Dyn1. This highly sensitive assay detects the release of low, nanomolar amounts of GDP and hence, accurately measures the intrinsic, basal rate of GTP hydrolysis, even at the low concentrations of dynamin and GTP necessary for HTS design and implementation. Previous high throughput screens using a less sensitive colorimetric malachite green assay to detect phosphate release were necessarily performed under conditions that stimulate dynamin’s GTPase activity, i.e. in the presence of dimeric GST-Grb2, which presumably aggregates dynamin, or with sonicated PS liposomes in low salt. Utilizing the Transcreener GDP FP assay, we conducted an 8000-compound pilot screen and identified several compounds that inhibit the basal GTPase activity of Dyn1. The most potent of these, compound 24, inhibited dynamin’s basal and lipid nanotubule assembly-stimulated GTPase activities with IC₅₀ values of \~0.6 μM and \~6 μM, respectively. However, this scaffold, identified as a potential “PAINS’ (Pan Assay Interference Compound) is not ideally suited for further refinement. Hence, we await results from a larger-scale screen to identify a lead scaffold appropriate for further development and optimization. Importantly, current commercially available dynamin inhibitors, Dynasore and Dyngo-4a, were also tested for their ability to inhibit dynamin’s basal GTPase activity in the Transcreener assay format. Although both Dynasore and Dyngo-4a could inhibit the NT-stimulated GTPase activity of Dyn1, neither was able to inhibit basal GTPase activity in our hands. Moreover, the reported IC₅₀ values we measured for Dynasore and Dyngo-4a NT-stimulated GTPase activity performed at physiological salt concentrations using 100 nM dynamin were much higher than those reported for assays performed in the presence of sonicated PS liposomes, under low salt conditions with 20 nM dynamin (0.4 μM and 12 μM, respectively). Given the reported off-target effects of Dynasore and Dyngo-4a and their uncertain mechanism of dynamin inhibition, a more robust and specific inhibitor of dynamin would be of immense value. As with any assay, fluorescence polarization has its limitations. Compounds that are either auto-fluorescent, or affect the affinity of the anti-GDP antibody for the tracer may be misinterpreted as potential hits. The hits must therefore be validated in secondary assays such as the malachite green assay and eventually for their ability to inhibit dynamin-dependent, clathrin-mediated endocytosis in intact cells. Having validated our assay using an 8000-compound pilot screen, we are currently expanding our search for more desirable chemical scaffolds and lead compounds to be used to develop more robust, specific, and cell-permeable dynamin inhibitors. We thank Dana K. Reed for helping with protein expression and purification. We also thank the members of the Schmid lab for thoughtful discussions. [^1]: MK, an employee of Bellbrook labs, assisted in early establishment of the Transcreener GDP FP assay conditions for dynamin. Otherwise, we received no financial or materials support from Bellbrook labs. This does not alter our adherence to PLOS ONE policies on sharing data and materials and the authors declare no competing interests.

# Introduction Hepatocellular cancer (HCC, hepatoma) is one of the most common solid tumors, fifth in incidence worldwide. Since HCC is refractory to currently available chemotherapeutics, surgical resection and liver transplantation remain the only potential curative treatment options for a limited number of suitable patients. Thus, new and effective therapeutic strategies for hepatoma are needed. The net balance of positive and negative intercellular signaling dictates tumor cell proliferation and survival, and proteins of the tumor necrosis factor (TNF) superfamily loom large in this array of regulatory signaling inputs. TNF superfamily proteins contribute to tissue homeostasis via effects on cell survival, death and differentiation. In the context of HCC and various other tumor types, two members of the TNF superfamily may have special significance, namely, the cell surface ligands TWEAK (TNF-related weak inducer of apoptosis) and TRAIL (TNF-related apoptosis ligand). TWEAK is expressed either as a type 2 membrane protein (with its carboxy-terminus oriented extracellularly) or as a soluble ligand by a range of cell types, including macrophage, monocytes, NK cells, and endothelial cells. The counter receptor for TWEAK, Fn14 (fibroblast growth –factor-inducible 14kD protein), is a membrane bound type 1 protein that is also expressed on a large variety of cells. Whereas hepatocytes do not normally express Fn14 mRNA or protein, HCC cells and hepatocytes of regenerating liver express significant amounts of them. Intriguingly, there are reports that HCC cells that have high levels of surface Fn14, also express TWEAK, in both membrane-anchored and soluble forms. In turn, this has led to evidence that the TWEAK:Fn14 signaling axis promotes HCC cell proliferation through autocrine and paracrine signaling. Studies with glioblastoma have indicated that TWEAK induces pro-survival signaling by activating the NF-kB transcription factor and upregulating the expression of the anti-apoptotic proteins BCL-X_L and BCL-W. TWEAK contributes to cancer by other mechanisms as well, for example, via its pro-angiogenic activity. TRAIL binds to several different cognate TNF superfamily receptors, some activating and others decoy. The activating receptors in humans are TRAIL-R1 (DR4), TRAIL-R2 (DR5), and osteoprotegrin. There is significant cross-reactivity between the human and mice TRAIL ligands. Importantly, TRAIL selectively induces apoptosis in a range of transformed cell lines, but not in normal tissues. In the case of HCC, there is constitutive expression of both TRAIL and its receptors. However, the data in the literature bearing on the functionality of this TRAIL receptor are contradictory, with most data indicating that HCC cells are resistant to TRAIL-induced apoptosis. This refractoriness has suggested there may be disturbances in the apoptotic pathway and/or over-activation of pro-proliferative and survival signals in HCC cells. In viewing TWEAK’s pro-proliferative and apoptosis-blockade activities in HCC, alongside the relative refractoriness of this tumor type to TRAIL-induced apoptosis, we posited that there may be functional cross-talk between these TNF pathways wherein TWEAK signaling contributes to TRAIL-resistance. In turn, this prompted the notion that blocking TWEAK:Fn14 signaling might restore death signaling through the TRAIL:TRAIL receptor axis in HCC cells. To test this hypothesis, we have here turned to Fn14•TRAIL, the recently-described fusion protein which bridges the TWEAK-Fn14 and TRAIL-TRAIL-R signaling axes. In particular, the Fn14 component of Fn14•TRAIL serves to block the binding of endogenous TWEAK to Fn14 on HCC cells, while the TRAIL component, once anchored to TWEAK-bearing cells or to soluble TWEAK via the Fn14 bridge, can direct intercellular and intra-cellular inhibitory signals to its cognate receptors on TRAIL-receptor bearing cells. Thus, Fn14•TRAIL is in essence converting a TWEAK pro-proliferative signal into a death-inducing one. In the present study, we assess Fn14•TRAIL’s ability to induce apoptosis of HCC cell lines *in vitro* and inhibit their growth as xenograft tumors *in vivo*. Our data establish a potent anti-tumor effect of this fusion protein, and point to its considerable therapeutic potential in bridging the TWEAK-Fn14 and TRAIL- TRAIL-R signaling axes. # Materials and Methods ## Materials Unless otherwise stated, all chemicals were obtained from SIGMA (Israel). DMEM medium, FBS, PBS, Trypsin-EDTA, penicillin, streptomycin and L-Glutamine were obtained from Biological Industries (Beit Haemek, Israel). ## Cell lines SK-HEP-1 (HTB-52; liver adenocarcinoma cell line) was purchased from ATCC (USA). HepG2, Huh7 and Hep3B HCC cell lines, originally from the ATCC, were kindly provided by the Hepatology Unit, Hadassah Hebrew University Medical Center in Jerusalem, Israel. Immortalized human hepatocyte cell lines FHB and NKNT3 were kindly provided by Prof. Eithan Galun at the Hadassah Gene Therapy Institute (Jerusalem, Israel). Cell lines were grown in 10% FBS DMEM supplemented with penicillin, streptomycin and L-Glutamine. The medium for FHB cells contained two additional components: hydrocortisone (5μM) and insulin (0.85 μM). NKNT3 cells were grown in Complete Serum-Free medium (CS-C; Cell Systems, Kyrkland, WA, USA). All cultures were tested periodically for mycoplasma contamination using EZ-PCR mycoplasma test kit (Biological Industries). ## Protein Production Fn14•TRAIL was produced and purified for us by Cobra Bio-manufacturing (Keele, UK). Chinese Hamster Ovary (CHO) cells were stably transfected with a DNA construct encoding a chimeric human Fn14•TRAIL sequence downstream of the CMV promoter. A stable CHO cell transfectant clone was isolated and expanded in serum-free medium. Fn14•TRAIL was purified from the medium using chromatographic methods to over 95% purity and stored at -80°C. The amino acid sequence of the expressed Fn14•TRAIL protein is shown in. The underlined sequence represents the signal-peptide of the human Urokinase protein, utilized here to promote secretion of Fn14•TRAIL, it is absent from the mature protein. The amino acid sequence of the extracellular domain of human Fn14 (amino acids 1-52 of the mature protein, designated in bold letters) are directly linked to the extracellular domain of human TRAIL (amino acids 53-217 of the mature protein). ## Western Blot and coomassie blue staining profiles of Fn14•TRAIL Fn14•TRAIL was separated on 12% SDS–PAGE, and Coomassie blue staining was performed with GelCode blue stain reagent (PIERCE, Rockford, IL, USA), according to manufacturer instructions. For western blot analysis, Fn14•TRAIL or whole cell lysate were separated on 10% SDS-PAGE and blotted according to standard procedures. Membranes were incubated with primary antibodies: anti-TRAIL (ALEXIS biochemicals, San Diego, CA, USA), anti-Fn14 (ALEXIS biochemicals), anti Caspase 3, 8 and 9, pNFkB, Bcl-2, pJNK, cIAP1, 2(Cell Signaling), anti- IkB (R&D), and anti cFLIP (ENZO, CA, USA). Secondary immunostaining was performed with HRP-conjugated Ab (Biorad, Hercules, CA, USA). ## Activity assay To asses Fn14•TRAIL's activity, 0.2 x 10⁶ cells/ml (for Huh7, 0.4 x 10⁶ cells/ml) were seeded as duplicates in 24-well plates (NUNC, Roskilde, Denmark) and cultured with or without varying concentrations of Fn14•TRAIL, soluble TRAIL (hTRAIL/Apo2L; PeproTech, Rocky Hill, NJ, USA, or BioVision, Mountain View, CA USA), Fn14-Fc (recombinant mouse; ALEXIS Biochemicals, San Diego, CA, USA), or a combination of the latter two. Soluble Fn14 (recombinant human Tweak receptor, PeproTech) was used instead of Fn14-Fc in some experiments. Cells were incubated with the indicated proteins for 24h, 48h or 72h for harvested and stained with trypan-blue. Live, unstained cells were counted in a grid chamber. ## Quantitative Real-Time PCR Analysis Total RNA was extracted from cell lines using TRIzol reagent (Invitrogen, Carlsbad, CA, USA). For cDNA synthesis, 2 μg of total RNA were reverse- transcribed using a High Capacity cDNA Reverse Transcriptase kit (Applied Biosystems, Foster City, CA, USA), according to manufacturer's protocol. Quantitive real-time PCR was performed using TaqMan Assay-on-Demand^TM (TWEAK - Hs00356411_m1, Fn14 - Hs00171993_m1, TRAIL - Hs00234355_m1, DR4 - Hs00269492_m1, DR5 - Hs01043171_m1, DcR1 - Hs00182570_m1, DcR2 - Hs01043162_g1, OPG - Hs00171068_m1) and TaqMan Gene Expression Master Mix (Applied Biosystems). All reactions were performed in triplicates using the 7900HT Fast Real Time PCR system (Applied Biosystems), and normalized against two endogenous control human genes, TBP (Hs99999910_m1) and Actin-B (Hs99999903_m1). Analysis of results was based on the formula 2^-ΔΔCt, using SDS v2.3 and data-assist v2.0 software (Applied Biosystems). ## Flow Cytometry To assess apoptosis, 0.2 x 10⁶ cells/ml (for Huh7, 0.4 x 10⁶ cells/ml) were seeded as duplicates in 24-well plates and incubated with or without varying concentrations of Fn14•TRAIL, soluble TRAIL, Fn14-Fc or the latter two in combination for 24 h. Apoptotic cells were detected by flow cytometric analysis, using the AnnexinV/PI MEBCYTO Apoptosis Kit (MBL, Nagoya Japan), according to the manufacturer’s protocol. 20,000 events per sample were counted using a FACSCalibur flow cytometer (Becton Dickinson, San Jose, CA, USA), and data were analyzed using CellQuest software (Becton Dickinson). In some experiments, Fn14•TRAIL was pre-incubated with one of the following proteins, prior to addition to the cultures: anti-human/mouse Fn14 (an IgA isotype Ab); anti-hTRAIL (an IgG-1 Isotype Ab); mouse IgA and mouse IgG, with all the latter reagents obtained from eBioscience, San Diego, CA, USA. To examine expression of TRAIL receptors, TRAIL, Fn14 and TWEAK on cell lines, adherent cells were retrieved using Accutase (Innovative Cell Technologies, San Diego, CA, USA), washed, and stained with phycoerythrin-conjugated mAb against TWEAK (CARL-1), Fn14 (ITEM-4), TRAIL (RIK-2), the TRAIL receptors DR4 (DJR1), DR5 (DJR2-4) and DcR1 (TRAIL-R3) and DcR2 (TRAIL-R4) (R&D, Minneapolis, Minnesota, USA) or their isotype control Abs (eBioscience). Presence of intracellular TWEAK was determined using a Fixation and Permeabilization Kit (eBioscience), according to the manufacturer's protocol. Flow cytometry was performed and analyzed as above. ## Tweak ELISA To determine TWEAK secretion to cell culture medium, cells (10⁵) were seeded in 24 well plates for 72h and conditioned medium was collected and centrifuged. Clear medium was concentrated x3 using a speed-vac centrifuge, and analyzed by Human TWEAK ELISA development Kit (PeproTech), according to manufacturer instructions. ## Establishment of Tumor Xenografts All experiments were approved by the Animal Care Committee of the Hebrew University. Athymic BALB/c nu/nu nude male mice (Harlan, Israel), 4-6 weeks of age, were maintained under defined flora conditions at the Hebrew University Pathogen-Free Animal Facility. HepG2 cells were grown to 80% confluence, harvested, washed with PBS, and injected subcutaneously (4 x 10⁶/mouse) into the right flanks of mice. Once palpable, tumors were measured for their widths and lengths using a micro-caliper, and tumor volumes were calculated (w² x length /2). Mice were treated daily with subcutaneous injections of Fn14•TRAIL (200 μg/mouse) for 8 days. Control groups were injected with Fn14•TRAIL dilution buffer. At the end of experiments, mice were sacrificed, and tumors were harvested, measured and weighed. For liver pathology assessment, Nude mice were treated for 8 days with Fn14•TRAIL and sacrificed one hour post last injection. Livers were removed, fixated in 4% formaldehyde, and embedded in paraffin. Paraffin sections (5μm) were stained with Hematoxylene and Eosin (H&E). Blood samples were taken at day 8, one hour post Fn14•TRAIL injection. Serum was kept at 4c. The next day, serum samples were diluted 1:5 or more with DDW. Aspartate aminotransferase (AST) and Alkaline phosphatase (ALK-P) serum levels were measured by the Clinical BioChemistry Analyzer Reflotron Plus Instrument and Starter Kit (Micglobal, Roche, London, UK), according to the manufacturer instructions. Serum Urea levels were determined by QuantiChromTM Urea Assay Kit (DIUR-500), (BioAssay Systems CA, USA), according to the manufacturer instructions. ## Immunohistochemistry To assess Fn14-TRAIL apoptosis-inducing effect *in-vivo*, HepG2 tumors bearing Athymic-Nude mice were sacrificed one hour post injection, at the 2^nd injection day (200ug Fn14-TRAIL per day). Subcutaneous tumors were removed, fixated in 4% formaldehyde, and embedded in paraffin. Sections (5μm) were deparaffinized in xylene (3x3') and rehydrated in graded alcohol (3x1' 100% ethanol; 3x1' 96% ethanol). Following 5' incubation in 3% H₂O₂ for endogenous peroxidase inactivation, slides were incubated in Citrate buffer (pH6; Invitrogen) and boiled in electric pressure cooker (BioCare Medical, CA, USA) for antigen retrieval. Samples were blocked for 30' in CAS-BLOCK (Invitrogen) prior to overnight incubation with the anti- cleaved caspase 3 primary antibody (Cell Signaling \#9661; 1:100 diluted in CAS- BLOCK) at 4°c in humidified box. Following washing (3x2' in Super Sensitive wash buffer, BioGenex), samples were incubated for 30' in RT with the Simple stain MAX PO (MULTI) anti-rabbit immune-peroxidase polymer (NICHIREI BIOSCIENCES INC.). Diaminobenzidine (DAB; UltraVision Detection System, Thermo scientific, MA, USA) was used as the chromogen according to manufacturer instructions, and 20'' incubation in hematoxylene (SIGMA-Aldrich) was used as the nuclear counter- stain. Following dehydration steps (2' 80% ethanol, 2' 96% ethanol, 2' 100% ethanol, 2' xylene) and mounting (Histomount mounting solution, Invitrogen), x20 pictures were taken by the Nikon ECLIPSE Ti light microscope and captured by the DSFI-1 camera (Nikon). ## Statistical analysis Data are presented as means ± SE. Statistical comparison of means was performed by a two-tailed unpaired Student's t test. Differences with a p<u>\<</u>0.05 were considered statistically significant. # Results ## Fn14•TRAIL protein characterization Fn14•TRAIL was produced as described in the materials and methods section. Protein molecular weight, purity and identity were verified by coommassie SDS- PAGE staining and western blot analysis with anti Fn14 and anti TRAIL antibodies. ## Fn14•TRAIL induces death of hepatoma cell lines To start, we evaluated Fn14•TRAIL’s cytotoxic activity against SK-HEP-1 hepatoma cells. As shown in, Fn14•TRAIL induces death of these cells in a dose- and time- dependent fashion. Cell death was detected at 24h and 48h, with only about 20% of live cells detectable at 30 ng/ml Fn14•TRAIL. Of note, significant cell death was detected at a concentration as low as 0.1 ng/ml, corresponding to an EC₅₀ of 0.4 nmol/l. We extended the analysis to other hepatoma cell lines, HepG2 and Huh7. Fn14•TRAIL exhibited a cytotoxic effect against these tumor lines similar to that for SK-HEP-1 cells, albeit with somewhat different kinetics. In the case of Huh7, the significant cytotoxicity was apparent only after 48 h. Importantly, non-malignant hepatocyte cell lines (NKNT3 and FHB) were resistant to death induction by Fn14•TRAIL, even at higher concentrations. ## *Fn14•*TRAIL induces apoptosis of hepatoma cell lines Given that TRAIL is an apoptosis-inducing ligand, we next assessed Fn14•TRAIL’s capacity to induce apoptosis in malignant versus non-malignant hepatic cell lines using a standard Annexin V/PI staining assay. Early (Annexin V⁺ only) and late (Anexine V⁺ and PI⁺) apoptosis are summed together. Fn14•TRAIL effectively induced apoptosis of SK-Hep1, HepG2 and Huh7 cell lines in a dose-dependent manner, with apoptosis apparent by the 24h time- point. Significantly, again, minimal apoptosis was detectable in the non- malignant cell lines treated with Fn14•TRAIL. In order to support the notion that Fn14•TRAIL's inhibitory effect is apoptotic, caspases 3, 8, and 9 cleavage was tested using western blots. Caspases 3 and 8 were significantly cleaved already at short incubation periods (0.5-2 h) in the presence of Fn14•TRAIL, and at 24 h incubation their shorter, active form appeared ( at 19 and 18 kD, respectively, not shown). Caspase 9 was spontaneously activated in this cell line, and therefore at earlier times no significant change was seen with Fn14•TRAIL treatment.. Of note, experiments were repeated with all cell lines mentioned above. Caspases activation was noted only in HepG2, SK-HEP-1 and Huh7, while no such activation was present in the other cell lines that did not underwent apoptosis when incubated with Fn14•TRAIL (not shown). We moved to look at anti-apoptotic signaling pathways. Fn14•TRAIL was found to decrease the expression of the anti-apoptotic proteins cFLIP and Bcl-2. Similar effects of Fn14•TRAIL were observed when SK-HEP-1 cells were incubated with the protein (not shown). However, in accordance with the findings that Huh7 cells were less affected by Fn14•TRAIL, no significant changes in cFLIP or BCL-2 expression were found after the cells were cultured in the presence of Fn14•TRAIL for up to 24h. Interestingly, a decrease in IκBα levels and increase in NFkB expression was evident when cells were incubated with Fn14•TRAIL. As it was previously shown that TRAIL promotes NFkB activation, this might be related to the TRAIL side of the molecule. No significant effect was found in cIAP1,2, JNK (total or phosphorilated) expression when cells were incubated with Fn14•TRAIL (not shown). ## Variable amounts of TRAIL, TRAIL receptors, Fn14 and Tweak mRNA and protein expression levels in hepatocyte cell lines We next wanted to test whether differences in expression levels of TRAIL, Fn14, TWEAK, DR4, DR5, DcR1, DcR2 and OPG can account for the different response of the various cell lines to Fn14•TRAIL. Real time PCR analysis revealed variable mRNA expression levels of the all examined genes. Interestingly, mRNA levels were not correlated with cells' sensitivity to Fn14•TRAIL, sTRAIL or Fn14, or with cell origin e.g. malignant or not. Therefore, we raised the possibility that protein expression might differ significantly between the various cell lines resulting in the observed response to Fn14•TRAIL. Cells in exponential growth were immunostained with fluorescent Abs directed against DR4, DR5, DcR1, DcR2, Fn14 and TWEAK and examined by flow cytometry. All cell lines were found to express TRAIL receptors (DR4, DR5, DcR1 and DcR2), albeit with varying surface levels. In accordance with the real time PCR results, there was no correlation between sensitivity to Fn14•TRAIL’s apoptosis-inducing effect and the levels of surface TRAIL receptors detected by flow cytometry. While Fn14 was also expressed on all of these cell lines, its ligand, TWEAK, was present only at minimally detectable levels on the cell surfaces. In order to examine TWEAK expression further, we proceeded to assess intracellular expression of TWEAK. Significant amounts of TWEAK were found intracellularly, but TWEAK expression level did not correlate with cells' sensitivity. Given the presence of TWEAK in the intracellular compartment, we next asked if it is secreted into the medium. To this end, cells were grown for varying time intervals up to 72 hours, and conditioned medium was analyzed by TWEAK ELISA. Secreted TWEAK was detectable in these cell cultures at low levels (not shown). Of note, again, no correlation was found between TWEAK levels and sensitivity to Fn14•TRAIL. ## *Fn14*•TRAIL binds to hepatoma cells We next determined Fn14•TRAIL’s ability to bind to cell surfaces. To that end, hepatoma cells were incubated in the presence or absence of Fn14•TRAIL, at 4°C prior to immunostaining with fluorescent-labeled anti-Fn14 Ab. All cell lines exhibited significant Fn14 immunostaining, consistent with Fn14•TRAIL binding to these cells. This finding suggests that the differential sensitivity of non- malignant and malignant hepatic cell lines to Fn14•TRAIL is not simply explained by differences in Fn14•TRAIL binding to the respective cell types. ## *Fn14*•TRAIL is more potent than soluble forms of TRAIL and Fn14, alone or in combination We next asked whether the robust efficacy of Fn14•TRAIL as an apoptosis-inducer can be achieved by simply deploying its component parts (Fn14 and TRAIL), alone or in combination. To address this question, the various hepatic cell lines were incubated in the presence or absence of either Fn14•TRAIL, soluble TRAIL (sTRAIL), soluble Fn14 (or Fn14-Fc), or the latter two in combination. sTRAIL is composed of 168 amino-acids of the carboxy-terminal of the extracellular domain of TRAIL (19.6 kD), and differs just in the 3 first amino acids in the amino side from the TRAIL domain of Fn14•TRAIL. sFN14 (5.6 kD), is composed of the 53 amino acids of the extracellular domain of Fn14, and differs from the Fn14 part of Fn14•TRAIL by one amino acid, the first from the carboxy-side. Fn14•TRAIL molecular weight is calculated to be 24.6 kD. As shown in, Fn14•TRAIL was substantially more efficient in inducing HCC cell death when compared not only to sTrail and Fn14-Fc, but also to a combination of the two. Of note, we repeated this set of experiments with soluble Fn14, instead of Fn14-Fc, as the later is a dimer and does not tend to form trimers as TNF-family proteins do, and the results were the same. Importantly, although comparison was made for the same dose of proteins, Fn14•TRAIL's molecular weight is heavier than that of sTRAIL or sFn14, and therefore there was actually more of the other proteins if the molar amount is considered, making the difference in efficacy more significant. ## *Fn14*•TRAIL's activity can be blocked using anti-TRAIL and anti-Fn14 Ab To investigate the importance of the Fn14•TRAIL fusion protein’s two ends, we turned to Ab blocking. Specifically, we incubated the HCC cell lines with Fn14•TRAIL in the presence or absence of anti-TRAIL or anti-Fn14 Ab for 24h, and assessed apoptosis by Annexine/PI staining and flow cytometry. Significantly, both anti-TRAIL and anti-Fn14 Ab completely abrogated the apoptosis induced by Fn14•TRAIL, indicating that both molecular domains of the fusion protein are critical for its activity. ## *Fn14*•TRAIL inhibits tumor growth in vivo We next moved *in vivo*, examining Fn14•TRAIL's ability to inhibit tumor growth. NUDE mice were injected subcutaneously with HepG2 cells and followed daily for tumor growth. When tumors were palpable in individual mice, they were treated with either Fn14•TRAIL (200 µg/day) or vehicle, both administered subcutaneously once daily, for 8 consecutive days. Starting on day 2 after the initiation of treatment, a clear difference in tumor growth rate became apparent, with tumors in the Fn14•TRAIL treatment group growing slower or even decreasing in size. Interestingly, the inhibitory effect on tumor growth persisted for at least 2-3 weeks after the 8 day treatment window. Tumors were excised and weighed at the end. The weights of excised tumors were consistent with the external measurements. In order to verify that Fn14•TRAIL induces apoptosis *in vivo*, in a different experiment, tumors from Fn14•TRAIL treated mice or vehicle treated mice were excised after two doses of treatment and were immuno-stained against cleaved caspase 3. Tumors taken from Fn14•TRAIL treated mice were stained positively, indicating that cells are undergoing apoptosis. Importantly, no morbidity, liver or renal toxicity were observed in the Fn14•TRAIL-treated mice based upon liver enzymes, histopathological examination and urea measurments as can be seen in. # Discussion In the present study, we explored Fn14•TRAIL’s unique properties as an anti- tumor agent. Main findings are: 1) Fn14•TRAIL strongly induces apoptosis in three different HCC cell lines; 2) While HCC cell lines express DR4, DR5 and Fn14 on their surfaces, substantial TWEAK is detected only intracellulary; 3) Non-malignant hepatocyte lines are insensitive to Fn14•TRAIL-induced apoptosis, notwithstanding their similar expression of the aforementioned molecules; 4) Fn14•TRAIL is significantly more effective than soluble Trail, Fn14-Fc or a combination of the two in inducing apoptosis; 5) Fn14•TRAIL's activity can be blocked by Ab directed against TRAIL and Fn14; 6) Fn14•TRAIL effectively inhibits the growth of HCC xenograft tumors in nude mice; and 7) Fn14•TRAIL is well tolerated by mice, with no detectable changes in liver histology. Taken together, these findings establish Fn14•TRAIL’s potential as anti-tumor agent, extending its therapeutic possibilities beyond treatment of autoimmunity. Soluble TRAIL has been considered for some time now for its potential as a cancer therapeutic. However, some malignant tumors are known to be resistant to the pro-apoptotic effect of soluble TRAIL. While recombinant TRAIL variants, as well as agonist Ab with specificity for the TRAIL receptors DR4 and DR5, were used for inhibiting the growth of various types of malignant cells both *in vitro* and *in vivo*, others, found the same HCC cell lines under study here, (SK-HEP-1, HepG2 and Huh7), to be highly resistant to TRAIL-induced apoptosis. This resistance was notwithstanding their expression of DR4 and DR5. Our observation that Fn14•TRAIL, a fusion protein derivative of this same protein, engenders robust apoptosis of the same malignant cells at extremely low concentrations (less than 3 ng/ml in the case of SK-HEP-1 cells) is especially notable. The basis for Fn14•TRAIL’s enhanced pro-apoptotic activity may be several-fold. One possibility is that it stems from the synergy attained by virtue of coordinate blocking of the TWEAK ligand and triggering of TRAIL receptors. However, our repeated observation that the fusion protein is consistently more effective than its soluble components (Fn14 and TRAIL) in combination suggests that there may be yet other explanations for Fn14•TRAIL’s superior activity. One of these explanations revolves around molecular structure, with the possibility that Fn14•TRAIL assumes a higher-order configuration that allows it to function as ’super-TRAIL’. For instance, this could result from stabilization of the TRAIL trimer via TWEAK-induced trimerization of the Fn14 end. TRAIL and other members of the TNF receptor family were shown to be more potent in the trimer form (18-20). The Ab-blocking experiments of the present study shed some light on these mechanistic possibilities. Anti-Fn14 Ab completely abrogated Fn14•TRAIL's pro- apoptotic activity. Possible explanation for this key observation is that the Ab interferes with Fn14’s binding to TWEAK, which in turn could impact both higher order structure of the chimeric protein and/or molecular arraying and signaling at the cell surface. The variable sensitivity of HCC cell lines to Fn14•TRAIL’s pro-apoptotic activity is of interest. However, we could not correlate this differential sensitivity with the protein and mRNA levels of TRAIL receptors, TRAIL, Fn14 and TWEAK in the targeted tumor cells. This is in agreement with previous reports indicating that wide range of tumors express TRAIL receptors, but these are not correlated with sensitivity to TRAIL–induced apoptosis, and post translation modifications of the receptors, influencing their activity has been proposed to explain this phenomena. We did observe that those HCC cells more sensitive to soluble TRAIL tended to be more sensitive to Fn14•TRAIL. Looking at the intracellular signaling pathways, we found that decreased expression of anti- apoptotic signals in parallel with activation of the pro-apoptotic ones was associated with higher sensitivity to Fn14•TRAIL. Decreased expression of the anti-apoptotic signals was not observed in non-malignant cells. The role of TWEAK in this system remains somewhat enigmatic. Whereas surface TWEAK could not be readily detected by immunofluorescence, this protein was readily detectable intracellularly and in conditioned medium. We could not show effect of Fn14•TRAIL on TWEAK:Fn14 signaling pathway in this study. However, most studies unfolding the signaling pathways involved in the Fn14:TWEAK axis were performed with recombinant TWEAK added to the experimental setting, and this is not the case in our study. TWEAK independent Fn14 signaling have been implicated in some tissues, however, it has not been described in HCC cell lines, and therefore it is not expected that Fn14•TRAIL will influence this signaling pathway. There is no consensus as to whether it is more beneficial to block as opposed to activate the TWEAK: Fn14 signaling axis in the context of cancer therapeutics. Arguing for blockade are studies indicating the importance of TWEAK in tumor cell survival, resistance to apoptosis and migration. Also pointing in this direction is the demonstration that enforced TWEAK over-expression enhances proliferation *in vivo*. This view is consistent with our suggestion of using Fn14•TRAIL as a TWEAK blocker. On the other hand, there is data in the literature arguing in the other direction, for a potential benefit in activating this pathway. These data encompass findings that TWEAK itself possesses pro-apoptotic activity, either directly through activation of the caspase cascade, or indirectly via inducing TNF-alpha. The indirect mechanism parallels that associated with other death domain-deficient receptors, such as CD30 and CD40. Also, a recent study has shown that agonistic anti-Fn14 Ab has anti-tumor activity, both *in vitro* and *in vivo*. However, caution is called for in interpreting the latter findings given that, as the authors themselves point out, at least some of anti-tumor activity of the agonistic anti-Fn14 Ab, was mediated by antibody-dependent cell cytotoxicity. Recently, we reported two other fusion proteins, CTLA-4•FasL and CD40•FasL, with cytotoxic activity against tumor cells. Interestingly, we demonstrated that these proteins function, at least in part, by bridging neighboring molecules on tumor cell surfaces. Whether Fn14•TRAIL shares mechanistic features with these other fusion proteins (for example, Fn14 binding to a surface molecule other than TWEAK, or to sub-threshold levels of TWEAK) remains to be determined. However, regardless of how this unfolds, the broader concept that fusion proteins with more than one functional component can be used for tumor cell targeting is now being reinforced, and in this way, opens up a new avenue for devising therapeutics that fit into the personalized onco-medicine paradigm. [^1]: MDE has a research grant and consulting fee from KAHR Medical LTD. SA is employed by KAHR Medical LTD, NS CEO of KAHR Medical LTD, and MLT is a share holder and inventor - KAHR Medical LTD. F n14-TRAIL is a fusion protein protected by patents issued and licensed to KAHR medical, MLT is an inventor. There are no further patents, products in development or marketed products to declare. This does not alter the authors' adherence to all the PLOS ONE policies on sharing data and materials. [^2]: Conceived and designed the experiments: AA MDE. Performed the experiments: AA TP SA KT. Analyzed the data: AA TP SA. Contributed reagents/materials/analysis tools: NS. Wrote the manuscript: AA JR MLT SA MDE.

# Introduction Microcystins (MCs) are a group of cyanobacterial toxins comprised of more than 80 variants. MC-LR is both one of the most common variants and one of the most potently toxic peptides, containing amino acids Leucine (L) and Arginine (R) in the variable positions. The outbreak of a cyanobacterial bloom induces the release of MCs into water and represents a serious threat to aquatic ecosystems. Previous studies have shown that the death of large numbers of fish during outbreaks of cyanobacterial bloom is associated with the production of MCs and with several special conditions, including high water temperature, high pH, high concentration of ammonia and nitrogen, and low dissolved oxygen. It is well known that MCs can cause a variety of toxic effects in fish through various pathways. MC exposure can cause histopathological changes in various organs, including the liver, kidney, gill, intestine, and heart. It can also alter the activity of various enzymes in the fish. In addition, MC exposure can affect growth rate, osmotic pressure, heart rate and behavior. Many laboratory and field studies have demonstrated that MCs can accumulate in various tissues and organs of fish (mainly in the liver, but they can also be detected in muscle tissues). The long-term accumulation of toxins in fish will definitely produce harmful effects, and may also affect human health through the food chain \[4-9\]. A number of exposure routes have been used to study the effects of MCs on fish, including intraperitoneal injection, feeding, immersion in water containing purified toxins (MC-LR or MC-RR), cyanobacteria crude extract, and whole cells of cyanobacteria. Intraperitoneal injection is the most commonly used technique, due to the rapid onset of toxicity. The mechanism behind the toxicity of MCs in fish is similar to that in mammals, causing irreversible inhibition of protein phosphatases PP1 and PP2A in fish liver cells. This leads to excessive phosphorylation of proteins, alterations in the cytoskeleton, loss of cell shape and subsequent destruction of liver cells, causing hepatic hemorrhage or hepatic insufficiency. MCs are also responsible for an increase in oxidative stress, which can subsequently trigger apoptosis. However, the association between intracellular ROS levels and other toxicities in fish remains unclear. The cytoskeleton consists of three major structural elements: microtubules, microfilaments, and intermediate filaments. These elements play an important role in maintaining cellular architecture and internal organization, cell shape, motility, cell division, and many other processes. It has been reported that microtubules can be disrupted by cyanobacteria extract or purified MCs in primary cultured rat hepatocytes and several non-hepatocyte cell lines. Ding et al. suggested that intracellular GSH plays an important role in MC-induced cytotoxicity and cytoskeleton changes in primary cultured rat hepatocytes. However, the role of the excessive production of ROS in this biological process has not been fully elucidated. In addition, it would be interesting to ascertain whether and how MC-LR could induce similar effects on the cytoskeleton system in fish liver cells, a question which has received little attention so far. Carp (*Cyprinus carpio* L.), a common fish that are widely distributed in Asia, including China, were chosen to study the toxic effects of MC-LR. In our present study, the effects of sublethal doses of MC-LR on the ROS level, HSP70 expression, cytoskeletal structure, and apoptosis in liver cells were investigated. The results obtained in this study help to reveal the association between intracellular ROS and other toxic effects induced by MC-LR and to further examine the detailed toxicological mechanisms behind MC-LR-induced toxicity. # Materials and Methods ## Ethics Statement This study was approved by the Animal Ethics Committee of the Nanjing Institute of Environmental Sciences, Ministry of Environmental Protection. The institute does not issue a number or ID to any animal study, but the ethical committee approved and helped guide the animal use in this study. ## Chemicals and reagents MC-LR (purity \> 96%) was purchased from Alexis Biochemicals (Läufelfingen, Switzerland). N-tert-Butyl-a-phenylnitrone (PBN, purity \> 98%) was purchased from J&K Chemical (USA). Paraffin, hematoxylin, eosin, glutaraldehyde, uranyl acetate and lead citrate were purchased from Sigma Chemical (St. Louis, MO, USA). Dimethyl sulfoxide (DMSO) and methanol were purchased from Tedia (Fairfield, OH, USA). Other reagents were analytical grade and obtained from chemical companies in China. ## Fish and experimental designs Six month old carp, with an average body length of 14.00 ± 1.08 cm and weight of 29.26 ± 5.09 g, were obtained from a pilot research station of the Freshwater Fisheries Research Center (FFRC), Chinese Academy of Fishery Sciences. These fish were acclimated to laboratory conditions for 14 days with dechlorinated tap water. Fish were fed commercial pellet food daily during the acclimation and test periods, except for the last two days of acclimation. During experiments, water temperature was 16.1 ± 0.2 °C, pH value was 7.20 ± 0.35, dissolved oxygen was 8.6 ± 0.5 mg/L, the photoperiod was 12 h/12 h, and total hardness was 129.7 ± 8.3 mg as CaCO₃ per liter. Water was constantly aerated during the acclimatization and test periods. Carp were randomly divided into three groups with 40 carp per group. Each group was treated with either 50 μg/kg of MC-LR, 120 μg/kg of MC-LR, or saline, by intraperitoneal injection (MC-LR was dissolved in saline). An equal volume of saline was administrated and used as a control. Each group was then subdivided into five groups with 8 fish per group and carp were sacrificed at 1, 5, 12, 24 and 48 hours after exposure to MC-LR. Livers were quickly taken out for immediate use or frozen in liquid nitrogen before storage at a temperature of -80°C for further analysis. ## ROS trapping and EPR measurement ROS levels were determined by electron paramagnetic resonance (EPR), according to the method described by Luo et al. and Jiang et al.. About 0.1 g of liver tissues were quickly homogenized with a cold glass homogenizer in 1.0 ml of 50 mM PBN dissolved in DMSO. The homogenates were transferred to quartz capillary tubes and then immediately stored in liquid nitrogen for EPR analysis. All operations were performed in a sealed box that was purged continuously with nitrogen gas. The EPR spectra were recorded with a Bruker EMX 10/12 X-band spectrometer (Bruker, Germany) under the same conditions described by Luo et al. and Jiang et al.. The height of the second peak of the EPR signals was interpreted as the intensity of •OH in liver tissues. All experiments were performed in quadruplicate. ## Immunohistochemistry Immunohistochemistry was performed according to the method described by Jiang et al. using Rabbit Anti-HSP70 (Fish) Polyclonal Antibody (Stressgen, USA) at a dilution of 1:800. Sections (4 to 5 μm) were mounted on silane-coated slides and stained with the SP-9001Histostain™-Plus Kit (Zymed, USA) according to the manufacturer's instructions. The mean integrated optical density (MIOD) of the HSP70 expression-positive area was calculated using Image-Pro Plus software. At least six fields were calculated per slide. ## Laser scanning confocal microscopy Cytoskeletal changes were evaluated based on the expression and distribution of beta-tubulin. Double fluorescence immunohistochemistry sections of 5 μm were cut from paraffin blocks and mounted on gelatin/chrome alum-coated glass slides. The paraffin sections were deparaffinized in xylene and rehydrated in graded ethanol and distilled water. The non-specific binding sites were blocked in 5% normal goat serum diluted in 1X PBS with 0.3% Triton X-100 for one hour at room temperature. Beta-tubulin was stained overnight with an Alexa Fluor® 488-conjugated monoclonal antibody against beta-tubulin (9F3, Sigma) at a 1:500 dilution at 4°C. The antibody was diluted in 1X PBS containing 1% BSA and 0.3% Triton X-100. DAPI (4’,6-diamidino-2-phenylindole) was used to stain the nucleus. The sections were measured using a Zeiss LSM 710 laser scanning confocal microscope (Zeiss, Oberkochen, Germany). The green fluorescence of Fluor® 488 was excited at 488 nm and the blue fluorescence of DAPI was excited at 405 nm. ## Real time quantitative PCR Total RNA was isolated from the liver samples using the TRIzol reagent (Invitrogen) according to the manufacturer’s instructions, and treated with DNAase to eliminate residual DNA prior to reverse transcription of total RNA to complementary DNA (cDNA). The concentration of total RNA was measured by absorbance at 260 nm. The purity was estimated by the 260/280 nm absorbance ratio. One microgram of total RNA was characterized by denaturing agarose gel electrophoresis. Primers for the evaluation of bcl-2 (forward primer: 5’-TTTCGCTCAGAAGTGACGGC-3’; reverse primer: 5’-GCAGTGCGGTGCTGAAAGAT-3’), JNKa (forward primer: 5’-TAAAACACCTCCACTCGGCG-3’; reverse primer: 5’-GCCAGACCGAAATCCAGGA-3’), and p38 (forward primer: 5’-TGGAAACGGCTCACGTATGAA-3’; reverse primer: 5’- TCTGGATGAAGGTCCTGGAGG-3’) gene expression were designed using Primer Express 2.0 (Applied Biosystems). The cDNA was synthesized from 0.5 μg of total RNA, using the PrimerScript RT reagent Kit (TaKaRa) on a GeneAmp® PCR System 9700 (Applied Biosystems) according to the manufacturer’s protocol. Quantification of gene expression was carried out on a 7500 Fast Real- Time PCR System (Applied Biosystems). The reaction mixture was composed of 10 μl of SYBR Green RT-PCR Master Mix (TaKaRa), forward and reverse primers (10 μM, 0.4 μl each), ROX Reference Dye II (50×, 0.4 μl), 7.8 μl of nuclease-free water, and the cDNA sample (1 μl). The PCR amplification protocol was 95°C for 15 seconds followed by 40 cycles of 95°C for 5 seconds and 62°C for 34 seconds. After PCR, a melting curve analysis was performed to demonstrate the specificity of the PCR product, which was displayed as a single peak. Every sample was analyzed in quadruplicate. Differences in expression levels were calculated using the 2^-∆∆Ct method. Amplification of 18S rRNA (Applied Biosystems) was used as an internal control for bcl-2, JNKa and p38 expression. Statistical significance was determined using a one-way ANOVA, followed by Duncan’s multiple range test (SPSS, Inc., Chicago, IL). Data are presented as means with standard errors (mean ± SE). A *p*-value \< 0.05 was considered statistically significant. ## Flow cytometry Cells were isolated from fresh carp liver. Briefly, the livers were cut into pieces and put into a cell separator containing 0.1 M phosphate buffer (pH = 7.4) and chopped for 1 min (the Becton Dickinson Medimachine, USA). The cells were then filtered through a membrane (STERIKING 75 mm, WIPAK Medical, Finland) and then centrifuged at 1000 rpm for 5 min. Cells were collected, washed, centrifuged and resuspended in 1% (w/v) paraformaldehyde in PBS (pH 7.4). They were then kept on ice for 1 hour, washed in PBS, centrifuged, and fixed in 70% (v/v) ice-cold ethanol at -20 °C for 12 hours, before being stained with terminal deoxynucleotidyl transferase (TdT) and FITC-labeled deoxyuridine triphosphates (FITC-dUTP). A commercial TUNEL kit (APO-DIRFCT^TM Kit, BD Biosciences) was used to perform the TUNEL staining of liver cells according to the manufacturer’s instructions. TUNEL is a method for detecting the 3'-OH ends of DNA exposed during the internucleosomal cleavage that occurs during apoptosis. TUNEL assays identify apoptotic cells by TdT-mediated addition of labeled (X) deoxyuridine triphosphate nucleotides (X-dUTPs) to the 3’-OH end of DNA strand breaks, which are subsequently visualized depending on the introduced label. Apoptosis of liver cells was analyzed using a FACSCalibur flow cytometer (BD Biosciences). ## Statistical analysis Data were expressed as mean ± standard error (SE) and analyzed using a one-way ANOVA. Significant differences between means were determined with the LSD-*t* test. Differences were considered to be significant at *p* \< 0.05 (\*) and *p* \< 0.01 (\*\*). # Results ## Effects of MC-LR on hepatic ROS levels in carp liver Typical EPR spectra of the carp hepatic ROS signal, induced by MC-LR, consisted of three groups with two hyperfine splitting peaks in each, with identical intensity. These six line spectra were identified as radical hydroxyls (•OH) with hyperfine splitting constants of g = 2.0057, a^N = 13.88 G, a^H = 2.35 G. The intensity of •OH in different groups is shown in. The intensity of •OH in the group treated with 50 μg/kg of MC-LR was significantly increased at 5h and 12 h after MC-LR exposure, compared to the control group (*p* \< 0.05, *p* \< 0.01). The intensity of •OH in the group treated with 120 μg/kg of MC-LR was significantly increased at 1, 5, and 48 hours after MC-LR exposure, compared to the control group (*p* \< 0.05, *p* \<0.01, *p* \< 0.01). Of note, signal intensity of •OH was also seen in the control group, indicating that •OH can also be produced under normal physiological conditions. ## Effects of MC-LR on HSP70 in carp liver Expression of HSP70 in the liver was measured using immunohistochemical methods. As shown in, the obvious yellow stain indicates the presence of HSP70. Image-Pro Plus software was used to calculate the average cumulative optical density (MIOD) based on HSP70 staining. The control group exhibited less HSP70 protein expression. As shown in, a 7.09 fold increase in HSP70 was observed in the group treated with 50 μg/kg of MC-LR after a 12 h exposure, as compared to the control group. A significant increase in HSP70 expression was observed in the 50 μg/kg group after 12 h and 24 h exposures to MC-LR, compared to the control group (*p* \< 0.01). A significant increase in HSP70 expression was observed in the 120 μg/kg treated group after 24 h and 48 h exposures to MC-LR, compared to the control group (*p* \< 0.05, *p* \< 0.01). A 6.27 fold increase in HSP70 was observed in the 120 μg/kg treated group after a 24 h exposure to MC-LR, as compared to the control group. ## Effects of MC-LR on the cytoskeleton in carp liver cells Immunofluorescence staining of beta-tubulin was used to evaluate changes in the cytoskeleton. shows the morphologies of hepatic cells stained for microtubules and nuclei. Cytoskeletal changes were dependent on the dose and the exposure time of MC-LR. In the control group, the nuclei were widely surrounded by highly organized microtubules with normal morphology, whereas in the 50 μg/kg treated group, cytoskeletal proteins were condensed around the nucleus, as evident after a 12 h exposure to MC-LR. Moreover, liver cells had a hollow nucleus with condensed chromatin and exhibited apoptotic properties as compared to the control group. After a 48 h exposure to 50 μg/kg of MC-LR, the cytoskeletal structure of most liver cells returned to normal. Similarly, the condensed cytoskeletal proteins could be clearly observed around the nucleus in the 120 μg/kg treated group after 5 h and 12 h exposures to MC-LR. The liver cells present in these two groups had more obvious hyperchromatic nuclei and nuclear pyknosis, with more irregular and hollow nuclei. After a 24 h exposure to 120 μg/kg of MC-LR, the cytoskeletal structure had been restored, however, the cytoskeletal protein content in the 120 μg/kg treated group was still lower when compared to the control group at 48 h. In addition, the effects of 200 mg/kg of N-acetylcysteine ​​(NAC, a GSH synthesis precursor) and buthionine sulfoximine (BSO, a GSH synthesis inhibitor) on the cytoskeleton were studied. As shown in, NAC pretreatment had a significant protective effect on cytoskeletal proteins, as early as 1 h after exposure to MC-LR. BSO treatment induced hepatocellular nucleus condensation and decreased the skeletal protein content – a less serious effect than that caused by BSO + MC-LR treatment. shows that the cytoskeletal structure of the liver cells had been destroyed, and a considerable number of nuclei had disappeared at 48 h in the 200 mg/kg BSO + 50 μg/kg MC-LR treated group. ## Effects of MC-LR on the expression of mRNA in carp liver Genes involved in apoptosis, such as p38, JNKa and bcl-2, were detected by real- time quantitative PCR. As shown in, although the expression of p38 appeared to be increased at 5 h and 48 h after exposure to 50 μg/kg of MC-LR, the increase was not statistically significant. Expression levels of JNKa at 5 h and bcl-2 at 24 h and 48 h after exposure to 50 μg/kg of MC-LR were significantly increased (*p* \< 0.05). In the group treated with 120 μg/kg of MC-LR, the expression of p38 was significantly increased at 5 h (a 5.99 fold increase) and 48 h compared to the control group (*p* \< 0.05). The expression trends of JNKa in the 50 μg/kg MC-LR treated group changed in a way that was parallel to that of the 120 μg/kg MC-LR treated group. The highest expression (6.23 fold increase) was found 5 h after exposure to 120 μg/kg of MC-LR, compared to the control group (*p* \< 0.05). Expression of bcl-2 was found to have significantly increased at 24 h and 48 h, compared to the control group (*p* \< 0.05). A 9.38 fold increase in Bcl-2 was found after 48 h of exposure to 120 μg/kg MC-LR. ## Effects of MC-LR on hepatocellular apoptosis The TUNEL stained hepatic cells suspension was analyzed by flow cytometry and the percentage of apoptotic cells was calculated. As shown in, a significant increase in apoptotic cells was found at 12- 48 h in different MC-LR treated groups (*p* \< 0.05). The highest percentage of apoptotic cells was found at 12 h in the 50 μg/kg treated group (*p* \< 0.05), which was about 156.6 percent of the control group. Moreover, with further prolongation of the exposure time, the percentage of apoptotic cells in the 50 μg/kg treated group tended to decline and a significant decrease appeared at 48 h, compared to that at 12 h (*p* \< 0.05). This phenomenon was, to some extent, supported by direct immunofluorescence and confocal microscopic studies, in which the evident hollow nuclei with condensed chromatin and the exhibited apoptotic properties were observed at 12 h. # Discussion As a liver-specific toxin, MC-LR is a potent inhibitor of protein phosphatase 1 (PP1) and protein phosphatase 2A (PP2A). In addition, previous studies have suggested that MC-LR might be able to induce excessive production of ROS. The association between the inhibition of protein phosphatases and the over- production of ROS is still controversial and under investigation. In this study, we use carp as a model to study crosstalk between these two potential mechanisms of MC-LR toxicity. Many natural or synthetic compounds can induce excessive production of ROS, such as O₂^•-, •OH and H₂O₂. Oxidative stress that is closely related to over production of ROS is an indicator of interference with the body's normal redox state. Hydroxyl radicals, the most harmful free radicals in the body, can cause lipid peroxidation, protein peroxidation and DNA oxidative damage by attacking proteins, unsaturated fatty acids, DNA and other macromolecules. It has been shown in many studies that MCs can cause oxidative stress both *in vitro* and *in vivo* \[3, 23-27\]. However, the levels of free radicals *in vivo* have never been determined, due to technical limitations involved in the measurement of these free radicals. As mentioned above, however, EPR has proven to be the most reliable method for measuring free radicals. In this study, we provide direct evidence of the production of hydroxyl radicals induced by MC-LR in carp liver, using the EPR technique. The hyperfine parameters obtained through computer-assisted fitting analyses are consistent with previous reports in which PBN captured •OH to generate PBN-•OH adducts. Thus, we believe the reactive oxygen species induced by MC-LR in carp liver is •OH. In addition, we observed that the •OH levels in the fish liver significantly increase during the early stage of MC-LR stress. In particular, the •OH level was significantly increased in the group treated with 120 μg/kg, one hour post MC-LR injection. ROS production in animal cells is mainly associated with the mitochondrial metabolism. Ding et al. observed a surge of the mitochondrial Ca²⁺ level in rat hepatocytes after a 10 min exposure to 1 μM MC-LR, resulting in the subsequent onset of membrane permeabilization transition (MPT). This led to the production of ROS, loss of the mitochondrial membrane, MPT, and the release of apoptotic factors, including cytochrome c, triggering apoptosis. ROS plays a critical role by serving as the second messenger in the MC-LR toxicity pathway. In addition, it has also been proven that MC-LR induces the production of ROS in the leaves of aquatic plants, leading to lipid peroxidation and resulting in ultrastructural damage. In the present study, we observed that MC-LR induced the generation of ROS in carp liver cells, which mediated oxidative damage, apoptosis and necrosis, as well as the destruction of the cytoskeleton. HSP70, a biomarker that is widely used to evaluate environmental stress in aquatic organisms, can protect the body from oxidative stress and apoptosis. Liver HSP70 induction may rely on the perturbation of the cellular redox status. In our previous study, the expression of HSP70 was dramatically increased following exposure to dissolved MC-LR at 1.0 to 10.0 μg/L, which indicates its important role as a molecular chaperone under oxidative stress and explains the high tolerance of *C. carpio* to dissolved MC-LR under common environmental concentrations. In the present study, a significant increase of the expression of HSP70 was also observed in the 50 μg/kg group after 12 h and 24 h exposures to MC-LR and in the 120 μg/kg group after 24 h and 48 h exposures to MC-LR. To some extent, this over-expression of HSP70 in fish liver might contribute to alleviate MC-LR toxicity, which could manifest itself by the reduction in symptoms of hepatic cytoskeletal disruption and apoptosis in the later stage of the trial, particularly apparent in the 50 μg/kg treated group. However, the expression of HSP70 cannot reverse the damage caused by MC-LR to liver tissue. MC-LR stress causes inhibition of protein phosphatases, resulting in over- phosphorylation of many proteins, which can further lead to rearrangement of the cell’s intermediate filament network and destruction of the cytoskeleton. Beta- tubulin, one of the major components of the cytoskeleton and an indicator of cytoskeletal damage, may be associated with oxidative stress. We observed that a noticeable reduction in the number of microtubules in liver cells and the appearance of densely aggregated microtubules retracted around the nucleus occurred consistently with a high level of •OH induced by MC-LR. In addition, NAC, a GSH precursor and a commonly used anti-oxidant, can protect the cytoskeleton of carp liver cells. On the contrary, BSO, which is specific to GSH synthesis, increased the amount of damage to the cytoskeleton. In addition, GSH is a major cellular nonprotein thiol reductant and participates in numerous cellular processes, such as intermediary metabolism and protection of cells against oxidative stress. Indeed, a noticeable depletion of intracellular GSH could be observed in the liver of fish exposed to 50 μg/kg and 120 μg/kg of MC- LR at 5 h – 12 h (results not shown), which could, in turn, alter the intracellular redox status. The decreased GSH level and the over-production of ROS could disturb the assembly of microtubules and destroy their stability. As mentioned above, the high level of MC-LR could induce ROS formation in fish hepatocytes within a relatively short exposure time (1 h). Therefore, we believe MC-LR-induced ROS formation may play an important role in the disruption of microtubule structure, as observed in the present study. MC-LR-induced apoptosis in mammalian cells has been widely confirmed. Gehringer reported that MC-LR could cause dual cellular effects (dualistic response). This means that at a low-dose (≤ 20 μg/kg *in vivo*) MC-LR successfully induced cell proliferation, while high doses (≥ 32 μg/kg *in vivo*) resulted in apoptosis/necrosis. In our present study, apoptosis was detected in MC-LR treated carp using immunohistochemistry and flow cytometry. We found that both low and high doses of MC-LR could induce significant apoptosis after a 12 h exposure. However, low dose effects on apoptosis were more obvious, as high doses of MC-LR induced necrosis. Multiple genes regulate apoptosis, such as the bcl-2 and the caspase families, as well as oncogenes like c-myc, and tumor suppressor genes like p53. Mitogen-activated protein kinase (MAPK) is an important eukaryotic signal transduction pathway and plays a key role in the regulation of gene expression and cytoplasmic activities. c-Jun amino-terminal kinase (JNK) and p38 play an important role in stress reactions, such as inflammation and apoptosis. The expression levels of JNK and p38 were significantly increased after exposure to low and high doses of MC-LR, indicating ROS mediated cell apoptosis. Oxidative stress has been shown to induce JNK activation, phosphorylation of bcl-2 and Bcl-xL, as well as to promote apoptosis. The JNK and p38 pathways usually present synergistic apoptotic signals, which we also found to be the case in our present study. Overexpression of bcl-2 was found in both low and high dose MC-LR treated groups. Bcl-2 is an anti-apoptotic gene and its overexpression is assumed to reduce production of oxygen free radicals and lipid peroxidation. Moreover, overexpression of bcl-2 can also increase production of GSH and other antioxidants. Although the over-expression of bcl-2 enhanced resistance to some degree, the damage to fish liver caused by MC-LR could not be reversed. In our present study, a significant induction of hydroxyl radicals (•OH) was observed in carp liver after exposure to MC-LR. This provides evidence for oxidative stress as the toxic mechanism induced by MC-LR. Excessive production of ROS and inhibition of protein phosphatase triggers a series of pathological effects, including destruction of the cytoskeleton. Pre-injection of the antioxidant NAC has a significant protective effect on the carp liver cytoskeleton. On the contrary, BSO exacerbates the damage to the cytoskeleton. ROS could induce the expression of apoptosis-related genes, including p38 and JNKa. A significant increase in apoptotic cells was observed 12 - 48 hours post- exposure. Apoptosis was reduced after 48 h, which may be related to the upregulation of bcl-2 and the previous over-expression of HSP70. Our study further supports the role of ROS in MC-LR induced liver damage in carp, and provides a basis for the ongoing study of the molecular mechanisms behind MC-LR toxicity. The EPR spectra of •OH *in vivo* were recorded by a Bruker EMX 10/12 X-band spectrometer at the Modern Analysis and Testing Center of Nanjing University. We thank the faculty for analyzing the ROS data. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: JJ ZS XW. Performed the experiments: JJ WX JZ DK JX. Analyzed the data: JJ WX ZS XW. Contributed reagents/materials/analysis tools: DK JX. Wrote the manuscript: JJ.

# Introduction Motion analysis systems are widely employed within both universities and clinical facilities, to explore the association between movement patterns and athletic performance or risk of injury, which is of great interest to coaches, physiotherapists and other medical professionals. However, while there are many studies that have examined movements using kinematic and kinetic measurements seeking to identify features (e.g. maximum knee flexion) that might be related to injury, there are no well established evidence-based guidelines that state what movement deficiencies are, what normative ranges of variability are or what a “normal” movement looks like. Additionally, there is little agreement as to which, if any, exercises or movement can expose patterns that could lead to injury or what properties an exercise should hold (single or double leg, movement in one, two or three planes, ecological validity and so on). For example, both Hewett et al., and Krosshaug et al., examined features extracted from a double leg drop jump (DLDJ) to assess their ability to predict the risk of sustaining an anterior cruciate ligament (ACL) injury. Both studies performed a prospective analysis on a large population ( = 205; = 782) of female athletes participating in field sports. While Hewett et al., concluded that specific features (describing knee motion and knee loading) were very sensitive in predicting ACL injury, Krosshaug et al., reported poor prediction abilities. The conflict in findings between movement analysis studies, provoked Bahr to state in a recent review: “To date, there is no screening test available to predict sports injuries with adequate test properties and no intervention study providing evidence in support for screening for injury risk”. A possible source of conflicting conclusions is the way features are extracted. When describing a movement, studies often extract features based on prior knowledge (previous research and / or personal and clinical experience) or post hoc analysis to reduce dimensionality within the highly multivariate datasets, across joints and time. These features are then used to compare the magnitude or timing between groups assuming that the extracted features capture the underlying function of a signal. While discrete points can be helpful in understanding movements, the selection of discrete points has the potential to: discard important information, to compare features that present unrelated neuromuscular capacities and to encourage fishing for significance—e.g. non- trivially biased non-directed hypothesis testing. Due to the apparent limitations in discrete point selection, other analyses have been introduced in recent years—statistical parametric mapping, (functional) principal component analysis, analysis of characterising phases, point by point manner testing and other techniques to improve the analysis of movement patterns. Another possible source for conflicting conclusions is the way extracted features are compared across groups. Comparisons are often made using statistical significance, and conclusions are developed by inferring properties about a population by testing hypotheses and deriving estimates by probability values (p-values). While, p values are useful they are frequently criticised due to the possibility of committing type 1 and 2 errors, that any difference can be statistically significant—with large enough sample size—that p values do not provide statistical precision and that conclusions do not account for multiple movement strategies (subgroups) within a dataset. As such, injury related features can be masked during an analysis. Differences in movement strategies within a dataset may be caused by differences in anthropometric measures as well as training, sporting and injury history and there is growing evidence for different movement strategies across individuals. When examining a human movement, the method of data analysis chosen needs to contend with a complex and multivariate system and any analysis using an inference test may not help to progress the understanding of movement as it does not account for differences in movement strategy or the interrelationship of segments. More suitable methods for movement analysis might be a data driven machine learning techniques that do not require expert knowledge, which have gained popularity in biomechanics and other fields as they have demonstrated an enhanced ability to understand complex and multivariate systems—see Barton et al., Halilaj et al., Rajpurkar et al., or Figueiredo et al.. While studies do not frequently use machine learning techniques in sport-biomechanics, machine learning techniques have been used in gait studies to classify movement data into pathological and able-bodied, to detect / judge severity of diseases or to predict surgical / therapy outcomes. For the interested reader Figueiredo et al. reviewed experiments that used machine learning in gait-biomechanics, while Halilaj et al. reviewed the usage of machine learning technique in human movement biomechanics stating best practices and common pitfalls. One of the few studies within sports-biomechanics that used a data-driven supervised learning approach to classify the skill level of athletes (novice and elite) using biomechanical data is Ross et al., and reported an accuracy of 70 to 80%. However, machine learning techniques have not been applied to differentiate movement patterns between non-injured and rehabilitating athletes (e.g. after ACL reconstruction). As such it is not known a) what the most appropriate machine learning technique for such a classification task is, b) what classification accuracies one could expect or c) if movement patterns differ on a group level (e.g. non-injured vs. rehabilitating) or limb level (normal limb vs. operated limb and limb contralateral to operated limb) as movement pattern difference might by driven by anatomical/morphological differences, which could affect both legs or just one limb in rehabilitating cohorts. The aim of this study was to develop and test a data driven framework (feature generation based on no expert or prior knowledge) to classify movement patterns of normal and rehabilitating athletes using only biomechanical data. # Materials and methods ## Subjects This study used a cohort of males recovering from ACL reconstruction as the rehabilitating cohort (ACL reconstruction should have altered kinematics and neuromuscular properties within the operated knee) and a cohort of healthy males as control / normal group. The ACL group (n = 156) was recruited from the caseload of two orthopaedic surgeons who specialise in knee surgery between January 2014 and December 2016. Inclusion criteria were: biomechanical assessment approximately 9 months after ACL reconstruction, intention of returning to full participation in multi-directional sport after surgery, bone patellar tendon bone graft or hamstring graft from the ipsilateral side, gender (male) and an age between 18 and 35. Exclusion criteria were: multiple or previous ACL reconstructions and any meniscal repairs. The control group (n = 62) contained males that participated in multi-directional sport (i.e. Gaelic Football, Soccer, Hurling, Rugby Union) that were free of injury in the 3 months prior to testing, had no previous knee surgery and were between 18 and 35 years of age. The study received ethical approval from Sports Surgery Clinic Hospital Ethics committee (25AFM010) and was registered on clinicaltrials.gov (NCT02771548). The examined data were fully anonymised and all subjects provided informed written consent to have their data used for research. The ACL group had an average age of 24.8 ± 4.8 years was 180 ± 8 cm tall and had a body mass of 84 ± 15.2 kg. The control group had an average age of 24.8 ± 4.2 years was 183 ± 6 cm tall and had a body mass of 82 ± 8.9 kg. ## Data capture and pre-processing The testing took place in the motion analysis laboratory using an eight-camera motion analysis system (200Hz; Bonita-B10, Vicon, UK), synchronised with two force platforms (1000Hz BP400600, AMTI, USA). Before data collection, all subjects undertook a standardised warm-up and wore their own athletic footwear with 24 reflective markers secured to the shoe or to the skin using tape, at bony landmarks according to the Plug-in-Gait marker set. Three trials of each limb for the following seven exercises were captured: double leg (DL) countermovement jump (CMJ;), single leg (SL) CMJ, DL drop jump (DJ;), SLDJ, Hurdle Hop (HuHo;), SL Hop (SLHop;), planned change of direction (CoDP) and unplanned change of direction (CoDU). Marker and force data were low-pass filtered using a fourth-order Butterworth filter—before computing kinematic and kinetic measures using Nexus (1.8.5; Vicon, UK). Data pre-processing (gap filling and waveform screening) was performed on examined biomechanical measures using a custom developed MATLAB program (R2015a, MathWorks Inc., USA). All kinetic variables were normalised to body mass. The start and end point of an exercise was defined using the force trace or a combination of center of mass (CoM) power and force trace and all measures were landmark registered using a dynamic time warping process to align the end of the eccentric phase across the all curves. Only the maximum trial of the captured (based in maximal jump height and shortest contact time) was chosen for data analysis. ## The framework The steps taken during data analysis can be described as follows: feature engineering, selection of a supervised learning technique, feature selection and final validation of minimal feature model. The performance of any generated classification model described in this study is the average classification accuracy of a 100 split stratified shuffle split cross-validation. A shuffle spilt cross-validation was used in preference to a k-fold cross validation because, although information leaking can occur between the splits, it allows the re-use of subjects that have been assigned previously to the trainings group and hence a number of splits and greater sample size during the training / fitting of the prediction model. The description of the generation of the framework is based on a single exercise and was done for every exercise separately. During the performed splits, false and correct classifications were recorded to allow the generation of a confusion matrix for every model generated on a group (e.g. ACL and NORM), limb (class: ACL^OP, ACL^{NO OP} and NORM) and subject level. This information is essential when examining the prediction model, in respect to errors (confusion or misclassifications) made and can also help in understanding the examined data—e.g. do features differ on a group or limb level (misclassifications of class but not group level), do some subjects behave differently than others within a class (misclassification of specific subjects) or if one model made misclassifications another did not (could to models complemented each other). In addition to recording false and correct classifications, during every split a “guess prediction” was made with each trial having a 1 in 3 chance of belonging to the class: ACL^OP, ACL^{NO OP} or NORM. The best guess, the highest guess classification accuracy observed within splits, was then used as a threshold to judge the meaningfulness of the generated models. A classification model was defined meaningful if the lowest observed accuracy during the splits was greater than the best guess. This study evaluated the created models on a number of levels: the number of features selected for inclusion (3 levels: all features, 20 features, minimal feature), the machine learning technique (7 widely used techniques were applied) and movement type (8 movement frequently used to assess the rehabilitation status of athletes with ACL and to predict future injury). Except of the minimal feature model, the evaluation was performed by assessing the average classification accuracy. Evaluation of the minimal feature model was based on average classification accuracy, classification errors and sensitivity. Other classification performance measures e.g. area under the curve, precision, recall, f1-score or specificity would very likely have resulted in different models and performance rankings. ### Feature engineering Traditional analyses, which use discrete measures, are dependent on the selection of features. To enable a data-driven feature engineering that does not require expert knowledge, this study reduced dimensionality of commonly examined motion capture time-series by examining variability within the time-series across examined in 100 random shuffle splits using analysis of characterising phases. This information was then used to reduce a time-series to just a few features that describe the variability of the signal across all participants. This process was performed on each examined time series separately. The following 82 time series were examined separately: ground reaction force (GRF; x, y, z), GRF impulse (x, y, z), center of mass (CoM) velocity (xy, xyz, z), CoM power (x, y, z), CoM in pelvis, hip, knee and ankle (x, y, z) as well as joint angles of the ankle, knee, hip, pelvis, thorax and thorax on pelvis in sagittal, frontal and transversal planes, joint angular velocities of the ankle, knee, hip, pelvis, thorax and thorax on pelvis in sagittal, frontal and transversal plane, joint powers, moments, work and impulse of ankle, knee, hip and pelvis in sagittal, frontal and transversal plane, time and the rotation foot angle to pelvis. Before performing the stratified shuffle splits, a matrix (memory matrix) was defined for every examined signal consisting of 101 rows (corresponding to length of the time-series) and 100 columns (corresponding to the number of splits) storing “0” values. The first step within every split was the random selection of 50 subjects from the ACL and NORM cohort to create a dataset of 200 trials—containing 50 ACL^OP limbs, ACL^{NO OP} limbs and 2x50 NORM limbs (—step 1). Analysis of characterising phases (described in) was then used to identify phases of variation within the created dataset. For every time point that was within a detected phase of variation, the default “0” value was changed to “1” for the corresponding row and column in the memory matrix. For example, if the described process identified the time points from 5 to 15 within a time series to be a phase of variation in the 7^th split, then the value of row 5 to 15 in column 7 in the memory matrix was changed to “1” (—step 2). After completing the 100 splits, the sum of the values in the memory matrix was computed along the split axis generating a 101x1 vector that stored the accumulated appearance as a phase of variation for each frame. A value of “0” indicates that this time point was never within a phase of variation, while the value 84 indicates that a time point was 84 times within a phase of variation (—step 4). Every collection of time points that occurred in 75% of the splits and spanned at least 5% of the movement cycle was considered robust phase of variation and used for the feature engineering. The investigators used a 75% threshold, instead of 95%, to maintain more information—i.e. allowing more phases to be classified as robust phases of variation. Features describing a phase were calculated as the average value of a phase and stored into a feature matrix that contained *i* rows / features (*i* is the number of detected phases of variation) and *n* = 436 columns / observations (156 operated limbs \[class: ACL^OP\], 156 limb contralateral to the operated limb \[class: ACL^{NO OP}\] and 124 control limbs \[class: NORM\]). For every double leg exercise, symmetry was also calculated as *limb* – *contralateral limb*—e.g. *left* – *right* for left leg trials or *right* – *left* for right leg trials. During the feature engineering the data examined for each exercise generated between 97 and 186 features for each of the 436 observations. The detailed description of results of this step is given in —Detected Phases. These features were used within the following steps as input features during the fitting (training) and validation (testing) of every generated model. ### Selection of supervised learning technique The second step in the model generation was the identification of the most appropriate machine learning technique. This process was done because the selection of the best learning technique can be challenging as there are multiple techniques and each technique has different abilities to learn relationships between the to be predicted classes and input features. This study examined only the ability of techniques included within the statistical toolbox of Matlab (R2015a, MathWorks Inc., USA): a decision tree (fitctree), an ensemble of decision trees (n trees = 50; TreeBagger), a discriminant analysis model (fitcdiscr), a naive Bayesian classifier (fitcnb), a k-nearest neighbour model (k = 5; fitcknn), a multi class model for support vector machines (fitcecoc), a linear regression model (mnrfit; in stepwise forward) and a neural network (patternnet). No grid search was performed at any time and is it very likely that accuracies in better-tuned models will outperform the presented findings. The random assignment of observations into training and testing dataset, during a split, was performed on athlete level to prevent data leakage from training into testing data—e.g. if a subject was selected to belong to the training dataset both the right and left limb were used within the training. Each learning technique was tested separately using raw values and normalised values (z-scores) of the feature matrix. During a split, the feature matrix built during the feature engineering step was split into training, testing and hold out dataset (—step 1). The sample size of the datasets was chosen based on the NORM sample size (training ≈ 70%, testing ≈ 25% and hold out ≈ 5%). Hence, the training dataset contained around 88 trials from each class (≈ 56% of ACL^OP and ACL^{NO OP}) resulting in a training matrix containing i features and ≈ 264 trials. The testing dataset contained around 30 trials from each class (≈ 20% of ACL^OP and ACL^{NO OP}) resulting in matrix containing i features and ≈ 90, while the hold out dataset contained the remaining trials. The training dataset was used to teach each machine learning technique to predict the three classes using the previously extracted features. After the training had been completed, class membership of the testing dataset was predicted and compared to the actual class to assess the performance of each learning technique (—step 3). When using normalised scores, the testing dataset was normalised using the mean and standard deviation of the training data. The purpose of the hold out dataset was increase the “interchange” of trials within splits, increasing the variation across datasets within the splits and facilitated balanced class distributions within the datasets. After the stratified shuffle split cross-validation was completed, the learning technique with the highest mean accuracy across the splits was judged most appropriate and used for subsequent analysis (—step 5). Findings during this step demonstrated small differences between classifications accuracies from raw and z-scores (1%). Based on the normalised scores (z-scores), this process identified that the neural network performed best on the DLDJ features (80%), the logistic regression performed best on the SLCMJ (57%), DLCMJ (72%), HuHo (69%) and SLHop (72%) features, while the discriminant analysis performed best on the SLDJ (60%), CoDP (68%) and CoDU (65%) features. The results of this step are described in and are visualised in detail in —Best Learning Technique. ### Feature selection The third step was the identification of the features that are most important in the classification. This process used only the best performing machine-learning technique and sought to reduce the effect of over-fitted models by identifying the minimal number of features required to capture most of the information within an exercise. To identify the most important features, a wrapper method (evaluate performance of subsets of features in a forward selection matter) in combination with a stratified shuffle split cross-validation with 100 splits was used. Again, no grid search was performed. Before starting the selection of important features, an empty matrix (model base) was defined. After the creating of this model base matrix a stratified shuffle split cross-validation with 100 splits was started assessing the classification accuracy of every feature within the feature matrix on its own. In every split the data were divided into training, testing and hold out datasets (—step 1). The training dataset was used to teach the previously selected machine-learning technique to predict the three classes (ACL^OP, ACL^{NO OP} and NORM). After the training phase had been completed, comparing the predicted class of the testing dataset to the actual class assessed the performance of the generated model. After the 100 splits, the feature with the highest mean accuracy was identified, added to the model base matrix and removed from the feature matrix (generated during the feature engineering). All features that correlated with the identified feature (greater than 0.70) were also removed from the feature matrix, to increase interpretability. The value of.70 was a subjective choice as this magnitude is often considered as strong. Subsequently, a second stratified shuffle split cross-validation with 100 splits was started assessing the classification accuracy of the model base matrix in combination with every remaining feature within the feature matrix. The feature that performed best was added to the model base matrix. The just added and any collinear feature where then removed from the feature matrix (—step 5). This process was repeated until the model base matrix grew to 20 features (this number was chosen to keep computing time to a reasonable duration; —step 6). The last step during the feature selection was determining how many features should be within a model that presents a good trade-off between efficiency and effectiveness (minimal feature model—e.g. the model with the smallest number of features that accounts for a large part of the contained information). To do this the Elbow method was used—described in Hastie and Tibshirani or Vapnik and Vapnik, with the “elbow” being defined as the point n where the differentiation of accuracy f improved less than 10% of its range. $$\begin{array}{r} {f(n + 1) - f(n) < \left\lbrack max(f) - min(f) \right\rbrack*.10,} \\ \end{array}$$ The detailed description of results of this step is given in —Best Features. ### Final validation of minimal feature model The number of parameters and models that were tested in this study is large and information could have leaked during the shuffle spilt cross-validation (e.g. possibly over fitting the presented models). To examine if a model was over fitted a 10 fold cross-validation was used to re-evaluate the minimal feature models. During the cross-validation the minimal feature model was trained using 156 subjects (n = 52 per class; ≈90% of Norm and ≈33% of ACL^OP and ACL^{NO OP}) and validated using 18 subjects (n = 6 per class; ≈9% of Norm and ≈4% of ACL^OP and ACL^{NO OP}) as testing data. Ten folds were chosen to balance the number of subjects in testing and training with an acceptable accuracy resolution (1 in 18 = 5.56%). The mean accuracy of the 10 fold were then compared to the mean accuracy of the previous used shuffle spilt cross-validation. Matlab (R2015a, MathWorks Inc., USA) was used for data processing and analysis. # Results Findings reported are based on normalised scores (z-scores), for two reasons: the difference between classifications accuracies for the raw and z-scores was small at less than 1% at most and z-scores remove any possible magnitude effect during the selection of features within the optimal model. The highest guess observed during the performed splits was 42%. ## Model performance evaluation Generated models were generally judged successful in classifying the classes (ACL^OP, ACL^{NO OP} and NORM). All 3 models generated within an exercise (all features, 20 features, minimal features) were judged meaningful except for the minimal feature for the SLCMJ and CoDU movements. The classification accuracies increased on average 6.1% when reducing the number of features from every extracted feature to 20 features. The classification accuracies of the minimal feature models were comparable to the full models (average difference -4.4%;). When comparing the full model to the minimal feature model: the performance of the DLCMJ, DLDJ, SLHop and HuHo stagnated or improved slightly (0 to 4%), while a slight decrease in performance (-4 to -5%) was observed in the SLCMJ, SLDJ and CoDP. The CoDU minimal feature model decreased its performance by 14% compared to the full model and 15% to the 20 feature model. ## Minimal feature models: Accuracies, errors and sensitivity The minimal feature models used between 2 and 5 features and the classification accuracies observed were (in decreasing order): DLDJ (81%), SLHop (74%), DLCMJ (70%), HuHo (69%), CoDP (63%), SLDJ (57%), SLCMJ (53%) and CoDU (52%). The detailed description of results of this step is given in —Minimal Feature Models. Results for the 10-fold cross-validation of the minimal feature model demonstrated differences in mean accuracy to the shuffle-split cross-validation (in decreasing order): DLCMJ (-11%), CoDU (+11%), SLHop (5%), DLDJ (-4%), CoDP (-3%), HuHo (+3%), SLDJ (0%) and SLCMJ (0%). Classification errors originated from misclassifications of the class ACL^OP and ACL^{NO OP} in the optimal SLHop and HuHo model. In these models the percentage of misclassification within the ACL group was higher (≈ 2 times) than misclassification of either ACL class with NORM, while the SLCMJ model tended to misclassify both ACL classes with NORM class. The DLCMJ and DLDJ results displayed increased misclassifications between NORM and either ACL class compared to between the ACL classes (≈ 1.8 and 4 times;). The SLDJ displayed increased misclassifications (≈ 2 times) between the ACL^{NO OP} and NORM class compared to ACL^OP with ACL^{NO OP} or NORM. In the CoDP and CoDU models, misclassifications were equally distributed between the classes. When assessing the performance of minimal feature models on a group level (ACL \[ACL^OP and ACL^{NO OP}\] and NORM) the single leg exercises demonstrate large improvements (12 to 24%), while the double leg exercises did not (2 to 8%;). The sensitivity for the classification of the NORM class was highest in the SLHop (81%), while for the sensitivity for ACL^OP and ACL^{NO OP} trials was highest for the DLDJ (84 and 85%; see). The percentage of the athletes within the NORM group that were never confused (’true’ NORMs) ranged from 27 to 71% depending on the exercise. The percentage of the athletes within the ACL^{NO OP} and ACL^OP group that were never confused ranged from 52 to 85% depending on the exercise. # Discussion This study examined the classification accuracy of a framework that combines a data-driven feature extraction procedure that did not require expert knowledge and a supervised machine learning technique to differentiate movement patterns from ACL operated class (ACL^OP), from a limb contralateral to an ACL operated limb (ACL^{NO OP}) and a non-injured control limbs (NORM) using only biomechanical data. Findings demonstrate that biomechanical data, regardless of the exercise, contained sufficient information to outperform guessing and suggest that it is possible to differentiate normal movement patterns from movement patterns after an ACL reconstruction on a group and limb level. ## Model evaluation ### Classification accuracy This study developed a minimal feature model in 3 steps that generated three prediction models: one that utilised a large number of features, one model utilised only 20 (most important) features and a minimal feature model. The minimal feature model contained 2 to 5 features and classification accuracies were similar between the three generated models. For the DLCMJ, DLDJ, HuHo and SLHop, using the minimal feature models reduced the information utilised by 94 to 98% (160 features vs. ≈ 4 features) but achieved performances similar (-2 to 2%) to models using all generated features. In contrast, the performance of the minimal feature model decreased for the SLCMJ, SLDJ, CoDP and CoDU in comparisons to the model using all generated features (-5 to -12%). The decrease in accuracy in the minimal feature model SLCMJ, SLDJ, CoDP and CoDU indicates an increased complexity in these exercises, compared to the DLCMJ, DLDJ, HuHo and SLHop, and suggests that these models need more information (features) to perform as well as they could. However, all exercises improved performances when reducing the number of features from 97 to 186 to only 20, which is a reduction of used features of ≈ 84% (138 features vs. 20 features), by ≈ 7%. This suggests that there was an effect of over-fitting to the training data in the models that used all extracted features and highlights that the importance of some kind of feature selection before model building using biomechanical analysis using comparable sample sizes. However, this finding also highlights that a relatively small number of biomechanical features (between 2 and 20) captures a large amount of the information content from a biomechanical assessment. Regarding the information contained within the examined movement data when assessing classification accuracies in a multi-class classifier, classification models achieved accuracies of (in decreasing order): DLDJ (81%), SLHop (75%), DLCMJ (73%), HuHo (69%), CoDP (58%), SLDJ (56%), SLCMJ (53%) and CoDU (52%). Little research exists currently that used biomechanical data to classify group / cohort membership. The closest research is that of Ross et al., who analysed the movements in elite and non-elite performers and found that a data driven model could classify an individual into novice and elite with an classification accuracies between 71 to 80% across 13 exercises. While the classification accuracies of the SLCMJ, SLDJ, CoDP and CoDU seem to be low in comparison to other exercises or the findings of Ross et al., it should be considered that their performance was assessed in an multi-class classifier. The classification accuracy of the SLCMJ, SLDJ, CoDP and CoDU in a binary or group classifier (ACL or NORM) was 67, 69, 77 and 70%, making them comparable to reported classification accuracies observed by Ross et al.. ### Classification errors Classification errors in the tested multi-class classifiers could have occurred due to a reduced / smaller difference in movement within the ACL classes compared to the ACL and NORM classes—e.g. because of anatomical / morphological differences between the groups (ACL and NORM). When assessing the performances on a group level (ACL and NORM) performances improved significantly, compared to the limb class level, for every single leg exercise (≈ 16%) but not for double leg exercises (≈ 5%), suggesting differences between the single and double leg exercises. These differences might be explained by the additional symmetry information that was contained in the double exercises but not in the single leg exercises. This assumption is supported by the fact that in the double leg exercises, symmetry features were selected frequently (4 out of 7). The magnitude of symmetry or asymmetry seems to be a useful feature within the generated models, which supports findings of Myer et al., and others. While symmetry features could have been included within the single leg models, it requires the intervention of the investigators—e.g. symmetry calculation as mean symmetry, symmetry between trial x and trial y and so on. Symmetry was not included in this study because it cannot be calculated without setting subjective rules (e.g. expert knowledge), as every execution presents different external and internal conditions. Nevertheless, findings suggest that symmetry measures are important and hence they should have been considered. Regarding errors that might have been caused by anatomical or morphological differences between the groups, a large part of error in the SLHop and HuHo classification errors originated from confusing ACL^OP and ACL^{NO OP}. In these models the percentage of misclassification within the ACL group was higher (≈ 2 times) than confusing either ACL class with NORM. This suggests that the movement of both limbs in the ACL group is affected by an ACL reconstruction—as both limbs can be differentiated from the NORM pattern but less from each other. As such, the ACL^{NO OP} may not be ideal reference when judging the rehabilitation status and readiness to return to play. Another reason could be that the differences between ACL^OP and ACL^{NO OP} are not well described in the examined features. The SLDJ, DLDJ, SLCMJ, CoDP and CoDU models demonstrated different error or misclassification patterns. The SLDJ would have benefited most from a binary- classifier that classified for ACL reconstructed limb (ACL^OP) and not ACL reconstructed limb (ACL^{NO OP} and NORM), as a large part of misclassifications were made by confusing ACL^{NO OP} and NORM (confusions within the ACL group were less common). This would suggest that the SLDJ is able to detect differences between a limb with ACL reconstruction and limb without ACL reconstruction—implying that the ACL^{NO OP} limb could be used as reference when judging the progress of rehabilitation. The interpretation of the selected features in the minimal feature model, however, suggests that the execution of the exercise was modified for within the ACL classes and between the groups. The vertical CoM velocity at impact was selected yet it should not hold any meaningful information, as it should be nearly the same (impact velocity should theoretically be 1.981 m/sec.) across all trails because of the controlled drop height. Inspection the model visually reveals: a) that the ACL^{NO OP} and NORM class demonstrate similar CoM resultant velocity at take off values, while values in ACL^OP are reduced, and b) that the vertical CoM velocity at impact values are different between ACL^{NO OP} and NORM and spanned over the range of classes for the ACL^OP class. Based on a post hoc analysis, the ACL^{NO OP} and some of the ACL^OP class have ‘changed’ the drop height by adapting a stepping down pattern, which was not detected and resolved during the biomechanical assessment, in spite of careful instruction and frequent lab quality assessments. This highlights is the importance of interrogating what the model has learned during a data driven process as other features should be been normalised to the ‘changed’ the drop height and highlights possible psychological differences between the group. In the DLCMJ and DLDJ, errors originated from misclassifying NORM with either ACL class or both ACL classes with NORM, with the DLDJ demonstrating a larger inequality in distribution (≈ 1.8 vs. 4 times). This might be explained by an increase of the physical demand of the DLDJ that poses an additional challenge to symmetry and reactivity. While the minimal feature SLCMJ model demonstrated classification errors that originated equally from confusion between the classes, the 20 feature or all feature SLCMJ model displayed that confusion between ACL^{NO OP} and NORM were ≈ 2 times higher compared to other classification errors. This suggests that models that utilised more features possibly detected some patterns that differ between ACL operated and not operated limbs (when being provided with information from many features). Similar to the SLCMJ the minimal feature CoDP and CoDU models demonstrated classification errors that originated equally from confusion between classes, while the 20-feature model displayed an increased error from misclassification within the ACL group (≈ 2 times). This would suggest that the ACL^OP and ACL^{NO OP} class share more similarity with each other than to the NORM class. ### Classification sensitivities Another source of error for the models is the misclassification of specific athletes within a class that does not present the ‘true’ class movement patterns. When examining the prediction errors on an athlete level, it was noted that some athletes are always classified correctly; some are occasionally classified correct, while some are always misclassified. This was true for all classes and exercises. The detection of members within a group that are not always classified correctly could help to understand group differences as they could be excluded from subsequent analysis (as done in to better understand the class pattern the model has learned). The existence of such athletes or the proportion of such athletes within a group may also explain why there are so many conflicting findings in inference based studies. However, knowing that such athletes exist also enables the development of our understanding of injuries as the probability of belonging to a desired class computed during a classification could be used when judging movement analysis and injury risk. Consequently, the probability of membership to a class (in this case the NORM class) could also allow the objective judgment of a movement or the rehabilitation progress. The probability of membership can give an objective measure of how close a trial is to a desired class and presents a clear criterion if an athlete has returned to normal. ## Practical implications Return to play after ACL reconstruction as well as the prevention of subsequent re-injury is not always guaranteed and this might be in part due to absence of clear criteria identifying if an athlete has returned to pre-injury levels or completed rehabilitation. Current clinical testing batteries often utilise biomechanics to assess a movement quality. However, there is little consensus on the appropriateness of biomechanical analysis and / or specific exercise tests and measures when differentiating between two specific groups. This study demonstrates that biomechanical data holds enough information to differentiate between ACL^OP, ACL^{NO OP} and NORM with classification accuracies above 70%, even if the ACL group was 9.4 ± 0.7 months post-op and most had completed their rehabilitation. Findings suggest that classification models from different exercises capture different pieces of information and hence a variety of exercises should be used. A biomechanical assessment should contain measures of symmetry and chosen exercises might be altered throughout the rehabilitation process to increase complexity, while familiarising the athlete to the demands of sports. The selection of a single exercise test cannot be recommended as all examined exercises demonstrated that they contain valuable information and we did not adjust parameters within the models that could lead to improvements in classification accuracy. Findings highlight problems with the assumption that the majority of a control group demonstrate healthy movement patterns, which could be a reason for conflicting findings in studies. Within the examined exercises only 27 to 71% of the limbs within NORM did present a true NORM pattern. ## Limitations ### Examined data Like most movement analysis studies, this study involved the recording of multiple trials but examined only the best trial. Examining the best trial is likely to be more valid than averaging multiple trials, which creates and examines an artificial movement where local peaks are altered in magnitude and temporal appearance and the intersegmental link (coordination) between joints might be lost. However, the approach used selects a unique instance (maximal performance) and could bias an analysis towards a non-realistic situation. No athlete will perform a task over and over with a maximal effort and consequently the sub-maximal efforts should not be discarded. An alternative approach is to utilise repeated random sampling, where the captured trials are selected at random and the analysis is run multiple times. This can overcome the ‘maximal effort bias’ and can also provide a measure of expected differences or accuracy within future studies. Further, such an approach can also overcome discrepancies between findings that are caused by the selection of a reference limb when comparing an abnormal (injured) to a normal or uninjured group. ### Examined features The use of waveform analysis and total exclusion of discrete points ignores the magnitudes and location of peak values that might be important. For example, angular velocity and GRF time series can be multimodal (multiple local maxima’s) and maximal or minimal might not be located in the same phases and the information of these features is hence lost. Such features and their temporal properties (position) might carry important information that has been discarded in this study. Further, the manner in which the data driven framework approached the engineering feature was based on no input of expert knowledge (e.g. complexity, stiffness, variability) or any advanced feature engineering techniques (principal component scores, non negative matrix factorisation or interactions between features). No additional engineering feature was used to keep the interpretable and meaningful for rehabilitation, while additional more complex features are likely to increase the reported classification accuracies. ### Model building This study tested machine learning techniques using only their default values and did not perform a grid search (optimise adjustable parameters within a technique; e.g. changing L1 and L2 regularisation in logistic regression or changing k in the k-nearest neighbour technique), which would very likely have resulted in higher classification accuracies and might impact the selected features in the optimal model. Additionally, for the DLCMJ and CoDU findings should be interpreted with care as findings from a 10-fold cross-validation possibly indicate over-fitting for the DLCMJ model and an under-fitting for the CoDU model. # Conclusion This study tested a data-driven framework that required no expert or prior knowledge and demonstrated that biomechanical data can predict with high accuracy (≈ 70%) if a movement was performed by a limb following ACL reconstruction, the contralateral limb and a limb of a healthy control group. Findings highlight that a few features can contain most of the information content, that symmetry measures are important, that it is important to seek to understand what a classification / prediction model has learned, that different exercises capture different movement characteristics and that not all subjects within a normative cohort utilise a ‘true’ normative movement pattern. # Supporting information We are thankful to our colleagues who provided expertise that greatly assisted the data processing or though process of this study. [^1]: The authors have declared that no competing interests exist.

# Introduction Coronavirus disease (COVID-19), caused by the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) virus, rapidly spread throughout the world and was declared a global pandemic as a public health emergency of international concern, which continues to have serious negative effects on health and the economy worldwide. In order to halt the spread of SARS-CoV-2 virus, several vaccines have been developed and implemented, but some of these efforts have been hindered by mutations leading to new virus variants. A recently emerged and rapidly spreading variant of SARS-CoV-2, named Omicron, has triggered worldwide concern, and the World Health Organization declared it a ’variant of concern (VOC)’ on November 26, 2021. VOC is the name given to a SARS-CoV-2 virus variant with mutations affecting the spike protein receptor-binding domain that increases the binding affinity within the RBD-hACE2 complex and increase the viral transmissibility. A total of 30 mutations have been identified on the spike glycoprotein of the Omicron variant, of which 15 are located on its receptor-binding domain, and some of these mutations are also present in other VOCs. As of today, the Omicron variant is the most genetically diverse strain seen in a large number of cases, which may increase the risk of reinfection, greater disease transmission, the severity of the disease and escape from diagnosis while decreasing vaccine effectiveness. A SARS-coV-2 infection begins with the binding of the viral spike protein to the Angiotensin-converting enzyme 2 (ACE2), followed by the proteolytic processing of the trimeric spike protein into S1 and S2 subunits by the serine protease furin. In turn, S2 aids in the fusion of the viral cell membrane with the host, leading to viral entry into the cell cytoplasm via endocytosis, where it blocks a number of antiviral pathways, eventually causing an upsurge in pro- inflammatory cytokine secretion and NF-\*B activation that causes cell death and hyper-inflammation. Therefore, the mutations in the spike protein would have a very large impact on the virulence of SARS-CoV-2, transmissibility, as well as the efficacy of the current vaccines and therapies. Currently, vaccine or neutralization antibody programs are primarily focused on RBD-ACE2 interactions. However, in addition to ACE2, a number of other host molecules are also reportedly involved in SARS-CoV-2 attachment to cells and act as entry factors. Therefore, most, if not all, of the current antibody strategies may not inhibit SARS-CoV-2 virulence or reduce the hyper-immune response. Additionally, higher mass vaccination rate in many countries compared to others is increasing the risk of SARS-CoV-2 mutating into a new strain that might be resistant to current vaccines. Highly infectious variants require greater vaccine penetration to build protective immunity therefore, it is vital to study the effect of the mutations on the immunologic properties for evaluating the current therapies and development of new vaccines to combat the SARS-CoV-2 pandemic. Several studies have reported a higher affinity for the ACE2-receptor in the mutant spike protein of Omicron than in wild-type SARS-CoV-2, as well as a greater ability to evade the immune system, which might result in higher viral transmissibility. To the best of our knowledge, none of the studies examined detailed effects on physicochemical, structural, or immunologic properties of the spike protein due to the large number of mutations found in the spike protein. Therefore, the purpose of our study was to use an in silico approach to comprehensively analyze the spike protein of the Omicron variant with respect to the reference variant to distinguish the differences from physicochemical, structural, immunologic and functional perspectives. Additionally, we have conducted protein-protein docking with the ACE2 receptor in order to investigate the effect of mutations on protein interactions. Finally, we performed atomistic molecular dynamics simulations of the RBD-ACE2 complex of Wild-type SARS-CoV-2 and Omicron variants for a detailed molecular analysis. # Methods and materials ## Data retrieval and annotation SARS-CoV-2 reference spike protein sequence was obtained from UniPort with accession number P0DTC2 and the first complete Omicron genome from GISAID with accession number EPI_ISL_6640916. The Omicron genome sequence was then annotated using the Cov-Seq program, followed by translation using the EMBOSS transeq tool. A pairwise alignment of spike protein sequences was later performed with Clustal Omega, followed by the analysis of mutations. Furthermore, the PDB structure of the RBD-ACE2 complex with accession 6M0J and the PDB structure of the whole spike protein with accession 7N1U were used as reference, downloaded from RCSB PDB database. ## Physicochemical parameter analysis The spike protein of the reference and Omicron variants were subjected to preliminary sequence analyses to distinguish their physicochemical differences. The amino acid composition, molecular weight, distribution of charged residues, hydropathicity, aliphatic index, instability index, and a few other parameters are calculated mainly using the online server EMBOSS Pepstats. Further verification of the results from Pepstats was conducted using ProtParam, Prosite and AA-prop. In addition, another web server called VOLPES was used to compare and visualize residue-level physicochemical properties. ## Structural properties analysis JPred4 was used to predict the secondary structure of spike proteins based on the JNett algorithm, which is one of the most advanced and accurate methods. NetSurfP-2.0 tool was used to evaluate further the prediction, which utilizes convolutional and long short-term memory neural networks to predict secondary structure, solvent accessibility, and residual disorder. Furthermore, the flDPnn server was used to predict intrinsically disordered regions and their functions in conjunction with NetSurfP-2.0. Following that, the PredyFlexy server was used to determine the flexibility of the structures, while consurf and predictprotein were used to predict conserved regions. Finally, AlphaFold2 was used to predict the tertiary structure of the Omicron spike protein. Following the prediction of 3D structure, we calculated the RMSD and TM-score between the predicted Omicron spike protein and our reference spike protein (7N1U) using the TM-align tool and calculated the overlap of common contacts using the CMView program. ## Functional properties analysis In order to determine the effect of mutations on protein stability, Dyanmut2 and DeepDDG were used, where Dyanmut2 used normal mode analysis and graph representations of protein structures, and DeepDDG used neural networks to predict the effect of mutations on protein stability. In addition, several tools, including SNAP2, PROVEAN and SIFT, were used to assess the impact of mutations on function. As a final step, we predicted how mutations would affect the propensity of SARS-CoV-2 to cause disease using the VarSite webserver. Whenever more than one tool was employed to predict the effects, the common outcomes were taken into consideration. ## Immunologic properties analysis First of all, CD8+ T-cell, CD4+ T-cell and B-cell epitopes were predicted in NetMHCpan-4.1, NetMHCIIpan-4.0 and BepiPred-2.0 servers. To analyze the immunologic properties in detail, all the epitopes passing the default threshold value of the servers were considered for downstream analysis. Then, the immunogenicity of the CD8+ T-cell epitopes was predicted using the Class I Immunogenicity analysis tool and immunogenicity for CD4+ T-cell epitopes was checked using the CD4+ T-cell immunogenicity prediction tool of the IEDB database while immunogenicity of B-cell epitopes was predicted with the iBCE-EL server. Antigenicity, pro-inflammatory and anti-inflammatory potentials for all types of epitopes were predicted using VaxiJen v2.0, PIP-EL and AIPpred webservers respectively. ## Molecular docking and protein-protein interaction analysis First, we retrieved the RBD-ACE2 protein complex with accession number 6M0J and separated the chains. Then, the Pymol mutagenesis wizard was used to introduce the specific mutations at the appropriate residues in the receptor-binding domain. After preparing the protein, Cluspro and HDOCK were used to dock the reference and Omicron RBD to the ACE-2 receptor (6M0J, chain A). We then used the PRODIGY webserver to calculate the binding affinity and the PIC server to investigate the interactions between RBD and hACE2. Finally, the Pymol graphical software was utilized for figure generation. ## Molecular dynamics simulation All-atom MD simulation was carried out in GROMACS 2021.2 software package, and ACE2-RBD protein complexes of both reference and Omicron variants were prepared using the CHARMM36 force fields. Each protein complex was then solvated with the TIP3P water model by adding 0.15mM sodium chloride within a dodecahedron box. The distance between the protein complex and the corner of the box was set to 1.2nm. The system energy was minimized with the Steepest Descent algorithm in 50,000 steps, followed by the system was equilibration in two phases. Firstly, 10ps NVT (constant number of particles, volume, and temperature) simulation was performed to equilibrate the temperature at 310.15K guided by V-rescale temperature coupling algorithm, followed by 100ps NPT (constant number of particles, pressure, and temperature) simulation to equilibrate the system at 1atm pressure and 310.15K by using Parrinello-Rahman barostat algorithm. Finally, the MD simulations of both reference and Omicron variant ACE2-RBD systems were run for 200ns with a time step of 2.0fs under NPT ensemble using GROMACS 2021.2 software and long-range electrostatic interactions were computed using Particle Mesh Ewald (PME) algorithm. The cutoff values of the electrostatic and Van der Waals interactions were set to 12 Å while the linear constraint LINCS algorithm was used to constrain all covalent bond lengths, including hydrogen. MD trajectories were analyzed using GROMACS’ integrated tools for computing root-mean-square deviation (RMSD) and difference root-mean- square fluctuation (RMSF). Lastly, we investigated hydrogen bond interactions and their relative frequencies with the VMD package, setting the hydrogen bond distance and angle to 3.0 Å and 20°, respectively, and calculated the binding energy using the Prime 3.0 MM-GBSA module. # Results SARS-CoV-2 Omicron variant has been designated as the variant of concern due to its rapid emergence worldwide, which includes 30 mutations in the Spike protein, and nearly half of them are in the receptor-binding domain. Due to sequence loss, the Omicron variant has 1270 amino acids instead of the reference spike’s 1273 amino acids. It was evident from a primary analysis of the protein sequence that this variant had more Arginine, Histidine, Lysine and Glutamic acid than the reference, indicating that the spike protein is more charged. Furthermore, these residues were exposed to a much greater extent and contribute to binding with receptors because their pKa’s are high enough with polar side chains, which can form hydrogen bonds. On the other hand, Isoleucine and Phenylalanine were also present in higher numbers within the protein’s core, making the spike protein more hydrophobic than the reference variant. These mutations would alter its physicochemical and structural properties, which will affect the transmission rate and pathogenicity within human populations by reducing antibody-mediated protection. ## Effects on physicochemical properties Despite having three fewer amino acids, the molecular weight of the Omicron variant (141328.11 Da) was higher than the reference variant of (141178.47 Da), and the mutations were biased toward the nonpolar amino acids; therefore, the hydrophobicity of the spike protein of the Omicron variant increased. While the number of hydrophobic residues increased, the GRAVY (Grand Average of Hydrophilicity) value indicates the protein has become slightly more hydrophilic intrinsically, which is indicative of the effects of mutations on the surface accessibility of the protein due to alteration of the secondary and tertiary structural properties. Furthermore, shows an increment of both acidic and basic residues in this variant; however, the increase in basic residues is higher, resulting in a net charge of 8, which is likely to facilitate the interaction with hACE2. There was less electronegativity observed among the residues closest to the recognition of the receptor-binding domain of spike protein where the ACE-2 receptor binds. In contrast, a high level of electronegativity was evident among the other residues of the domain and the complete protein of the Omicron variant. Moreover, the Omicron variant’s isoelectric point (pI) is 7.18, meaning the protein was slightly alkaline, whereas the reference variant was acidic in nature with a pI value of 6.62. Then, another important physicochemical parameter is the extinction coefficient, which is the measure of how much light is absorbed by the polypeptides. The extinction coefficient of the Omicron variant was calculated to be 1.036, while 1.037 was that of the reference variant assuming all cysteine residues are reduced. Moreover, we found that both variants of the spike proteins were stable with scores of 33.01 and 34.57 respectively for reference and Omicron, with reference being more stable. Finally, both the reference and Omicron variants had higher aliphatic index values of 84.67 and 84.95, respectively, indicating that both variants are thermostable over a wide temperature range. ## Effect on structural properties Mutations in the spike protein were predicted to affect its structural properties. First of all, according to the secondary structural analysis, this variant had a higher fraction of alpha-helix (23.46%) than the reference (21.52%), while the beta-strand structure was decreased. The T470-Q474 residues of the receptor-binding domain transitioning into the alpha-helix structure would increase the RBD’s stability, making the variant more transmissible and pathogenic since hACE2 interacts with the T470-F490 loop. Overall, ten residues of beta-strands and coils were predicted to be transformed into alpha-helix, but the opposite was not observed. There were, however, fourteen beta-strand residues predicted to be transformed into random coils, while seven random coil residues may be transformed into beta-strand residues. Then, the mutations influenced the solvent accessibility of 154 residues and made the variant more hydrophilic because a higher number of residues were exposed. Among 154 residues, 61 were exposed from the buried or intermediate state, while 54 were buried. Additionally, mutations in the spike protein changed the residual flexibility and increased the rigidity of the protein, which would affect its functionality. In the reference variant, 361 residues were predicted to be flexible, but the number decreased to 353 while rigid residues increased from 304 to 311. While some flexible residues gained an intermediate state without becoming rigid, very few rigid residues developed a flexible state directly, and all rigid transitions occurred among residues of the intermediate state. In this variant, flexibility predictions showed that the transmembrane domain and heptapeptide repeats (1213–1237, 912–984 and 1163–1213 residues) of the S2 subunit are highly flexible, which could affect the viral cell fusion with host cells. Despite the mutations affecting the protein’s residual flexibility, we did not observe any significant changes in the residual disorder of the protein implying several algorithms. Finally, the Omicron variant of the spike protein has 703 conserved residues for structure and function, compared to 675 residues in the reference variant. It was found that the structural residues of this variant increased from 354 to 394, which would likely increase the stability of the protein. There was, however, a decrease from 321 to 309 functional residues, which might have an impact on the viral fatality. In addition, both protein structure and conformation dynamics are associated with biological functions, so we analyzed the reference and the Omicron variant’s tertiary structure further to find the structural variations. We observed an RMSD of 0.20 and a TM-score of 0.99780 between the spike proteins, indicating a higher degree of structural similarity. In contrast, a contact map overlapping analysis yielded 90.5% common contacts, with the reference variant having 89 unique contacts among the residues, whereas the Omicron variant had 438. The contact map analysis indicated some differences among the functional residues of the protein despite the RMSD value and TM-score indicating no major structural changes caused by the mutations. The superimposition of the receptor-binding domain structures yielded an RMSD value of 1.07, indicating higher structural differences in atomic coordinates than the reference variant; however, the TM-score of 0.96355 suggests no major structural differences occurred. A total of 91.3% of residue contacts were common between the reference and Omicron variant, with 50 and 34 unique contacts respectively, which designated differences in the functional residues of the RBD that may affect viral transmissibility. ## Effect on stability, functionality and disease propensity Using a combination of deep learning neural network algorithms and structure- based predictions, the effect of the mutations on the spike protein stability was predicted. There was a decrease in structural stability for all amino acid changes, but the S142H, N764K and P681H are predicted to have a significant impact. The functional effect analysis revealed that the E484A, Y505H, T547K, N764K, N856K, and N969K mutations impair the spike protein’s function, and the rest are neutral. One of the five mutations, E484A, is located in the receptor- binding domain, so this mutation would likely influence viral transmission. The rest of the mutations in the RBD were predicted not to affect protein function but to reduce its structural stability, which could affect it either way. Additionally, sixteen mutations were predicted to increase disease propensity and thirteen to decrease it. Together with the other twelve mutations, it was predicted that the E484A mutation would decrease the probability of diseases induced by the protein, which was predicted to affect the protein function. On the other hand, the other four protein function impairing mutations are predicted to increase the likelihood of disease. ## Effects on immunologic properties A series of analyses using conventional and neural network-based tools revealed changes in antigenicity and immunogenicity of the spike protein, but no significant changes in both pro-inflammatory and anti-inflammatory properties. In general, we found that the overall antigenicity and immunogenicity of the spike protein was increased, however in-depth studies were conducted on each prospective epitope to get better ideas. To begin with, 29 of the 133 epitopes predicted in the reference spike proteins against B-cells found to be altered in the strain Omicron. In addition, five new epitopes were observed, while six were lacking due to the higher number of mutations compared to the reference strain. Meanwhile, eight mutations were associated with an increase in antigenicity where nine mutations resulted in a significant loss of it. Using S375F as an example, the antigenicity score of the SVLYNSASFSTFKCYG epitope increased from 0.1864 to 0.8321 while T547K decreased the antigenicity score of TGTGVLTESNKKFLPF from 0.9925 to 0.5818. On the other hand, S375F reduced immunogenicity by 25% but T547K did not have any effect on the respective epitopes while, S375F increased pro-inflammatory activity by 3%, and T547K decreased it by 4%. In the case of S375F, the anti-inflammatory properties of the epitope would be increased by 10%, where, T547K was predicted to have no effect. Overall, compared to mutations altering antigenicity properties, there were relatively few mutations that could affect immunogenicity, pro-inflammatory or anti-inflammatory properties of the B-cell epitopes. When it came to epitopes for CD4+ T-cells, 49 of 253 were affected by the mutations where majority of them were predicted to increase the antigenicity of epitopes, while few were predicted to have negative impacts. The increase in antigenicity, however, often had the adverse effect of decreasing immunogenicity, anti- inflammatory and pro-inflammatory properties of the epitopes. As an example, the S371L mutations increased the antigenicity of the epitope YSVLYNSASFSTFKC by 69% while decreasing the immunogenicity and anti-inflammatory potential of the epitope by 74% and 7% respectively relative to the reference spike protein. While the S371L, S373P, and S375F mutations have previously been found to cause immune evasion, they were predicted to decrease immunogenicity by 117%. However, unlike B-cell epitopes, the immunogenicity of several CD4 T-cell epitopes were predicted to be highly increased by the mutations but high level of fluctuation was not observed in inflammatory properties. For instance, T547K and G339D both increased the immunogenicity of their respective epitopes by 165% and 127%, but the pro-inflammatory potential decreased by only 2% and 3% while the anti- inflammatory potential declined by 4% and 7% respectively. The utmost pro-inflammation enhancing mutation for any epitope was D796Y, resulting in an increase of only 6% in epitope PIKDFGGFNFSQILP over the reference variant while N764K increased the maximum anti-inflammatory potential by 4% in the epitope LLLQYGSFCTQLNRA. Finally, only 30 of the 259 epitopes targeting CD8+ T-cells exhibited altered antigenic, immunogenic, inflammatory properties due to mutations. Sixteen of these epitopes showed an increase in antigenicity and thirteen suggested a reduction. Two mutations S373P and S375F in the epitope NSASFSTFK enhanced immunogenicity, antigenicity and pro- inflammatory properties by 42%, 121%, and 11% respectively while the anti- inflammatory potential was anticipated to be reduced by 11%. In addition to enhancing antigenicity and immunogenicity, these mutations improved the pro- inflammatory properties of the epitope, a phenomenon that was not seen in other epitopes for any mutation therefore, this epitope needs further study. As a whole, a smaller number of mutations were associated with increasing the immunogenicity and inflammatory properties than antigenicity against CD8 T-cell epitopes. Additionally, eight of the perspective epitopes predicted from the reference sequence lacked in the Omicron variant while, nine new ones emerged. Our analysis showed that the missing epitopes in Omicron had significantly higher immunogenicity, antigenicity, and inflammatory properties than the newly appeared epitopes. The contains details of epitopes and the effects of mutations on them. ## Effect on binding interaction with hACE2 Infectivity, transmission, and pathogenesis are largely determined by the binding affinity of the virus towards the receptor, so mutations in the receptor-binding domain of the virus can greatly influence its activities. Since SARS-CoV-2 interacts with hACE2 through its C-terminal domain, mutations at key residues would affect the interaction with hACE2. In this part of the study, we found that the binding affinity of the Omicron variant differs from the reference variant. Our results from the HDOCK server showed that the docking score and binding affinity for the Omicron variant were -343.56 and -11.8 kcal/mol, compared to -310.19 and -11.5 kcal/mol for the reference variant. Moreover, the ClusPro server provided us with the docking score and binding affinity for Omicron variants -703.8 and -14.7, respectively, compared to -639.3 and -11.9 for the reference variant. Thus, it was evident that the Omicron variant exhibits a stronger binding affinity to the hACE2 than the reference variant, implying a potential for higher viral transmission. There are significantly more interactions with hACE2 than in the reference, with 48 versus 25 interfacial contacts in the Omicron variant, most of which are hydrogen bonds. Our study of protein-protein interaction in the reference variant found four hydrophobic interactions involving Y489 and F486 residues. These interactions were not present in the Omicron variant; however, six new hydrophobic bonds involving Y446, K452, I469, A481 and F487 were discovered. Pi- Cation interactions were observed at three new sites with Y446, Y486 and Y498 residues in the Omicron variant, but there were no Pi-Cation interactions in the reference variant, while the only aromatic bond of the reference was absent in Omicron. Additionally, the number of ionic interactions increased from three to seven when the RBDs of Omicron spikes bind with hACE2. However, the maximum discrepancy was observed regarding hydrogen bond formation, where the only hydrogen bond between the main chains of the reference RBD and hACE2 was absent from the Omicron variant. Meanwhile, the number of bonds formed between the main chain and side chain increased dramatically from 5 to 14, and the number of bonds formed between side chains rose from 11 to 18. Six of the fifteen mutations found in the RBD of the spike protein in the Omicron variant directly affected its interaction with hACE2. For instance, E484 was in ionic contact with K31 in the reference structure, but the Omicron variant (E484A) lost this ionic interaction; instead, A484 formed two hydrophobic interactions with L79 and M82 residues of hCAE2. The most drastic effect on the interactions took place due to the Q498R mutation, which is located near the hACE2 recognition loop structure of RBD. There was a hydrogen bond present between the Q498 of the reference RBD and the Q42 residue of hACE2; however, the Q489R mutation impaired that bond and led to the formation of several new hydrogen bonds among the residues of the main chains and side chains, as well as ion interactions with E208 and D216 residues of hACE2. In addition, the N501Y mutations, which were previously reported for the alpha variant, also increased the binding affinity for the Omicron variant because two hydrogen bonds and one Pi-Cation bond were formed instead of one hydrogen bond in the reference variant. In the supplementary file, we provide a full comparison of the protein-protein interactions of Omicron RBD and reference RBD with hACE2. To summarize, it is evident that mutations increased the binding affinity between the receptor-binding domain and hACE2, which is a key factor in the transmissibility and pathogenesis of SARS-CoV-2. ## MD simulation result on RBD-ACE2 complex Using molecular dynamics simulation, we observed new interactions formed with hACE2 while retaining many of the old ones found in the reference sequence. These interactions were attributed to additional residues at the interface (Q474, G476, N477, K478, A484, F486, and Y489). Despite this, we found several interaction disruptions; for instance, the salt bridge between E484 and K31 was lost, the hydrogen bond between K417 and D30 was damaged, and other interactions were disrupted at residues F456, A475, and G502. According to our analysis of 15 mutations that have been reported to occur in the RBD of the Omicron variant, 6 of these mutations were found to have additional effects on the binding of the Omicron RBD with hACE2. Furthermore, on average, we found 6.89±1.28 hydrogen bonds between Omicron RBD and hACE2, as compared to 5.52±1.26 hydrogen bonds with the reference variant. This indicates that the Omicron variant enhances the interaction between RBD and hACE2 compared to the reference variant. Interestingly, mutations of Q498R and N501Y were found to form new hydrogen bonds with occupancy higher than 15%. In contrast, mutations of K417N and Y505H led to the loss of hydrogen bonds, and some mutations did not affect hydrogen bonding, for example, the K493-E35 bond resulting from the Q493K mutation. After that, we calculated the binding energies of the hACE2 complex in comparison to either the RBD of Omicron and the reference, and it was found that the binding energy of hACE2 is lower when binding to Omicron RBD (-98.02 kcal/mol) as opposed to the reference (-83.7 kcal/mol). In addition, the root-mean-square-deviation (RMSD) for the reference ACE2-RBD complex was 2.61 Å. In comparison, the value was 2.05 Å for the Omicron RBD-ACE2 complex; a finding that indicates the mutations did not lead to significantly reduced stability of RBD but instead increased the stability of the Omicron RBD- ACE2 complex over the reference variant. Additionally, we calculated the root- mean-square fluctuations (RMSF) from the trajectory data and found that the RBD of the Omicron variant is more rigid than the reference variant and that the RMSF values averaged to 1.7 Å and 2.2 Å, respectively. Interestingly, the mutated residues at the interface of the hACE2 showed reduced fluctuations compared to the reference variant at the interface. Thus, when taking into account the number of hydrogen bonds, binding energies, RMSD, RMSF, and RMSD, Omicron RBD-ACE2 appears to be more stable than the reference complex. # Discussion The World Health Organization deems Micron a variant of concern due to its rapid transmission rate and the unusually large number of amino acid mutations in its spike protein. In this study, we investigated how the mutations affect the spike protein from physicochemical, structural, functional and immunologic standpoints and how they modulate its interactions with the host protein ACE2. To begin with, despite a decline in overall hydrophobicity resulting from changes in surface accessibility of several residues due to mutations, the hydrophobic residues are increased in number in the spike protein of the Omicron variant, which is required for the stability of the protein and make the protein’s core. In addition, changes in the hydrophobicity of amino acids may alter the structure of the epitope in the receptor-binding domain of the spike protein, which is likely to affect the immune response to the virus. In Omicron, the number of positively charged amino acids in the spike protein has increased by nine in comparison with the reference variant. Especially, a number of changes occurred in the S1/S2 domain where a basic amino acid replaced a negatively charged amino acid, indicating a selective advantage to bind to host cells. Moreover, the electrostatic potential analysis indicated that L452R, T478K, Q493R, Q498R, and Y501H mutations increased the number of cationic residues near the receptor-binding motif, which may strongly interact with negatively charged residues in ACE2. In addition, the structure of the spike protein of the Omicron variant consists of a higher percentage of alpha-helix structures, which are known for increasing conformational stability of proteins; therefore, it is expected that the stability of the spike protein of the Omicron variant will also be increased. We found that T470-Q474 residues of the receptor-binding loop might undergo a coil to alpha-helix transition, which might facilitate the formation of strong binding with hACE2 and enhance transmissibility. However, no significant change was detected in the residual disorder property, despite a reduction in overall residual flexibility. Possibly, the reduction in flexibility will affect its stability and activity, as well as its ability to be recognized by current vaccines and therapies. On the other hand, the transmembrane domain of the S2 subunit of the spike protein was predicted to be more flexible, which may facilitate viral fusion with host cells. Then, the Omicron variant is predicted to contain fewer conserved functional residues. Contact map overlapping analysis also supported the prediction since only 90.5% of common contacts with reference protein was observed, suggesting differences among the functional residues. The receptor-binding domain of the proteins, however, showed 91.3% common contacts with an RMSD value of 1.07, and a TM-score of 0.96355, while the respective values for the whole protein were 0.2 and 0.99780. Based on these results, it is evident that the greater part of the changes took place in the RBD region, which can also be explained by the presence of half the mutations there. The functional RBD domain is essential for recognition of and binding to ACE2, so any mutation in the RBD domain could influence its function as well as its affinity for binding to the ACE2 receptor. Interestingly, we observed that three mutations in the RBD region can affect its function; fourteen mutations can exacerbate the severity of the disease; and all mutations will decrease its stability, which will, in turn, affect the transmission and virulence potential of SARS-CoV-2. Furthermore, the spike glycoprotein is exposed antigen of SARS-CoV-2, which triggers humoral as well as cell-mediated immunity therefore, most of the current vaccines and therapies target the spike RBD-ACE2 interaction. However, SARS-CoV-2 can, like other viruses, develop immune evasion strategies through antigenic variations in order to escape these immune responses. There have previously been several studies showing that SARS-CoV-2 variants can escape the immune system because of key mutations. Consequently, identifying and analyzing the effects of mutation on immunologic properties is critical for a better understanding of pathogenesis and transmission. Based on our extensive in silico analysis of large sets of epitopes, we found that most of the previously reported mutations, which helped in immune escape, would reduce antigenicity and immunogenicity. As an example, K417N mutations cause immune escape against class I antibodies, which we found to decrease humoral immunity by 7%. The mutation E484A had been reported to confer resistance to multiple antibodies, and our prediction suggested that it could reduce the antigenic and immunogenic potential of the corresponding B-cell epitope by 6% and 5% respectively. In addition to humoral immunity, several mutations have been predicted to decrease the immunogenic potential of T-cell epitopes. For example, S371L, S373P, and S375F mutations have previously been reported to cause immune evasion, and our analysis showed that they decreased immunogenicity by 117% of a + CD+ T-cell epitope. On the other hand, the mutation K417N might decrease the immunogenic and antigenic potential of an epitope predicted to produce CD8+ T-cell mediated immunity by 20% and 18%, respectively. Despite the fact that some mutations might lower the immunogenicity of the epitopes, our analyses illustrated that most mutations tended to increase immunogenic potentials. In addition, we observed that the mutations had no substantial impact on the pro-inflammatory or anti-inflammatory properties of the epitopes. In light of this, we assume that the combination of higher immunogenicity and antigenicity as well as lower inflammatory potentials of the epitopes account for the lower disease severity. Additionally, several mutations in the Omicron have previously been reported in other VOCs, and all of these mutations correlated with immune evasion, increased transmissibility, and stronger binding to hACE2. Therefore, this variant is also expected to exhibit higher transmissibility and increased immune evasion. The extent of viral transmissibility is largely determined by its ability to bind to its receptor, and we found that the Omicron variant binds to hACE2 more strongly than the reference strain. T478K and N501Y mutations have been reported to enhance the interaction with hACE2 in the Delta variant, which are also present in this variant. The N501, Q493, Q498 residues are crucial in RBD-hACE2 interactions since they form polar contacts with the ACE2 residues K31 and K353. Q498R mutation led to a dramatic change in the protein-protein interactions, forming several hydrogen bonds with V209, D216, E208 and Q210 residues, and ion interactions with E208 and D216 residues of hACE2. Moreover, the Omicron variant carries the D614G and P681H mutations, previously described as critical mutations that enhance the transmission and infectivity of SARS-CoV-2. Overall, the number of interfacial contacts in the Omicron variant is significantly increased and the majority are hydrogen bonds, which explains the increased binding affinity. Lastly, the molecular dynamics simulations revealed that residues Q474, G476, N477, K478, A484, F486, and Y489 at the interface enhanced the interactions by forming salt bridges and hydrogen bonds while maintaining the previous interactions. Moreover, on average, we observed 6.89±1.28 and 5.52±1.26 hydrogen bonds with binding energies of -98.02 kcal/mol and -83.7 kcal/mol for the Omicron and reference variants, respectively when RBD was bound with hACE2. An increase in interactions due to the mutations of the receptor-binding domain of spike protein induced a higher binding affinity with hACE2 and stabilized the RBD-hACE2 complex. Moreover, the RMSD and RMSF values for the hACE2-RBD complex of the Omicron variant were 2.05 Å and 1.7 Å, while, 2.61 Å and 2.2 Å were for the reference variant, signifying the higher integrity of the complex. Finally, we observed that when Omicron spike protein was bound to hACE2 there were fewer fluctuations among the atoms than when reference spike protein was bound. Hence, the Omicron spike protein exhibits better binding affinity and greater structural stability to ACE2-receptor than the reference protein. # Conclusion In light of the large number of mutations observed in the spike protein of the SARS-CoV-2, Omicron, a new variant of the virus poses serious concerns. Therefore, in the following work, we used *in silico* computational methods to study the dynamic changes in spike protein of the SARS-CoV-2 caused by the large number of mutations. The results of our analysis revealed significant differences with respect to physicochemical, structural, functional, and immunologic changes, the enhanced binding affinity of RBD with hACE2, and a lower residual fluctuation, which might facilitate the greater transmission. The case reports about the rapid spread of this variant in different parts of the world are consistent with our observations. Therefore, this variant should be subjected to more comprehensive research. # Supporting information We would like to express our appreciation to the scientists, researchers, and all health workers around the world for their valuable contribution to combating this pandemic. [^1]: The authors have declared that no competing interests exist.

# Introduction Particulate matter (PM) refers to a heterogenous mixture of particles and droplets in the ambient atmosphere. These may include organic and inorganic particles, such as tobacco smoke, metals, dust, and pollen. Particles of 10 μm (PM₁₀) or less (PM_2.5), when inhaled into the lungs, can cause detrimental health effects. In particular, PM is significantly associated with cardiovascular and respiratory diseases. There is growing evidence that air pollution is significantly associated with skin aging and inflammatory skin diseases such as atopic dermatitis, acne, and psoriasis. The mechanisms by which PM may affect human skin involve skin barrier disruption and oxidative stress by reactive oxygen species, leading to the activation of inflammatory cascades. Acne is a common inflammatory disease involving pilosebaceous units and results from increased sebum production, follicular keratinization, inflammation, and colonization by *Cutibacterium acnes* (*C*. *acnes*). Although there is a lack of data, the detrimental effects of air pollutants on the skin and the worsening of acne severity have been recognized by most skin experts. Several studies have investigated the effects of pollution on acne vulgaris; however, it has not been proven whether PM aggravates acne inflammation. Therefore, we aimed to investigate whether PM exposure exacerbates acne-like inflammation and elucidate the underlying mechanisms. # Materials and methods ## Chemicals and preparation The standard reference materials (SRM) 1649b were purchased from the National Institute of Standards and Technology (Gaithersburg, MD, USA) and dispersed in phosphate-buffered saline (PBS). *C*. *acnes* (KCTC 3314) was obtained from the Korean Collection for Type Cultures (KCTC, Daejeon, Korea), and peptidoglycan (PGN) from a *Staphylococcus aureus* cell wall component was purchased from Sigma-Aldrich (MO, USA). Specific antibodies against TLR4 and glyceraldehyde 3-phosphate dehydrogenase (GAPDH) were purchased from Santa Cruz Biotechnology (Santa Cruz, CA, USA). Antibodies against phospho-NF-κB p65, NF-κB p65, and cyclooxygenase (COX)-2 were purchased from Cell Signaling Technology (Danvers, MA, USA) and Abcam (Cambridge, UK). Human interleukin (IL)-6, IL-8, and tumor necrosis factor (TNF)-α enzyme-linked immunosorbent assay (ELISA) development kits were purchased from R&D Systems (Minneapolis, MN, USA). ## Cell culture Primary neonatal human epidermal keratinocytes (HEKn, Cascade, Invitrogen, Carlsbad, CA, USA) were cultured in EpiLife medium (Cascade Biologics, Portland, OR, USA) supplemented with human keratinocyte growth supplement (HKGS, Cascade Biologics) at 37°C in a humidified atmosphere with 5% CO₂. When the cultures reached 80% confluence, cells were dissociated into single cells using TrypLE Select Enzyme (Gibco, Waltham, USA) for 5 min at room temperature. The HEKn cells used in the experiments were between passages three and five. ## Bacterial culture Reinforced clostridial liquid and solid medium (RCM, Difco Laboratories, Detroit, MI, USA) were used to grow *C*. *acnes* for 48–72 h at 37°C under anaerobic conditions (5% H₂, 5% CO₂, and 90% N₂). *C*. *acnes* suspensions were centrifuged at 4,000 rpm for 10 min at 4°C and then washed three times with cold PBS. Finally, the cell number was estimated by measuring the optical density of the suspension at 600 nm using a spectrophotometer. As we previously observed that an OD₆₀₀ = 1.0 is equivalent to 5.0 × 10⁸ colony forming units (CFU) per 1 mL, the number of bacterial cells was adjusted with PBS to 5×10⁸ CFU/mL. To obtain heat-killed bacteria, a *C*. *acnes* suspension (5 × 10⁸ CFU/mL) was heated at 80°C for 30 min. Heat-killed *C*. *acnes* were maintained at 4°C and centrifuged at 13,200 rpm for 10 min before use. ## Cell viability assay Cell viability was assessed using a modified 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyl-2H-tetrazolium bromide (MTT, M5655, Sigma) assay. The cells were seeded at 70% confluence in each well of a 96-well plate. After 24 h, the cells were treated with several concentrations of heat- killed *C*. *acnes* (50, 100, 500, and 1000 Multiplicity of infection (MOI)) and PM (5, 10, 25, and 50 μg/cm²). After incubation for 3, 6, 12, and 24 h, 10 μL of MTT solution (5 mg/mL in PBS) was added to each well. Cells were incubated for a further 3 h at 37°C. Supernatants were removed and 100 μL of dimethyl sulfoxide (DMSO) solution was added to each well. Finally, spectrophotometric absorbance was measured at 570 nm. ## Ribonucleic acid (RNA) isolation and real-time quantitative polymerase chain reaction (qRT-PCR) Total ribonucleic acid was extracted from HEKn cells using the TRIzol reagent (Invitrogen) according to the manufacturer’s instructions. Next, cDNA was synthesized from total RNA using the RevertAid First Strand cDNA Synthesis Kit (Thermo Scientific, Waltham, USA). Moreover, qRT-PCR assays were performed with a real-time thermal cycler (Applied Biosystems, Foster City, CA, USA) using PowerUp SYBR Green Master Mix (Applied Biosystems). The PCR data were normalized according to GAPDH levels, and relative quantitation was performed using the comparative 2-ΔΔCt method. ## Western blot analysis Cell lysates were centrifuged, and protein concentration was determined using the bicinchoninic acid assay (BCA) method. Equal concentrations of protein were loaded for 8–10% sodium dodecyl sulfate-polyacrylamide gel electrophoresis and transferred to nitrocellulose membranes. After blocking with 5% skimmed milk for 1 h, the membrane was incubated with primary antibodies (24 h) and anti-rabbit horseradish peroxidase-conjugated antibodies (1 h). The protein expression of NF-κB, TLR2, and COX2 was detected using the EzWestLumi plus system (ATTO, Tokyo, Japan) and ChemiDoc™ XRS image analyzer (Bio-Rad, Hercules, CA, USA). ## ELISA After exposure to heat-killed *C*. *acnes* or PGN and PM 1649b for 24 h, the supernatants were collected and stored at -70°C until assays were performed. The protein expression levels of IL-6, IL-8, and TNF-α in HEKn cells were measured using specific ELISA kits, according to the manufacturer’s recommended protocols. ## Statistical analysis We compared the resulting PM effects in the treatment groups and controls using one-way analysis of variance and Tukey’s multiple-comparison post hoc test. Differences between groups were considered significant at *P* \< 0.05, and statistical analyses were performed using GraphPad Prism 8.0 (GraphPad Software, Inc., CA, USA). # Results ## Effect of PM and *C*. *acnes* on the viability of HEKn cells To choose a suitable concentration and time point for HEKn cell treatment, we investigated the cytotoxicity of several concentrations of *C*. *acnes* (50, 100, 500, and 1000 MOI) and PM (5, 10, 25, and 50 μg/cm²) and time points of 3, 6, 12, and 24 h. The data from the MTT assay showed no significant change in the viability of HEKn cells treated with *C*. *acnes*. In contrast, significant cytotoxicity was observed after PM treatment in a dose- and time- dependent manner. As PM-treated cells exhibited \< 70% cell viability at concentrations above 25 μg/cm² for 24 h, the cells were treated with PM at a concentration of 10 μg/cm² in further experiments. ## Proinflammatory cytokine expression induced by *C*. *acnes*, PGN, and PM Proinflammatory cytokines such as IL-1β, IL-6, IL-8, and TNF-α are responsible for the follicular keratinization and inflammation of acne. Quantitative RT-PCR and ELISA analyses were performed to investigate the effect of *C*. *acnes*, PGN, or PM on the expression of proinflammatory cytokines. HEKn cells were treated with several concentrations of heat-killed *C*. *acnes* (100, 500, 1000 MOI), PGN (1, 10, and 25 μg/mL), and PM (5, 10, 25 μg/cm²). The relative mRNA levels of IL-1α, IL-1β, IL-6, IL-8, and TNF-α increased in a dose- dependent manner after treatment with *C*. *acnes*, PGN, and PM. The results of the ELISA assay also confirmed that treatment with *C*. *acnes*, PGN, and PM significantly increased the expression of IL-6, IL-8, and TNF-α. Since IL-8 has a role in the pathogenesis of acne, we selected the concentration of *C*. *acnes* of 500 MOI for further experiments based on our ELISA data. ## Effect of PM on proinflammatory cytokine expression in *C*. *acnes* (500 MOI) or PGN (10 μg/mL) treated HEKn cells We examined whether PM could upregulate the expression of proinflammatory cytokines (IL-1α, IL-1β, IL-6, IL-8, and TNF-α) in *C*. *acnes-* or PGN-treated HEKn cells. HEKn cells were seeded in six-well plates, allowed to proliferate until 70% confluence, and subsequently subjected to HKGS starvation for 4 h in EpiLife medium. HEKn cells were co-treated with PM (10 μg/cm²) and heat-killed *C*. *acnes* (500 MOI) or PGN (10 μg/mL) at 37°C. After 3 h (mRNA) and 24 h (protein), the supernatants were collected, and the mRNA and protein expression levels of proinflammatory cytokines were determined using qRT-PCR and ELISA. The data revealed that PM further increased the expression of proinflammatory cytokines, such as IL-1α, IL-1β, IL-6, IL-8, and TNF-α, in HEKn cells treated with *C*. *acnes* or PGN. This suggests that PM amplifies *C*. *acnes-* and PGN-induced inflammation. ## Effect of PM on the COX2 expression in *C*. *acnes* (500 MOI) or PGN (10 μg/mL) treated HEKn cells COX2 is an enzyme associated with inflammation and control of cell growth. In particular, several studies have demonstrated that COX2 expression is selectively upregulated in inflammatory acne lesions. Therefore, we investigated whether COX2 expression in HEKn cells co-treated with *C*. *acnes* and PM increased compared to cells treated with *C*. *acnes* alone. Our results showed that *C*. *acnes* and PGN induced COX2 expression in HEKn cells. In addition, PM treatment significantly increased the mRNA and protein expression levels of COX2 . ## PM intensified the inflammation induced by *C*. *acnes* and PGN through activation of the NF-κB signaling pathway The nuclear factor kappa-B (NF-κB) pathway, which is activated in response to various stimuli, has been considered a key transcriptional modulator of inflammatory processes. After activation, NF-κB penetrates the nucleus and upregulates COX2 gene expression. Recent studies have reported that *C*. *acnes* may trigger inflammation through activation of toll-like receptors (TLR2 and TLR4) in keratinocytes, leading to the activation of specific signaling cascades, including the NF-κB and mitogen-activated protein kinase pathways, resulting in the induction of immune response genes. Thus, to elucidate the possible effects of PM on *C*. *acnes*-induced inflammatory signaling pathways, we evaluated the effects of PM on the expression of TLR/NF-κB pathway signaling molecules. The phosphorylation of NF-κB and TLR4 was increased by *C*. *acnes* and PGN. As shown in, PM upregulated the phosphorylation of NF-κB1, NF-κB2, and TLR4 expression in HEKn cells. These data suggest that PM amplifies the inflammatory reaction by increasing the activity of phospho-NF-κB in HEKn cells induced by *C*. *acnes* or PGN. # Discussion Inflammation has been suggested as a key factor involved in the development and aggravation of acne vulgaris. For example, IL-1 can trigger remodeling of the pilosebaceous units and promote comedogenesis by follicular hyperkeratinization. In addition, prostaglandin E2, which is mediated by COX2, has been shown to cause sebaceous gland hyperplasia and sebum overproduction. Several studies have pointed out that PM can negatively affect the skin barrier by inducing inflammation and oxidative stress. Since little is known about the effects of PM on acne, we aimed to investigate whether PM aggravates acne inflammation by assessing the proinflammatory cytokine expression in human epidermal keratinocytes. Previous reports demonstrated that PGN from *S*. *aureus* can activate the NF-κB pathway and induce inflammatory cytokine production via TLR activation. Thus, we used PGN to mimic *C*. *acnes*-induced cellular responsiveness *in vitro* and observe expression of inflammatory cytokines. Our results showed that PM increased the production of proinflammatory cytokines (IL-1α, IL-1β, IL-6, IL-8, and TNF-α), and of COX2, and promoted the phosphorylation of NF-κB induced by *C*. *acnes* in HEKn cells. The group treated with both *C*. *acnes* and PM showed increased expression of phospho-NF- κB and TLR4 compared to the groups treated only with *C*. *acnes* or PGN. Taken together, these results demonstrate that PM has proinflammatory properties, with an emphasis on the activation of the TLR/NF-κB pathway, thus aggravating the *C*. *acnes*-induced inflammatory response. It has been proposed that TLR2 may play a pivotal role in the induction of *C*. *acnes*-induced inflammation, including cytokine production. The expression of IL-6 and TNF-α is significantly inhibited by a TLR2 inhibitor in human keratinocytes or TLR2-deficient murine keratinocytes. In addition, TLR2 and TLR4 have been reported to be involved in the inflammation caused by *C*. *acnes*. Nagy et al. showed that *C*. *acnes* increased the hBD2 and IL-8 mRNA levels in keratinocytes and that *C*. *acnes*-induced upregulation in gene expression is dependent on both TLR2 and TLR4. In our study, *C*. *acnes* significantly induced the expression of TLR4 in keratinocytes, which was further increased by PM. Our findings reveal a potent role of TLR4 in *C*. *acnes*-induced inflammation and suggest that *C*. *acnes* mediates different regulatory factors in response to different TLR signaling pathways in keratinocytes. However, this study has several limitations. While it includes several key factors in the pathogenesis of acne, further *in vitro* research using sebocytes is required, and additional factors are needed to mimic the *in vivo* conditions of acne. Although *C*. *acnes* is the most well-known pathogen involved in the pathogenesis of acne, recent studies suggest that other bacteria can directly and indirectly contribute to the inflammatory process in acne. For example, *Staphylococcus epidermidis* produce virulence factors that inhibit the growth of *C*. *acnes*, and *Cutibacterium granulosum*, which is highly abundant in acne lesions, also generates virulence factors. Moreover, it has been shown that *Malassezia restricta* and *Malassezia globosa* are abundant in young patients with acne. Therefore, the presence of other microorganisms and their interactions should also be considered. To the best of our knowledge, no previous reports have been published on the direct effect of PM on acne inflammation. Our results suggest that PM may potentially aggravate acne by amplifying the inflammatory response via upregulation of the TLR/NF-κB pathway. Based on our findings, the inhibition of TLR/NF-κB signaling may be a promising target for the prevention and treatment of PM-induced acne exacerbation. Further studies with TLR/NF-κB pathway inhibitors and acne microenvironment models are needed to investigate the *in vivo* effects of PM on acne inflammation and to elucidate the underlying mechanisms. # Supporting information [^1]: The authors have declared that no competing interests exist.

# Introduction The annexins are a family of Ca²⁺-dependent phospholipid-binding proteins with various membrane-related functions. The name “annexin” is derived from the Greek “annex,” which means “bring/hold together,” and was chosen to describe the principal property of all, or at least nearly all, annexins — the binding to and possibly holding together of certain biological structures, in particular membranes. At least 20 members of the family have been described to date. Annexin A2 (ANXA2), also called Annexin II, is one of the best characterized of the Annexins. ANXA2 is composed of two main structural domains: the 33-kDa C-terminal conserved core domain, which contains the Ca²⁺- and membrane-binding sites, ; and the 3-kDa N-terminal variable domain, which contains the protein binding sites and phosphorylation sites. Otherwise, the N-terminus harbors a high affinity hydrophobic interaction site for the EF-hand Ca²⁺ binding protein S100A10 (p11). Two molecules of ANXA2 and two molecules of p11 form a heterotetrameric complex (A2t) that has been suggested to be involved in exocytosis, endocytosis and membrane vesicle trafficking. ANXA2 was first discovered as a substrate of the Rous sarcoma virus-encoded tyrosine protein kinase. Subsequent studies have implicated ANXA2 in several biological functions including mitogenic signal transduction, fibrinolysis, immune response, proliferation, carcinogenesis and tumor progression,. Large- scale genomic and proteomic studies have begun to accumulate evidence regarding the association and possible involvement of ANXA2 with benign and malignant neoplasms of diverse origins. Increased expression of ANXA2 has been described in a large number of spontaneous neoplasms, including pancreatic cancer, gastric carcinoma, colorectal cancer, breast cancer, high-grade gliomas and kidney cancer (reviewed) and is positively correlated with tumor invasion and migration. In contrast, the expression of ANXA2 is lost or reduced in prostate cancer, and the role of ANXA2 in prostate cancer appears contradictory. The differential expression of ANXA2 in HCC and normal liver tissue has been reported, but a more detailed functional assessment is lacking. Although published data support a crucial role for ANXA2 in tumor progression, the detailed mechanisms underlying this role have yet to be fully elucidated. Breakdown of the extracellular matrix (ECM), which is mediated by a variety of proteases, endows malignant cells with the ability to penetrate through tissue barriers and is believed to play a major role in tumor migration and invasion. ANXA2 has been found to be a putative co-receptor for both plasminogen and tissue-type plasminogen activator (tPA). Cell surface ANXA2 acts as a platform for plasmin activation, where inactive plasminogen is cleaved by tPA to yield the active serine proteinase, plasmin, thereby facilitating the migration and invasion of malignancies. Studies have also demonstrated that ANXA2 may regulate the production and activation of matrix metalloproteinases (MMPs). CD147 is a widely distributed cell surface glycoprotein that belongs to the immunoglobulin superfamily. It was first identified as a factor shedding from the surface of tumor cells that is responsible for stimulating the production of MMP-1 by fibroblasts. Accumulating evidence indicates that CD147 is a major mediator of the malignant phenotypes of various tumors. CD147 induces angiogenesis by stimulating the production of VEGF, invasiveness by stimulating the production of MMPs and multidrug resistance via hyaluronan-mediated up- regulation of ErbB2 signaling and the activity of cell survival pathways. Induction of MMP production through cell interactions is one of the most important functions of CD147 — thus the derivation of its other name: **<u>e</u>**xtracellular **<u>m</u>**atrix **<u>m</u>**etallo**<u>pr</u>**oteinase **<u>in</u>**ducer (EMMPRIN). CD147 may serve as its own counter-receptor in homotypic cancer cell interactions and cancer cell-fibroblast interactions, thereby stimulating the production of MMPs via a homophilic interaction with other CD147 proteins. In addition, MT1-MMP, MMP-2, and MMP-9 have been reported to cleave and release a shorter form of soluble CD147 that lacks the C-terminus, thereby modulating the expression of MMPs. Interestingly, recent studies have provided evidence that membrane microvesicles shed from tumor cells carry full-length CD147 and play a role in tumor–stromal interactions through the upregulation of the production of MMPs. Previous studies have demonstrated that CD147 promotes the invasion and metastasis of human hepatoma cells by stimulating both tumor cells and peritumoral fibroblasts to produce elevated levels of MMPs, although the modulation of fibroblasts is the more critical part of the process. Although the overexpression of ANXA2 in HCC has been shown, the role of ANXA2 in the migration and invasion of HCC cells remains obscure. In the present study, we knocked down the expression of ANXA2 in HCC cells to explore its role in HCC cell migration and invasion. To further investigate the mechanisms of ANXA2 in tumor progression, we introduced CD147, which has been hypothesized to interact with ANXA2 but which has not yet been shown to do so. Based on the involvement of ANXA2 in exocytosis and membrane vesicle trafficking, and the role of CD147-harboring membrane microvesicles on tumor progression, we hypothesized that ANXA2 is involved in the shedding of CD147-harboring microvesicles from tumor cells, thereby regulating tumor migration and invasion. # Materials and Methods ## Cell lines Two highly invasive human HCC cell lines, SMMC-7721 and FHCC-98, were cultured in DMEM containing 10% fetal bovine serum (FBS). SMMC-7721 cells were obtained from the Shanghai Institute of Biochemistry and Cell Biology, Chinese Academy of Science. The FHCC-98 cells were purchased from the Cell Engineering Research Centre, Fourth Military Medical University, China. Human embryo pulmonary fibroblast-1 (HPF-1) cells were purchased from the Chinese Academy of Medical Sciences and were cultured as described above. ## RNA interference si-ANXA2 was purchased from Santa Cruz Biotechnology, Inc. (sc-29199). si-CD147 (5′-GUUCUUCGUGAGUUCCUCtt-3′, 3′-dTdTCAAGAAGCACUCAAGGAG-5′) was synthesized by Ambion, Inc. HCC cells were transfected with siRNA using LipofectAMINE 2000 according to the manufacturer's instructions (Invitrogen, USA). Silencer negative control siRNA (snc-RNA) (Ambion, USA) was used as a negative control under identical conditions. ## Reverse transcription-PCR Forty-eight hours after transfection with siRNA, total RNA was extracted from the cells with TRIzol (Invitrogen, USA) and reverse transcribed into cDNA using a ReverTra Ace-α-™ kit (TOYOBO, Japan). GAPDH was used as an internal control. All primers were synthesized by Shanghai Sangon Co. as follows: ANXA2, forward primer 5′-GAGGATGGCTCTGTCATTGATT-3′; reverse primer 5′-CTGGTAGTCGCCCTTAGTGTCT-3′; GAPDH, 5′-ACCACAGTCCATGCCATCAC-3′ and 5′-TCCACCACCCTGTTGCTGTA-3′.The conditions for PCR were one cycle at 94°C for 4 min, 40 cycles at 94°C for 30 s, 60°C for 30 s, and 72°C for 30 s. PCR products were electrophoresed on 1% agarose gels. All PCR reactions were performed in triplicate. ## Western blot Cells were harvested and lysed in lysis buffer. A BCA Protein Assay Kit (Pierce Biotechnology, USA) was employed to determine the concentration of total protein. Equal amounts of protein were separated by SDS-PAGE (12%). Proteins were transferred to a polyvinylidene fluoride (PVDF) microporous membrane (Millipore, USA) and the blots probed with ANXA2 mAb (Santa Cruz, USA) or CD147 mAb (Santa Cruz, USA). Tubulin was chosen as an internal control, and the blots were probed with mouse anti-tubulin mAb (Santa Cruz, USA). ## *In vitro* migration/invasion assays The assays were performed using chambers with polycarbonate filters (pore size, 8 µm). Some filters were coated with Matrigel (Becton Dickinson Labware, USA), and some were not. Twenty-four hours after being transfected with siRNA, HCC cells were harvested. Equal numbers (5×10⁴) of transfected and HPF-1 cells were then placed in the upper chamber in 300 µL of 0.1% serum medium. Cells transfected with snc-RNA were used as a negative control. The lower chamber was filled with 0.1% fetal bovine serum medium (200 µL) and serum-free conditioned medium collected from HPF-1 cells (200 µL). After 8 h (migration assay) or 24 h (invasion assay) of incubation, the cells in the upper chamber were carefully removed with a cotton swab. The cells on the underside were then fixed in methanol, stained with H&E, and counted under a microscope. ## Gelatin Zymography Serum-free conditioned medium collected from siRNA-transfected SMMC-7721 and FHCC-98 cells was added to HPF-1 cells. Fifteen hours later, the conditioned medium was collected and separated using a 10% acrylamide gel containing 0.1% gelatin. The gels were incubated in a 2.5% Triton X-100 solution at room temperature with gentle agitation and were then soaked in reaction buffer \[0.05 mol/L Tris-HCl (pH 7.5), 0.2 mol/L NaCl, and 0.01 mol/L CaCl₂\] at 37°C for 18 h. After the reaction, the gels were stained for 6 h with Coomassie brilliant blue and de-stained for 0.5 h. Areas of gelatinolytic activity were evident based on their negative staining. ## Immunocytochemistry and confocal microscopy SMMC-7721 and FHCC-98 cells were harvested and allowed to attach to pre-coated glass coverslips for 24 hours. They were then fixed in 3.7% formaldehyde in PBS, permeabilized with 0.1% Triton X-100 and blocked with 1% BSA (Fraction V) in PBS for 1 h. Coverslips were incubated with goat anti-CD147 antibody (1∶100; Santa Cruz, USA) and ANXA2 mAb (1∶100) in PBS overnight at 4°C. The cells were then washed and incubated with Alexa 594-conjugated goat anti-mouse (Invitrogen, USA) or donkey anti-goat IgG-FITC (Santa Cruz, USA) secondary antibody at a dilution of 1∶400 for 1 h at room temperature. Cell nuclei were stained with DAPI (Vector, USA). The slides were then washed and mounted onto glass slides. Anti- fade was added to prevent quenching of the fluorophores. The proteins were visualized with a FluoView™ FV1000 confocal microscope (Olympus, Japan). ## Extraction and co-immunoprecipitation of total cellular membrane proteins Total cellular membrane proteins (TMP) were extracted from SMMC-7721 and FHCC-98 cells using a Plasma Membrane Protein Extraction Kit (BioVision, USA), according to the manufacturer's instructions. SMMC-7721 and FHCC-98 cells (5×10⁷) were collected, centrifuged and washed with 1 ml ice-cold PBS. The cells were then re-suspended in 1 ml Homogenize Buffer Mix in an ice- cold Dounce homogenizer 30–50 times. The homogenate was centrifuged at 700 g for 10 minutes at 4°C. The supernatant was collected and centrifuged at 10,000 g for 30 minutes at 4°C. The cytosol fraction (supernatant) and total cellular membrane proteins (pellet) were collected. The samples were either used immediately or stored at −20°C for further study. The interaction of ANXA2 with CD147 in the plasma membranes of SMMC-7721 and FHCC-98 cells was detected using a ProFound™ Mammalian Co-Immunoprecipitation Kit (Pierce, USA) according to the manufacturer's instructions. Briefly, TMP extracted as described above were collected onto a Coupling gel pre-bound with 200 µg anti-ANXA2 mAb, anti-human CD147 mAb, or anti-JEV mAb (mouse IgG, kindly provided by the Department of Microorganism, Fourth Military Medical University and used as a negative control), followed by four washes with the co- immunoprecipitation buffer. The coupling gel was washed with elution buffer, and aliquots of the eluent were used for Western blotting with anti-ANXA2 mAb and anti-CD147 mAb. ## Isolation and electron microscopy of membrane microvesicles Microvesicles were isolated as previously described. Briefly, medium conditioned for 24 hours by subconfluent, healthy SMMC-7721 cells was centrifuged at 600 g for 15 minutes and then at 1500 g for 15 minutes to remove cells and large debris. The supernatant was ultracentrifuged at 100,000 g for 1 hour at 4°C. Pelleted microvesicles were re-suspended in PBS (pH 7.4). Isolated vesicles were quantified based on protein concentration measurements using the Bradford method (Bio-Rad, Milan, Italy), with BSA (Sigma, St. Louis, MO) used as a standard. Samples were either frozen at −20°C or used immediately for electron microscopy studies. Aliquots of vesicles were applied to 200 mesh nickel parlodion-coated grids and allowed to settle. Samples were negatively stained with 1% phosphotungstic acid (pH 6) before being analyzed with an electron microscope. # Results ## Specific siRNA effectively down-regulates ANXA2 expression in HCC cells To investigate the role of ANXA2 in the invasion and migration of HCC cells, RNA interference was used to downregulate the expression of ANXA2 in SMMC-7721 and FHCC-98 cells. Both cell lines were transfected with si-ANXA2 and snc-RNA. Forty-eight hours after transfection, the expression of ANXA2 mRNA and protein was examined using RT-PCR and Western blot, respectively. RT-PCR showed that si- ANXA2 could effectively downregulate the expression of ANXA2 mRNA in SMMC-7721 and FHCC-98 cells, with an inhibition rate of 63.39±3.87% and 62.80±2.98%, respectively, compared to snc-RNA (*p\<0.001*). These results were consistent with Western blot assays. The protein expression of ANXA2 was notably reduced in si-ANXA2 transfected cells, with an inhibition rate of 64.12±1.46% and 58.80±1.36%, respectively, compared to snc-RNA-transfected cells (*p\<0.001*). These data show that ANXA2-specific siRNA can effectively reduce the expression of ANXA2 in HCC cells. ## Effects of ANXA2 on the migration and invasion of HCC cells To mimic *in vivo* tumor-stroma interactions within local microenvironments, HCC cells were transfected with si-ANXA2 and then co-cultured with an equal number of HPF-1 cells in the upper chamber. The migration assay showed that the number of cells that migrated through the filter was drastically reduced in the SMMC-7721 and FHCC-98 cells transfected with si-ANXA2, with an inhibition rate of 59.22±2.43% and 55.94±2.76%, respectively, compared to snc-RNA-transfected cells (*p\<0.001*). Using an *in vitro* invasion assay, the number of invading cells was also shown to be reduced in si-ANXA2-transfected SMMC-7721 and FHCC-98 cells, with an inhibition rate of 60.63±1.65% and 59.49±3.19%, respectively, compared to snc-RNA-transfected cells (*p\<0.001*). Both *in vitro* and *in vivo*, increased tumor aggression has been reported to be correlated with high expression levels of MMPs, which are mainly secreted by stromal cells. A critical question that arises from our study is whether the expression level of ANXA2 in HCC cells affects the production of MMPs by peritumoral fibroblasts. To address this issue, we performed a gelatin zymography assay, which indicated that the secretion of MMP-2 was significantly reduced in HPF-1 cells cultured in supernatant collected from si- ANXA2-transfected HCC cells, with an inhibition rate of 46.75±3.29% and 35.60±2.88% in SMMC-7721 cells and FHCC-98 cells, respectively, compared to snc- RNA-transfected cells (*p\<0.01*). These data demonstrate that ANXA2 may facilitate the migration and invasion of HCC cells *in vitro* by regulating the production of MMP by peritumoral fibroblasts. ## ANXA2 and CD147 co-localize on HCC membrane structures To study the possible interaction between ANXA2 and CD147 and to explore the hypothesis that ANXA2 is involved in the shedding of CD147-harboring microvesicles from tumor cells, we performed immunofluorescent double-labeling on SMMC-7721 and FHCC-98 cell slides. As shown in, ANXA2 (red) and CD147 (green) were co-localized in both cell lines. We focused on the strong double-staining of the membrane. To further confirm the results and explore the possibility that ANXA2 is involved in CD147 membrane microvesicle trafficking, we extracted the TMP from SMMC-7721 and FHCC-98 cells. Cytoskeletal protein (tubulin), membrane protein (VEGF-R2), and ANXA2 were detected in whole cell lysates and TMP and cytosol fractions using Western blot. We then performed co-immunoprecipitation of the TMP extracted from HCC cells. These results showed that ANXA2 and CD147 co-immunoprecipitated with each other in the TMP extracted from both SMMC-7721 and FHCC-98 cells, indicating that ANXA2 and CD147 co-localize on HCC membrane, and there may be interaction effects due to membrane-associated events between the two molecules. ## ANXA2 affects the invasiveness of tumor cells by regulating the transportation of CD147-harboring membrane microvesicles Data have shown that ANXA2 is involved in microvesicle trafficking and that CD147-harboring microvesicles play an important role in tumor-stroma cross-talk by stimulating the production of MMPs. The co-localization of ANXA2 and CD147 described above indicated that ANXA2 may be involved in CD147 membrane microvesicle trafficking, and could therefore affect the invasiveness of tumor cells. To address this issue, we isolated microvesicles shed from SMMC-7721 cells and analyzed the harvested pellet using Western blot. Electron microscopy of the ultracentrifuged pellet revealed that SMMC-7721 cells shed spherical or ellipsoid-shaped microvesicles (∼200–500 nm), consistent with previous reports on plasma membrane-derived microvesicles. A representative image is shown in. The microvesicles were then analyzed using Western blot, which showed that the microvesicles shed from SMMC-7721 cells carried both ANXA2 and CD147. To further investigate the effect of ANXA2 on the shedding of CD147-harboring microvesicles, SMMC-7721 cells were transfected with si-ANXA2 or si-CD147 and the microvesicles then isolated and analyzed as described above. The results showed that the expression of both ANXA2 and CD147 in the isolated microvesicles was downregulated when SMMC-7721 cells were transfected with si-ANXA2, with an inhibition rate of 64.49±3.56% and 36.16±1.49%, respectively, compared to the snc-RNA-transfected group (*p\<0.01*). However, when SMMC-7721 cells were transfected with si-CD147, only the expression of CD147 in the microvesicles was significantly downregulated, with an inhibition rate of 54.78±0.64% compared to the snc-RNA-transfected group (*p\<0.01*). We then focused on the effect of the isolated microvesicles on the secretion of MMPs by fibroblasts. HPF-1 cells were treated with microvesicles isolated from untreated or si-RNA-transfected (snc-RNA, si-ANXA2 or si-CD147) SMMC-7721 cells. A gelatin zymography assay indicated that microvesicles isolated from si-ANXA2- or si-CD147-transfected SMMC-7721 cells were less efficient in the induction of MMP-2 compared to microvesicles isolated from untreated or snc-RNA-transfected SMMC-7721 cells (*p\<0.01*). These results suggest that ANXA2 is involved in the trafficking of CD147-harboring microvesicles derived from tumor cells, which regulates the production of MMP-2 by fibroblasts, thereby facilitating the progression of HCC. # Discussion ANXA2 is thought to be involved in the transduction of cellular signals associated with inflammation, differentiation and proliferation. Emerging evidence indicates that changes in the expression and/or subcellular localization of ANXA2 contribute to the development and progression of a variety of malignancies. Frohlich et al. reported that ANXA2 was up-regulated in HCC. Recently, Mohammad et al. confirmed the upregulation of ANXA2 in HCC and provided a more detailed description. They reported that ANXA2 is almost undetectable in normal liver and chronic hepatitis tissue, but the expression of ANXA2 is abundant in non-tumorous cirrhotic tissue at both the transcriptional and translational levels. Furthermore, the expression of ANXA2 is greater in tumorous tissue than in non-tumorous cirrhotic tissue. These data strongly indicate that ANXA2 may be involved in the malignant transformation and progression of hepatocellular carcinoma. In the present study, we explored the role of ANXA2 in the invasion and migration of HCC cells. Two HCC cell lines, SMMC-7721 and FHCC-98, were co- cultured with HPF-1 cells to mimic the tumor-stroma interaction system *in vitro*. A migration assay, invasion assay, and gelatin zymography assay were employed. Our results show that knocking down the expression of ANXA2 in HCC cells inhibits the migratory and invasive potential of these tumor cells and significantly attenuates the production of MMPs by fibroblasts, which were previously reported to be involved in human hepatic tumorigenesis and metastasis. All these findings suggest that ANXA2 may play an important role in the progression of HCC, including migration, invasion, and enzyme degradation. Previous studies have illustrated that ANXA2 functions in the invasion and migration of various tumors. However, the exact molecular mechanisms underlying this function remain largely unknown. Sharma and Sharma proposed a mechanistic cascade for the role of AXNA2 in tumor progression. They suggested that ANXA2 at the cell surface of tumor/endothelial cells provides for the mechanical assembly of plasminogen activator and plasminogen, which locally activates plasminogen to plasmin. Plasmin then induces the degradation of ECM, which in turn facilitates endothelial/tumor cell invasion and migration. In this study, we found that the treatment of HCC cells with ANXA2-specific siRNA significantly reduced the production of MMP-2 by HPF-1 cells cultured in supernatant collected from HCC cells, suggesting that certain factors may exist in the supernatant that regulate the production of MMPs by HPF-1 cells. Accordingly, we focused on CD147, an important molecule responsible for stimulating the production of several MMPs (MMP-1, MMP-2, MMP-3, MMP-9, MMP-14, and MMP-15) by fibroblasts and endothelial cells. Thus, a possible interaction between ANXA2 and CD147 was considered. In this study, CD147 was found to co-localize with ANXA2 in HCC cells. The two molecules were also found to co-immunoprecipitate with each other in TMP extracted from SMMC-7721 and FHCC-98 cells. These results demonstrate that ANXA2 and CD147 are in close proximity, if not directly associated, and most likely interact in HCC cells. If we assume that CD147 is the bioactive factor hiding in the supernatant, how and in what form is CD147 released by tumor cells? Vesicle shedding has been observed in normal cells under certain physiological conditions and is present at much higher rates in tumor cells. The shedding of tumor surface antigens in membrane vesicles has been implicated as an important feature of malignant transformation. It is likely that vesicle shedding and, more importantly, the factors released by vesicle shedding, are vital to tumor survival and growth because it is by the release of such factors that tumors condition their microenvironment, regulate metastasis and evade immune surveillance. Sidhu provided evidence of a form of tumor-stromal interaction. He showed that the degradation of the ECM by fibroblasts is controlled by the microvesicular release of CD147 from NCI-H460 cells. In addition, ANXA2 has been shown to participate in the aggregation and transportation of membrane microvesicles. Our present data support the possibility of an interaction between ANXA2 and CD147. We then hypothesized that CD147 carried by membrane microvesicles may be the soluble bioactive factor affecting the production of MMPs in the supernatant collected from HCC cells and that ANXA2 may be involved in the trafficking. Our subsequent study confirmed this hypothesis. Membrane microvesicles were successfully isolated from the supernatant of SMMC-7721 cells using ultracentrifugation. Both CD147 and ANXA2 were detected in the isolated microvesicles using Western blot. When HPF-1 cells were treated with microvesicles, the production of MMP-2 was significantly increased. To further investigate the effect of ANXA2 on the shedding of CD147-harboring microvesicles, SMMC-7721 cells were transfected with si-ANXA2 or si-CD147 and microvesicles then isolated. These results indicate that the downregulation of ANXA2 in tumor cells reduces the expression of CD147 in isolated microvesicles, but the expression of ANXA2 in the microvesicles is not affected when cells are transfected with si-CD147. It has been reported that ANXA2 is involved in exocytosis and membrane vesicle trafficking, which may be the reason why the down-regulation of ANXA2 affects CD147 protein levels. There is no evidence that CD147 regulates the production and release of microvesicles, which is consistent with our result that the down-regulation of CD147 did not affect ANXA2 protein levels. The induction effect of microvesicles on the production of MMP-2 was weakened when SMMC-7721 cells were transfected with either si-ANXA2 or si-CD147. All of these findings show that the shedding of microvesicles from tumor cells acts as an efficient vehicle for CD147 trafficking and that ANXA2 regulates the transportation of CD147-harboring microvesicle, thereby contributing to the progression of HCC, although there may be other molecules in the microvesicles besides CD147. In conclusion, we report that ANXA2 promotes the migration and invasion of HCC cells co-cultured with fibroblasts *in vitro* by regulating the shedding of CD147-harboring microvesicles from tumor cells, which contributes to tumor- stroma crosstalk and sheds light on the mechanisms of ANXA2 in tumor progression. ANXA2 may be a potential target for the development of effective therapeutic strategies for the treatment of HCC. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: KFD ZNC WPZ. Performed the experiments: WZ PZ XLX LC DYC. Analyzed the data: KFD WZ PZ ZNC WPZ. Contributed reagents/materials/analysis tools: XLX ZSS KST. Wrote the paper: KFD WZ PZ.

# Introduction The seasonality of tourism demand is an important research topic in the academic fields and industries of tourism and hospitality. Seasonality of demand is widely recognized in economic fields and has long been studied in tourism research. The temporal imbalance of tourism demand causes various problems in the tourism and hospitality industries. For example, problems include unstable employment, inefficiency of investment, and exhaustion of local communities. Although recent tourism studies dealing with tourism demand have mainly focused on forecasting, research focusing on seasonality of tourism has also been widely conducted for the development of regional economies, tourism marketing, and policy suggestions. There are various kinds of data to observe tourism demand. These include, for example, visitors to tourism facilities, air passengers, and room occupancy rate. If the number of analysis subjects (e.g., tourist facilities, hotels, tourism destinations, countries, etc.) is too high, then it is difficult to compare their seasonal patterns and organize data because the data for analysis subjects are time-series data. Therefore, for the problem, many measurements of seasonality have been proposed in tourism studies. By utilizing a seasonal measurement, time-series data can be summarized as a real number; nevertheless, we need to choose an appropriate measurement from the proposed seasonal measurements in accordance with our research object. One of the basic analysis approaches for seasonality is to observe the seasonal patterns of analysis subjects. When conducting exploratory analysis as a first step to gaining an overview, it might be natural to check patterns of seasonality of the analysis subjects and compare the patterns, remembering that the number of the analysis subjects is typically greater than one. If the sample size of the seasonality datapoints is moderate, then we can analyze a simple line graph with time-series data expressing seasonal patterns. This is called the graphical analysis in this study. The graphical analysis allows for the direct verification of seasonal peaks, troughs, and shoulder seasons. This method is simple, but a powerful tool for intuitively capturing seasonal patterns of analysis subjects. The seasonal index is a useful seasonal measurement and is often used in tourism studies. The seasonal index consists of averages of seasonal factors by month and is derived from a time series decomposition method and denotes the pure seasonality element of time-series data that does not include trend and residual factors. Although the seasonal index summarizes the seasonal factors into a 12-dimension real number vector, the measurement does not include information on changes in seasonality because the measurement is presented in terms of averages of seasonal factors by month. This implies that using a seasonal index to determine the seasonal pattern means treating the seasonal pattern as a deterministic one. Even though it has been reported that changes in seasonal patterns tend not to be generally considerable, seasonality changes commonly occur. In order to check seasonal patterns by considering their stability over the years, a few tourism studies use seasonal plots or cycle plots that express monthly seasonality and the changes in seasonal factors for each month by using a bar graph or line graph. However, if the number of unstable seasonal patterns that we need to analyze increases, it becomes an even more onerous task to compare and organize these seasonal patterns because the number of generated graphs becomes large. To tackle this problem, this study suggests a new seasonal measurement that is based on a seasonal index and interval-valued data, which are one of the symbolic data. Symbolic data are a complex data type that expresses the uncertainty of data, and the interval-valued data are presented by interval numbers but not real numbers. The interval-valued seasonal measurement includes information on fluctuations in seasonality over multiple years. In addition, this study introduces a hierarchical clustering method based on Ward’s method for the measurement proposed by Ogasawara and Kon. A dendrogram generated by the clustering method reveals the whole picture of analysis subjects, which are presented by interval-valued seasonal measurements. From the new measurement and the clustering method, we can intuitively classify seasonal patterns and compare among them by considering fluctuations of the seasonal patterns. This study also shows interval-valued seasonal indices computed from Japanese overnight data from 2011 to 2019 and clustering results obtained by the clustering method suggested by Ogasawara and Kon, as a case study. Japanese overnight data consist of domestic and inbound guests by prefecture in Japan. As a result, changes in seasonal patterns of domestic guests are negligible and uniform for all prefectures, although the seasonal amplitude of domestic guests in each year is larger than that of inbound guests. Furthermore, the diversity of seasonal patterns for inbound guests among prefectures is higher than that for domestic guests. From the results obtained from the cluster analysis, clusters for domestic guests were divided by slight differences in seasonal patterns with strong stability. Classified results for inbound guests are strongly affected by the locations of prefectures, and the difference in seasonal patterns among the clusters is larger than that of domestic guests. Hence, the stable and uniform patterns of seasonality of domestic guests might be determined by institutional factors. The seasonal patterns of inbound guests might be determined by the existence of major ski resorts and marketing policies of destination marketing/management organizations for the East Asia tourism market. Fuzzy theory and rough set theory have been applied in tourism studies dealing with forecasting tourism demand to consider uncertainty. Similarly, this study employs interval-valued data, a kind of symbolic data used to express uncertainty and to deal with the stability of seasonal patterns of tourism demand. The Japanese case study showed that seasonal patterns of Japanese inbound guests by prefecture are less stable than those of domestic guests. This phenomenon has also been detected in Italy. If we need to analyze mutable seasonal patterns for many subjects over multiple years, then both the interval- valued seasonal index and hierarchical clustering method introduced in this study can be helpful tools to present a complete picture of the unstable seasonal patterns. In this study, an interval-valued seasonal index is suggested as a new seasonal measurement and we show its effectiveness through a Japanese case study. The study also reveals characteristics including instability of seasonality of Japanese overnight demand especially relating to unstable inbound demand. The interval-valued seasonal index can capture a cluster that cannot be observed with the traditional seasonal index. The cluster consists of prefectures in the Tohoku area in Japan which was devastated because of a 9.0 magnitude earthquake and a tsunami that followed in March 2011. In this area, inbound demand rapidly increased because of the first worldwide destination campaign organized by Japan in 2016. In the next section, seasonality in tourism demand and its measurement suggested in tourism studies are introduced. Further, seasonal patterns expressed by seasonal measurements and their changes are discussed. In the third section, a new seasonal measurement, called the interval-valued seasonal index, is defined and explained. The hierarchical clustering method for interval-valued data suggested by Ogasawara and Kon is briefly introduced in the fourth section. In the fifth section, overnight data in the case study and its overview are shown. Interval-valued seasonal indices by prefecture in Japan and clustering results are also indicated, and the results are discussed. # Seasonality A number of tourism studies have dealt with seasonality of demand in tourism industries. It is well known that the intensity fluctuation of seasonal demand leads to an increasing degree of difficulty in managing the suppliers’side. In the accommodation sector with a fixed capacity, the difficulty of management is directly related to revenue. From the viewpoint of revenue management, which is a field of management aimed at increasing revenue in industries with fixed capacity, large fixed products, and perishable products, such as airlines, hotels, and rental cars, it is natural that the supplier’s side pays attention to seasonality of tourism demand because the effect of booking control on expected revenue is magnified by increasing demand. The difference between the peak and trough of tourism demand is also important for management and investment in tourism industries because it is not generally easy to increase the capacity of facilities or transport infrastructure to handle tourism demand. If a facility cannot fully cover tourism demand in a certain period, then the manager of the facility might consider adding capacities. However, it often leads to an increase in excess capacity during the off-season. An excessive difference in demand between peak season and off season is undesirable for tourism-related facilities or local regions because it causes inefficiency in development or investment in tourism industries and instability in labor markets. ## Seasonal measurement If an analysis subject has an occupancy rate and we can access the data, then we may directly use the value that is standardized between 0 and 1 or 100. However, if given data also have scale or a trend, tourism studies often use the time- series decomposition method to isolate and quantify seasonal components (see for details). In this method, it is assumed that the time-series data *O*_*t* at time *t* ∈ {1, ⋯, *T*} consist of trend-cycle *TC*_*t*, seasonal *S*_*t*, and residual *R*_*t* components. There are two models: an additional model and a multiplicative model, owing to the differences in relationships among the components. In this study, a multiplicative model $$\begin{array}{r} {O_{t} = TC_{t} \cdot S_{t} \cdot R_{t},t = 1,\cdots,T} \\ \end{array}$$ is employed. The seasonal component *S*_*t* is called the seasonal factor and is used as an indicator for seasonal measurements. *S*_*t* is interpreted as an amplifier for demand, that is, *S*_*t* = 1, *S*_*t* \> 1, and *S*_*t* \< 1 corresponds to no seasonality, positive seasonal effect, and negative seasonal effect in *t*, respectively. This study assumes that time-series data are monthly data to simplify notations. Let *S*_*i*,*y*, *i* = 1, ⋯, 12, *y* = 1, ⋯, *Y* denote seasonal factors driven from the time-series data, where *i* and *y* are indices for month and year, respectively. Note that the seasonal factor is still a time- series formula. It can be summarized as some measurements focusing on the seasonal pattern or amplitude of demand, to easily make graphs and apply it to various data analysis methods. The definitions of seasonal measurements using seasonal factors are, for examples, as following, seasonal index: ${\hat{S}}_{i} = \sum_{y = 1}^{Y}S_{i,y}/Y,i = 1,\cdots,12$, seasonal peak: $S_{y}^{max} = \text{max}_{i \in \{ 1,\cdots,12\}}S_{i,y},y = 1.\cdots,Y$, seasonal ratio: $S_{y}^{ratio} = \text{max}_{i \in \{ 1,\cdots,12\}}S_{i,y}/\text{min}_{i \in \{ 1,\cdots,12\}}S_{i,y},y = 1,\cdots,Y$ and seasonal range: $S_{y}^{range} = \text{max}_{i \in \{ 1,\cdots,12\}}S_{i,y} - \text{min}_{i \in \{ 1,\cdots,12\}}S_{i,y},y = 1,\cdots,Y$. A popular seasonal measurement that does not include the seasonal factor is the Gini coefficient, which has been used in economics as a traditional measurement. ## Seasonal pattern Seasonal patterns used in tourism studies are often identified by seasonal peak and seasonal trough. Furthermore, a simple and effective way to intuitively capture seasonal patterns is to observe the shape of the seasonal index, which is intrinsically the same as graphical analysis by line graph with seasonal factors. Koenig and Bischoff use graphical analysis to indicate seasonal patterns although they use indicators provided by principal component analysis and not seasonal factors. The definitions of seasonal measurements mentioned in this study imply that seasonal patterns shown by seasonal factors can change over time. While the change in the seasonal pattern is not generally significant, it is a common occurrence. Sørensen reported a change in seasonality for hotels’ overnight demand in Denmark from 1970 to 1996 by unit root tests. As a more short-term case, Cuccia and Rizzo showed a change in seasonality in Sicily from 1998 to 2006 by using the F-test for dynamic seasonality. The major causes of seasonality in tourism demand are climate, weather, social customs, holidays, business customs, calendar effects, and supply side constraints. Considering that these occurrences are subject to global warming, changing social structures, and changing business environments, it is not surprising that seasonal demand patterns change over time. In addition, the increasing use of retrieval services and online services also affects seasonality. If we need to find a seasonal pattern for a region by taking into account changes in the pattern over the years, seasonal plots and cycle plots, which show transitions of seasonal factors for each month, have been suggested as graphical analysis tools. However, analyses using these graphs do not allow easy observation and organization of multiple seasonal patterns at once. If we use seasonal indices to see patterns for a number of analysis subjects, then we can classify them by cluster analysis with the seasonal indices as input values and reveal the derived clusters on a map, which leads to capturing a complete picture of seasonality for the subjects. In fact, Koenig and Bischoff adopted a similar approach. However, we cannot include information on changes in seasonal patterns in the procedure because the information is lost by calculating seasonal indices even if the seasonal patterns are unstable. Therefore, it is not easy to express many seasonal patterns with their changes and mechanically classify the patterns by using seasonal measurements that have been suggested in tourism studies. # Interval-valued seasonal index Interval-valued data are a type of symbolic data that are extended from classic data to treat complex conditions. For instance, let us assume that we obtain one’s contact address and a city of the address is “Tokyo.” Then, the city name corresponds to a classical categorical value “Tokyo.” However, if one’s residential city is “Tokyo” or “Yokohama” because of the fact that one has two residences and lives in Yokohama only on weekends, then we may need to treat the city data as a set of values: {*Tokyo*, *Yokohama*}. The set of values is called multi-valued data, which are a type of symbolic data. In another case, when one’s weight is between 60 kg and 61.5 kg, we can define that the data is a closed set \[60, 61.5\]. The closed set is called the interval-valued data, which is defined as $$\begin{array}{r} {X = \left\lbrack \underline{X},\overline{X} \right\rbrack = \left\{ x \in \mathbb{R}:\underline{X} \leq x \leq \overline{X} \right\}} \\ \end{array}$$ where $\mathbb{R}$ denotes the set of real numbers. The new seasonal measurement based on interval-valued data and a seasonal index is presented as follows: **Definition 1**. *For the given seasonal factors S*_*i*,*y*, *i* = 1, ⋯, 12, *y* = 1, ⋯, *Y*, $$\begin{array}{r} {I_{i} = \left\lbrack \underline{S_{i}},\overline{S_{i}} \right\rbrack,i = 1,\cdots,12} \\ \end{array}$$ *is called an interval-valued seasonal index for a month i, where* $\underline{S_{i}} = \text{min}_{y \in \{ 1,\cdots,Y\}}S_{i,y}$ *and* $\overline{S_{i}} = \text{max}_{y \in \{ 1,\cdots,Y\}}S_{i,y},i = 1,\cdots,12$. *In addition*, ***I*** = (*I*₁, ⋯, *I*₁₂) *is called an interval-valued seasonal index*. From Definition 1, an interval-valued seasonal index for an analysis subject is addressed using 12-dimensional interval-valued data. The interval-valued seasonal index has maximums and minimums of seasonal factors for each month. *I*_*i* in indicates the range of observed seasonal factors for a month *i* from year 1 to *Y*. D’Urso and Leski introduce that interval-valued data are used in three situations: (1) data cannot be shown as a traditional real value for example, daily temperatures or blood pressures, (2) we do not have a true single valued data but an interval value including the true value because of a lack of knowledge, and (3) we have only aggregated data whose values are intervals. The interval-valued seasonal index is defined in situation (1), which means that a seasonal pattern shown by the seasonal index over multiple years needs to be expressed by interval-valued data because the pattern can be changed through various motions. shows an image of transforming seasonal factors into an interval-valued seasonal index and a seasonal pattern addressed by an interval-valued seasonal index. In, the number of months is only five, and *Y* = 2 for simplification. Note that the measurement does not provide an increase or decrease in seasonal factors, although the interval-valued seasonal index can present the change of seasonal factors for each month as intervals, as seen at months 3 and 4 in. Let *w*(*I*_*i*) be the width of the interval-valued seasonal index for a month *i*; that is, $w\left( I_{i} \right) = \overline{I_{i}} - \underline{I_{i}}$. In, we can find *w*(*I*₂) \> *w*(*I*₃), which can be interpreted as the stability of seasonality in month 3 being smaller than that in month 2 through years 1 and 2. # Clustering method for interval-valued seasonal indices When we conduct an exploratory analysis of seasonal patterns expressed by interval-valued seasonal indices without any assumptions, a dendrogram given by a hierarchical clustering method is helpful to look at complete picture of those patterns. Clustering methods for interval-valued data have been studied in the field of pattern recognition. Ogasawara and Kon suggested a hierarchical clustering method for interval-valued data based on Ward’s method. This study briefly introduces a hierarchical clustering method suggested by Ogasawara and Kon and uses the method for classifying Japanese seasonal patterns as a case study addressed in the next section. Let ***I***_*u* = (*I*_1,*u*, ⋯, *I*_12,*u*), *u* ∈ {1, ⋯, *N*} be the interval-valued seasonal index for subject *u*. In the clustering method introduced in this study, the squared Euclidean Hausdorff distance, which is extended from the squared Euclidean distance, is employed instead of the squared Euclidean distance used in Ward’s method. The squared Euclidean Hausdorff distance between the interval-valued seasonal indices ***I***_*u* and ***I***_*v* is expressed as follows: $$\begin{array}{r} {H^{2}\left( \mathbf{I}_{u},\mathbf{I}_{v} \right) = \sum\limits_{i = 1}^{12}\left( \text{max}{\{|} \right.\underline{I_{i,u}} - \underline{I_{i,v}}|,|\overline{I_{i,u}} - \overline{I_{i,v}}{{|\}})}^{2}.} \\ \end{array}$$ Using the distance function, the sum of the squared deviations of a cluster *G*_*a* is $$\begin{array}{r} {ESS\left( G_{a} \right) = \sum\limits_{\mathbf{I} \in G_{a}}H^{2}\left( \mathbf{I},E\left( G_{a} \right) \right)} \\ \end{array}$$ where $E\left( G_{a} \right)\left. = 1/ \right|G_{a}|\sum_{\mathbf{I} \in G_{a}}\mathbf{I} = 1/\left| G_{a} \right|\left\lbrack \sum_{\mathbf{I} \in G_{a}}\underline{I},\sum_{\mathbf{I} \in G_{a}}\overline{I} \right\rbrack$, and \|*G*_*a*\| is the number of individuals in *G*_*a*. *E*(*G*_*a*) is called the interval-valued mean of *G*_*a*. For two clusters *G*_*a* and *G*_*b* such that *G*_*a* ∩ *G*_*b* = *ϕ*, dissimilarity is defined as $$\begin{array}{r} {\Delta ESS\left( G_{a},G_{b} \right) = ESS\left( G_{a} \cup G_{b} \right) - ESS\left( E_{a} \right) - ESS\left( E_{b} \right).} \\ \end{array}$$ The form of corresponds to a dissimilarity in Ward’s method. We employ the dissimilarity and obtain the result of cluster analysis for interval-valued seasonal indices as a dendrogram by applying the dissimilarity to an algorithm of Ward’s method. # Japanese case study Japan is ranked as a top-ten tourism destination in the world, as reported by the World Tourism Organization of the United Nations (UNWTO) in 2020. The number of international tourist arrivals grew at about 3.7 times the rate over 2010 to 2019. However, there have been few studies on inbound tourism in Japan. Although we can find research that analyzes the influence of climatic and economic variables on seasonality of tourism demand in several major Japanese tourism destinations as a reference, little attention has been paid to the seasonality of tourism demand in Japan. In this section, actual interval-valued seasonal indices derived from Japanese overnight data of Overnight Travel Statistics Survey over the period January 2011 to December 2019, which were collected by the Japan Tourism Agency are demonstrated. The overnight data consist of two items: domestic guests and inbound guests. In addition, each item has two sectors: 1) the number of overnight stays in hotels where more than half the guests stay for tourism or recreation purposes and 2) a number of overnight stays in hotels where more than half the guests stay for business purposes. Therefore, there are four items that are a combination of domestic or inbound guests and the two purposes. In this study, hotels corresponding to 1) and 2) denote “hotels for tourism” and “hotels for business,” respectively. shows a stacked graph for the four items from 2011 to 2020. It is obvious that seasonality exists on all items of overnights but vanishes in 2020 because of the COVID-19 pandemic. Hence, this study removed overnight data from the analysis subject in 2020. Further, while inbound visitors increased until 2019, as reported by the UNWTO, we can observe that domestic guests still dominate overnight stays in Japan. ## Japanese seasonality Even though we can clearly find Japanese seasonal patterns from, a comparison of patterns of seasonality among those items is unclear. Figs and show the seasonal factors for each item from 2011 to 2019, which are computed by X13-ARIMA-Seat with a multiplicative model and X11 option to obtain the results of statistical seasonal analysis. Although the scale of the seasonal factors between hotels for tourism and those for business are different among domestic guests, patterns of seasonality are similar to each other. For inbound guests, seasonal patterns between hotels for tourism and those for business are slightly different; specifically, there is a small peak in February in, but not in, while the scales are almost the same. The main peak was in August, and shoulder peaks were observed in March, May, and October in the case of domestic guests. In terms of inbound guests, peaks were observed in hotels for tourism in February, April, July, and October, and hotels for business in April, July, and October. Note that the seasonal peaks of domestic guests and inbound guests do not overlap. The seasonal factors of domestic guests for tourism in August declined from 2011 to 2019 as seen in. We can also see changes in seasonal patterns in inbound guests for tourism, reflected by the increase in seasonal factors in February. In comparison with changes in seasonal factors for tourism, changes in seasonal factors for business are relatively small. However, observing the results of the F-test of moving seasonality, we can obtain different impressions from the figures. It is reported that the results of Freedman’s test on the null of the absence of seasonality are that all items in Figs and are significant at 5%. The results of the F-test on the null of the absence of moving seasonality derived from X13-ARIMA-Seat with the X11 option are that domestic guests in hotels for business and inbound guests in hotels for tourism are significant at 5% and the F-values for those are 2.48 and 3.02, respectively (the critical value at the 5% is *F*(8, 88) = 2.05). Therefore, even though Figs and shows that there is moving seasonality for domestic guests in hotels for tourism, we can find that it is difficult to claim that from the viewpoint of the moving seasonality test. In addition, it is difficult to find clear moving seasonality for domestic guests in hotels for business, although the item is significant in the F-test. Furthermore, the seasonal pattern in inbound guests for tourism obviously changed from 2011 to 2019, and the movement was significant on the F-test. Given the resemblance of seasonal patterns between overnight guests in hotels for tourism and business, this study shows results by prefecture obtained only from overnight stays in hotels for tourism. The seasonal factors in are calculated from the total sum of overnight guests in hotels for tourism in Japan. However, there are time series data for overnight guests by 47 prefectures in Japan, and seasonal factors for each prefecture can be derived, which means that 47 interval-valued seasonal indices and graphs illustrating them can be generated. Figs and provide interval-valued seasonal indices of domestic and inbound guests for tourism in Hokkaido, Tokyo, and Okinawa as examples. Hokkaido is the northernmost prefecture in Japan, and Okinawa is the southernmost prefecture in Japan. Interval-valued seasonal indices are illustrated as error bars, and the mid-points of the ranges are connected with a line in the figures. The ranges of the interval-valued seasonal index in Tokyo are smaller than those in Hokkaido and Okinawa as a whole, which is not surprising because Tokyo is the most developed city in Japan and has abundant tourism resources that are immune to seasonal fluctuations. Comparing intervals among months, combinations of months with wider ranges and months with narrower ranges differ by prefecture. Moreover, the ranges of intervals for inbound guests tend to be wider than those for domestic guests in these prefectures. In terms of the shape of the graph, the shapes of lines among the prefectures in are more similar than those in. This suggests that the diversity of seasonal patterns in inbound guests is stronger than that in domestic guests in Japan. We find that the impression is not wrong by viewing Figs and, which illustrate ranges between maximum of seasonal factors and minimum of seasonal factors among prefectures on each month over 2011 to 2019 in the case of domestic guests and inbound guests for tourism. This study also provides results of the F-test of moving seasonality by prefecture. Prefectures with significant F-values of 5% were Hokkaido (7.07), Aomori (3.36), Hukushima (2.50), Ibaraki (2.16), Tochigi (3.25), Gunma (2.38), Saitama (2.79), Chiba (2.67), Yamanashi (3.03), Nagano (2.15), Aichi (2.11), Hyogo (2.14), Nara (3.31), Wakayama (2.24), Shimane (2.18), Kagawa (2.51), Kumamoto (4.95), and Okinawa (2.66), 18 prefectures. With regard to inbound guests, the prefectures whose F-values are significant at 5% are Akita (2.74), Tochigi (2.75), Tokyo (3.08), Ishikawa (3.42), Fukui (2.84), Yamanashi (3.48), Gifu (5.10), Shiga (3.49), Osaka (2.38), Shimane (2.54), Hiroshima (2.25), Tokushima (3.13), Kochi (2.07), Saga (3.22), Nagasaki (4.51), Kumamoto (3.54), Miyazaki (3.37), and Okinawa (6.00), 18 prefectures. The values in parentheses correspond to the F-values. From these results, the seasonality of domestic guests in Hokkando and Okinawa shown in changes at the 5% significance level. Contrarily, the seasonality of inbound guests in Hokkando does not move at the 5% significance level, although its seasonal pattern seems not to be stable, as shown in. The seasonality of inbound guests in Tokyo and Okinawa shown in significantly changes at the 5% level. An F-value does not necessarily reflect an impression obtained by the shape of the graph denoting the interval-valued seasonal index because the F-value is affected by not only the change in seasonal factors but also the change in residual components in. ## Clustering patterns of Japanese seasonality ### Using interval-valued seasonal indices The clustering method mentioned in the previous section is applied to the interval-valued seasonal indices of 47 prefectures to classify seasonal patterns. The dendrograms obtained from interval-valued seasonal indices in domestic guests and inbound guests are shown in Figs and, respectively. The seasonal patterns are divided into four clusters in both domestic and inbound guests. The four clusters for domestic guests are named D1, D2, D3, and D4, and the inbound guests are named I1, I2, I3, and I4. The cluster names are sequentially allotted to divided clusters from top to bottom in the dendrograms, as shown in Figs and. Traditional analysis of variance cannot be used to observe the characteristics of the clusters in this study because the applied data are interval-valued data, that is, non-traditional data. The interval-valued mean of each month is illustrated in a simple manner to see the characteristics. Figs and show the interval-valued mean of each month for the clusters. Clusters D1 and D2 differ from D3 and D4 based on the strength of the main seasonal peak in August. Clusters D2 and D3 are different from D1 and D4 in the form of off-seasons, which are in January and February. In addition, the upper bounds of the intervals in May in D2 and D3 are slightly smaller than lower bounds in March in D2 and D3, and in D1 and D4. Further, the magnitude relationship between March and May is reversed. Hence, clusters D1, D2, D3, and D4 may be determined by the combination of the sizes of these interval-valued seasonal indices. However, all forms of the graphs for clusters D1, D2, D3, and D4 are close to each other, even considering the stability of the seasonality. It is difficult to divide them into well-balanced clusters in terms of inbound guests, as shown in the dendrogram in, because there are few prefectures with significantly diverging seasonal patterns from others. I1 and I2 are almost the same clusters, and the seasonal peaks occur in January and February. The peak of seasonality of I3 occurs in October, and the interval of the peak is relatively wide. Hence, the strength of the seasonal peak of I3 is not stable. We find that the seasonal peak of cluster I4 is in April; however, the seasonal scale of I4 is relatively weak. Figs and show locations of the prefectures belonging to the clusters for domestic guests and inbound guests in a map of Japan, respectively. The figures are derived by a package, choroplethr (ver. 3.7.0) on R (ver. 4.0.3). The locations of prefectures belonging to same clusters for domestic guests are scattered across Japan and we cannot find obvious seasonal features from the map. presents clusters’ positional features, I1 and I2, being almost the same clusters, are located near each other. Further, prefectures of I3 are agglomerated at a part of the north area of Japan. ### Using traditional seasonal indices We can also use another combination of seasonal indices and Ward’s method to generate dendrograms and make clusters. As a comparative study, dendrograms obtained through the seasonal indices of domestic guests and inbound guests for Japanese prefectures and by the traditional Ward’s method are shown in Figs and, respectively. The clustering computation are conducted by the hclust function on R (version 4.0.3). Figs and show the mean of each month for the clusters. The seasonal patterns are divided into four clusters to better compare them with the clustering results obtained by interval-valued seasonal indices. The clusters for domestic guests seem to be distinguished by the strength of seasonality in the main season (August) in, similar to the clustering results obtained by interval-valued seasonal indices. Cluster WD3 displays a shoulder season in January and February, which is a different characteristic from the clustering results obtained through seasonal indices. Clusters WI1 and WI2 are the same as I1 and I2, respectively. However, we cannot find a cluster for inbound guests for whom the main peak season is October. ## Discussion The seasonal peak of domestic guests is in August, which includes summer holidays, and graphical forms of interval-valued means for domestic guests by prefectures broadly resemble each other. In addition, the patterns were stable from 2011 to 2019. This indicates that domestic demand, occupying the majority of all Japanese tourism demand, is stably concentrated in limited periods, and this phenomenon occurs across Japan. From these results, it can be considered that this problem might be caused by institutional factors that affect an entire nation’s tourism demand. Hence, bold institute building for affecting institutional factors and customs, for example, staggering holidays might be necessary to more evenly distribute demand into other demand peaks throughout the year. The Japan Tourism Agency has already been tackled for staggering vacations and equalizing tourism demand since 2009 through a working team. In addition, the Japan Tourism Agency has implemented a project that aims to disperse tourism demand by time and place to avoid congestion in order to reduce the prevalence of COVID-19 in 2020. After fully containing the COVID-19 pandemic, it is possible that this project could be useful in reducing the concentration of domestic tourism demand. Prefectures with similar seasonal patterns of inbound guests are adjacent to each other in a map of Japan. Prefectures belonging to clusters I1 and I2, whose peak is in the snow season of January and February, have famous ski resorts that are frequented by inbound tourists (see, p.138 in). Hokkaido prefecture not belonging to I1 and I2 also has famous ski resorts areas, such as the Niseko- Hirafu district. Indeed, shows that there is a seasonal peak of inbound guests in January and February in Hokkaido from 2011 to 2019. Hokkaido is one of the most famous prefectures for tourism in Japan and includes large cities and famous tourism destinations other than ski resorts, such as the Shiretoko peninsula, registered as a UNESCO World Heritage Site. These results might lead to exclusion from clusters I1 and I2. A part of the north of Japan, which is a part of the Tohoku area, belongs to cluster I3. The Tohoku area was devastated by a magnitude 9.0 earthquake and a tsunami that followed it in March 2011. This disaster caused inbound guests to depart from the Tohoku area. Based on these circumstances, the Japan National Tourism Organization in Japan Tourism Agency started the Tohoku destination campaign in 2016, which increased the inbound demand in the Tohoku region for earthquake restoration support. This campaign was the first worldwide destination campaign in Japan. Important targeted markets for the campaign are mainly China, Taiwan, Thailand, and Korea in 2016 and 2017. Consequently, the campaign’s goal of achieving 1,500,000 inbound overnight guests in the Tohoku area by 2020 was attained in 2019. In Aomori, Iwate, Miyagi, and Akita prefectures belonging to cluster I3, the number of Chinese tourists was especially high. Specifically, the percentage of Chinese overnight guests in these four prefectures were 65.7%, 80.7%, 60.9%, and 62.3%, respectively (in 2019). Thus, a peak of I3 in October may stem from a long holiday in China, the National Day of the People’s Republic of China. While modifying seasonal patterns of domestic demand seems to be difficult in Japan, it might be possible that some neighboring prefectures execute a marketing strategy for inbound tourism demand to change their seasonal patterns and build a common characteristic seasonal pattern. If the concentration of inbound tourism demand in the area needs to be reduced, then it might be effective to control tourism demand by each origin country through a marketing promotion in the whole area. This implies that seasonal patterns of inbound demand in Japan may change if the demand structure of the origin countries is reconstructed after the COVID-19 pandemic. We find that characteristics of several clusters obtained by traditional seasonal indices and Ward’s method differ from ones derived from interval-valued seasonal indices. Cluster WD3, which consisted of Niigata and Nagano, shows that these prefectures have a small peak in winter season as a slight seasonal feature. These prefectures are famous across Japan for their many ski resorts. However, there is no major difference in domestic demand. This may be due to the insignificance of seasonal changes in domestic demand. In terms of inbound demand, the two clusters: WI1 and WI2 which are the same as I1 and I2 are generated. This means that these prefectures have a robust characteristic seasonality for inbound demand. In addition, it is worth noting that a cluster which includes prefectures in the Tohoku area and has a main peak in October cannot be observed using seasonal indices. This may be because the increase in inbound demand in the Tohoku region is rapid, as already mentioned. Although this rapid change cannot be captured by a seasonal index, an interval-valued seasonal index can capture it, as seen in. # Conclusions This study proposes a new seasonal measurement, namely the interval-valued seasonal index, which is based on interval-valued data in symbolic data. Additionally, a hierarchical clustering method for interval-valued data based on Ward’s method, which was suggested by Ogasawara and Kon, is introduced to grasp the whole picture of a dataset comprising many interval-valued seasonal indices and classify their seasonal patterns. By using the seasonal measurement and clustering method, the computational results obtained from Japanese overnight guest data from 2011 to 2019 are shown as a case study. These results demonstrate that there are differences in not only seasonal patterns but also their stability between domestic and inbound demand. In domestic tourism demand, seasonal patterns are almost the same among prefectures and their stability is stiff even though more than one-third of prefectures have significant F-values on the seasonality moving test at the 5% level. In contrast, seasonal patterns of inbound tourism demand among prefectures are more diverse than those of domestic tourism demand. We find characteristic seasonal patterns as clusters consisting of adjusted prefectures with ski resort developed regions and the Tohoku area in Japan. In particular, the cluster of the Tohoku area having rapidly recovered from the earthquake can be obtained with the proposed new measurement, the interval-valued seasonal index, and not with the traditional seasonal index. A tendency of inbound overnight guests moving more easily than domestic overnight guests, especially during peak months, can be found in Italy from 1988 to 2004. When we deal with the seasonality of inbound demand with low stability, an interval-valued seasonal index might be a useful seasonal measurement. Furthermore, there are data-analysis methods for interval-valued data other than the hierarchical clustering method introduced in this study. Thus, the interval-valued seasonal index can be utilized to analyze seasonal patterns by various other methods. In terms of limitations of this study, an interval-valued seasonal index does not have transition information, specifically going up or down for each month. The results of the Japanese case study cannot include this information. Furthermore, an interval-valued seasonal index corresponds to a prefecture that covers a large area, several markets, and various tourism resources, which limits the interpretation of analysis results. If an analysis subject corresponding to an interval-valued seasonal index becomes smaller and finer, then the procedure presented in this study is expected to provide more specific insights for tourism policy decisions or building marketing strategies. However, the computational cost of conducting the cluster analysis suggested in this study greatly increases due to the large number of given interval-valued data, as the clustering method does not use the squared Euclidean distance. Regardless of the numbers shown in the Japanese case study, we need to choose a moderate sample size to apply the procedure introduced in this study. While this study focuses on seasonal patterns and expresses their stability by interval-valued data, tourism studies have also focused on amplitude of seasonality, using seasonal range and ratio. From the viewpoint of symbolic data analysis, the seasonal range is equivalent to the width of an interval-valued data whose lower and upper boundaries correspond to the minimums and maximums of the seasonal factors in a year for an analysis subject. Therefore, we can define further seasonal measurements to analyze tourism seasonality. As a future study, we could consider reviewing contribution factors for seasonality, although we might need to develop further statistical models. I appreciate the helpful comments from reviewers. They greatly improved this manuscript. 10.1371/journal.pone.0267453.r001 Decision Letter 0 V E Sathishkumar Academic Editor 2022 Sathishkumar V E This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 15 Feb 2022 PONE-D-21-39279New seasonal measurement with stability and clustering seasonal patterns: a case study in Japan from 2011 to 2019PLOS ONE Dear Dr. Ogasawara, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. Please submit your revised manuscript by Apr 01 2022 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <plosone@plos.org>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript:A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols>. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at <https://plos.org/protocols?utm_medium=editorial- email&utm_source=authorletters&utm_campaign=protocols>. We look forward to receiving your revised manuscript. Kind regards, Sathishkumar V E Academic Editor PLOS ONE Journal Requirements: 1\. When submitting your revision, we need you to address these additional requirements. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at  <https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main _body.pdf> and  <https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_titl e_authors_affiliations.pdf> 2\. Thank you for stating in your Funding Statement:  (This work was supported by JSPS KAKENHI Grant Number JP20K20080.) Please provide an amended statement that declares \*all\* the funding or sources of support (whether external or internal to your organization) received during this study, as detailed online in our guide for authors at [http://journals.plos.org/plosone/s/submit- now. ](http://journals.plos.org/plosone/s/submit-now. ) Please also include the statement “There was no additional external funding received for this study.” in your updated Funding Statement.  Please include your amended Funding Statement within your cover letter. We will change the online submission form on your behalf. 3\. Please update your submission to use the PLOS LaTeX template. The template and more information on our requirements for LaTeX submissions can be found at <http://journals.plos.org/plosone/s/latex>. 4\. We note that Figures 12 and 13 in your submission contain map images which may be copyrighted. All PLOS content is published under the Creative Commons Attribution License (CC BY 4.0), which means that the manuscript, images, and Supporting Information files will be freely available online, and any third party is permitted to access, download, copy, distribute, and use these materials in any way, even commercially, with proper attribution. For these reasons, we cannot publish previously copyrighted maps or satellite images created using proprietary data, such as Google software (Google Maps, Street View, and Earth). For more information, see our copyright guidelines: <http://journals.plos.org/plosone/s/licenses-and-copyright>. We require you to either (1) present written permission from the copyright holder to publish these figures specifically under the CC BY 4.0 license, or (2) remove the figures from your submission: a\. You may seek permission from the original copyright holder of Figures 12 and 13 to publish the content specifically under the CC BY 4.0 license.   We recommend that you contact the original copyright holder with the Content Permission Form (<http://journals.plos.org/plosone/s/file?id=7c09/content- permission-form.pdf>) and the following text: “I request permission for the open-access journal PLOS ONE to publish XXX under the Creative Commons Attribution License (CCAL) CC BY 4.0 (<http://creativecommons.org/licenses/by/4.0/>). Please be aware that this license allows unrestricted use and distribution, even commercially, by third parties. Please reply and provide explicit written permission to publish XXX under a CC BY license and complete the attached form.” Please upload the completed Content Permission Form or other proof of granted permissions as an "Other" file with your submission. In the figure caption of the copyrighted figure, please include the following text: “Reprinted from \[ref\] under a CC BY license, with permission from \[name of publisher\], original copyright \[original copyright year\].” b\. If you are unable to obtain permission from the original copyright holder to publish these figures under the CC BY 4.0 license or if the copyright holder’s requirements are incompatible with the CC BY 4.0 license, please either i) remove the figure or ii) supply a replacement figure that complies with the CC BY 4.0 license. Please check copyright information on all replacement figures and update the figure caption with source information. If applicable, please specify in the figure caption text when a figure is similar but not identical to the original image and is therefore for illustrative purposes only. The following resources for replacing copyrighted map figures may be helpful: USGS National Map Viewer (public domain): <http://viewer.nationalmap.gov/viewer/> The Gateway to Astronaut Photography of Earth (public domain): <http://eol.jsc.nasa.gov/sseop/clickmap/> Maps at the CIA (public domain): <https://www.cia.gov/library/publications/the- world-factbook/index.html> and <https://www.cia.gov/library/publications/cia- maps-publications/index.html> NASA Earth Observatory (public domain): <http://earthobservatory.nasa.gov/> Landsat: <http://landsat.visibleearth.nasa.gov/> USGS EROS (Earth Resources Observatory and Science (EROS) Center) (public domain): <http://eros.usgs.gov/#> Natural Earth (public domain): <http://www.naturalearthdata.com/> \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Partly Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 2\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 3\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: No Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 4\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: No Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 5\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: This paper study the seasonal patterns of Japanese overnight data from 2011 to 2019. The measurement is based on a seasonal index and expressed interval-valued data, which are a kind of symbolic data. This paper is a study paper and the contributions are minimal, whereas the work and results are promising and fit for our journal. The organization of the paper is well. However, there are certain corrections required before consider this paper for publication. 1\. The literature of this article is poor. It is recommended to provide the literature on the topic. The motivation behind the topic also needed to include. 2\. It is recommended to summarize the contributions in the introduction. 3\. Section 3 is too small, it is recommended to elaborate through detailed discussion and importance of each equation. 4\. There are several performance metrics in the literature, whereas the authors considered only few. It is recommended to evaluate through multiple metrics. 5\. It is recommended to provide the futuristic prediction about the seasons. 6\. What are the limitations identified during your study? Reviewer \#2: 1. Need to add a Comparison study. 2\. Need to update the limitation. 3\. How to deploy the research idea, detailed information required. 4\. In many figures the X-Axis name was inverted, kindly reupdate it. 5\. Overall the idea was good. 6\. The dataset is available in public, easily we can able to downloadable. That's great work. \*\*\*\*\*\*\*\*\*\* 6\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No Reviewer \#2: **Yes: **ANANDHAN K \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <figures@plos.org>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0267453.r002 Author response to Decision Letter 0 31 Mar 2022 Dear Dr. Sathishkumar V E Thank you for inviting me to submit a revised manuscript. I appreciate the time and effort you and each of the reviewers have dedicated to providing insightful feedback to my manuscript. The comments have helped significantly improve the manuscript. The following is a point-by-point response to the reviewers' comments. Response to Reviewer \#1: I wish to express my appreciation to the reviewer for insightful comments on our manuscript. The comments have helped significantly improve the paper. The following is a point-by-point response to the reviewers' comments. 1\. The literature of this article is poor. It is recommended to provide the literature on the topic. The motivation behind the topic also needed to include. Thank you for your thoughtful suggestion. I added some references on this topic to express more details and clarify the motivation. Specifically, I added sentences on lines 76-85, 144-147, 151-159 and 170. 2\. It is recommended to summarize the contributions in the introduction. I agree with your suggestion. I added a paragraph to mention the contribution of this study on lines 86-94. This contribution includes a result of the comparison study which is newly added in the revised manuscript. 3\. Section 3 is too small, it is recommended to elaborate through detailed discussion and importance of each equation. Thank you for your suggestion. I explained Eq. (3) on lines 182-184 and added sentences and a figure on lines 192-201 to discuss for details. 4\. There are several performance metrics in the literature, whereas the authors considered only few. It is recommended to evaluate through multiple metrics. Since the individual data (prefectures) in the Japanese case study does not have correct answer labels, it is difficult to derive performance metrics for an evaluation. Hence, I additionally conducted a clustering analysis with traditional seasonal index and Ward’s method to compare with the results of a suggested method and evaluate by obtained insights. I explained these results in the section named “Using traditional seasonal indices” in page 16 and on lines 411-416 and 426-439. 5\. It is recommended to provide the futuristic prediction about the seasons. I appreciate your comment on this point. Accordingly, I have added sentences on lines 390-392 and lines 423-425. I have mentioned the futuristic prediction of Japanese seasonality, based on the results of this study. 6\. What are the limitations identified during your study? I thank the reviewer for this comment. I additionally mentioned the limitations on lines 466-468 and 474-478. Response to Reviewer \#2: I wish to thank the reviewer for thoughtful comments on my manuscript. The comments have helped significantly improve the paper. The following is a point- by-point response to the reviewers' comments. 1\. Need to add a Comparison study. I wish to thank for this comment. The results of the additional work are shown in the section named “Using traditional seasonal indices” with Fig 15, 16, 17 and 18 in page 16. The discussion for the results were mentioned on lines 411-416 and 426-439. 2\. Need to update the limitation. Thank you for this comment. I additionally mentioned the limitations on lines 466-468 and 474-478. 3\. How to deploy the research idea, detailed information required. I appreciate the comment on this point. I added a future issue and its details on lines 479- 487. 4\. In many figures the X-Axis name was inverted, kindly reupdate it. Thank you for the review’s suggestion. I confirmed that many figures were automatically rotated when making a combined pdf file in the submitting process. Specifically, Fig. 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 17 and 18 were rotated at 90 degrees in the resubmitted pdf file. I consider that this problem might be caused by the submitting system. 5\. Overall the idea was good. 6\. The dataset is available in public, easily we can able to downloadable. That's great work. Thank you so much for these comments. As the review’s pointing out, all results in this study are derived from Japanese government open data. I am glad to hear that. Sincerely, Yu Ogasawara 10.1371/journal.pone.0267453.r003 Decision Letter 1 V E Sathishkumar Academic Editor 2022 Sathishkumar V E This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 11 Apr 2022 New seasonal measurement with stability and clustering seasonal patterns: a case study in Japan from 2011 to 2019 PONE-D-21-39279R1 Dear Dr. Ogasawara, We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements. Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication. An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at <http://www.editorialmanager.com/pone/>, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to- date. If you have any billing related questions, please contact our Author Billing department directly at <authorbilling@plos.org>. If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact <onepress@plos.org>. Kind regards, Sathishkumar V E Academic Editor PLOS ONE Additional Editor Comments (optional): Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation. Reviewer \#1: All comments have been addressed Reviewer \#2: All comments have been addressed \*\*\*\*\*\*\*\*\*\* 2\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 3\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 4\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: No Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 5\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 6\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: The authors addressed all the recommended comments and this version may be considered for publication in this journal. Reviewer \#2: The author addressed all the reviewer comments. The manuscript is written well in standard English. All the graphs are plotted now in a proper manner. \*\*\*\*\*\*\*\*\*\* 7\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No Reviewer \#2: **Yes: **ANANDHAN K 10.1371/journal.pone.0267453.r004 Acceptance letter V E Sathishkumar Academic Editor 2022 Sathishkumar V E This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 13 Apr 2022 PONE-D-21-39279R1 New seasonal measurement with stability and clustering seasonal patterns: a case study in Japan from 2011 to 2019 Dear Dr. Ogasawara: I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department. If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact <onepress@plos.org>. If we can help with anything else, please email us at <plosone@plos.org>. Thank you for submitting your work to PLOS ONE and supporting open access. Kind regards, PLOS ONE Editorial Office Staff on behalf of Dr. Sathishkumar V E Academic Editor PLOS ONE [^1]: The authors have declared that no competing interests exist.

# Introduction Plants interact with a broad range of endophytic microorganisms, ranging from fungi, such as arbuscular mycorrhizal fungi (AMF), to bacteria, including rhizobia. Most of these interactions are neutral or mutualistic, with only a small percentage being pathogenic. However, endosymbionts are not limited to the rhizosphere but are also detected in above-ground plant organs such as stems and leaves. The term ‘endophyte’ as defined by Partida-Martínez and Heil is used in this study to refer to endosymbionts occurring in plant tissue without negatively affecting the host plant. Bacterial leaf nodule symbiosis is an intimate endosymbiosis in which endophytes are housed in galls or ‘nodules’ in the leaves of several species in the flowering plant families Dioscoreaceae, Primulaceae, and Rubiaceae. In the latter two families, the endosymbiosis is suggested to be unique due to the presence of vertical transmission, the obligate nature, and the high specificity of the interaction. The presence of vertical transmission in combination with possible ecological advantages may transform beneficial interactions into long- term and possibly obligate mutualisms. Bacterial leaf nodule symbiosis evolved into a highly specific and obligate endosymbiosis in which one specific host plant species interacts with one single bacterial species. This leaf symbiosis is the most prevalent in Rubiaceae, where the endophytes were identified as *Burkholderia* species. Papers like von Faber, Lersten and Horner, and Miller are the most extensive microscopic reports on the nodulated species *Pavetta zimmermanniana*, *Psychotria bacteriophila*, and *Psychotria kirkii* (the latter two being synonyms of *Psychotria punctata*). These studies not only detected bacterial microorganisms in leaf nodules, but also found them in other plant structures such as vegetative and reproductive buds, sepals, and pyrenes (i.e., the stones of fleshy drupes containing the seeds). Although, vertical transmission of the endophytes to the next generation was suggested by von Faber–based on the detection of bacterial microorganisms in the gynoecium (micropyle region) and the pyrenes (cavity surrounding the embryo)–to date, strong evidence is still lacking. Van Oevelen et al. was the first to molecularly identify the endophyte in the leaf nodules of *Psychotria punctata* as *Candidatus* Burkholderia kirkii and Lemaire et al. detected endophytic DNA in seeds, flowers, shoots, and leaves. Despite this progress, the exact location of *Candidatus* Burkholderia kirkii still remains unclear. Here, we combine two methods (i.e., molecular screening and fluorescent *in situ* hybridisation) to study the location of the host specific endophyte, *Candidatus* Burkholderia kirkii, in the seeds of the nodulated host plant *Psychotria punctata*. We hypothesize that, if the bacterial endophytes are vertically transmitted, they should ideally be located in close proximity to the embryonic apical shoot meristem as this facilitates their transmission to the vegetative and reproductive structures of the next generation of host plants. In addition to the localisation of the endophytes in the seeds, we also screened other suggested locations of the bacterial endophytes within the host plant using the same approach. # Materials and methods ## Plant material Leaves, vegetative buds, flower buds, flowers, pyrenes, twigs, and roots of *Psychotria punctata* were obtained from plant specimens grown at the Botanic Garden Meise (accession numbers BR19536779, BR20010513-92, BR20001943-58, BR19951273-22, and BR20021526-47). ## Molecular analysis For the molecular analysis, all plant structures, sampled in three-fold, were surface sterilised. Leaves, one-year-old and older lignifying twigs, roots, flowers, vegetative and flower buds were surface sterilised with 70% (v/v) ethanol and 1.6% (w/v) sodium hypochlorite for ten seconds each and the samples were subsequently rinsed with sterile water. Pyrenes were treated with 70% (v/v) ethanol for three minutes and 1.6% (w/v) sodium hypochlorite with 0.1% (v/v) Tween20 for 20 minutes before rinsing them with sterile water. Anthers and gynoecia were dissected from the flowers, whereas endosperm, sclerified endocarp and embryos were prepared from the pyrenes. To remove the embryo, an incision was made along the ventral intrusion at the opposite side of the funicle. This incision facilitated the rupture of the pyrene and the isolation of the embryo from the embryonic cavity. For smaller plant structures, multiple samples of the same plant individual were combined, i.e., five gynoecia, ten anthers, and 15 embryos. Plant tissues were ground in liquid nitrogen with a tissue homogeniser and their total genomic DNA was extracted using a modified cetyltrimethylammonium bromide (CTAB) protocol. Polymerase chain reactions (PCR) were performed using 16S rDNA, *recA* and *gyrB* bacterial primers, specifically designed for *Burkholderia* species. A positive control with *Burkholderia caledonica* and a negative control were included in each PCR run. The amplified products were visualised using gel electrophoresis and the positive results were purified and sequenced in both directions by Macrogen sequencing facilities (Macrogen Europe, Amsterdam, Netherlands). The raw sequences were assembled and screened for potential sequencing errors in Geneious v9.1.8 (Biomatters, Auckland, New Zealand). After assemblage, the consensus sequences were compared to sequences present on GenBank with the BLAST program ([www.ncbi.nlm.nih.gov/BLAST](http://www.ncbi.nlm.nih.gov/BLAST)) and were attributed to species level based on 99% or more sequence similarity. ## Microscopic analysis The samples (i.e., three leaves, five embryos, three vegetative buds, and three flower buds) for microscopic analysis were fixed in 4% (v/v) formaldehyde in PEM buffer (100 mM 1,4-piperazinediethanesulfonic acid, 10 mM MgSO₄, and 10 mM ethylene glycol tetra-acetic acid, pH 6.9) for two hours under vacuum. After washing in phosphate-buffered saline (PBS, Na₂HPO₄ 0.148 g, KH₂PO₄ 0.043 g, NaCl 0.72 g, NaN₃ 0.9 g in 100 mL distilled water, pH 7.1), the samples were dehydrated using a graded ethanol series (30, 50, 70, 85, 100% (v/v)). Subsequently, they were gradually impregnated with and embedded in LR White acrylic resin (medium grade, London Resin Company, UK) using polypropylene embedding moulds and polymerized at 37°C for three days. Semi-thin sections of 350 nm were cut using a diamond knife mounted on a Leica UC7 ultramicrotome (Leica Microsystems, Vienna) and of each biological sample between five and ten sections were collected on polylysine- adhesion slides (Carl Roth, Germany). As the FISH probe can only detect endophytes at the surface of sections, we also employed a ‘bacteria isolating’ technique to collect endophytes from the outer surface of the embryo. To this end, three freshly excised embryos were submerged for five minutes in one droplet of sterile water on polylysine-adhesion slides. After air-drying the slides, the residue left by the embryos was fixed in 4% (v/v) formaldehyde in PEM buffer for 30 minutes and rinsed with PBS and sterile water. This ‘bacteria isolating’ technique was replicated three times. Fluorescent *in situ* hybridisation (FISH) was performed as outlined by Daims et al. using the 5’-Alexa-555-labelled Burkho primer (Thermo Fisher Scientific, US), a specific primer designed for *Burkholderia*. The slides were mounted in Citifluor AF2 anti-fade agent (Agar Scientific, UK) and observations were made with a Nikon Eclipse Ni-U microscope equipped with a Nikon DS-Fi1c camera (Nikon, Japan) and the following filter cubes: FITC (excitation 480/20 BP; dichroic mirror 505 LP; emission 410 LP) referred to as the green channel and TRITC (excitation 535/30 BP; dichroic mirror 565 LP; emission 580 LP 541/572) referred to as the red channel. The green channel was used to observe plant tissue autofluorescence, while the red channel visualised the 5’-Alexa-555-labeled probe (excitation 555 nm, emission 580 nm). The use of merged epifluorescence images of both channels facilitated interpretation of the tissue-specific localisation of the endophyte. The hybridisation protocol of Daims et al. with the Burkho probe was successfully tested on leaf cross sections through a bacterial leaf nodule. After observation of FISH labelling, cover slips were carefully removed, and the slides were thoroughly rinsed with demineralised water to remove the anti-fade agent. Subsequently, the slides were stained with 1% (w/v) toluidine blue O (Merck, Germany) (TBO) in 1% (w/v) Na₂B₄O₇ for 20 seconds at 50°C. After rinsing with demineralised water, slides were mounted with DePeX (VWR international, Belgium). Observations were made with a Nikon Eclipse Ni-U bright field microscope equipped with a Nikon DS-Fi1c camera. # Results ## Molecular analysis Endophytic DNA was detected in the leaves, vegetative buds, flower buds, anthers, gynoecia, embryos, and twigs with the specific *Burkholderia* primers 16S rDNA, *recA*, and *gyrB*. *Burkholderia* DNA was not detected in the roots, the endocarp, nor in the endosperm of the seeds. All the sequences were identified with BLAST (99% threshold) as *Candidatus* Burkholderia kirkii. Based on these results, plant structures for detailed microscopic analysis were determined, which included leaves, vegetative buds (*inter alia* shoot apical meristem), flower buds (*inter alia* gynoecium and anthers), and embryos. ## Fluorescent *in situ* hybridisation The bacterial leaf nodule symbiosis is characterised by macroscopically visible bacterial leaf nodules that are, in case of *P*. *punctata*, randomly distributed in the lamina of the leaves. The nodules, located in the spongy parenchyma of the mesophyll, are lined by two or three cell layers of compressed mesophyll cells. FISH labelling allowed identification of the nodule bacteria as *Burkholderia*. This result indicated that the used protocol and probe allows for *in situ* detection of *Burkholderia* endophytes. The endophytic colony was interspersed with strands of parenchymatous cells. No endophytes were detected in the extracellular spaces of the spongy parenchyma of the lamina. In vegetative buds, the leaf enclosed chamber (LEC) is formed by the interpetiolar stipules positioned above the shoot meristem that protect the developing leaf primordia. This LEC is further protected by decussately arranged pairs of leaves and their still enclosing interpetiolar stipules. In addition to the leaf primordia, multicellular, dendroid trichomes or colleters were detected in the LEC. These structures were distinguishable from neighbouring cells due to their high level of autofluorescence and their densely stained cytoplasm, which suggests a high cytoplasmic activity, potentially correlated with mucus production. At the abaxial side of the enclosing leaves, a different multicellular type of trichomes was observed, characterised by the absence of dense cytoplasmic staining. *Burkholderia* bacteria were detected in the mucus that surrounds the colleters in the LEC. Furthermore, endophytes, yet at a lower abundance compared to the LEC, were also detected in the mucus between the trichomes at the abaxial side of the enclosing leaves. To analyse the presence of the endophytes in flowers, longitudinal sections through developing flower buds were made. Flowers of *P*. *punctata* are pentamerous, with an inferior gynoecium, bilocular ovaries, uniovulate locules and anatropous ovules. The observed flower buds were in a late developmental stage, judging by the differentiation state of sepals, petals, and anthers. Endophytes were not found in the locules (only one of the two locules can be observed) of the gynoecium (data not shown). The developing stamens and style (not shown) are still enclosed by the petals. FISH labelling only detected endophytes in a mucus-filled space between the sepals and petals. At the base of the latter space, colleters with intensely TBO stained cytoplasm were observed. Although molecular techniques indicated the presence of endophytic DNA in the anthers, no endophytes were detected in these structures by FISH (data not shown). The fruits of *P*. *punctata* contain one or two hemispherical pyrenes that are characterised by a T-shaped intrusion at the ventral side following the longitudinal axis. When the pyrene is cut in half, a cavity is observed above the ventral intrusion close to the funiculus in which the embryo is located. Mucus was observed in this embryonic cavity and between the cotyledons of the embryo. Despite the presence of mucus in the embryonic cavity, endophytic DNA was only found in embryos and not in the endosperm or sclerified endocarp samples. To assess the presence of endophytes in close association with the embryo, we employed two methods. The first method is based on the collection of bacteria from mucus covering the embryo surfaces (‘bacteria isolation method’). FISH labelling of the extracted residue allowed detection of the *Burkholderia* endophyte. In case the endophytes are transmitted vertically, one would expect to detect them in close proximity of the embryonic shoot apical meristem, flanked by the cotyledons. Therefore, longitudinal sections (perpendicular to the surface of the cotyledons) of embryos were investigated using FISH. The second method allowed detection of endophytes–which were also stained with TBO–in the intercotyledonary space of several embryos. In addition to the intercotyledonary space, *Burkholderia* bacteria were also observed at the outer surface of the embryo. Endophytes were not detected in the embryonic tissue nor randomly dispersed over the slide. # Discussion This study, which combined FISH and PCR techniques, is the first to report and identify the host specific endophyte *Candidatus* Burkholderia kirkii in several locations in the host plant. Detection of *Burkholderia* bacteria in the pyrene of *P*. *punctata* demonstrates how these endophytes are vertically transmitted to the next plant generation. The detection of the endophyte in the leaf enclosing chamber (LEC) above the shoot apical meristem is in line with the findings of previous studies. The location of the latter endophytic colony is an essential part of the symbiotic cycle of the bacterial leaf nodule symbiosis as it enables the transmission of endophytes to the leaves as well as to (axillary) vegetative and reproductive buds. Besides the presence of endophytes, mucus-producing colleters were also present in the LEC. The production of mucus most likely facilitates the passive infiltration of the endophytes towards the leaves during the early leaf developmental stages via stomatal pores. Endophytes were also detected in mucus residue between the trichomes at the abaxial side of the leaves forming the LEC. This can be explained by the formation of successive LECs above the shoot apex. In the LEC, newly developed leaf primordia are completely surrounded by mucus and endophytes. As the latter leaves elongate and cover the LEC, a small amount of mucus and endophytes may still be present at abaxial leaf surfaces. The low abundance of endophytes may be an indirect result of the absence of mucus- producing colleters at the leaf surface and the decreasing mucus production of the colleters at the enclosing stipules, as mucus is necessary to sustain the endophytic colony. This process might also explain why endophytic DNA was detected in the twig samples, as internodal stem segments also develop in the LEC and endophytic colonies cannot be maintained as mucus-producing trichomes are absent. The detection of the endophyte in the flower buds, between the sepals and the petals, supports the findings of earlier studies. Moreover, endophytic DNA was confirmed in the gynoecia. Although the endophytes were not observed during the microscopic analysis of multiple biological replicates, their presence was suggested by molecular detection. This molecular detection corroborates the observations of von Faber and the current hypothesis, which states that the endophytes infiltrate the locules of the gynoecium during early floral development, when the carpels are not yet fused. This mode of transmission may be possible as Figueiredo et al., in an ontogenetic study of *Psychotria carthagenensis* flowers, observed a (temporary) opening in developing gynoecia with carpels that are not completely fused. There are two hypotheses that explain the infiltration of the ovule by the endophyte: von Faber suggested that the bacteria infiltrate the ovule together with the pollen tube when it fuses with the ovule, while Miller, on the other hand, added that, if the embryogenesis of *Psychotria* was apomictic, placental secretion could also facilitate the transmission of the endophyte to the ovule, which in fact has been confirmed for *Ardisia* (Primulaceae). In other mutualistic or parasitic interactions throughout the angiosperms, endophytes may infiltrate the seeds via other pathways including the gametes, the shoot meristem, vascular tissue in the funiculus, the micropyle, or through the seed coat. Our results do not unequivocally refute either of these hypotheses, but the absence of endophytes in vascular tissue seems to suggest that the endophytes infiltrate the ovule via the micropyle. *Candidatus* Burkholderia kirkii was also genetically identified in embryos. FISH labelling supported these results as endophytes were detected in close proximity to the embryo, more specifically in the intercotyledonary space. These findings provide strong proof for the presence of endophytes near the embryonic shoot meristem, as suggested by earlier studies. Although Miller was not able to detect bacterial microorganisms with a transmission electron microscope, we were able to detect *Burkholderia* in the embryonic cavity of five biological replicates. The presence of endophytes above the embryonic shoot meristem may be a first and crucial step towards developing successive endophyte-housing LECs as soon as the first leaves appear after germination and vegetative growth is initiated. In other mutualistic or parasitic interactions, seed microbiota is mostly detected near the seed coat and present in grooves, while other bacteria are detected in the endosperm or embryonic tissue. In contrast to the endophytes located near the seed coat, the bacteria present in the embryonic tissue are unlikely to be contamination of external microbiota and are vertically transmitted microorganisms. In bacterial leaf nodule symbiosis, the close proximity of the endophytes to the embryonic shoot meristem validates the vertical transmission of the endophyte and the obligatory character of the endosymbiosis. Although endophytic DNA was found in the anthers, complementary FISH experiments on transverse sections through several flower buds did not confirm the presence of endophytes. Transmission of endophytes via pollen (both internally, enclosed within the pollen wall, as externally, attached to the exine) has been reported in other symbiotic interactions, e.g., the bacterial endophyte *Enterobacter cloacae* in Mediterranean pines, and endophytic fungi (*Alternaria alternata* and *Cladosporium sphaerospermum*) in forbs. Other studies suggested the possibility of (vertical) endophyte transmission via pollen for bacterial leaf nodule symbiosis as well, but they did not provide conclusive evidence. Besides transmission via pollen, other modes of transmission, through herbivorous insects or free-living soil bacteria, have been proposed. Our study can neither confirm nor reject these modes of transmission. However, transmission via free- living soil bacteria seems to be less plausible as no endophytic DNA was detected in the roots of the host plant. Even though the number of endophytes in some plant structures appears to be low (Figs), these observations are reliable due to the specificity of the FISH-probe and the reproducibility of the observations in multiple biological replicates. Comparison of the microscopically observed abundances among the analysed plant structures indicated a decrease from the vegetative apex to the developing flower and the embryo. This is in line with findings of previous studies investigating the location and abundance of endophytes in *Psychotria* using molecular techniques as well as those of other vertically transmitted symbionts, such as *Burkholderia phytofirmans* in *Vitis vinifera* and *Xanthomonas fuscans* in *Phaseolus vulgaris*. Based on the above-mentioned decline in endophyte abundance, a low amount might be expected in gynoecia and anthers and this may explain why they were not detected in the latter structures using FISH. The occurrence of vertical transmission of the endophyte combined with the high specificity and the obligatory nature of the endosymbiosis underline the significance of this interaction for the host species. In other interactions, the endophytes are vertically transmitted due to their importance for seed germination and growth of the seedlings, which increases their survival chance. Genomic analysis of *Candidatus* Burkholderia kirkii demonstrated the absence of genes or pathways for nitrogen fixation or hormone synthesis that could influence plant growth. However, a specific pathway was detected in the endophytic genome for the production of a C₇N aminocyclitol, i.e. kirkamide, which protects the host plant against insects. Although our results did not contribute to the functional aspects of the endosymbiosis, the obligate nature of the interaction is reinforced by the confirmation of vertical transmission. To conclude, our study is the first to unveil the location of the *Burkholderia* endophytes in the seeds, unequivocally confirming vertical transmission. The host specific endophytes were detected in close proximity to the shoot apical meristem of the embryo. This is the first step in the establishment of a key colony in the LEC above the shoot apical meristem, which enables the transmission of endophytes to new leaves and inflorescences. We obtained strong evidence by combining two independent but complementary techniques. The molecular screening enabled the localisation of endophytes in different plant structures, even in not earlier suggested locations. Fluorescent *in situ* hybridisation enabled determination of the *in situ* location of *Burkholderia* endophytes in several plant structures. The results of this study show that adopting a similar approach to other interactions would generate new insights in how endophytes are transmitted. # Supporting information We thank Frank Van Caeckenberghe and Marc Reynders of Botanic Garden Meise for the opportunity to sample in the living plant collection. [^1]: The authors have declared that no competing interests exist.

# Introduction Alzheimer’s disease (AD) is a neurodegenerative disease characterized by abnormal accumulation of amyloid-β (Aβ) and progressive impairments of cognitive abilities. When amyloid precursor proteins are cleaved by β- and γ-secretases on the membrane of neurons, Aβ peptides are released in the extracellular regions and begin to misfold into soluble oligomers and insoluble plaques. Among several Aβ aggregate types, oligomers are considered to be responsible for major biochemical causes of neurodegeneration, such as neuroinflammation, synaptic/neuronal injury, neuronal ionic homeostasis breakdown, oxidative stress, and tau abnormalities in AD brains. It is well known that increasing levels of neurotoxic Aβ oligomers in the brain impair cognitive behaviors and synaptic plasticity. As AD is a chronic disorder requiring timely accumulation of Aβ, transgenic animal models overexpressing human amyloid precursor proteins are often preferred in preclinical investigations of Aβ-associated pathophysiology. However, these models are not suitable for Aβ oligomer-specific experiments due to the simultaneous and uncontrolled production of diverse pathogenic species, such as amyloid precursor proteins, Aβ (1–42) monomers, Aβ (1–40) monomers, Aβ oligomers, Aβ plaques, and enzymatic cleavage byproducts. To conduct in vivo studies in an Aβ-species-controlled manner, Aβ-infused animal models are suggested. Pathogen-induced rodent models are commonly used in neurodegenerative studies. Injection of scopolamine or 1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine are well established methods to induce acute onsets of cognitive deficits and movement disorders, respectively, in rodents. Unlike these chemical pathogens that are injected intravenously, Aβ is injected directly into the intracerebroventricular (ICV) region of the brain with controlled aggregate species and peptide concentrations. Aβ injection rodent models acutely exhibit hippocampal-dependent learning and memory deficits in Y-maze, passive avoidance, fear conditioning, Morris water maze, and novel object recognition tests and Alzheimer-like pathological alterations such as decreased long-term potentiation, increased inflammation, decreased acetylcholine levels, activated astrocyte/microglia, and amyloid deposition. In addition to the controllable Aβ conditions, benefits to bypass the ageing process of transgenic models for AD-like symptom and pathology onsets have allowed researchers to shorten the in vivo efficacy evaluation step of AD drug candidates directly regulating Aβ. The Aβ-infused animal models are mostly investigated for less than a month since the peptide injection because the onset of AD-like phenotypes are promptly made, a recent study reported that memory deficits and synaptoxicity of mice became gradually worse in 40 days since the single injection of Aβ(1–42). Currently, multiple behavioral assays are available for testing hippocampal- dependent spatial memory (e.g., the Morris water maze), contextual memory (e.g., fear conditioning) or working memory (e.g., the Y-maze) functions in AD animal models. These assays focus on memory loss, which is one of the most common signs of AD. However, the loss of memory is not the only AD symptom, and overreliance on these assays can contribute to the failure of translational studies across species. Other clinical symptoms related to brain functions, such as sensory processing, have not been investigated fully by the AD mouse research community due to the difficulty in assessing these functions in a mouse model. One potential research strategy is the use of test assays that are applicable in both humans and mice through the delivery of cross-species comparative parameters. In this regard, the use of electrophysiological responses might be a suitable means to assess cognitive deficits in a translational study of an AD mouse model. Electrophysiological signals generated by neural circuitry seem to be preserved among species. One of the most commonly used paradigms for measuring neural function is the presentation of sensory input, such as the auditory oddball paradigm. Event-related potentials (ERPs) have been investigated as biomarkers of cognitive decline and disease severity in patients with mild cognitive impairment (MCI) and Alzheimer’s disease. Auditory ERP traces have several positive and negative peaks after sound onset, representing a well-defined brain response to a sensory process. Generally, early ERP components, appearing as peaks within \~ 200 ms, index rapid and automatic brain processes, while later components represent slower and more complex functions. The early components have been presumed to be suited for cross-species investigation since they represent automatic brain functions involving auditory discrimination without consciousness. Previous studies suggested that rodents show auditory ERP components similar to humans with a systematically short latency. Umbricht et al. presented the same auditory stimuli to both humans and mice and generated a formula for calculating the difference between latencies (Latency (human) = 1.67 \* Latency (mouse) + 37.61 ms). The early sensory component of the auditory response includes an initial positive peak at approximately 50 ms and 20 ms (P1), an initial negative peak at approximately 100 ms and 40 ms (N1), and a second positive peak at approximately 200 ms and 120 ms (P2) in both humans and mice, respectively. The function of each ERP component has been investigated in a human study. P1 is related to sensory gating, which is suppressed during repetitive auditory stimulation. N1 is linked to early attention and sensory memory-related variables and is related to detecting changes in sensory input. N1 overlaps with the auditory mismatch negativity (MMN) specific to the oddball sound deviant from expected stimulus, which shows exaggerated and delayed negativity associated with the discrimination of deviant stimuli. The MMN classically is calculated as a negative deflection near the N1 period; it is a differential waveform obtained by subtracting the ERP for the repetitive tone from the ERP for the deviant tone. P2 is considered to reflect the endogenous neural mechanism linked to initial conscious awareness. Anomalies in these components have been reported in various brain disorders and their animal models. Although consciousness-related late components of ERPs have been the main focus of investigations in Alzheimer’s patients, several studies have shown a correlation between AD pathology and alterations in early components. Compared to that in normal older control subjects, the amplitude of the MMN has been shown to be attenuated in AD, MCI and pre-symptomatic individuals who have mutations in an AD-related gene. In mild AD, the N1 amplitude also been shown to decrease in response to both standard and deviant auditory stimuli. Based on these previous results of MMN and N1 reductions, we hypothesized that novel sounds may be hard to perceive in Alzheimer’s disease, resulting from impaired automatic attention to the deviant stimulus. Clinical measurements of such ERP alterations have not yet been assessed in a mouse model of AD, whereas diminished ERP components have been observed in the mouse model for schizophrenia. In the AD mouse model, EEG studies investigating brain dysfunction have been limited to analyses of spontaneous EEG signals. Moreover, spectral analysis of spontaneous EEG signals has yielded contradictory results in the AD mouse model, showing either an increase or decrease in the amplitude of delta or subdelta frequency signals in 5xFAD, PLB1 and APP/PS1 transgenic mice. In this study, we aim to evaluate early components of the auditory ERP that represent novelty discrimination abilities in AD model mice. We characterized the auditory ERP response using an oddball paradigm in a mouse model mimicking AD pathology a result of Aβ injections into the ventricle to study its neurophysiological influence. # Materials and methods ## Animals For the generation of Aβ (1–42) infusion mice, 6-week-old Imprinting Control Region (ICR) male mice were purchased from Orient Bio Inc. (Seoul, Korea) and habituated for 4 days. Mice were maintained in a sterile laboratory animal breeding room at the Korea Institute of Science and Technology with stable temperature and humidity. Mice were exposed to a controlled 12-hour light-dark cycle with food and water ad libitum. Nineteen mice were purchased for the study (n = 7 for vehicle group, n = 12 for Aβ group). After all experiments, animals were euthanized using carbon dioxide gas for over 10 minutes under deep anesthesia induced by intraperitorial injection of 2% avertin (250mg/kg). All animal experiments were performed in accordance with the National Institutes of Health guide for the care and use of laboratory animals (NIH Publications No. 8023, revised 1978). The animal studies were approved by the Institutional Animal Care and Use Committee of Korea Institute of Science and Technology (Approval Number: 2016–035). ## Aβ-infused mice model We produced the model according to previously demonstrated method. For each injection, 100 μM Aβ (1–42) was prepared (10% DMSO \[dimethyl sulfoxide\] in PBS). All 100 μM Aβ samples were incubated at 37°C for 7 days. After incubation, mice were anaesthetized with 4% avertin(250mg/kg) and assessed the foot or tail pinch response to confirm adequate anesthesia. While anesthetized, mice were placed on a warm mat to maintain its body temperature. 5 μL of vehicle (10% DMSO in PBS), or Aβ solution was injected into the lateral ventricle by intracerebroventricular (ICV) injection. Briefly, using their thumb and index finger, we tightly hold down the mouse skin of the forehead and drag the skin behind to minimize skull movement under skin. Without removing the skin, we positioned and inserted the needle into the lateral ventricle (3.8 mm depth). Then we slowly inject 5 μL of vehicle or Aβ solution over 5 s and wait 3 to 5 s before removing the syringe for diffusion. After injection, mice were placed in cage above the warm pad and watched for abnormalities until it regains consciousness. We postoperatively monitor the mouse behavior for several days. ## Y-maze test The Y-maze apparatus was constructed with black plastic and composed of three equally spaced arms (40 L × 10 W × 12 H cm). The mice were placed at the end of one arm and allowed to freely explore their surroundings for a 12-min session. An arm entry was defined as the tip of the tail of the mouse entering the arm entirely. Entries of all mice were recorded manually. A different entry than the previous two entries was defined as an alternation, and the following equation was used to calculate spontaneous alternation behavior. <img src="info:doi/10.1371/journal.pone.0230277.e001" id="pone.0230277.e001g" /> % alternation = 100 × \[ ( number of alternations ) / ( total number of arm entries \- 2 ) \] ## Surgery for EEG electrodes To implant the microscrew electrodes, surgical procedures were performed under deep anesthesia with intraperitoneal injection of a ketamine (120 mg/kg) and xylazine (6 mg/kg) cocktail. Five minutes after i.p. injection, mouse toe and tail pinch was done to check whether the anesthesia is sufficient. When mouse did not show any reaction to the toe and tail pinch, the animal placed on to the stereotaxic apparatus (Kopf Instruments, Tujunga, CA, USA). Sterilized microscrew electrodes (0.8 mm diameter, Asia Bolt, South Korea) were fixed onto the skull surface of the frontal (anteroposterior, 2 mm; mediolateral, 1 mm) and parietal cortex (anteroposterior, 2 mm; mediolateral, 2 mm), with ground/reference electrodes on the occipital bone above the cerebellum. The electrode coordinates were determined according to the mouse atlas. Dental cement (VertexTM Self-Curing, Vertex-Dental, Netherlands) was applied to secure the position of the electrodes. After surgery, the incision was closed with sterile suture and antibiotic ointment (sodium fusidate, 20mg/g) was applied to prevent infection. Mice were treated with analgesic ointment (ketoprofen, 30mg/g) and underwent one week of recovery. ## Auditory oddball test The test was performed in a transparent cylinder (10 cm in diameter, 25 cm in height), which was placed in the center of Faraday cage (55 cm x 60 cm x 65 cm, light and sound proof). A 60 dB white noise was presented throughout the experiment as background noise (White noise generator, San Diego Instrument, CA, USA). For acclimation, we placed the bedding from the home cage at the cylinder floor and left the animal for at least 15 min prior to the experiment. The auditory oddball test was composed of two pure tone sound stimuli (2 kHz as the standard tone and 4 kHz as the deviant tone, 80–85 dB, 10 ms) presented via four surround speakers (Dongguan edifier technology, China). The temporal sequence and frequency were predetermined by voltage output based on a custom-built MATLAB program (Mathworks, Inc. Natick, MA, USA). The digital output was sent to the speaker and high-level input port in the amplifier via a digital-to-analog converter (NI 9263 measurement system, National Instruments, TX, USA). The ratio of standard to deviant tones was 9 to 1, where the deviant tone was presented randomly with the restriction of having \>3 standard tones between two deviant tones. The whole session consisted of 1000 stimulus presentations with variable interstimulus intervals (1–1.5 s) and lasted approximately half an hour. ## EEG acquisition EEG data were collected in a Neuroscan SynAmps2 amplifier system (Compumedics, Charlotte, NC, USA) at a 2 kHz sampling rate. The impedance of each electrode was kept below 300 kΩ, and online filtering (60 Hz notch and 1–200 Hz bandpass) was applied. ## ERP analysis Data was analyzed using the MATLAB (Mathworks Inc., Natick, MA, USA). We obtained ERP by following procedure: First, we filtered the EEG with a band-pass filter (cut off frequencies = 0.5 and 59 Hz, FIR filter type). Next, we extracted 1 s of epochs from -0.4 s to 0.6 s with respect to the auditory stimulus. Any epoch with high voltage with absolute voltage over 1 mV was considered to be contaminated and then excluded in further analysis. Then, we corrected the baseline by subtracting the mean EEG values in the prestimulus period from -100 to 0 ms. Lastly, the ERP was obtained by average individual epochs of EEG separately for the standard and deviant stimuli. To balance the numbers of two conditions, we used only the epoch with the standard stimulus preceding the epoch with deviant stimulus, of which number is roughly 100 epochs per mouse. The peaks of ERP were obtained based on peak-detection algorithm. Previous studies reported that the auditory ERP in mice typically show peaks at latencies within 120 ms after stimulus onset. Each peak was considered equivalent to human P1, N1, and P2 components. We detected the ERP peaks for individual mice using peak detection algorithms based on max-min amplitude in the window of interest. Specifically, P1, N1 and P2 peaks were defined at maximum points between 10 and 25 ms, minimum locations between 25 and 45 ms, and maximum locations between 45 and 200 ms, respectively. The ranges were modified from the method used in previous mice studies. All detected peaks were visually confirmed. The peak detection results are summarized in. ## Statistical analysis A Lilliefors normality test was used to test the null hypothesis that the data exhibited a normal distribution. ERP waveforms and detected ERP components were compared between groups or conditions by a Student’s t-test or Kruskal-Wallis test where appropriate. A two-way ANOVA was performed to test effect of group (vehicle vs. Aβ-infusion) and stimulation conditions (standard vs. deviant) and their interaction. A sign test was used to test whether the negative deflection near N1 of the differential ERP waveforms showed significant negativity compared with 0. An alpha level of 0.05 was considered a significant result. # Results ## Behavioral performance Prior to electrophysiological recordings, vehicle (6-week-old males, N = 7) or Aβ (6-week-old males, N = 12) injected mice were subjected to Y-maze spontaneous alternation tests to assess working memory ability on the third days of injection. Total arm entries and percentage of alternations were calculated to assess locomotion and spatial working memory abilities, respectively. We found that the alternation rate was significantly reduced in the Aβ group compared to the vehicle group (Student’s t-test, p = 0.007), while locomotion was not affected (p = 0.45). The working memory of 6-week-old ICR male mice was confirmed to be reduced by the ICV injection of Aβ. ## ERP waveforms At one month after ICV injection (32.1 ± 2.1 days, ranging from 29 to 36 day after injection), EEGs were recorded from frontal and parietal cortex during presentation of auditory oddball paradigm. Due to implantation of EEG electrodes and recovery procedures, there was a three-week gap between behavioral and EEG experiments. presents the event-related potentials (ERP) elicited by standard (2 kHz, 90%) and deviant (4 kHz, 10%) sound stimuli averaged over vehicle (N = 6) and Aβ-infused (N = 7) mice. The time windows showing significant differences between responses to standard and deviant sound presentations are marked by a gray shadow overlapping the ERP plots (Student’s t-test, p \< 0.05). In the vehicle group, the deviant ERP was significantly different from the standard ERP. In the frontal cortex, there was a significant difference between the two types of ERP in the very early (10–22 ms), early (27–39 ms), intermediate (73–102 ms), and late periods (150–300 ms), which correspond to time windows for the P1, N1, P2, and P3 peaks in rodents. In the parietal cortex, the two ERPs were significantly different during the very early (12–22 ms) and early periods (27–35 ms). Although the averaged ERP traces for the Aβ-infusion group were enhanced in response to the standard tone and reduced in response to the deviant tone compared to the vehicle group, the difference was not statistically significant between the groups. (, Student’s t-test, p \> 0.05). The MMN classically is calculated as a negative deflection near the N1 period of differential waveform obtained by subtracting the ERP for the repetitive tone from the ERP for the deviant tone. We compare the differential ERP waveforms for groups. presents the differential ERP traces obtained by subtracting the standard ERP from the deviant ERP in frontal and parietal cortex averaged over vehicle (N = 6) and Aβ (N = 7) infused mice. The grand-averaged differential ERP of the Aβ-infusion group was reduced in both frontal and parietal cortex. However, the differential ERP waveforms were not significantly different between groups (Student’s t-test, p \> 0.05). ## Peak amplitude of early components and MMN-like activity The amplitude and latency of the P1, N1, and P2 peaks were determined by maximum or minimum peak detection and are summarized in. Consistent with the ERP wave forms, a significantly different amplitude in the early components between responses to the deviant and standard tones was observed in the control group, while this difference was destructed in both the frontal and parietal ERP of the Aβ-infusion group. In order to analyze the effects of stimulus type and groups, we performed a two-way ANOVA on the peak values of P1, N1, and P2, and groups from the frontal ERP. We found a main effect in the stimulus condition (deviant vs standard, F(1,22) = 6.25, 19.07, and 7.03 for P1, N1, and P2, respectively, all p \< 0.05). On the other way, neither the group effect nor the stimulation type x group interaction was found to be significant. The differently elicited ERP between standard and deviant sound were known to relate with auditory sensory memory which important in ability to automatically attend to novel environment. To report the marginal tendency of the reduced differential ERP in Aβ-infused mice, we calculated the amplitude changes to deviant tones in early components, subtracting the amplitude of early components detected in the standard ERP from that of the deviant ERP. Since the shape and latency of the ERP responses to standard and deviant tones are identical in mouse, the MMN-like component in mice is supposed to be the same as the difference in N1 amplitude between deviant and standard ERPs. The waveform of the vehicle group produced robust negative deflection via sign tests (difference from zero) performed on the peaks near the N1 period (20–80 ms) in both frontal and parietal cortex (sign test, p = 0.03 same in frontal and parietal cortex), while the Aβ-infusion group failed to show statistically significant negative peaks in both cortices (sign test, p = 0.16 for frontal cortex, p = 0.22 for parietal cortex). In frontal cortex, the maximum negativity was -36.5 ± 4.4 and -25.2 ± 9.6 μV in the vehicle and Aβ-infusion groups, respectively. In parietal cortex, the negative peak was -45.2 ± 7.7 and -25.0 ± 9.1 μV in the vehicle and Aβ-infusion groups, respectively. The differences in early components between the responses to standard and deviant tones were smaller in Aβ-infused mice than in controls in both frontal and parietal cortex. However, a significantly reduced amplitude was found only in the differential N1 component in the parietal cortex (p = 0.03, Kruskal-Wallis test). # Discussion Recent studies shed light on neuronal dysfunction induced by amyloid-beta levels. In accordance with this perspective, we inspected whether the increase of amyloid β disrupts neural activity related to sensory processing in a mouse model of Aβ infusion. Our aim was to determine whether the auditory oddball paradigm suggested in human patients to diagnose AD, could indicate the severity of AD pathology in the mouse. In our AD model, Aβ-infused mice, the auditory ERP did not discriminate deviant tones from the standard tones. Compared to the control group, Aβ-infused mice tended to exhibit a reduced N1 component in response to deviant sound stimuli in both frontal and parietal regions. Only the parietal region showed a significantly attenuated N1 differential response between deviant and standard tones. The amplitude of the N1 component was correlated with impaired Y-maze test performance. The shape and latency of auditory ERPs in this study are consistent with previous oddball studies conducted in rodents. The shape of ERP components in the rodent is similar to that in humans, with a systematically short latency. Additionally, researchers have debated whether the ERP component specific to novel sound is comparable across species. The mismatch negativity (MMN), which indexes novelty in the auditory environment, is represented by a negative peak around the N1 latency in the differential waveform generated by subtracting the standard tone ERP from the deviant tone ERP. In human, MMN component is distinct from N1 although N1 contribute to shape and amplitude of difference waveform between ERP to standard and deviant tone. Without the adaption process of N1 by repetitive stimuli, MMN is presented when expected stimulus does not matched to the presented one. During oddball test, participants was thought to create a model of the regularities in the acoustic environment by repetitive sound, and expect continuation of same stimulus. MMN induced by oddball sound is regarded as a signal for the regularity violation or the prediction error. Although the amplitude of N1 is larger in response to the deviant tone, the shape and latency of the ERP responses to standard and deviant tones are identical in mouse studies. Therefore, the MMN in mice is supposed to be the same as the difference in N1 amplitude between deviant and standard ERPs. In this study, we calculated the N1 difference and considered it to be analogous but not exactly the same as the human MMN since the N1 difference has been known to partially account for the MMN wave in humans. We found significantly reduced N1 differences in the parietal cortex in the AD mouse model. Additionally, although not statistically significant, the standard-deviant differences at all components in both frontal and parietal regions tended to be attenuated in the AD mouse model, which can be inferred from the lack of significant differences in the ERP responses to standard and deviant tones in terms of the whole trace in both frontal and parietal regions. Similar to the reduced N1 difference in our mouse model of AD, MCI and Alzheimer’s patients generally exhibit significantly lower MMN amplitudes than do healthy controls. Previous MMN studies with contradictory results reveal that the interstimulus interval could be important in whether a significantly reduced MMN is observed between AD and control groups. When a short interstimulus interval (\~ 1 s) was adopted, the MMN amplitude was intact compared to the control group. A longer interstimulus interval systematically decreases N1 adaptation to repetitive sound and therefore reduces MMN amplitude. The MMN has been shown to disappear when the interval was 8–10 s long. Pekkonen et al. used various interstimulus intervals and reported that the MMN amplitude was normal in AD patients when the interstimulus interval between deviant and repetitive tone was less than 1 s; however, it more sharply decreased as a function of the interstimulus interval in AD. An intact MMN at short intervals was interpreted as the normal formation of the auditory sensory memory trace in AD patients, but the retention period of the sensory memory was as short as 1 s to detect an environmental change in sound stimulation. In this study, we used a maximum of 1.5 seconds for the interstimulus interval (1–1.5 s), which is longer than previous oddball studies in rodents that adopted a 500 ms interval. Due to the lack of previous investigations of the oddball response in rodents, we cannot determine whether the interstimulus of 1.5 s in mice is optimal for dissociating AD and the control group. Our results show a trend in which the differential response to deviant and standard tones were attenuated in both frontal and parietal cortex in Aβ-infused mice. However, only the N1 difference in parietal cortex showed a significant decrease. The marginally significant results suggest that the 1.5 s interval is not long enough to observe sensory memory deficits in the AD mouse model. The deviant-specific response itself vanishes if the interval is too long to retain sensory memory in the brain, whereas the deviant-specific response is not altered in AD if the interstimulus interval is too short. Therefore, further research is necessary to determine the optimal interval to differentiate AD model mice from control group. An impaired oddball response could be induced by Aβ-related synaptic dysfunction. N-methyl-D-aspartate receptors (NMDAR) and the induction of long- term potentiation (LTP) are required for auditory discrimination. Reduction of NMDA receptors on the neural surface and blocking of LTP both in vitro and in vivo have been reported as synaptic impairments induced by Aβ. Our AD mouse model was produced by direct injection of soluble oligomeric Aβ into the cerebrospinal fluid (CSF) of the ventricle. Because soluble oligomeric Aβ is able to easily transfer via brain parenchyma and neuronal connections, Aβ in CSF may affect global brain tissue due to CSF circulation. We speculate that neural circuits generating an oddball response would be globally affected by the soluble Aβ, resulting in NMDAR receptor abnormalities. Previous studies showed that the integrity of the NMDAR system could be well represented by MMN component in both human and rodents. MMN amplitude was attenuated by NMDA antagonist, ketamine, and correlated with severance of psychotic responses after ketamine injection. Studies in schizophrenic patients extensively investigated the link between NMDAR hypofunction and MMN attenuation since they are prominent features in schizophrenia. Conversely, modulator of NMDAR function increases MMN in patient. The MMN-like component was also altered by NMDR blocker such as MK-801 or ketamine during the oddball paradigm in rat and mice. Similar to human, the NMDR blockers generally reduced MMN-like component in rodents. Moreover, direct neurotoxicity of Aβ has been confirmed in brain regions relevant to sensory processing and memory consolidation. Calcium-dependent neurotoxic events and oxidative injury following Aβ administration may result in acute cognitive impairments. In addition to behavioral characterization of AD, mouse ERP profiles of sensory processing may serve as valuable cognitive measures comparable to human cognition, facilitating diagnosis and assessment of drug effects and disease progression in the AD mouse model. # Supporting information 10.1371/journal.pone.0230277.r001 Decision Letter 0 Jo Dong-Gyu Academic Editor 2020 Dong-Gyu Jo This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 5 Jan 2020 PONE-D-19-32332 Destruction of ERP responses to deviance in an auditory oddball paradigm in amyloid infusion mice with memory deficits PLOS ONE Dear Dr. Choi, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. We would appreciate receiving your revised manuscript by Feb 19 2020 11:59PM. When you are ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. To enhance the reproducibility of your results, we recommend that if applicable you deposit your laboratory protocols in protocols.io, where a protocol can be assigned its own identifier (DOI) such that it can be cited independently in the future. For instructions see: <http://journals.plos.org/plosone/s/submission- guidelines#loc-laboratory-protocols> Please include the following items when submitting your revised manuscript: A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). This letter should be uploaded as separate file and labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. This file should be uploaded as separate file and labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. This file should be uploaded as separate file and labeled 'Manuscript'. Please note while forming your response, if your article is accepted, you may have the opportunity to make the peer review history publicly available. The record will include editor decision letters (with reviews) and your responses to reviewer comments. If eligible, we will contact you to opt in or out. We look forward to receiving your revised manuscript. Kind regards, Dong-Gyu Jo, Ph.D Academic Editor PLOS ONE Journal Requirements: 1\. When submitting your revision, we need you to address these additional requirements. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at <http://www.journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_m ain_body.pdf> and <http://www.journals.plos.org/plosone/s/file?id=ba62/PLOSOne_f ormatting_sample_title_authors_affiliations.pdf> 2\. To comply with PLOS ONE submissions requirements, in your Methods section, please provide additional information on the animal research and ensure you have included details on (1) methods of sacrifice, (2) methods of anesthesia and/or analgesia, and (3) efforts to alleviate suffering. Please ensure that you have provided this information for all parts of your study, including the icv injections. 3\. PLOS requires an ORCID iD for the corresponding author in Editorial Manager on papers submitted after December 6th, 2016. Please ensure that you have an ORCID iD and that it is validated in Editorial Manager. To do this, go to ‘Update my Information’ (in the upper left-hand corner of the main menu), and click on the Fetch/Validate link next to the ORCID field. This will take you to the ORCID site and allow you to create a new iD or authenticate a pre-existing iD in Editorial Manager. Please see the following video for instructions on linking an ORCID iD to your Editorial Manager account: <https://www.youtube.com/watch?v=_xcclfuvtxQ> 4\. Thank you for stating the following in your Competing Interests section:  I have read the journal's policy and the authors of this manuscript have the following competing interests Please complete your Competing Interests on the online submission form to state any Competing Interests. If you have no competing interests, please state "The authors have declared that no competing interests exist.", as detailed online in our guide for authors at <http://journals.plos.org/plosone/s/submit-now> This information should be included in your cover letter; we will change the online submission form on your behalf. Please know it is PLOS ONE policy for corresponding authors to declare, on behalf of all authors, all potential competing interests for the purposes of transparency. PLOS defines a competing interest as anything that interferes with, or could reasonably be perceived as interfering with, the full and objective presentation, peer review, editorial decision-making, or publication of research or non-research articles submitted to one of the journals. Competing interests can be financial or non-financial, professional, or personal. Competing interests can arise in relationship to an organization or another person. Please follow this link to our website for more details on competing interests: <http://journals.plos.org/plosone/s/competing-interests> 5.  Thank you for stating the following financial disclosure: The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. a\) Please provide an amended Funding Statement that declares \*all\* the funding or sources of support received during this specific study (whether external or internal to your organization) as detailed online in our guide for authors at <http://journals.plos.org/plosone/s/submit-now>. b\) Please state what role the funders took in the study.  If any authors received a salary from any of your funders, please state which authors and which funder. If the funders had no role, please state: "The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript." Please include your amended statements within your cover letter; we will change the online submission form on your behalf. \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 2\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 3\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 4\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 5\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: The authors conducted an auditory oddball test on the Alzheimer’s disease mouse model with infusing amyloid-beta intra-ventricle. They present data showing amyloid beta eliminated the difference of ERP signature, and showing the difference of MMN in A-beta group for the first time. They suggest this study may provide a markers of sensory memory in mouse AD model. This manuscript describes a technically sound data and the experiments conducted rigorously. The conclusion is well drawn from data. This manuscript can be re-evaluate to publish after some minor revisions performed. Minor revision 1\. affiliation has typo. 2\. line7, ‘Alzheimer’s mouse model’ may change to “Alzheimer’s disease mouse model” 3\. line 138, “method3” may be typo. 4\. line 204, “studies27” may be type. 5\. line 210, character “beta” is missing 6\. recommend to be edited by native speaker. Reviewer \#2: The manuscript reports an interesting study of an Aβ infusion mouse model of Alzheimers Disease (AD). They investigate behavioural measures of spatial memory using the Y-maze task and electrophysiological measures of sensory memory using the oddball paradigm comparing deviant and standard responses. They report impaired performance of the AD mouse model on the Y- maze task as well as evidence that deviant responses in the oddball paradigm and perhaps mismatch like response are also reduced in the AD mice compared with the control mice. 1\. The manuscript is clearly written, however this could be developed further in particular to tie together the interaction of the abeta oligomers with the likely interaction point at the level of the neuron that is then assayed by the electrophysiological tests – and how this correlates with assaying. 2\. Was any work done to corroborate the distribution of the Abeta oligomers are ICV (whether within this study of previously)? 3\. How do the concentrations and timings and electrophysiological findings correlate or differ from work done using abeta oligomer incubations in brain slice (rodent), or biopsy derived human material? This is important in the overall experimental rationale. \*\*\*\*\*\*\*\*\*\* 6\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No Reviewer \#2: No \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files to be viewed.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email us at <figures@plos.org>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0230277.r002 Author response to Decision Letter 0 9 Feb 2020 Reviewer \#1 1\. affiliation has typo. We deleted the typo. 2Division of Bio-Medical Science & Technology, KIST School, Korea University of Science and Technology, Seoul, Republic of Koreaparentheses 2Division of Bio-Medical Science & Technology, KIST School, Korea University of Science and Technology, Seoul, Republic of Korea 2\. line7, ‘Alzheimer’s mouse model’ may change to “Alzheimer’s disease mouse model” We changed the sentence as you suggested. an Aβ-infused Alzheimer’s mouse model an Aβ-infused Alzheimer’s disease mouse model 3\. line 138, “method3” may be typo. It was a style error for referencing. We corrected the error. method3 was changed to method \[4\]. 4\. line 204, “studies27” may be type. studies27 was changed to studies \[27\] 5\. line 210, character “beta” is missingWe fixed the error. A�- was changed to Aβ- Reviewer \#2: 1\. The manuscript is clearly written, however this could be developed further in particular to tie together the interaction of the abeta oligomers with the likely interaction point at the level of the neuron that is then assayed by the electrophysiological tests – and how this correlates with assaying. \[Reply\] We agree with the reviewer’s opinion of adding value by collimating the pathological and electrophysiological changes by A infusion. Since the pathological data can be obtained after the experiment, it would be desirable to record EEG as close as possible to the injection time point. But the 4-week delay in EEG recording was inevitable to reach sufficient mouse welfare and health conditions required for obtaining reliable EEG responses. For instance, differential ERP is known to be reduced in inattentive or fatigued or unhealthy animals. Hence, we endowed animals 3 weeks before implantation surgery and another week before recording, for full recovery from infusion and implantation surgery, respectively. We believe that the Aβ oligomer injection influences the central nervous system in a devastating way. It is known that Aβ infusion induces an irreversible change in brain tissue: Cardinal features of Aβ infusion include synapse loss, tau hyperphosphorylation, astrocyte and microglial activation, and most importantly, neurofibrillary tangle formation, which is irreversible (Forny-Germano et al., J Neurosci, 2014). Our study shows an irreversible influence of ICV injection fingerprinted in the ERP results, which has a translational value because of its availability for cross-species comparison between human patients and the mouse model. 2\. Was any work done to corroborate the distribution of the Abeta oligomers are ICV (whether within this study of previously)?\[Reply\] We have reported a brain section-level evidence of A infusion in brain tissue in our previous methodology paper (Kim et al., J. Vis. Exp., 2016). However, the propagation of infusion effect over time through various brain regions has not been studied yet. We expect that the devastating effect of A infusion will not be restricted to the injection site since it affected differential ERP, engaging both bottom-up and top-down cortical circuits. 3\. How do the concentrations and timings and electrophysiological findings correlate or differ from work done using abeta oligomer incubations in brain slice (rodent), or biopsy derived human material? This is important in the overall experimental rationale. \[Reply\] We appreciate this comment. As far as we know, a direct comparison between biopsy-derived human material and brain slice of infusion model has not yet been reported. Notwithstanding, the Aβ-ICV injection model has been widely investigated with variations of Aβ aggregate species, Aβ isomers, injection numbers, and base animal models accompanied by behavior tests. However, it is hard to find complete datasets of pathology and behaviors in terms of timing and concentration. Nonetheless, several research articles (cited in our manuscript) used similar experimental conditions to ours (\~ 2 ug concentration, one-month delay after injection in observation). Here are the corresponding papers. Scientific Reports (2015) 5, 11708 reports that memory deficits and synaptoxicity of mice became gradually worse over 3 and 40 days since the single injection of Aβ(1-42) 3 ug. Journal of neuroscience research (2010) 88, 2923-2932 reports long-term potentiation changes with fear conditioning and CaN activity results by CV injection of Aβ oligomers in lower amounts than those of our experimental design. Neuroscience letter (1994) 170, 63-66 reports behavioral alterations in water- maze by varied concentrations of Aβ monomers in lower amounts than those of our experimental design. 10.1371/journal.pone.0230277.r003 Decision Letter 1 Jo Dong-Gyu Academic Editor 2020 Dong-Gyu Jo This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 26 Feb 2020 Destruction of ERP responses to deviance in an auditory oddball paradigm in amyloid infusion mice with memory deficits PONE-D-19-32332R1 Dear Dr. Choi, We are pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it complies with all outstanding technical requirements. Within one week, you will receive an e-mail containing information on the amendments required prior to publication. When all required modifications have been addressed, you will receive a formal acceptance letter and your manuscript will proceed to our production department and be scheduled for publication. Shortly after the formal acceptance letter is sent, an invoice for payment will follow. To ensure an efficient production and billing process, please log into Editorial Manager at <https://www.editorialmanager.com/pone/>, click the "Update My Information" link at the top of the page, and update your user information. If you have any billing related questions, please contact our Author Billing department directly at <authorbilling@plos.org>. If your institution or institutions have a press office, please notify them about your upcoming paper to enable them to help maximize its impact. If they will be preparing press materials for this manuscript, you must inform our press team as soon as possible and no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact <onepress@plos.org>. With kind regards, Dong-Gyu Jo, Ph.D Academic Editor PLOS ONE Additional Editor Comments (optional): Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation. Reviewer \#1: All comments have been addressed Reviewer \#2: All comments have been addressed \*\*\*\*\*\*\*\*\*\* 2\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 3\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 4\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 5\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 6\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: (No Response) Reviewer \#2: Authors have addressed all concerns and provide additional information that support the conclusion of their study. \*\*\*\*\*\*\*\*\*\* 7\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No Reviewer \#2: No 10.1371/journal.pone.0230277.r004 Acceptance letter Jo Dong-Gyu Academic Editor 2020 Dong-Gyu Jo This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 2 Mar 2020 PONE-D-19-32332R1 Destruction of ERP responses to deviance in an auditory oddball paradigm in amyloid infusion mice with memory deficits Dear Dr. Choi: I am pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department. If your institution or institutions have a press office, please notify them about your upcoming paper at this point, to enable them to help maximize its impact. If they will be preparing press materials for this manuscript, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact <onepress@plos.org>. For any other questions or concerns, please email <plosone@plos.org>. Thank you for submitting your work to PLOS ONE. With kind regards, PLOS ONE Editorial Office Staff on behalf of Dr. Dong-Gyu Jo Academic Editor PLOS ONE [^1]: The authors have declared that no competing interests exist.

# Introduction Idiopathic inflammatory myopathies (IIMs) are a group of clinically heterogeneous, autoimmune inflammatory muscle disorders characterized by muscle weakness and multisystem involvement. The main clinical subtypes of IIMs among Chinese populations are polymyositis (PM) and dermatomyositis (DM). The presence of myositis-specific autoantibodies (MSAs) is one hallmark of the disease. The clinical importance of MSAs has gradually been recognized in recent years. More and more studies show that different MSAs are characteristic of different clinical subtypes. For example, anti-melanoma differentiation-associated gene 5 (anti-MDA-5) antibodies are characteristic of a distinctive clinical subgroup with interstitial lung disease, and anti-transcription intermediary factor 1 (anti-TIF1) antibodies are characteristic of a subgroup of IIM patients who are at a high risk of developing cancer. The anti-3-hydroxy-3-methylglutaryl coenzyme A reductase (anti-HMGCR) autoantibody was first reported by Christopher-Stine and colleagues as “anti-200/100.” Given the strong association of this novel autoantibody and associated necrotizing myopathy with statin use in 63% of the patients, an autoantigenic target in the cholesterol synthesis cascade was sought. Mammen and colleagues in the same group later identified the anti-200/100 autoantibody as anti-HMGCR.. However, other studies have revealed that the majority of anti- HMGCR antibody-positive IIM patients from European cohort were not exposed to statins. The prevalence of anti-HMGCR antibodies has never been investigated among Chinese IIM populations, and the clinical features of Chinese anti-HMGCR antibody-positive patients were previously unknown. To address these questions, we measured serum anti-HMGCR antibodies levels in 405 IIM patients and 311 controls, and compared the differences between the clinical features of the anti-HMGCR antibody-positive and -negative patients. The anti-HMGCR antibody- positive patients were also followed-up to analyze how this subgroup responded to therapy and whether levels of anti-HMGCR antibody could predict disease activity or disease prognosis. # Materials and Methods ## Ethics statement All samples were obtained for research purposes. In the retrospective study, patients’ consents were impossible or impractical to obtain. All patients data was anonymously used. This study was approved by the Research Review Committee (RRC) and the Ethical Review Committee (ERC) of the China-Japan Friendship Hospital. ## Patients All IIM patients (n = 405) fulfilled the Bohan and Peter criteria for PM and DM; 117 of these patients had PM and 288 had DM. Their sera were obtained between April 2009 and March 2015. The following data were obtained from their medical records: age, sex, past medical history, statin exposure, muscle strength, lung function, chest computed tomography (CT) images, CK levels, lactate dehydrogenase (LDH), hydroxybutyric acid dehydrogenase (HBDH), the presence of other MSAs, and levels of immunoglobulin (IgA, IgG, IgM), C-reactive protein (CRP), Complement 3 (C3) and Complement 4 (C4) as well as erythrocyte sedimentation rate (ESR). Muscle biopsy specimens of patients who carried anti- HMGCR antibodies were reviewed. Patients demonstrated muscle weakness, according to their Manual Muscle Testing 8 (MMT-8) scores, the maximum score being 80. The Modified Myositis Disease Activity Assessment Tool (MYOACT)–which assessed constitutional, cutaneous, skeletal, gastrointestinal, pulmonary, and cardiac conditions—was used to evaluate disease activity. Patients who carried the anti-HMGCR antibody received follow-up, and their response to treatment was evaluated. Improvement was defined as an improvement of more than 20% on at least two of the study’s core measurement tools (e.g. MMT-8, MYOACT and serum CK levels), with the score on the third tool decreasing no more than 25%. Our study controls included blood from 90 healthy donors and from 221 patients with other different inflammatory/autoimmune diseases, including 30 patients with primary Sjögren’s syndrome (pSS), 20 with ankylosing spondylitis (AS), 55 with systemic lupus erythematosus (SLE), 50 with rheumatoid arthritis (RA), 10 with systemic sclerosis (SSc), 30 with idiopathic hyper-creatine kinase-emia, 10 with muscular dystrophy, 6 with myasthenia gravis, 5 with motor neuron disease, and 5 with mitochondrial myopathy. Basic clinical data, including age and sex, were recorded for these controls samples. Sera from all groups were collected and stored at -80°. ## Quantifying anti-HMGCR antibody by ELISA ELISA plates coated with recombinant HMGCR were incubated with diluted patient samples. Assay procedures followed the standard protocol of the QUANTA Lite assays (INOVA Diagnostics). Mean absorbance (OD) of the samples was plotted as a standard curve against their values in unites. A linear Cubic Spine (third order polynomial) fit or a fourth order polynomial fit was used to draw these curves. Unknown HMGCR concentrations were determined by finding unit values (U) on the Y axis and reading their corresponding absorbance values on the X axis. Negative values corresponded to Y-axis values 0–20 U. Positive values corresponded to Y-axis values greater than or equal to 20 U. ## Statistical analysis All analyses were completed using SPSS 17.0, and *P* values less than 0.05 were considered significant. Correlation analyses were performed using the Spearman test. Quantitative variables are reported as means and compared using a nonparametric test. Categorical variables are reported as numbers and/or percentages and were compared using the Chi-square test or, when appropriate, Fisher’s exact test. # Results ## 1. The frequency of anti-HMGCR antibodies in IIM patients In total, 22 of the 405 patients (5.4%) with IIM carried anti-HMGCR antibodies. Anti-HMGCR antibodies were not present in healthy controls or in disease control groups—i.e., patients with AS, RA, SLE, SSc, muscular dystrophy, myasthenia gravis, motor neuron disease, and mitochondrial myopathy. However, one pSS patient carried the anti-HMGCR antibody. ## 2. Clinical phenotype ### 2.1 Clinical features of anti-HMGCR antibody-positive patients The anti-HMGCR antibody was found in 2.8% (8/288) of DM patients and 12% (14/117) of PM patients. There were 16 (73%) females and six (27%) males among the anti-HMGCR antibody-positive patients. The average age and average disease duration of anti-HMGCR antibody-positive patients are similar to those of antibody-negative patients (both *p \>* 0.05). In 3 of 20 patients (15%), the anti-HMGCR antibody was produced following exposure to statins (complete clinical data was not available for every patient). Two patients were administered by atorvastatin, rosuvastatin was administered to another one. Two patients had statin exposure in the five anti-HMGCR antibody-positive subjects over the age of 50 (40%), the prevalence of statin used in patients with anti- HMGCR antibody was higher than 28 of the 133 (21%) anti-HMGCR antibody-negative IIM patients over 50 years old, but the difference was not statistically significant (*p* \> 0.05). No other relevant myotoxin exposures were identified upon review of the patient records. Of these 20 patients, five (30%) experienced a subacute onset (\< 12 months), 14 (70%) experienced a progressive onset (\> 12 months). Of these same 20 patients, 15 (75%) patients presented with muscular weakness, and 14 (70%) patients suffered myalgia for the whole duration. Twelve patients experienced symmetric proximal muscle weakness at the time of the initial serum collection; the remaining three patients had a history of muscular weakness. Half (50%) of these patients had dysphagia, and this incidence was higher than dysphagia incidence in the antibody-negative group (20%) (*p* \< 0.01). Of the ten patients with dysphagia, 4 patients characterized by difficulty into solid food, 3 patients choking cough when drinking water and one of them had difficulty in speaking concurrently, and the other three patients required nasogastric tube and two of them had difficulty in speaking concurrently. Of the anti-HMGCR antibody-positive patients, five (25%) had arthralgia and three (15%) had ILD. But these numbers were not significantly different compared to those in the negative group (*p* \> 0.05). Osteosarcoma was observed in one patient. ### 2.2 Laboratory tests of anti-HMGCR antibody-positive patients Eighteen of 21 (85.6%) anti-HMGCR antibody-positive patients had a history of elevated serum muscle enzyme levels, with mean CK levels of 2538.7 ± 3047.6 IU/L during their initial admission. Serum muscle enzyme levels—including CK, LDH and HBDH—in anti-HMGCR antibody-positive patients were significantly higher than those in antibody-negative patients (all *p* \< 0.05). Although IgG, IgA and IgM levels in anti-HMGCR antibody-positive patients were all lower than those in the antibody-negative group, the difference was not statistically significant (all *p* \> 0.05). Levels of CRP, C3 and C4 as well as ESR were also not significantly difference between the two groups. Anti-Jo-1 antibody was detected in three (14%) of the anti-HMGCR antibody-positive patients. Anti-SRP antibodies were not detected in all anti-HMGCR antibody-positive patients. ### 2.3 Pathological features of muscles of antibody-positive patients Muscle biopsies were available for 12 of the 22 anti-HMGCR antibody-positive patients. Eight patients (67%) suffered from significant myofiber necrosis with little or no inflammatory cell infiltration that could be diagnosed as immune- mediated necrotizing myopathy (IMNM), according to European Neuromuscular Centre criteria (ENMC). Two patients had necrotic myofibers along with obvious inflammation. Two subjects did not suffer from myofiber necrosis but did experience some inflammatory cell infiltration. Of the eight patients who experienced myofiber necrosis, four (50%) expressed MHC-I on the surface of regenerative or necrotic myofibers. Of the muscle biopsies from the three patients who had experienced statin exposure, two showed myofiber necrosis with little inflammation, and the remaining one only had inflammatory infiltrates. ## 3. Correlation analysis between disease activity and anti-HMGCR antibody levels Anti-HMGCR antibody levels, serum CK levels, MMT-8 score, and MYOACT score were assessed in anti-HMGCR antibody-positive patients at the time of their initial evaluation. Anti-HMGCR antibody levels had a strong correlation with MYOACT scores (*r* = -0.611, *p* = 0.02) but not with CK levels and MMT-8 scores (both *p* \> 0.05) in patients without statin exposure. From an overall analysis of all anti-HMGCR antibody-positive patients, there was a statistically significant association between anti-HMGCR antibody levels and serum CK levels (*r* = 0.486, *p* = 0.026) and MYOACT scores (*r* = -0.67, *p* = 0.003), but there was no significant correlation between anti-HMGCR antibody levels and muscle strength (MMT-8 score) in these patients (*r* = -0.16, *p* = 0.545). ## 4. Follow-up study of anti-HMGCR antibody-positive patients Among anti-HMGCR positive patients, 11 patients were followed longitudinally, including two patients with previous statin exposure. The median follow-up time was 9 months, with a range of 2.5 months to 2 years. Anti-HMGCR antibody levels, serum CK levels, MMT-8 scores, and MYOACT scores were recorded at each visit for monitoring. Of these 11 patients, eight (73%) responded well to glucocorticoids and/or other immunosuppressive agents; either their MMT-8 scores and MYOACT scores improved or their CK levels declined. The average glucocorticoid dosage was reduced from 63.1 mg/day to 20.5 mg/day for these patients. Furthermore, three patients—including two with prior statin exposure—had near-complete remission after treatment; they recovered normal muscle strength and normal CK levels, and suffered very little extramuscular disease activity. Three of these 11 patients (27%) did not experience improvement according to our criteria for improvement. Patients who deteriorated were younger on average than those who experienced improvement (mean ages: 27.7 years VS 41.9 years, respectively). The longitudinal data points representing anti-HMGCR antibody levels in seven patients. Despite the decline in anti-HMGCR antibody levels over time in some patients after treatment, the antibody levels had not normalized in six (85.7%) of the seven patients by the last study visit, despite some patients regaining normal muscle strength. The antibody levels of only one patient—whose anti-HMGCR antibody titer was 22 U—decreased to fall within the normal range during the observation period. After treatment, antibody levels changed, but the change in serum antibodies levels did not correlate with changes in CK levels, MMT-8 scores and MYOACT scores in the seven patients (all *p* \> 0.05). # Discussion This is the first study to describe the prevalence of anti-HMGCR antibodies in a large sample of Chinese IIM patients. We found that anti-HMGCR antibody-positive patients exhibited a slow onset of disease and presented with muscle weakness and dysphagia, which were seen in patients with and without statin exposure. Most patients experienced myofiber necrosis with little or no inflammatory cell infiltration in their muscle biopsies. Most of the patients responded well to immunosuppressive treatment and achieved a good prognosis, but serum levels of the anti-HMGCR antibody did not correlate with disease activity. The present study shows that the anti-HMGCR antibody is present in 5.4% of patients with IIM, similar to findings from two other large sample studies. But this antibody is also present in patients with other autoimmune diseases. We detected the antibody in one pSS patient, but that titer was lower. Similar results were also reported by Musset L et al. In our study, the pSS patients with anti-HMGCR antibody did not have muscular weakness, shin rash, elevated CK level and other MSAs from 2009 to July of this year. In our study, the mean age of anti-HMGCR antibody-positive IIM patients was similar to the age of antibody-negative patients. Mammen et al showed that the mean age of anti-HMGCR antibody-positive patients without statin exposure was lower than that of statin-exposed antibody-positive patients. Because of the small number of patients who had prior statin exposure in this study, it was difficult to make comparisons between antibody-positive patients with statin exposure and those who had never been exposed. Only 15% of anti-HMGCR antibody- positive patients in the current study had previous statin exposure. This percentage is lower than the 44% and 72.7% reported in studies by Allenbach et al and Werner JL et al, respectively. Since hyperlipidemia increases with age, the prevalence of statin use increases with age. So the Hopkins group compared statin exposure prevalence of anti-HMGCR antibody-positive patients with age- matched antibody-negative patients. They found that the prevalence of statin use in anti-HMGCR patients was significantly higher than the rates of statin use in age over 50 years-old groups of PM, DM, and IBM patients. But we did not find the association in older patients, may be due to the small number of antibody- positive patients in our study. This should be confirmed by large-sample studies. However, the reason for this lower percentage is not that statins are prescribed less often in China, In fact, tens of millions of Chinese patients suffer from hyperlipidemia, and the majority of them are given statins to lower cholesterol. The lower percentage in our study may suggest that different genetic backgrounds may trigger the occurrence of anti-HMGCR antibodies in different populations. Regarding the type of disease onset, most of the patients in our study experienced a muscular weakness whose slow progression ranged from 12 months to a few years. In contrast, subacute onset was the main onset type in European patients and in Japanese patients. This difference may suggest that the onset of anti-HMGCR antibody-associated IIM varies between different races and ethnicities. The main clinical features of muscular weakness and myalgia were observed both in our cohort and in other study cohorts. However, another prominent clinical feature of antibody-positive patients in our cohort was dysphagia. Christopher- Stine et al also noted dysphagia in 63% of their patients with the anti-HMGCR antibody. But, Allenbach Y et al reported that 26.7% of their European patients presented with dysphagia, and Watanabe Y et al observed no swallowing disorders in their Japanese patients. This difference may show that heterogeneity of dysphagia in different group. We observed that three patients carried both anti-HMGCR and anti-Jo-1 antibodies. This suggests that the anti-HMGCR antibody could coincide with other MSAs. Mammen et al also reported that one patient with anti-HMGCR antibody had Jo-1 antibodies with ILD and another patient had scleroderma with positive anti- Pm/Scl titers and ILD. Interestingly, two of the three patients with the anti- Jo-1 antibody did not have ILD in our study. Because anti-Jo-1 antibodies were closely associated with ILD, so we couldn’t rule out the possibility of anti- Jo-1 antibody false positivity in the other two patients. We reviewed the muscle biopsies of anti-HMGCR antibody-positive patients, most of them associated with the IMNM We also found necrotic myofibers along with mild inflammation or only inflammation in muscle biopsies of other patients. This finding shows that muscle histopathology varies among anti-HMGCR antibody- positive patients. This variation was also observed by Allenbach Y et al. MHC-I expression was detected in half of the antibody-positive patients, consistent with findings from the Johns Hopkins cohort, and four of the eight available biopsy specimens included sarcolemmal MHC-I staining of myofibers. Werner JL et al demonstrated that anti-HMGCR antibody levels had a statistically significant association with CK levels and muscle strength when statin-exposed and statin-unexposed patients were examined together during their initial visit. When analyzed separately, these associations were confirmed in the antibody- positive patients who had prior statin exposure, but not in antibody-positive patients with no prior statins exposure. Our cross-sectional study shows that anti-HMGCR antibody levels do not correlate with muscle strength but are significantly associated with serum CK levels. In patients with no statin exposure, anti-HMGCR antibody levels only correlated with MYOACT scores. Because only three anti-HMGCR patients were exposed to statins in our study, we were unable to define the relationship between antibody levels and CK levels or muscle strength in statin-exposed patients. Furthermore, we studied some patients longitudinally to see how antibody- positive patients responded to treatment. In the present study, we found that most anti-HMGCR antibody-positive patients respond well to treatment, especially patients with previous statin exposure. This findings is similar to that in the Watanabe Y et al study, in which seven patients who were followed for two years all responded well to immunotherapy. Additionally, this study shows that patients whose conditions did not improve after treatment were younger than those who improved after treatment. This phenomenon may show that anti-HMGCR antibody-positive patients with early onset are refractory to treatment. In order to understand whether the anti-HMGCR antibody reflects disease activity, we detected serial changes in the antibody levels of seven patients. We found that the antibody levels did not decrease significantly in patients who got improvement. Some patients had elevated antibody levels even after complete remission. In our cohort, changes in antibody levels did not significantly correlate with lower CK levels or improved muscle strength in the follow-up period. Therefore, anti-HMGCR antibody levels may not be a useful indicator of disease activity. This finding differed from those from the Australian cohort; they reported that immunosuppressive therapy caused antibody and CK levels to decline, with a corresponding improvement in muscle strength during long-term follow-up. We recognize that this study has several limitations. First, the study involves a relatively small number of patients for follow-up, and some of whom were only followed for a short time. Second, because this was a retrospective study, treatment plans and the time intervals between each visit were not standardized. Furthermore, some patients had already received immunotherapy before admission to our hospital, so our initial visit data do not reflect these patients’ conditions prior to treatment. Lastly, complete clinical data was not available for every patient at every visit. In conclusion, our study shows that the major clinical features of anti-HMGCR antibody-positive Chinese IIM patients are muscle weakness and dysphagia, which were seen in patients with and without statin exposure. Our study also shows that these patients respond well to immunosuppressive treatment and achieve good post-treatment prognoses. All of these findings suggest a phenotypic difference between anti-HMGCR antibody-positive patients and antibody-negative patients. Anti-HMGCR antibody-associated IIM may be a distinct clinical phenotype. However, additional studies involving larger numbers of patients over longer time periods are required to confirm this conclusion. # Supporting Information We express our sincere thanks to all physicians, clinical fellows, and nurses of the hospital for their invaluable support and contribution. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: YPG XL GCW XMS. Performed the experiments: YPG QLP. Analyzed the data: YPG. Contributed reagents/materials/analysis tools: YPG. Wrote the paper: YPG XL GCW.

# Introduction Prochloraz (\[N-propyl-N-2-(2,4,6-trichlorophenoxy)-ethyl\]-imidazole-1-carboximide; CAS67747-09-5) is a broad-spectrum fungicide belonging to the class of imidazoles. It was confirmed that the active substance prochloraz and fungicides based on prochloraz are particularly suitable for protecting the wood and wooden materials against attacks or destruction by lignivorous fungi. Prochloraz was also used on some cereal crops, fruits and vegetables to control eyespot disease and powdery mildew through the inhibition of ergosterol biosynthesis. Prochloraz-manganese is a good alternative to prochloraz due to its similar function and relatively low toxicity and stability. An efficacy test of -prochloraz-manganese indicated that a small amount of the pesticide has significant inhibitory effects on the black spot caused by *Alternaria alternata* and wet bubble disease of white mushroom and could also be used to extend the preservation time of mango and persimmon fruit\[–\]. Therefore, prochloraz-manganese has been widely used in agricultural industries with high dosages since the early 1980s, especially in the mushroom industry. The overuse of prochloraz-manganese is likely to cause considerable pollution problems in soils. As a result, there is an urgent demand to explore an efficient way to solve the problem. The source of pesticide residue in the environment is mainly agricultural practices through a few routes such as spray drift, surface runoff and field drainage. The fate of residual pesticides was determined by various mechanisms including volatilization, chemical degradation, photodegradation, absorption, hydrolysis, enzymatic degradation and bacterial/fungal degradation. The use of microorganisms to degrade pesticides or even macromolecular substances was dependent on their high metabolic diversity and adaptability. In addition, microbial degradation rates were strongly influenced by a wide range of environmental factors such as nutrient availability, the presence of alternative carbon substrates, pH, temperature, and salinity. Therefore, the isolation of a new bacterial strain capable of degrading pesticides and the evaluation of its degradative properties under various laboratory conditions would provide the first step for the practical application of microorganisms in bioremediation. However, due to various environmental restrictions such as low temperature and high alkalinity, the biodegradation efficiency was significantly affected and suppressed. As a result of the long-term application of prochloraz-manganese, a considerable amount of the pesticide has accumulated in plants, soil and water, which might have some adverse impacts on the environment. Research on environmental toxicology and the metabolism of the residues has been carried out domestically and throughout the world since the early development and application of prochloraz. Chen et al. isolated a Bacillus sp. designated DG-02, which could degrade 95.6% of 50 mg L-13-Phenoxybenzoic acid within 72 h at a pH of 7.7 and 30.9 °C. Liu et al. successfully screened an endophytic quinclorac-degrading bacterial strain (Q3), and the results indicated that the use of this strain to degrade quinclorac for 7 d resulted in a 93% decrease from the initial concentration of 20 mg/L. In the treatment in which glucose was used, degradation percentages of methyl parathion and chlorpyrifos of 98% and 97%, respectively, were obtained in 120 h. This treatment also achieved the highest percent reduction in toxicity when monitored with a high-intensity light source. Moreover, a high-efficiency bacterial strain that was capable of degrading organochlorine pesticides (OCPs) such as aldrin, 4, 4’-DDT, dieldrin, endrin and endrin aldehyde simultaneously was isolated and screened by Belal, E.S.B. et al., while its biodegradation rates ranged between 24.4% and 98%. These results showed that many types of pesticide residues could be effectively degraded by bacteria, even in extreme environments. However, there have been few studies on the bacteria that could effectively degrade prochloraz-manganese. Because of the slow metabolic degradation of prochloraz-manganese in the environment, a bacterial strain that could efficiently degrade the pesticide is now urgently needed and is the focus of this study. The goal of this study is to isolate and characterize a bacterial strain capable of degrading prochloraz-manganese. Furthermore, the growth and degradation characteristics of this strain were also investigated under various conditions. The research results enrich the prochloraz-manganese-degrading microbial species resource and provide a valuable technical reference for the biodegradation of this pesticide. # Materials and methods ## Soil collection Soil samples (obtained from the top 20 cm of soils) were taken from a prochloraz-manganese-contaminated field in Tianjin, China(117.123,39.24). After drying, pulverizing, and sieving the soil with a 2 mm sieve, 100 g of sample soil was taken, and the strain was isolated in quadruplicate. The study was carried out on private land, and we confirm that the owner of the land gave permission to conduct the study on this site and the field studies did not involve endangered or protected species. ## Bacterial isolation and medium Analytical-grade prochloraz-manganese (98% purity) was obtained from the Kangbaotai Fine Chemical Company (Hubei, China). Soil samples (5.0 g) were added into sterilized triangular flasks (250 mL), which contained 100 mL of mineral salt medium (MSM) and 5.0 mg/mL of prochloraz- manganese. The bacterial suspension was cultivated at 37°C with a rotation speed of 150 rpm and transferred into fresh MSM every 3 d for continuous enrichment with 5% inoculum. After 4 transfers, the concentration of prochloraz-manganese in the medium was gradually increased to 500 mg/L, and the suspension was then cultivated for 3 d under the same conditions as described above. A total of 0.1 mL of the suspension was taken and spread on beef cream peptone medium, GAUZEˊs medium No.1 and potato medium (for the cultivation of bacteria, fungi, and actinomycetes, respectively) and cultivated for 3 d. Several well-grown strains were selected, isolated and purified. The purified strains were transferred to the MSM media (containing 500 mg/mL of prochloraz-manganese) to determine their degradation ability. Then, the strain with the best degradation ability was selected and stored at 4 °C. ## Characterization of the bacteria ### The morphological characteristics of the isolated bacteria The preserved degrading strains were inoculated on the solid medium by the spread plate method and cultured at 37°C for 2 d. The colony characteristics and the morphological structure of the cells were observed under a Nikon 80i microscope. Then, gram staining was performed on the strain capable of degradation for physiological and biochemical identification. ### 16S rDNA sequence analysis DNA extraction of the isolated strain was completed by the Cetyltrimethylammonium bromide (CTAB) method. The 16S rDNA amplification procedure and phylogenetic analyses were performed as described by Weisburg W G et al.. 16S rDNA primers were used to conduct PCR amplification using the DNA extracted from the isolated strain as a template. The forward primer for the bacterium was 8f (`5’-AGAGTTTGATCCTGGCTCAG-3’,` 20 bp), and the reverse primer was 1492r (`5’-GGTTACCTTGTTACGACTT-3’`, 19 bp). PCR products were transported to Shanghai Sangon Biological Engineering Technology and Services Co. Ltd. (Shanghai, China) for sequencing. The sequences were analysed using the BLAST program (Altschul et al., 1990) of the National Center for Biotechnology Information (NCBI) and constructed with the MEGA 6.0 software package and neighbour-joining algorithm. The evolutionary distances were computed using the Kimura 2-parameter method\[–\]. Bootstrap analysis was performed by means of 1000 replicates. ## Growth curve of the isolated bacteria A 0.5 mL aliquot of bacterial suspension was added to a sterilized conical flask containing 100 mL of LBM and incubated in a shaking incubator at different temperatures (10°C, 18°C, 28 °C and 37 °C) at 150 rpm. The cell density was measured at 600 nm (OD₆₀₀) at intervals of 5 h. The optimal culture time of the degraded strains at different temperatures was obtained, and the growth curve of the strains was drawn. ## Effects of different culture conditions on bacterial growth and the degradation rate of prochloraz-manganese The strain was cultivated to a logarithmic stage in Luria-Bertani (LB) media and then inoculated (by a 5% inoculation amount) into the inorganic salt medium with a high prochloraz-manganese concentration of 500 mg/L. After incubation at a speed of 150 rpm under various culture conditions, samples were separately collected to determine the OD₆₀₀ and the degradation rate of prochloraz-manganese of each culture medium. The medium was cultured at 10 °C, 18 °C, 28 °C and 37 °C to assess the effect of temperature. Glucose, sucrose, maltose, and soluble starch (concentration of 1 g/L) were added, and the medium was cultured at 37 °C to assess the effect of the carbon source. A concentration of 1 g/L of nitrate, urea, ammonium chloride, peptone was added individually to cultures at 28 °C to assess the effect of the nitrogen source. The pH of the culture solution was 4.0, 5.0, 6.0, 7.0, 8.0, 9.0, or 10.0 respectively, and culture was performed at 28 °C. The initial salinity of the culture solution was set to 0%, 1%, 3%, 5%, 7%, and 9% respectively, cultured at 28 °C. The inorganic salt media were prepared separately with different prochloraz-manganese concentrations of 100 mg/L, 200 mg/L, 300 mg/L, 500 mg/L, 750 mg/L and 1000 mg/L respectively, cultured at 37 °C. Each treatment was repeated 3 times. ## Detection of prochloraz-manganese One millilitre of the liquid medium sample was mixed with 1 mL of dichloromethane as well as 1 mL of 2.0% sodium chloride solution. After thirty- minute oscillation extraction and stratification, the combined organic phases from the substratum were dried with anhydrous Na₂SO₄ and then evaporated using a rotary vacuum evaporator. The mixtures were dissolved in methanol to a final volume of 5.0 mL. Waters e2695 high performance liquid chromatography (which is equipped with a Waters 2489 UV detector and Empower3 data processing system); Chromatographic column: Agilent ZORBAX Extend-C18, 250 mmX x 4.6 mm (i.d.), 5 μm. Methanol/water = 9/1(v/ v, adjust pH to 3.0 with H₃PO₄) was used in the mobile phase, and the flow rate was 0.8m·min^-1. The column temperature was 30°C. The detection wavelength was 210 nm. Sample quantity was 20μL. Retention time (Rt) was 5.384 min, and its detection limit ranged from 0.24 to 10.8 mg/kg. The limits of detection (LOD) and limits of quantification (LOQ) were calculated on the basis of signal-to- noise ratios (S/N) of 3 and 10, respectively. The LOD and LOQ values were 0.036 and 0.120mg/kg. The recovery of prochloraz-manganese was in the range of 87.6–108.2%, with a relative standard deviation of 2–13%. # Results and analyses ## Identification of the prochloraz-manganese-degrading strain Through repeated separation and purification, the strain with the highest prochloraz-manganese degradation rate, designated WD-2, was isolated from the pesticide-contaminated soil. After 24 h of culture in the MSM medium, the colonial morphology was as follows: the bacteria were white, moist and circular and had a smooth, convex surface as well as a neat edge. SEM images of strain WD-2 showed a short rod shape with a size of (0.6–0.8) μm × (1.1–1.4) μm in. Biochemical characteristics assayed for the strain WD-2 are shown in. According to the results from Shanghai Biological Engineering Co., Ltd., the length of the specific 16S rDNA sequence of strain WD-2 was 1250bp (GenBank accession number KJ526821). A BLAST analysis of the bacterial strain WD-2 was performed, and the results are shown in. The phylogenetic tree was constructed by MEGA 6.0 software. The phylogenetic tree of WD-2 is shown in. The results indicated that the sequence of the 16S rDNA gene of strain WD-2 showed a high sequence similarity (more than 99%) to the genus *Bacillus*. Based on the physiological and biochemical characteristics and the 16S rDNA of the strain, strain WD-2 could be preliminarily identified as *Bacillus cereus*. The degradation of pesticides by the genus *Bacillus* has been widely reported by previous studies. Tang et al. isolated a *Bacillus sp*. (TAP-1) strain capable of degrading triazophos, and Ali M et al. isolated a *Bacillus pumilus* strain (WI) capable of degrading organophosphate. In addition, Zhang et al. successfully isolated *Bacillus cereus* strain Y1, which could efficiently decompose deltamethrin\[–\]. These results indicated that the genus *Bacillus* plays an essential role in the biodegradation of pesticides. ## Growth curves of the strain WD-2 Growth curves of strain WD-2 at different culture temperatures are shown in. When the culture temperature was between 20°C and 37°C, the OD₆₀₀ increased with increasing temperature to a maximum at 45 h. When the culture temperature was above 16°C, strain WD-2 grew slowly and could barely grow at a low temperature of 10 °C. The results showed that the optimal temperature for strain WD-2 growth was 37°C. When the culture temperature was approximately 37°C, the lag, logarithmic, stationary and death phases of the strain were from 15 to 40 h, 40 to 45 h, 45 to 50 h, and 50 h and later, respectively. ## Effects of different culture conditions ### Effects of temperature The effect of temperature on bacterial growth and the degradation rate of prochloraz-manganese are represented in. Both the degradation rate and bacterial growth significantly increased at 10–37°C and decreased after 37 °C, indicating that the degradative ability of WD-2 was positively correlated with the OD_600。 When the temperature was 37 °C, both the growth and the degradation rate of strain WD-2 reached their maximum, and the degradation rate of prochloraz-manganese reached 90.6%. In addition, when the temperature was approximately 18–42°C, the strains were able to grow well and significantly degrade prochloraz-manganese. This result indicated that the strain WD-2 could survive under a wide range of temperatures, which will enable the strain to work effectively in practical applications. ### Effects of carbon sources The effects of different carbon sources on strain WD-2 are depicted in. When the carbon sources differed, the growth of WD-2 and the degradation rate of prochloraz-manganese were positively correlated, and the effect of different carbon sources was glucose\>sucrose\>maltose\>starch. When glucose was used as the carbon source, the growth of strain WD-2 was maximal, and the OD₆₀₀ value reached 0.905. The degradation rate of prochloraz- manganese increased by 11.5% compared with that of the control group, reaching 90.2%. In addition, when starch was chosen as the carbon source, the growth and the degradation ability of the strain were limited. The reason for this result may be that glucose is a small molecular substance that is easily utilized by the strain WD-2, and starch is not easily utilized since it is a macromolecular substance. Based on these results, glucose was used as the optimum carbon source in further experiments. Malghani et al. reported that a bacterial strain capable of degrading profenofos preferentially utilized glucose as a carbon source in the medium. The degradation rate of prochloraz-manganese increased by 8.0% when sucrose or maltose was used as the carbon source. ### Effects of nitrogen sources The effects of different nitrogen sources on strain WD-2 are depicted in. The OD₆₀₀ value of strain WD-2 was between 0.83–0.86 under different nitrogen sources (peptone, ammonium chloride, ammonium nitrate and urea), and the degradation rate of prochloraz-manganese was 81.2–83.5%. Compared with the control group, the experimental groups showed no notable difference in the growth of the strain and the degradation rate of the prochloraz-manganese. The results indicated that the nitrogen source had no significant impact on the growth of strain WD-2 and its ability to degrade prochloraz-manganese. The reason for the analysis may be that the strain WD-2 mainly uses carbon sources as nutrients, and the utilization rate of nitrogen sources is low. Considering the cost of practical application, urea was chosen as the nitrogen source in further studies. ### Effects of pH The effects of different pH values on strain WD-2 are depicted in. The growth rate of strain WD-2 and the degradation rate of prochloraz-manganese increased in the pH range of 4.0\~8.0. The degradation rate reached more than 88% in the pH range of 7.0\~8.0 and reached a maximum of 90.7% at a pH of 8.0. When the pH was greater than 10.0 or less than 4.0, the prochloraz-manganese could barely be degraded. When the pH was above 8.0, the degradation efficiency began to decrease. The effects of pH conditions on bacterial growth and degradation activity have been reported in previous studies. For example, Zhao et al. reported that slightly acidic conditions were favourable for the growth of the strain ZHJ6 and its degradation of methamidophos. Singh D P et al. reported that the optimal pH for the growth of the strain PUPCCC 64 and its degradation of chlorpyrifos was 7.0. According to the results, strain WD-2 had the best degradation under weakly alkaline conditions, and the optimum pH for degradation was 7.0–8.0. WD-2 was more suitable for repairing the pollution of weak alkaline environments in practical applications and should be controlled within a certain pH range. ### Effects of salinity Salinity may have considerable effects on biomass and microbial species, microbial physiological changes, and microbial molecules and cells. As shown in, the bacterial growth,as well as degradation rate, decreased with salinity varying from 1% to 9%. The growth inhibition at high salinity may have been caused by the extreme dehydration of the cells of strain WD-2. Although the optimal salinity was observed at 1%, there was a slight difference in bacterial growth and degradation rate at salinities between 0% and 1%, indicating that the strain WD-2 may have a certain degree of tolerance to low salinity. A. Aziz et al. isolated three benzo\[a\]pyrene-degrading bacterial strains from sea sediments, and approximately 41% of an initial 50 mg/L benzo\[a\]pyrene concentration was successfully decomposed by the bacterial consortium after 8 days of incubation in a simulated seawater environment (28 ppm of NaCl). These results provide theoretical support for the application of microorganisms under different salinity conditions. In conclusion, the optimum salinity for growth of WD-2 was identified as 1%. Under low salinity conditions, the strain grew well and was highly effective at degrading prochloraz-manganese. ## Biodegradation kinetics By studying the kinetics equations of the degradation process, the removal of contaminants could be predicted, which is necessary for the modification of the treatment method as well as the optimization of the biochemical disposal process. The first-order kinetic model has been widely applied for the prediction of bacterial treatment results. The prochloraz-manganese concentrations were measured, and time curves of the concentrations of prochloraz-manganese were drawn, as shown in. It was assumed that strain WD-2 follows the first-order reaction kinetics equation for prochloraz-manganese biodegradation, and the kinetic equation was as follows: $$ln(C/{mg}/L) = - kt + A$$ In formula, k—kinetic constant of the prochloraz-manganese degradation rate; C—the concentration of prochloraz-manganese (value C/mg/L); t—biodegradation time, h; and A—constant. Based on the results shown in, the kinetic equation can be obtained by applying the formula. When the initial concentration of prochloraz-manganese was less than 500 mg/L, the half-life of WD-2 degradation of prochloraz-manganese was similar, which corresponds to the degradation kinetics equation lnC = −0.061t + A, and the half-life was 11.06 h. The correlation coefficients (r) of each k reached 0.9830, 0.9850, 0.9945, 0.9976, 0.9956 and 0.9954. When the initial concentration became higher than 500 mg/L, the degradation rate constant decreased with the initial concentration of prochloraz-manganese, indicating that a high concentration of prochloraz-manganese might inhibit the strain’s degradation ability of the strain. Furthermore, the degree of influence may also increase with the initial concentration of prochloraz-manganese. # Conclusions 1. The strain with high degradation activity was isolated from soils contaminated with prochloraz-manganese by enrichment culture. The strain was preliminarily identified as *Bacillus cereus* and named *Bacillus cereus* WD-2 (GenBank accession number JX114949). 2. The optimal carbon source, temperature, pH, and salinity scope for the growth of WD-2 were glucose, 35–40°C, 7.0–8.0, and within 1%, respectively. When glucose was chosen as the carbon source, the growth ability (OD₆₀₀ value of 0.905) and degrading abilities (90.2%) both reached a maximum. However, the difference in nitrogen sources seemed to have little influence on the growth of WD-2. Based on the study of the growth characteristics of WD-2, it can be concluded that strain WD-2 is adapted to a wide range of temperature and pH conditions. The kinetics study indicated that the degradative ability of the strain was negatively correlated with the initial concentration of prochloraz-manganese. Thus, the strain has a substantial degree of potential application for the bioremediation of soils contaminated by prochloraz-manganese, especially soils with low alkalinity and salinity. [^1]: The authors have declared that no competing interests exist. [^2]: ‡ These authors also contributed equally to this work.

# Introduction Digestive processes along the gastrointestinal tract are aided by acidic and basic secretions by a number of epithelia. In particular, the pancreas and the stomach are the most avid base (HCO₃^-) and acid (H⁺) secretors, respectively. The gastric H⁺ secretory mechanisms are well established, however, the cellular mechanism by which pancreatic duct cells secrete almost isotonic HCO₃⁻ fluid has long been a challenge to epithelial physiologists. The current ion transport model for pancreatic HCO₃⁻ secretor, the duct cell, involves two machineries on the two epithelial membranes: first, cells accumulate cellular HCO₃⁻ with the help of a basolateral Na⁺-HCO₃⁻ cotransporter (pNBC, NBCe1) and a Na⁺/H⁺ exchanger (NHE1) together with carbonic anhydrase; second, HCO₃⁻ efflux occurs via co- operation between Cl⁻ channels and Cl⁻/HCO₃⁻ anion exchangers from the SLC26A6 family, e.g. SLC26A6 on the luminal membrane. The Cl⁻ channels are the cystic fibrosis transmembrane conductance regulator (CFTR) Cl⁻ channels, which may have some HCO₃⁻ permeability and the Ca²⁺-activated Cl⁻ channels, such as TMEM16A/ANO1. Furthermore, K⁺ channels (e.g. K_Ca3.1, K_Ca1.1, KCNQ1) maintain the membrane potential and provide the driving force for anion secretion together with the Na⁺/K⁺-ATPase. Na⁺ and water follows passively. Nevertheless, this model can only explain production of 80–100 mM NaHCO₃ in secreted fluid, yet human pancreas can secrete up to 140 mM NaHCO₃. In addition to NBCs and NHE, earlier studies have shown vacuolar H⁺ ATPase (V-ATPase) activity on the basolateral membrane of pancreatic ducts by intracellular pH (pH_i) measurements and use of the V-ATPase inhibitor bafilomycin A₁ Nevertheless, whether the V-ATPase plays a significant role in pancreatic HCO₃⁻ secretion is not clarified, as for example in guinea pig pancreatic ducts bafilomycin A₁ could not inhibit agonist- stimulated HCO₃⁻ and fluid secretion. Therefore, in the present study we have focused on the function of H⁺/K⁺-ATPases (pumps), which are pharmacologically approachable and physiologically relevant. Such H⁺/K⁺-pumps have not been proposed for HCO₃⁻ secreting tissues, except for our earlier study on rat pancreatic ducts; rather they have well-established roles in acid secretion, and in H⁺ and K⁺ homeostasis in other tissues. The H⁺/K⁺-ATPases are classified into two subfamilies, gastric and non-gastric (latter also called colonic), coded by *ATP4A* and *ATP12A*. The gastric H⁺/K⁺-ATPase is expressed in stomach parietal cells, kidney distal nephrons and cochlea, where they are responsible for H⁺ secretion, K⁺ absorption and K⁺ recirculation, respectively. The non-gastric H⁺/K⁺-ATPase is present in several epithelial tissues including colon, kidney, skin, placenta, and prostate, and here it is associated with acid-base or K⁺ and Na⁺ homeostasis. Each pump is composed of two catalytic α-subunits and two regulatory β-subunits. The gastric α-subunit (HKα1) assembles with the gastric β-subunit (HKβ), while the non-gastric α-subunit (HKα2) can borrow the gastric β-subunit, and β3/β1-subunits of the Na⁺/K⁺-ATPase. The gastric H⁺/K⁺-ATPase is the primary target for treatment of peptic and duodenal ulcers and reflux diseases. Proton pump inhibitors (PPIs), such as omeprazole, are activated in acid environment of secretory canaliculus of the parietal cells and bind covalently to cysteines of the ATPase. Another experimental class of ATPase inhibitors are potassium-competitive acid blockers (P-CABs), such as SCH28080, though at high concentrations they may also inhibit the non-gastric H⁺/K⁺-ATPase. Our hypothesis is that the H⁺/K⁺-ATPases may be important in supporting pancreatic function, which may be of particular relevance in human pancreas. In a most simplistic way, one could envisage that these ATPases would pump H⁺ out towards the interstitium and provide HCO₃^- for luminal transport and thus fluid secretion. Thus the aim of this study was to establish whether human pancreatic ducts express functional gastric and/or non-gastric H⁺/K⁺-ATPase and whether H⁺/HCO₃^- transport and whole pancreatic secretion is sensitive to proton pump inhibitors (PPIs). For this purpose we have used human cells/tissue and performed *in vivo* studies on the rat pancreas where H⁺/K⁺-ATPases are expressed, as already established in our earlier study. We show that human duct cells express subunits of both gastric and non-gastric H⁺/K⁺-ATPases and these exhibit unusual localization patterns. We propose that these pumps have physiological functions in pancreatic H⁺/HCO₃^- transport and fluid secretion, and speculate on their additional role in mucosal protection and K⁺ recirculation. Most importantly, the present studies show that proton pump inhibitors inhibit pancreatic secretion and we speculate about consequences of using these drugs as treatment therapies in several pancreatic diseases. # Materials and Methods ## Ethical Approval The permission for animal experiments and protocols were approved by the Danish Animal Experimentation Inspectorate (license no. 2011/561-56 and 2012-15-2934-00693). The experiments were performed on male Wistar rats weighing between 250 and 400 g. Data from 22 rats treated with omeprazole, 6 rats treated with SCH28080, and 21 matched controls were included in the study. For details see below. ## Cell Culture Pancreatic cell lines (human pancreatic ductal adenocarcinoma) were purchased from ATTC (Manassas, VA, USA). PANC-1 (#CRL-1469) was grown in Dulbecco’s Modified Eagles Medium (DMEM), CFPAC-1 (#CRL-1918) and Capan-1 (#HTB-79) were grown in Iscove’s Modified DMEM (IMDM). Cell culture media contained Glutamax, 10% (20% for Capan-1) fetal bovine serum (FBS), 100 U/ml penicillin and 100 μg/ml streptomycin. Cells were grown at 37°C in a humidified atmosphere with 5% CO₂. Cells from passage 12 to 25 were used in this study. All standard chemicals were purchased from Sigma-Aldrich unless otherwise stated. ## pH_i Measurements Methods for pH_i measurements in Capan-1 cells are based on the one described elsewhere. Briefly, Capan-1 cells grown on standard Ibidi μ-Dish^35mm were loaded with 2 μ<span class="smallcaps">M</span> BCECF/AM (Invitrogen) for 20–30 min; thereafter they were superfused at 2 ml/min, at 37°C. Control perfusate had the following composition (in mM): Na⁺ 145, Cl⁻ 145, K⁺ 4, Ca²⁺ 1.5, Mg²⁺ 1, phosphate 2, HEPES 10 glucose 5; pH was 7.4. For ammonium pulses 20 mM Na⁺ was replaced with NH₄⁺. In Na⁺-free solutions, N-methyl-D-glucamine was used for replacement. Intracellular pH (pH_i) was estimated from changes in the fluorescence emission (at 510 nm) from 15–20 cells after excitation at 495 and 440 nm. Signals for each batch of cells were calibrated *in situ* with 1 μ<span class="smallcaps">m</span> ionophore carbonyl cyanide m-chlorophenyl hydrazone, CCCP, and the fluorescence ratios and pH_i were fitted to a calibration curve. A standard method of ammonium pre-pulse was used to study H⁺ transport. Tissues were exposed to ammonium pulses (2–3 min), then ammonium was removed, and pH_i recovery rates from acidosis were determined from the initial slopes of pH_i changes and expressed as *dpH/dt* (*i*.*e*. pH units/min). The following common representative acid blockers (PPIs and P-CABs) were used: omeprazole (10 μM) and SCH-28080 (10 μM) (Sigma Aldrich). Acidified ethanol (0.15 M HCl in 75% ethanol) was used to prepare omeprazole stock solutions for these experiments, as omeprazole needs acid environment to be activated. ## Reverse transcription polymerase chain reaction (RT-PCR) and real-time PCR RT-PCR and real time PCR were carried out as detailed in our recent work. Briefly, cells were cultured to confluence and then RNA was isolated with RNeasy Mini Kit (Qiagen 74104). DNase 1 (RNase free DNase Set, Qiagen 79254) was used to avoid any DNA contamination. 1 μg RNA per reaction was used in OneStep RT-PCR Kit (QIAGEN 210212) with amplification parameters as follows: one cycle at 50°C for 30 min and one cycle at 95°C for 15 min followed by 40 cycles at 95°C for 30 s, 55°C for 30 s, 72°C for 1 min, and finally, one cycle at 72°C for 10 min. For real-time PCR, cDNA was synthesized based on 5 μg RNA template per reaction using the RevertAid First Strand cDNA synthesis kit (Fermentas \#K1622) with oligo (dT)18 primers and RevertAid M-MuLV reverse transcriptase, and then purified using GenElute PCR Clean-Up Kit (Sigma, NA1020). The purified cDNA was quantified by absorbance at 260/280 nm, and 100 ng was used as template for each PCR reaction. The PCR reactions were run using LightCycler 480 SYBR Green I Master (Roche, 04707516001) with parameters as follows: pre-incubation for 5 min at 95°C followed by 45 amplification cycles of 10 s at 95°C, 1 min at 55°C, and 30 s at 72°C. A melting curve was performed following the PCR by 5 s at 95°C, 1 min at 65°C, and subsequent heating up to 97°C, and then cooling down for 10 s at 40°C. Reactions were performed as triplicates and were repeated four times. shows primers used; these were synthesized by MWGBiotech or TAG Copenhagen A/S (Copenhagen, Denmark). Four house-keeping genes, 18S ribosomal RNA (18SrRNA), β-actin, β-glucuronidase (GUSB) and glutaminyl-tRNA synthetase (QARS) were used for normalization. These genes have relatively stable expression in both normal and cancerous pancreas. ## Western Blot Analysis Pancreatic cells, stomach and colon from mice were lysed by adding 5x diluted lysis buffer (50 mM Tris-base, 0.25 M NaCl, 10 mM EDTA, 0.5% NP40, 1% TritonX-100, 4 mM NaF, pH 7.5). The final lysates were centrifuged at 15,000 g, 4°C for 15 min. Protein concentration was estimated using Coomassie protein assay (Thermo Scientific). Mouse stomach and colon were used as positive control. All solutions contained 1× Sigma protease inhibitor (Sigma S8820). Protein samples were loaded on 10% polyacrylamide precast gels (Invitrogen), separated by electrophoresis, and blotted to PVDF membranes (Invitrogen). Membranes were blocked with 5% skim milk solution in TBS-Tween (0.1%) buffer for 1 h at room temperature and incubated overnight at 4°C with primary antibodies against gastric HKα1(1:1000 rabbit monoclonal, Abcam, EPR12251 or polyclonal Calbiochem 119101), HKβ (1:1000 mouse monoclonal 2G11, Sigma, A274), non-gastric HKα2 (1:1000 rabbit polyclonal, Sigma, HPA039526) or antibody (C384-M79), kindly donated by J. J. H. H. M. De Pont and H. G. P. Swarts, and β-Actin (1:1000 mouse monoclonal C4, Santa Cruz, Sc-47778). Blots were then incubated with appropriate HRP-conjugated antibodies (DAKO or Santa Cruz), developed with EZ-ECL (Biological Industries) and visualized on Fusion FX (Vilber Lourmat) and band intensity was calculated using Bio1D software. ## Immunocytochemistry Human duct cells (Capan-1) were cultured to confluence on coverslips or Transwells (Costar, 3407), and then fixed in 4% paraformaldehyde for 15 min at room temperature. Human pancreatic sections (GeneTex) were deparaffinized according to standard procedures, and antigen retrieval was performed with 1x citrate buffer in 98°C for 20 min. After washing with PBS, monolayers or pancreatic sections were treated with 0.1 M TRIS-glycine (pH 7.4) for 15 min; washed in phosphate buffered saline, PBS, with 0.3% TritonX-100 and blocked with 10% BSA for 30 min. Subsequently, preparations were incubated with 1:50 to 1:300 dilutions of the primary antibodies recognizing the gastric HKα1, HKβ, or non- gastric HKα2 overnight at 4°C. Then preparations were incubated with secondary antibodies conjugated to Alexa 568 or Alexa 488 (Invitrogen, 1:400). 4’, 6-diamidino-2-phenylindole (DAPI, 1:400) was used for nuclear staining, and a fluorescent dye Texas Red-X phalloidin (Invitrogen; 1:40) was used for actin staining. Fluorescence was examined with 40x 1.3 NA or 63x 1.2 NA objectives in Leica SP 5X MP confocal laser scanning microscope, CLSM (Leica Microsystems, Heidelberg). Images and overlays were analyzed in Leica software and exported as TIFF files to CorelDraw for composite picture. Except for cropping, no image manipulation was used. ## In vivo animal experiments The experiments were performed on male Wistar rats and surgical procedures were similar to those described earlier for mice. Briefly, animals were anaesthetised with gas isoflurane and placed on a heated surgical table and maintained at 38^o C. The jugular vein was cannulated and thereafter anaesthesia was maintained with intravenous injection of 2 mg/100 g animal pentobarbital hourly or as needed. The abdomen was opened, and the proximal end of the bile duct was ligated. The pancreas and the common pancreatic bile duct was located and cannulated with a polyethylene cannula. The pancreatic juice was collected every 15 minutes. First, the basal secretion was collected for 30 min, and then secretion was stimulated with constant intravenous infusion (0.03 ml/min) of secretin (10 pmol/min/animal). Pancreatic juice samples were collected on ice into weighted vials; secretion rates were calculated and corrected for the animal weight. Pancreatic juice samples were stored at -20^o C for further analysis. At the end of the experiment, animals were euthanized with pentobarbital overdose and the samples of stomach contents were collected. ## Administration of PPI In acute experiments a single dose of omeprazole and SCH28080 were administrated by the intravenous injection through the jugular vein. Omeprazole was dissolved in 40% polyethylene glycol (PEG 400) and SCH28080 was dissolved in 0.4% methylcellulose-saline suspension. The doses were chosen according to the previous studies and injected two hours before collecting pancreatic secretion. Omeprazole was given in doses of 5 mg/kg animal and 20 mg/kg animal; SCH28080 was given in 10 mg/kg animal. In matched controls, animals were injected with appropriate vehicle solutions. In the long-term treatment study, animals were treated daily by subcutaneous injections of either 5 mg/kg omeprazole or vehicle (40% PEG) for 30 ± 2 days. The final dose of omeprazole was given a day before the planned operation. ## pH and ion concentrations in pancreatic juice To avoid contaminations from basal and bile secretion for pH and ion measurements, the first 3 samples were excluded from analysis. The pancreatic juice samples were equilibrated with 5% CO₂/air for 30 min and the pH was measured using a glass pH combination electrode (Hanna Instruments, nr. HI 1083). HCO₃^- concentrations were calculated using the Henderson-Hasselbalch equation. In order to determine whether PPIs were working as predicted in rats, the pH of stomach contents was also measured, similar to published studies. Stomach contents were centrifuged to obtain the liquid fraction, which was diluted with distilled water at 1:1 ratio and the pH was measured. The following analyses were performed on pancreatic juice samples. Concentrations of Cl^- were determined using QuantiChrom Chloride Assay Kit according to the supplier guidelines (Bioassay Systems). Samples were diluted 10x, transferred to 96 microwell plate together with Chloride Assay reagent, absorption at 610 nm was measured and Cl^- concentrations were calculated from the standard curve. Na⁺ and K⁺ concentrations were measured using FLM3 flame photometer (Radiometer Copenhagen). Lactate concentrations were measured using Lactate Assay kit (Sigma Aldrich) and phosphate concentrations were measured using QuantiChrom Phosphate Assay Kit (Bioassay Systems), following manufacturer’s instructions. ## Statistics For real-time PCR, relative quantification (2^-ΔΔCt) was used, where Ct denotes the threshold cycle. The level of transcripts were normalized to house-keeping genes (ΔCt) and then normalized to the expression in Capan-1 cells (ΔΔCt). The protein level was normalized to β-Actin and then to the Capan-1 protein level. The differences in gene and protein expression were tested using one way analysis of variance (ANOVA) in Sigma Plot 11 and P\<0.05 accepted as statistically significant. Data from functional measurements (pH_i and pancreatic secretion) are presented as original recordings and summaries showing the mean values ± standard error of mean, SEM. Control and test pH_i measurements were made on the same cell, and *n* refers to measurements on different batches of cells. Paired Student´s *t* test was applied. For ion concentration graphs, raw data was bound into 1 μl/min-kg intervals and each bullet shows the mean value ± SEM in secretion rates (x-axis) and ion concentrations (y-axis). Statistical analysis on data obtained from animal experiments was performed using Student′s *t*-test and ANOVA, and *P* \< 0.05 was accepted as statistically significant and denoted with *asterisks*. # Results ## Human duct cells express gastric and non-gastric H⁺/K⁺-ATPases Human pancreatic duct adenocarcinoma cell lines Capan-1, CFPAC-1 and PANC-1 are commonly used as human pancreatic duct models to study the expression and function of different ion transporters. Capan-1 and PANC-1 cells express functional CFTR, while CFPAC-1 cells have F508 deletion in CFTR and thus the protein expression and function are defect. We used the three cell lines for RT- PCR and results are shown in. We found expression of the gastric H⁺/K⁺-ATPase α subunit (HKα1, 200 bp) and the β subunit (HKβ, 136 bp), as well as the non-gastric H⁺/K⁺-ATPase α subunit (HKα2, 339 bp) in all cell lines. Real time PCR analysis is also shown in. Expression levels of transcripts for HKα1, HKβ and HKα2 subunits among different cell lines were compared with respect to Capan-1 cells. Expression of gastric and non-gastric H⁺/K⁺-ATPases on protein levels was determined in the three cell lines and with protein extracts from the mouse stomach and colon as positive controls. The HKα1 band at \~115 kDa was detected in cell lysates of the three cell lines and in the stomach. In addition, in Capan-1 cells there was a noticeable band at \~100 kDa. The HKα2 as well as HKβ were also detected in all cell lines. For HKα2, a band at \~100 kDa was detected in colon and in the three duct cell lines. A weaker band at \~115 kDa was also detectable. Similar results were also observed in the rat pancreas. For the HKβ subunit the expected band size for the core protein is 35 kDa, and bands at higher sizes indicate glycosylated subunits, most prominently glycosylated in the stomach sample. In pancreatic duct cell lysates, we detected a band with highest intensity at about 40 kDa, similar to the stomach, and also seen in the rat pancreas, and this may indicate lower degree of glycosylation of the subunit. Additionally, higher bands from 50 to 65 kDa were also observed in all cell lines. ## Localization of H⁺/K⁺-ATPases in human pancreatic duct cell lines and human pancreas The expression and localization of H⁺/K⁺-ATPases were further analyzed using immunofluorescence and confocal microscopy. Immunoreactivity for HKα1, HKα2 and HKβ subunits was observed in the three duct cell lines (data not shown), but here we focus on Capan-1 cells, which form pancreatic duct epithelia with characteristic ion transporters when grown on permeable membranes. shows images of H⁺/K⁺-ATPase α1, β or α2 subunits on non-stimulated cells. For all subunits some immunoreactivity was detected intracellularly, e.g. in vesicles, but we also observed expression of the pumps on the plasma membranes. The gastric HKα1 subunit was most strongly expressed on and close to the luminal membranes. The gastric HKβ subunit was predominantly found on the luminal side of the epithelium. The non-gastric HKα2 subunit localized to the luminal, and importantly also to the lateral membranes of the epithelium. Whole human pancreas sections showed similar staining to Capan-1 cells. shows images of ducts of various sizes, and surrounding pancreatic acini, which did not stain with HK antibodies. The gastric HKα1 localized on or close to the luminal membrane and sub-membrane vesicles in human pancreatic ducts. The gastric HKβ subunit was clearly localized to the luminal membrane, and more diffusely, possibly in vesicles, proximal to the basal plasma membrane of pancreatic duct cells. The non-gastric HKα2 subunit was detected intracellularly, as well as on the plasma membranes. ## Intracellular pH in Capan-1 cells is sensitive to proton pump inhibitors In order to evaluate the function of H⁺/K⁺-ATPases, we applied a method common to study HCO₃^-/H⁺ transport, i.e. monitoring pH_i recovery from an acid load, i.e. the NH₄⁺/NH₃ pre-pulse technique (Fig and). In order to eliminate the contribution of Na⁺ and/or HCO₃^- transporters to pH_i recovery, we used Na⁺- and/or HCO₃^—free solutions for bath perfusion. Capan-1 cells were continuously stimulated with secretin (10^–9 M) during the experiments, to imitate stimulated ductal epithelium. The pH_i recovery rate of secretin-stimulated Capan-1 cells in HCO₃^—free physiological buffers was 0.313±0.018 pH units/min, and it was significantly reduced to 0.036±0.003 pH units/min in the absence of extracellular Na⁺ (n = 15). However, Capan-1 cells were still able to defend pH_i changes even without HCO₃^- transporters and Na⁺/H⁺ exchangers. Importantly, this Na⁺ independent pH_i recovery was reduced by H⁺/K⁺-ATPases inhibitors (Fig). The gastric H⁺/K⁺ pump inhibitor omeprazole inhibited 75% of the Na⁺ independent pH_i recovery (n = 6), while SCH28080 reduced the Na⁺ independent pH_i recovery by 52% (n = 5). In addition, PPIs also reduced pH_i recovery when cells were returned to control Na⁺ containing buffer ( phase III vs. I). ## Proton pump inhibitors reduce pancreatic secretory rates The crucial question to answer now was whether the H⁺/K⁺-pumps contribute to pancreatic secretion. Therefore, the following acute and long-term *in vivo* experiments with proton pump inhibitors were performed on rats. In the first series of experiments, animals had free access to food prior to surgery. Omeprazole was given intravenously two hours before pancreatic secretion was induced with secretin. Two doses of omeprazole (5 mg/kg and 20 mg/kg animal) were tested and the results are shown in. Clearly, 5 mg/kg omeprazole had no significant effect on pancreatic secretion compared to the vehicle infusion (n = 4 and 6, respectively). High dose of omeprazole (20 mg/kg animal) had a tendency to reduce pancreatic secretion rate by about 30% (n = 6), however, statistical significance was not reached with these number of experiments. Pancreatic juice contained relatively high content of secreted proteins, i.e. 29 g/l, which would indicate enzyme secretion from acini. It is well recognized in pancreatic physiology that non- fasted animals have higher fluid and enzyme secretions, due to the endogenous hormones/transmitters that can activate acinar and duct secretions. Therefore, in all following experiments animals were fasted overnight. shows the effect of low and high doses of omeprazole, which now caused significant effects on secretion apparent 30 minutes after stimulation and maintained throughout the experiment. Integrated secretion in the first and second hour after secretin stimulation were 663±42 and 804±39 µl/h/kg in control animals (n = 4), and significantly lower in test animals in the same sample periods, i.e. 484±51 (*P* = 0.018) and 563±68 µl/h/kg (*P* = 0.014) after low dose omeprazole treatment (n = 4), and 487±58 (*P* = 0.027) and 581±60 µl/h/kg (*P* = 0.013) after high dose omeprazole treatment (n = 4). Interestingly, both 5 and 20 mg/kg omeprazole had similar effects, indicating that the maximal effective dose was reached at 5 mg/kg. This dose of omeprazole also effectively inhibited basal gastric acid secretion in rats, though 20 mg/kg dose was required for inhibition of pentagastrin stimulated gastric secretion. In our experiments, we also checked that omeprazole was working on gastric acid secretion by measuring stomach pH. In control animals pH_stomach was 4.16±0.08 and in omeprazole-treated animals it was 5.05±0.13 (n = 15, *P* = 1.00×10^–5). In order to check the possible contribution of non-gastric H⁺/K⁺ pump to pancreatic secretion, the acute effects of SCH28080 were tested. SCH28080 inhibits gastric pump and it has been reported that in high doses it can also inhibit non-gastric H⁺/K⁺ pumps. In it can be seen that administration of 10 mg/kg SCH28080 resulted in a significant reduction of secretory rates. During the first hour of secretin stimulation, the secretion was reduced from 739±50 to 438±65 µl/h/kg (*P* = 0.002) and during the second hour from 1174±92 to 501±117 µl/h/kg (*P* = 0.0009), comparing control and treatment groups (n = 7; 6). There seems to be more pronounced inhibition of secretion with SCH28080 than with omeprazole compared to their respective controls. Interestingly, given the dose of inhibitors used in our studies, we would have expected weaker inhibition by SCH28080, due to higher expected ED₅₀ values. That is, ED₅₀ of 0.8 mg/kg for omeprazole and 3 mg/kg of SCH28080 was determined for rodent gastric function. Above experiments show that acute treatment with both types of blockers (for simplicity denoted here PPIs) reduced pancreatic secretion. In order to imitate PPI treatment in humans, it was relevant to investigate the outcome of long-term omeprazole treatment. Animals were treated with omeprazole (5 mg/kg) or vehicle for 30 days and subsequently pancreatic secretion was monitored in anaesthetized animals. The results of the long-term study are represented in. Comparing control and treatment groups, during the first hour of secretin stimulation the secretion was reduced from 906±130 to 356±159 µl/h/kg (*P* = 0.019) and during the second hour from 876±49 to 403±179 µl/h/kg (*P* = 0.036) (n = 4; 4). Together, these data on short- and long-term treatment with PPIs show that they have significant inhibitory effect on pancreatic secretion. Do they also have effects on electrolyte composition of pancreatic juice and can that reveal how the secretion is formed? ## Effect of proton pump inhibitors on pancreatic juice electrolytes Results of sample analyses for acute and long-term omeprazole experiments compared to control experiments are given in. shows that HCO₃^- concentrations depend on secretory rates. Without inhibitors, in the secretory rate range 10 to 15 µl/min-kg body weight, HCO₃^- concentrations were between 60 to 80 mM. With acute omeprazole treatment, secretion rates decreased and in the secretory range 7.5 to 12.5 µl/min-kg, HCO₃^- concentrations spread between 45 to 90 mM. Prolonged omeprazole treatment resulted in low secretory rates, below 7 µl/min-kg, and HCO₃^- concentrations were around 10 to 20 mM. Also with SCH28080, HCO₃^- concentrations decreased in low secretory rates. Nevertheless, it appears that all data fall into a HCO₃^- excretion curve that follows a characteristic pattern, which is valid for several animal species and humans. shows that Cl^- excretion pattern in control and acute omeprazole samples is inversed to that for the HCO₃^- pattern, as expected. However, in samples from the long-term omeprazole treated animals, Cl^- concentrations were unexpectedly low and independent of secretory rates. There seemed to be an anion deficit in pancreatic juice, which could have been due to one or more organic or inorganic anions. We determined lactate in the juice and found that concentrations were low in both types of experiments: in control samples they were 0.79 ± 0.07 mM (20 samples form 13 animals) and between 0.60 ± 0.04 mM in PPI samples (40 samples from 18 animals). In addition, we estimated inorganic phosphate concentrations, which were between 1.5 and 3.5 mM in both control and PPI samples and given pH values in secreted pancreatic juice, we estimate that they contribute 2.5 to 6 mequivalents/l. shows typical Na⁺ excretory pattern that is independent of secretory rates. Most importantly, shows that K⁺ excretion has a positive linear correlation with the secretory rates (r = 0.642 and *P*\<0.0001). Notably, at higher secretion rates pancreatic juice K⁺ concentrations are higher than the plasma values. # Discussion This is the first study that shows the expression of H⁺/K⁺-ATPase α and β subunits in human pancreatic duct cells on both molecular and functional levels. Furthermore, substantial effect of proton pump inhibitors on the whole pancreas level confirms that H⁺/K⁺-ATPases contribute to pancreatic secretion, though finding of luminally placed ATPases is surprising. These findings may mark a shift in paradigm in our understanding of acid/base transport in the pancreas. ## Various pancreatic duct cells express H⁺/K⁺-ATPases subunits A number of experimental approaches in the present study show that several human duct cell lines, as well as human pancreatic sections and the whole rat pancreas express gastric and non-gastric H⁺/K⁺-ATPases. First, we observed the expression of both gastric and non-gastric H⁺/K⁺-ATPases on mRNA and protein levels in three different types of human duct cell lines, and these findings agree with those made on the rat pancreas. The differential expression of pumps in the three cells lines may reflect the fact that they are cancer cells with different properties. Nevertheless, Capan-1 cells, which can be grown as monolayers are good models for transepithelial ion transport in human ducts, and we used these for further functional and immunolocalization studies. The non-gastric pumps have relatively wide distribution and function in H⁺ and cation homeostasis. The targeting of the non-gastric HKα2 subunit to plasma membranes (apical or basolateral) depends on coding motifs and interactions within the subunit, and possibly association with gastric HKβ or β-subunit of the Na⁺/K⁺-ATPase, probably explains localizations to the apical and/or basolateral membranes of epithelia. The gastric pump has been reported for H⁺ secreting epithelia and its presence in the HCO₃^- secreting epithelium, the pancreas, seemed somewhat peculiar. From a number of expression studies it is known that specific motifs in the HKα subunit and the glycosylated HKβ subunit are required for targeting and functional assembly. Notably, in Western blot analysis of human material, bands of about 40\~65 kDa were detected for gastric HKβ subunit, which might indicate low glycosylation of the protein, also detected in rat pancreatic ducts. Nevertheless, we still observe similar localization of the gastric HKβ subunit with α subunits close to the plasma membrane (Figs). Thus, from molecular considerations one might expect both the gastric and non-gastric H⁺/K⁺-ATPase to be functional. The next question is how their cellular localization explains pancreatic secretion. ## Localization of the H⁺/K⁺-ATPases—possible functions on the duct epithelium The immunohistochemical data (Figs) shows that the two types of pumps have somewhat different localizations on the pancreatic epithelium. The results for human duct cell lines in particular show that the H⁺/K⁺-pumps (predominantly non-gastric type) are expressed intracellularly and on the lateral membrane of pancreatic ducts. The gastric HKα1 subunits are detected on the luminal membranes and in adjacent intracellular vesicles, resembling those in parietal cells. Also gastric HKβ subunits show most distinct placement on or close to the luminal membranes, indicating that functional gastric H⁺/K⁺ ATPase would be formed. This differential pump distribution is similar to that observed in the rat pancreatic ducts. Although immunolocalization shows some overlap between gastric and non-gastric pump distribution, the most interesting question is whether the basolateral and luminal pumps would have different functions. Let us address the basolateral pumps. The HCO₃^- driven fluid secretion can be achieved by basolateral HCO₃^- transporter loading duct cells with HCO₃^- or transport of H⁺ across the basolateral membrane and intracellular HCO₃^- generation. Thus, the H⁺/K⁺-ATPases localized on the lateral and basal membranes would fulfill HCO₃^- secretion by pumping H⁺ to the basolateral side, leaving HCO₃^- for luminal transport. Both the pH_i data and pancreatic secretion data are consistent with above considerations (Figs). The second observation that H⁺/K⁺-pumps (notably the gastric type) are also on the luminal membrane of pancreatic ducts seems at first perhaps unusual. However, similar pumps (vacuolar type proton pumps) are found in fish intestine and insect midgut that also secrete HCO₃^—rich fluid. Also airway epithelia express proton pumps (and show sensitivity to vacuolar, gastric and non-gastric pump inhibitors) on the luminal membrane, though compared to pancreas, their net HCO₃^- secretion is lower and pH of airway surface liquid layer is \< pH 7.4. The function of luminal proton pumps is unclear. Let us consider possible functions for the luminal proton pumps. We propose that the luminal H⁺/K⁺-ATPases participate in a protection mechanism in the base secreting epithelium. One may draw inspiration from the mucus-bicarbonate barrier found in stomach and duodenum, where the barrier provides a near-neutral pH and protection at the epithelial surfaces. In pancreas we have the reverse situation. Human pancreatic duct cells are able to secrete up to 140 mM HCO₃^- and luminal pH values are above 8. Secretion of H⁺ may provide protection of luminal epithelial surface. Additionally, several mucin genes have been identified in pancreatic duct cells and pancreatic mucins would be relevant in epithelial protection, and altered expression pattern of mucins is one of the important factors in development and drug resistance in pancreatic cancer. In physiological context, we propose that H⁺ secretion and mucus could provide protection against luminal base, a so called “pancreatic mucus acid barrier”. In addition, H⁺/K⁺-ATPases would serve to recirculate secreted K⁺ (see below). ## Pancreatic secretion—H⁺/K⁺-ATPases on integrative level and effect of proton pump inhibitors The observation that PPIs and P-CABs inhibit secretin-evoked pancreatic secretion in *in vivo* rat studies shows that H⁺/K⁺-pumps are in general involved in pancreatic duct secretion. Rat pancreatic secretion is sensitive to omeprazole, as even the low doses were effective. Omeprazole, the acid-activated pro-drug, would inhibit the gastric pump, while SCH28080, which competitively binds to the K⁺ site, could, most likely, inhibit both gastric and non-gastric pumps and seems more effective given the dose used in our experiments and published ED₅₀ values. However, clear functional and pharmacological distinction between the two types of pumps was not possible in our study. Nevertheless, the most important observation is that PPIs had significant effects. In pH_i studies, we acidified the pro- drug and thus the activated form of omeprazole would be formed. However, in animals, omeprazole would have to be activated by the acid environment, and the simplest theory is that this would be by the pancreatic H⁺/K⁺-ATPases directly. This might be somewhat analogous to what happens in the stomach, and thus then provides further evidence for the pumps. The long-term experiments on rats were performed to find whether omeprazole had cumulative effect on inhibition of pancreatic secretion, or whether animals adjusted to the treatment and regained normal secretory rates. Clearly, the first was the case and pancreatic secretion was further reduced by long-term PPI treatment. In addition to secretory rates, the excretory electrolyte curves indicate underlying secretory mechanisms. First of all, HCO₃^- excretory curve follows predicted relation with secretory rates. The fact that PPIs do not significantly alter the curve/form indicates that H⁺/HCO₃^- transport drives the fluid transport and thus secretion. Interesting point to recall here is that rodents, other animals and humans do show similar HCO₃^- excretory curves, indicating similar underlying mechanisms. The Cl^- excretory curves are inversed to those of HCO₃^-, which may be due to Cl^-/HCO₃^- exchange, or other more complex mechanisms, and we cannot explain anion deficit at low secretory rates. Regarding the cation excretion, Na⁺ is not dependent on secretory rates. Similar was thought to be the case for K⁺ excretion. However, our data shows that there is a positive relation between K⁺ and secretory rate, and moreover that K⁺ concentrations are higher than plasma values. Similar higher K⁺ concentrations were also reported in a few early studies. Increased K⁺ concentrations in pancreatic juice are most likely due to K⁺ secretion into duct lumen via luminal K⁺ channels, such as K_Ca3.1, K_Ca1.1 or KCNQ1. These channels are expressed on luminal membranes and their activation would provide increased driving force for secretion as already predicted earlier. In that context, it is reasonable to envisage that the luminal H⁺/K⁺-ATPase could thereby operate and in fact serve to recirculate or salvage K⁺ from the lumen, and would explain the observation that H⁺ and K⁺ transport across the luminal membrane is inversely related. ## Clinical implications Proton pump inhibitors (PPIs) are used widely in clinical practice as one of the treatments for various acid-peptic disorders. As adjuvant therapy PPIs are also prescribed to patients with pancreatic diseases, such as cystic fibrosis and pancreatitis. Moreover, they are also suggested as an adjuvant therapy for patients with type 2 diabetes, supposedly because they elevate plasma gastrin and thereby improve insulin secretion. Our acute and long-term experiments on rats show that pancreatic secretion is significantly inhibited by PPIs. Therefore, considering the parallels between H⁺/K⁺-ATPases expression in rodent and human pancreatic epithelia, it would be important to re-consider effect of PPIs on human pancreatic function. Furthermore, our recent data mining analyses indicate that on the mRNA level *ATP4A* and *ATP4B* are down-regulated in pancreatic cancer. Therefore, the role of H⁺/K⁺-ATPases, alongside with other acid-base transporters, should be considered in potential development of deregulated acid- base homeostasis in pancreatic ductal adenocarcinoma. ## Conclusions In conclusion, we find that pancreatic ducts express both gastric and non- gastric H⁺/K⁺-ATPases. These pumps are functional in human duct cells and in rat pancreas on the whole organ level where they contribute to secretion as demonstrated by PPIs. The laterally and basolaterally expressed H⁺/K⁺-ATPases would export H⁺ out of the cell, leaving HCO₃^- for luminal transport and duct secretion. Localization of the H⁺/K⁺-ATPases to the luminal membrane is a change in paradigm, and we propose that they may contribute to the protective buffer zone and K⁺ recirculation. Lastly, the solid evidence that we provide for the H⁺/K⁺-ATPases in pancreas calls for re-evaluation of the use of PPIs as they will not only affect stomach secretion but also pancreatic secretion—contrary to what may be wished. # Supporting Information We are grateful to J. J. H. H. M. DePont and H. G. P. Swarts for providing us with the non-gastric pump antibody. Greatly appreciated are the fruitful discussions with A.M. Heegaard, and technical assistance of Pernille Roshof and Nynne M. Christensen. Special thanks to Y. Gao, P. Hyttel and H. Pilegaard for providing the Real time PCR machine. Images were taken at the Center for Advanced Bioimaging (CAB) the University of Copenhagen. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: IN JW DB MT. Performed the experiments: JW DB MT AG IN. Analyzed the data: JW DB MT AG CES. Contributed reagents/materials/analysis tools: IN. Wrote the paper: IN JW DB CES. [^3]: Current address: National Institute for Viral Disease Control and Prevention; Chinese Center for Disease Control and Prevention, Chang-Bai Rd 155, Beijing, China

# Introduction Dendritic cells (DCs) play a pivotal role in the induction of anti-tumor immunity. Immunization approaches employing DCs pulsed with tumor lysates or apoptotic tumor cells have been examined in murine models. Vaccination with tumor antigen-pulsed DCs can induce anti-tumor T cell proliferation, cytotoxicity, and regression of established tumors. Vaccination with tumor antigen-pulsed DCs has been shown to be safe and effective at inducing anti- tumor T cell responses in patients with advanced cancers. We have previously demonstrated upregulated macrophage receptor with collagenous structure (MARCO) expression in DCs pulsed with LPS or tumor lysates. MARCO is a member of the class-A scavenger receptor family and binds to Gram-positive and negative bacteria. MARCO was initially characterized on macrophages in the marginal zone of the spleen and in the medullary cord of the lymph nodes. MARCO has been shown to be important for immune responses to bacterial infections by mediating the binding and phagocytosis of pathogens. MARCO has also been implicated in the formation of lamellipodia-like structures and dendritic processes in macrophages and dendritic cells. In this study, we further examined the role of MARCO in DCs. We have developed MARCO knockout mice and examined the phenotype and efficacy of DCs generated from the bone marrow of these mice at inducing anti-tumor immunity in a murine B16 melanoma model. # Materials and Methods ## Animals Six-to twelve-week-old female C57BL/6 mice were purchased from Charles River Laboratories, Inc. (Wilmington, MA) and Harlan Laboratory (Indianapolis, IN). MARCO-deficient mice (MARCO^-/-) were developed on the C57BL/6 background in our laboratory. This study was carried out 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 reviewed and approved by the Institutional Animal Care and Use Committee at the University of South Florida (# A4100-01). Mice were humanely euthanized by CO₂ inhalation according to the American Veterinary Medical Association Guidelines. Mice were observed daily and humanely euthanized if a solitary subcutaneous tumor exceeded 1.5 cm in diameter or mice showed signs referable to metastatic cancer. All mice were maintained in the Animal Research Facility at the H. Lee Moffitt Cancer Center and Research Institute. ## PCR To perform PCR, tail snips were collected from mice that were less than 21 days old. DNA from tail snips was isolated with Wizard SV Genomic DNA Purification System (Promega). The genotype of MARCO^-/- mice was confirmed by PCR using the following primers: MARCO-F: 5’ GGT TGG TTT GGT GGC TAT GTA GAG 3’ (Intron 13) MARCO-R: 5’ CCG GAC GCG TTG GAA AGA TT 3’ (Exon 16) neo: 5’ CAA AAC CAC ACT GCT CGA CA 3’ (*neo*^*r*). MARCO-F and MARCO-R primers were used to verify the WT allele. MARCO-F and neo primers were used to verify the targeted allele. Either combination of primers results in a 2.0 kbp PCR product. ## Reverse Transcriptase Polymerase Chain Reaction (RT-PCR) For detection of MARCO mRNA, DC were unpulsed, pulsed with 1 µg/ml LPS (from *E. coli* 0111: B4; Sigma), or B16 tumor lysate for 24 hours. To isolate mRNA from DC, RNeasy Micro Kit (Qiagen, Valencia, CA) was used according to the supplier’s instructions. For RT-PCR reactions, 100 ng mRNA was used to synthesize cDNA with Ready-to-Go RT-PCR beads (Amersham Biosciences, Buckinghamshire, England). The cDNA synthesis reaction was performed at 37°C for 60 min followed by 95°C for 5 min. After the cDNA reaction, 400 nM primers were added into the reaction mixture. The following primers were used for murine MARCO PCR reactions; Sense: 5’-GCA CTG CTG CTG ATT CAA GTT C-3’, Anti-sense: 5’-AGT TGC TCC TGG CTG GTA TG-3’ (205 bp product). ## Cell Line and Culture Medium The B16-BL6 (denoted B16) melanoma cell line was derived from a spontaneous melanoma in a C57BL/6 mouse and is considered poorly immunogenic. B16 was cultured in complete medium (CM) and maintained by serial *in vitro* passage. CM consists of RPMI1640 medium (Mediatech, Inc., Herndon, VA) supplemented by 10% heat-inactivated fetal bovine serum (Atlanta Biologicals, Lawrenceville, GA), 1 mM sodium pyruvate (Mediatech, Inc.), 0.1 mM non-essential amino acids (Mediatech, Inc.), 100 units/ml penicillin, 100 µg/ml streptomycin (Mediatech, Inc.), 50 µM 2-mercaptoethanol (Sigma, St. Louis, MO), 0.5 µg/ml fungizone (Cambrex, Walkersville, MD), and 10 µg/ml gentamicin (Cambrex). ## Generation of DC and Pulsing with Tumor Lysate Bone marrow cells were collected from the femurs and tibias of C57BL/6 wild-type and MARCO^-/- mice under sterile conditions. Erythrocytes were lysed with ACK lysing buffer (0.15 M NH₄Cl, 1 mM KHCO₃, and 0.1 mM EDTA in sterile water). Erythrocyte-depleted bone marrow cells were then washed twice with Dulbecco’s phosphate-buffered saline (PBS) (Mediatech, Inc.) and resuspended in CM containing 20 ng/ml of recombinant mouse granulocyte/macrophage colony-stimulating factor (GM-CSF) and recombinant mouse interleukin-4 (IL-4) (both from R&D Systems, Minneapolis, MN) at the concentration 1 x 10⁶ cells/ml, and then incubated at 37^oC and 5% CO₂. On day 5, non-adherent cells were collected and DCs were enriched by density centrifugation over OptiPrep (Axis-Shield PoC AS, Oslo, Norway). Analysis of collected cells by flow cytometry revealed that the DC population was \>80% positive for MHC class-II, CD80 and CD86, and \>70% positive for CD11c (data not shown). To generate B16 lysate, B16 melanoma cells were collected using trypsin/EDTA solution (Mediatech, Inc., Herndon, VA), washed twice with PBS and resuspended at 30 x 10⁶ cells/ml in PBS. The cells were subjected to four cycles of rapid freeze and thaw exposures. DCs were pulsed with tumor lysates at a ratio of 1 DC to 3 cell lysate equivalents for 18-24 hours. ## Flow Cytometry For flow cytometric experiments, the following anti-mouse antibodies were used: Fluorescein isothiocyanate (FITC)-conjugated anti-mouse I-A^b, CD8, CD11b, CD14, CD86, Ly6C and NK1.1; phycoerythrin (PE)-conjugated anti-mouse CD4, CD11b, CD11c, CD18, CD40, Ly6G and MARCO; APC-conjugated anti-mouse CD3, Gr-1 and CCR7 (all from BD Biosciences, San Jose, CA). For cell surface marker staining, cells were washed with flow buffer (0.01% NaN₃, 2% fetal bovine serum in PBS) and Fcγ II/III receptor blocking was performed by purified anti-mouse CD16/32 antibody (BD Biosciences). The blocking antibody (1µg/1 x 10⁶ cells) was added and cells were placed on ice for 10 min. After the blocking procedure, antibodies (1µg/1 x 10⁶ cells) for cell surface staining were added into each sample and placed on ice for 30 min protected from light. After two additional washes with buffer solution, all cells were fixed with 1% paraformaldehyde. Data acquisition was performed by flow cytometry (FACScan or FACSCalibur, BD Biosciences) within 24 hours after sample fixation. Data analysis was performed with CellQuest software (BD Biosciences). ## ELISA To measure cytokine secretion, day 5 DC were not pulsed, cultured with B16 tumor lysate, or cultured with LPS. After 24 hours, culture supernatants were harvested for measurement of cytokine production by standard ELISA (BD Biosciences). To measure IFN-γ release, spleens were collected from mice receiving PBS, WT DC or MARCO^-/- dC.T cells were purified, then restimulated with CM, WT DC or MARCO^-/- DC. Supernatants were collected 48 hours later and tested for IFN-γ release by standard ELISA (BD Biosciences). ## Microscopy Tumor lysate was stained with PKH26 red according to the manufacturer’s protocol (Sigma-Aldrich, St. Louis, MO). After 24 hours, the cells were harvested and washed with PBS. Lysate-pulsed DCs were stained with rat anti-MARCO mAb followed by the staining with Alexa Fluor 594 chicken anti-rat IgG (Invitrogen Corp.). Lysate pulsed-DC were washed twice with PBS, fixed with 1% PFA, and spun onto glass slides by a Shandon Cytospin-2 (International Medical Equipment, Inc., San Marcos, CA) at 800 rpm for 5 min. The slides were mounted with Vectashield mounting medium containing DAPI according to the manufacturer’s instructions (Vector Laboratories, Inc., Burlingame, CA). Slides were viewed with a fully automated, upright Zeiss Axio-ImagerZ.1 microscope (Carl Zeiss MicroImaging, Inc., Thornwood, NY). Images were produced using the AxioCam MRm CCD camera and Axiovision version 4.5 software suite (Carl Zeiss MicroImaging, Inc.). ## *Migration* In Vitro and In Vivo DC pulsed with tumor lysate were used as responders at a concentration of 3 x 10⁶ DC cells/ml. Assays were performed in 24-well plate format with 5 µm pore polycarbonate Transwell inserts (Corning Inc, NY). Secondary lymphoid tissue chemokine (SLC, CCL-21) was added to the lower chambers at 100 ng/ml in CM. DC was added to the top chamber at 100 µl and incubated at 37^oC with 5% CO₂ for 5 hours. A 1:5 dilution of the cells was also directly added to the lower chamber of two wells for determination of the input amount. The inserts were removed and the cells were harvested, washed twice in PBS and stained with MHC Class II (I-A^b) and CD11c to detect the DC. To quantify the number of migrated cells, polystyrene beads (Bangs Laboratories, Fishers, IN) were added to all samples and analyzed by flow cytometry (FACScan or FACSCalibur, BD Biosciences). The number of cells in each sample was determined by the equation: (number of cell events/number of bead events) x 10⁴ beads/sample. The percentage of migration in each sample (% input) was determined by the equation: (number of cells in sample/(number of cells in input x 5)) x 100. For *in vivo* determination of migration of DC, BM-DC were pulsed with tumor lysate, stained with PKH26 red, and 5 x 10⁶ cells were injected into WT mice. After 60 hours, the vaccine-draining lymph nodes were collected. For confocal microscopy, the lymph nodes were fixed with 3.7% formaldehyde. For flow cytometry analysis, cells were stained with MHC Class II (I-A^b) and CD11c. Polystyrene beads were added to all samples and analyzed by flow cytometry (FACScan or FACSCalibur, BD Biosciences). The cell number was calculated according to the following: 200,000/counted micro beads x number of cells positive for MHC Class II, CD11c, and PKH26 red. ## Treatment of Established Subcutaneous Tumor Wild type and MARCO^-/- mice received 1 x 10⁵ viable B16 melanoma cells subcutaneously (s.c.) in the right flank. Three days after tumor injection, mice were treated with 1 x 10⁶ DC pulsed with tumor lysate or PBS in the left flank. Additional DC treatments were given on days 6 and 9 for a total of 3 DC injections. Tumor sizes were measured every 2-3 days after tumor injection. ## Statistical Analysis All results for continuous variables are reported as mean ± SEM. Statistical analysis was performed by one way t-test. P values of 0.05 or less were considered statistically significant. Tumor treatment models were analyzed with GraphPad Prism (GraphPad Software, San Diego, CA). # Results ## Generation of MARCO^-/- Mice A 228 bp probe containing exons 16 and 17 of the MARCO gene was used to screen a B6 LAMBDA DASH library (Stratagene, La Jolla, CA). As shown in , the wild-type allele containing the MARCO gene was replaced using a targeting vector containing a Neomycin resistance (*neo*^*r*) gene. Briefly, the targeting vector was used to transfect B6 embryonic stem (ES) cells. Cells that had undergone homologous recombination with the targeting vector were resistant to both neomycin and acyclovir. Cells that randomly inserted the targeting vector were resistant to neomycin but sensitive to acyclovir. Gene insertion in positive clones selected in both neomycin and acyclovir was verified by PCR and Southern Blot (not shown). Positive ES clones were selected and micro-injected into albino C57BL/6 blastocysts. Blastocysts were implanted into pseudo-pregnant B6 females. Chimeras were bred to B6 albino mice and black offspring were screened by PCR to determine if the MARCO gene exons 16-17 had been deleted. Mice heterozygous for the MARCO gene knockout were bred to generate a homozygous strain of MARCO knockout mice (MARCO^-/-). shows verification of the absence of the MARCO gene at the DNA level. MARCO gene knockout was also verified by Southern Blot analysis (not shown). ## MARCO Expression in DC We first compared the expression of MARCO in DCs generated from the bone marrow of wild-type (WT) and MARCO^-/- (K/O) mice by gene expression analyses. As shown in, WT DCs express MARCO and MARCO expression is upregulated after exposure to either tumor lysate or LPS. Expression of MARCO in MARCO^-/- DCs was absent, even after pulsing with tumor lysate or LPS. confirms the loss of MARCO cell surface expression in DCs generated from MARCO^-/- mice. DCs from WT and MARCO^-/- mice were pulsed with tumor lysate, stained, and examined by microscopy. ## Comparison of Cell Surface Marker Expression in Lymph Node, Spleen, and DCs Lymph node cells ( and splenocytes ( from WT and MARCO^-/- mice were compared for surface expression levels of a selected battery of ancillary immune markers by flow cytometry. In contrast to MARCO expression, there were no significant differences detected between them. DCs generated from the bone marrow of WT and MARCO^-/- mice were unpulsed ( or pulsed with LPS ( or tumor lysate (not shown). No difference in expression levels of co- stimulatory molecules on the surface of DCs was measured. ## Comparison of Cytokine Production by DCs We also compared cytokine production of IL-12, IL-10 and TNF-α by unpulsed or LPS-pulsed DCs from WT and MARCO^-/- mice. As shown in, the levels of cytokine production measured were comparable. ## Comparison of In Vitro and In Vivo Migration of DC We next examined the effect, if any, of loss of MARCO on DC migration *in vitro*. DCs were unpulsed or pulsed with tumor lysate before their use in a micro-chemotaxis assay. DC migration was evaluated using the CCR7 ligand, CCL21, which is a known chemokine for DC migration. As shown in, MARCO^-/- DCs that were pulsed with tumor lysate demonstrated a significantly higher migratory ability than WT DC. WT DC and MARCO^-/- DC expressed the same level of CCR7 expression on their respective cell surfaces (data not shown). For *in vivo* analysis, DCs from WT and MARCO^-/- mice were stained with PKH26 red dye and injected s.c. into WT mice. The draining lymph nodes were collected 60 hours later and examined for DC accumulation by confocal microscopy. As shown in, the lymph nodes of WT mice injected with MARCO^-/- DCs showed a dramatic accumulation of these injected cells, which was confirmed by flow cytometry (p\<0.05 WT compared to K/O). In this latter study, single cell suspensions of the draining lymph nodes were prepared, depleted of erythrocytes, stained with MHC Class II and CD11c antibody, and microbeads added to enumerate the number of cells infiltrating the lymph node. ## IFN-γ Production by Stimulated T Cells We also examined the capacity of antigen loaded DCs to effectively prime T cells *in vivo* following s.c. injection. As shown in cells from WT mice immunized with tumor lysate-pulsed MARCO^-/- DCs and then restimulated *in vitro* with tumor lysate-pulsed MARCO^-/- DCs produced the highest levels of IFN-γ (p\<0.001 compared to all other groups). ## Treatment of Established B16 Melanoma with WT and MARCO^-/- DCs To test the therapeutic efficacy of MARCO^-/- DC treatment on tumor growth, we first injected B16 tumor cells s.c. into C57BL/6 mice. Mice were then vaccinated on days 3, 6 and 9 with WT or MARCO^-/- DCs. As shown in, mice treated with MARCO^-/- DCs demonstrated delayed tumor growth (p\<0.05 compared to other treatment groups). These mice also exhibited prolonged survival as compared with the mice that received WT DCs or PBS (, p\<0.05 compared to other groups). # Discussion The current study has shown that loss of expression of the scavenger receptor MARCO enhances the trafficking and anti-tumor efficacy of DCs pulsed with tumor lysates. DCs generated from the bone marrow (bm) of MARCO^-/- mice were phenotypically identical to DCs generated from the bm of wild-type mice. DCs expressed high levels of MHC and co-stimulatory molecules and responded to LPS by producing IL-12 and TNF-α. No difference in phagocytic capacity was measured between MARCO^-/- and wild-type DCs after 24 hours of lysate pulsing (not shown). MARCO^-/- DCs were morphologically normal. This was surprising as MARCO has been implicated in actin cytoskeleton rearrangements and our previous studies demonstrated that targeting MARCO with a monoclonal antibody led to a loss of dendritic-like processes on the surface of DCs. While the role of scavenger receptors in innate immunity is well documented, the role of these receptors in T cell priming and adaptive immunity is less defined. Scavenger receptors are pattern recognition receptors with broad ligand-binding specificities. Scavenger Receptor A (SRA) and MARCO are members of the Class A Scavenger Receptor family and are expressed in macrophages and dendritic cells. Both act as phagocytic receptors of multiple ligands including pathogenic bacteria and apoptotic cells and participate in cell adhesion and migration. While these receptors cannot initiate inflammatory responses by themselves, SRA and MARCO may play a role in attenuating innate immune responses to ligands. Blockade of SRA and MARCO has been shown to enhance inflammation in response to LPS. Recent studies have demonstrated the ability of scavenger receptors to modulate the induction of T cell responses. It has been reported that vaccination with DCs generated from the bone marrow of SRA knockout mice leads to enhanced T cell priming and anti-tumor efficacy. In our study, vaccination with tumor lysate pulsed, MARCO-deficient DC led to enhanced anti-tumor T cell activity and regression of B16 melanoma. We performed additional studies in which mice were vaccinated with tumor lysate-pulsed DC generated from the bone marrow of mice deficient in both MARCO and SRA. In these studies, there were no differences in anti-tumor T cell activity or tumor regression observed between vaccination with MARCO^-/- DCs and the double knock-out DCs (not shown). These results may indicate a redundant role of SRA and MARCO in DCs for the induction of adaptive immunity. Vaccination with tumor antigen-pulsed DCs has been shown to be safe and effective at inducing anti-tumor T cell responses in patients with advanced cancers, although these responses rarely translate to clinical efficacy. As injected DCs must migrate to a draining lymph node to stimulate antigen-specific CD4⁺ and CD8⁺ T cells, strategies to improve the migration of DCs have been explored. Intradermal injection of DCs results in small numbers of DC trafficking to draining lymph nodes with the majority of DCs localized to the vaccination site. Direct intranodal injection of DCs to enhance anti-tumor T cell activation by increasing the number of DCs localized in the draining lymph node has also been explored. Several studies have shown that although higher numbers of DCs can be detected in the lymph node after intranodal injection, the quality of T cell responses were inferior to those induced by intradermal vaccination of DCs. It has been suggested that migration of DC is required for maximum maturation and potency resulting in the induction of strong, durable anti-tumor immune responses. We have previously shown that exposure of tumor lysate-pulsed DCs with an anti-MARCO antibody enhanced DC migration to draining lymph nodes. In this study, we have verified that loss of MARCO expression in DC enhances migration *in vitro* and *in vivo*. We believe this enhanced migration led to the enhanced anti-tumor T cell activity and tumor regression that we observed. Our studies have confirmed the role of MARCO on the migration of injected DCs from a subcutaneous injection site to the draining lymph node. We have previously demonstrated the expression of MARCO on the surface of human monocyte-derived DCs. Strategies to target MARCO on the surface of human DCs may improve responses in clinical trials employing DC-based vaccination. We thank the staff of the shared resources (analytical flow cytometry, biostatistics, flow cytometry, and mouse model cores) at the Moffitt Cancer Center for excellent technical assistance and support. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: HK NM JJM SPT. Performed the experiments: HK NM LK SPT. Analyzed the data: HK NM LK JJM SPT. Contributed reagents/materials/analysis tools: JJM SPT. Wrote the manuscript: HK LK JJM SPT.

# Introduction Selenium (Se) is an essential trace element, primarily to protect against oxidative stress. Current dietary recommendations include a safety margin to avoid loss of antioxidant protection at low concentrations and the risk of Se toxicity at high concentrations. In addition, data on the importance of Se intake on long-term health are emerging that give rise to other concerns. There are indications that Se is involved in the prevention of diseases such as cancer and cardiovascular diseases and impaired immune function, etc. On the other hand, the Se and vitamin E cancer prevention trial (SELECT) reported an increased risk of diabetes mellitus type 2 in humans with a high Se intake (200 μg selenomethionine/day). The effects of selenium intake on the health and longevity of companion animals such as the dog is now also coming under scrutiny. The current recommended allowance for Se in dog food takes into account a bioavailability factor, but is not diet type specific, despite several studies reporting a large variation in Se digestibility between diet types. We have previously shown in 20 canned and 23 kibble diets, the average amount of Se was higher in canned than kibble diets (34.8 vs. 22.5 μg/MJ, respectively). Also, apparent crude protein (CP) digestibility had a significant effect on *in vitro* Se accessibility, i.e. the dietary Se fraction in the filtrate after *in vitro* digestion. The reasons for the differences in Se digestibility between canned and kibble diets may be due to the way the diets are processed or the raw materials used during manufacture. The raw materials used for pet foods vary greatly in their Se concentration and availability. Canned diets typically contain higher concentrations of meat and/or meat by-products, whilst kibble diets contain mainly cereal grains and cereal grain by-products. Wedekind *et al*. tested the availability of Se in several pet food ingredients and found a range of 9% in mackerel to 38% in beef spleen when compared to the availability of sodium selenite. Due to the differences in ingredients, it is likely that the chemical form of Se in the diet will also be different between canned and kibble diets, which is known to be an important factor for the digestibility of Se. Canned and kibble diets also differ in the way of processing. Kibble diets are extruded and canned diets are retorted, resulting in differential effects of heat, pressure and shear which may also impact on Se digestibility. The differences between canned and kibble diets may not only cause a variation in Se digestibility between diet types, but also the bioactivity, defined as the amount of Se that can be used for the incorporation into selenoproteins. The antioxidant glutathione peroxidase (GPx) is often used as a measure of Se bioactivity. In addition, serum isoprostanes (IsoPs) are a measure for lipid peroxidation and are used as an indicator of total antioxidant status. Selenium is not only incorporated into GPx, but also into iodothyronine deiodinases, of which type I and II are involved in the transformation of thyroxine (T4) into triiodothyronine (T3). Olivieri *et al*. has shown that selenium status was positively correlated with the ratio of T3:T4 in humans and Wedekind *et al*. have reported similar results in puppies. Based on the *in vitro* results, it was hypothesized that canned diets have a lower Se digestibility than kibble diets and that CP concentration (as the main factor for the ingested amount of digestible protein) is positively associated with Se digestibility in kibble diets and negatively in canned diets *in vivo*. To test this hypothesis, two different diet types (canned and kibble) and four different concentrations of CP per diet type were selected. Diets were selected to reflect the processing conditions used in the pet food industry, including the characteristics that are associated with diet type and dietary CP concentration. # Experimental Methods ## Study design The study consisted of four feeding periods with six transition days and experimental periods of between 29–43 days. Four groups of each six dogs were selected. Each dog group was fed four of the eight experimental diets (2 canned, 2 kibble) in a randomised incomplete cross-over design. Blood, urine and faecal samples were taken at the end of every feeding period. This study (HO0647) was approved by the WALTHAM Centre for Pet Nutrition Animal Welfare and Ethical Review Body and was conducted under Home Office Project License authorisation. ## Dogs Twenty-four adult Labrador retrievers, of which 15 female (14 neutered, 1 entire) and 9 male (all neutered), were selected for this study. All dogs used in this study were bred at WALTHAM or sourced from Home Office approved breeders for research purposes. At the start of the study, the average age of the dogs was 4.2 years (range 2.0–8.1) and the average body weight (BW) was 27.9 kg (range 23.7–32.7 kg). The dogs were fed twice daily (8:30 and 15:00) to maintain BW and body condition score (BCS) at ideal (D) on the S.H.A.P.E^™ (Size, Health And Physical Evaluation) BCS-scale. BW and BCS were recorded weekly and food intake daily. Dogs had access to fresh drinking water at all times. Dogs were divided into four treatment groups of six dogs, with age, BW and gender as blocking factors. Dogs were housed in triplets with others of the same diet group indoors with continuous access to a concrete outside pen. Between 9:00 and 14:30, the dogs also had access to a concrete outside paddock and they were socialized at least once a day (e.g. toy play or walk). During the last six days of each feeding period, dogs were housed individually with access to the outside paddock during the day. The average age, body weight and energy intake in kJ/kg BW^0.75 per dog group are shown in. ## Diets Eight commercially available diets were selected that varied in diet type (canned or kibble) and protein concentration. Commercial diets were chosen for practical relevance of the study, and the difference in Se concentration and other parameters (e.g. ingredients, processing conditions, Se species) between the canned and kibble diets are considered inherent to the diet types. For every diet type, diets with an estimated (from the pet food labels) crude protein concentration of 9.6, 14.3, 19.1 and 23.9 g CP/MJ ME (40, 60, 80 and 100 g CP/1000 kcal ME, resp.) were included. Four different protein concentrations were selected, rather than diets with different protein digestibility coefficients. This resulted in a range of the absolute amount of digestible protein, as the limited differences in protein digestibility in commercially available diets are usually overruled by the protein concentration *per se*. All diets were single-batch, to prevent differences in nutrients over time due to variations in ingredients. Selected canned diets were Royal Canin^® Canine Veterinary Diet Hepatic, Sensitivity Control Duck & Rice, and Recovery and Canine Veterinary Care Nutrition Senior Consult Mature. Selected kibble diets were Royal Canin^® Canine Veterinary Diet Renal, Gastro Intestinal Moderate Calorie, and Satiety Weight Management and Canine Veterinary Care Nutrition Pediatric Junior Large Dog. The canned diets contained Se solely from the ingredients, which is primarily organically-bound Se, whereas Se in the form of sodium selenite was supplemented in all kibble diets before processing (standard, not specifically for this study). Supplemented amounts of sodium selenite were 0.05, 0.08, 0.08, and 0.08 mg/kg, respectively. As veterinary diets were used in this study, some of the diets were supplemented with additional nutrients in order to be nutritionally complete and balanced for healthy adult dogs with energy requirements of 397 kJ (95 kcal)/kg BW^0.75. Details of the supplementation are included in. Each dog was fed individually to maintenance requirements. ## Blood samples At the end of every feeding period, blood samples (13ml) were taken at 13:00 h from each dog by jugular venipuncture using a 21 gauge needle. Blood was put into 2 Microvette^® 500 μl lithium-heparin tubes, one 300 μl fluoride- heparin tube, one 200 μl Tri-Kalium-EDTA tube and 2 Vacuette^® z serum clot activator tubes (1× 9 ml and 1× 4 ml). One of the lithium-heparin tubes was used for whole blood GPx analysis immediately after collection. The remaining heparin tubes were centrifuged (accuSpin^™ Micro R, Thermo Fisher Scientific Inc.) immediately after collection at 1680 g and 4°C for 10 min. Lithium-heparin plasma was analysed for general biochemistry parameters and fluoride-heparin plasma for glucose (Spectrophotometry, Olympus AU400). Tri- Kalium-EDTA tubes were placed on a roller at room temperature until analysis for haematology parameters (Mythic 18 Vet analyser, Orphée S.A.). Biochemistry and haematology measurements were used as a general health check to confirm that all dogs were in good health. Serum tubes were incubated for 30 min on ice and then centrifuged (Sigma 6K15, rotor 11150, cups 13550, Sigma GmbH) for 10 min. at 2000 ×g and 4°C. Serum samples were divided into centrifuge tubes and stored at -80°C for analysis of Se, IsoPs and triiodothyronine (T3) and thyroxine (T4) at the end of the study. ## Urine samples Free catch urine was collected in the section of the day between meals at the end of each feeding period, using a Uripet urine collection device (Rocket Medical plc., Watford, England). 1 ml of urine was stored in at -80°C and analysed for creatinine (CT) within one month after sampling (IDEXX laboratories, UK). The rest of the sample was stored at -20°C and analysed for Se content (ICP-MS). ## Faeces samples During the last six days of each feeding period, titanium dioxide (TiO₂) was added to the diets as a digestibility marker at a concentration of 0.45 g/day (mean 1.5 g/kg DM, range 0.67–2.14 g/kg DM). Total faeces were collected during four consecutive days at the end of each diet phase. Faeces were homogenised with a mixer (Kenwood professional PM900, Kenwood LTD) for 1 minute and a sample of approximately 350 g (mean 346, range 245–410 g) was taken. Samples were freeze-dried (SuperModulyo^®, Thermo Fisher Scientific Inc.) until a stable weight, at -50°C. Faeces were analysed for TiO₂, DM, ash, nitrogen (N) and Se. N and Se were used for the calculation of apparent Se and CP (N×6.25) digestibility coefficients, respectively, with the use of TiO₂ as a marker by means of the following formula: $$\text{Apparent~digestibility~\%} = 100 - (100\ \times \left( {\frac{\text{\%~marker~in~diet}}{\text{\%~marker~in~faeces}}\ \times \ \frac{\text{\%~nutrient~in~faeces}}{\text{\%~nutrient~in~diet}}} \right))$$ ## Chemical analyses Biochemistry and glucose analyses were carried out using spectrophotometry (Olympus AU400, Olympus Inc.) with Beckman Coulter reagents (Beckman Coulter Biomedical LTD) within 20 mins of sampling. Whole blood GPx was also analysed within 20 mins of sampling using the Ransel kit (Randox laboratories LTD) on the Olympus AU400 spectrophotometer. A 4-point calibration curve was used as control. Whole blood GPx analysis had an average CV of 1.4% in the samples and the 4-point calibration curve showed a recovery of 83.1, 93.7, 98.5, and 96.4%, respectively. Haematology parameters were analysed using a Mythic 18 Vet analyser (Orphée S.A.). Thyroid hormone analyses (T3 and T4) were performed according to the method of Darras et al. and the IsoPs were analysed using an enzyme-linked immunosorbent assay (8-isoprostane EIA kit, Cayman Chemical Co.). Serum, urine and faeces samples were prepared for total Se analyses with closed vessel microwave destruction as described in van Zelst *et al*.. Se was analysed using inductively coupled plasma-MS (ICP-MS, Elan DRC-e, PerkinElmer), as described by Lavu *et al*.. Urine CT was determined using a creatinine kit based on the Jaffe reaction (OSR6178, Beckman Coulter Biomedical LTD, IDEXX Laboratories, UK). Faeces samples were analysed for DM and crude ash by drying to a constant weight at 103°C and combusting at 550°C, respectively. The Kjeldahl method (ISO 5983–1, 2005) was used to determine CP (N×6.25). TiO₂ was analysed according to the method of Myers *et al*.. ## Statistical analyses Data were analysed using linear mixed effect models in RStudio (version 0.98.507, RStudio,Inc.) using the nlme package. Two primary response variables were measured: GPx and urinary Se:CT ratio relative to the Se intake (relative urinary Se:CT ratio). To account for two primary response variables, the statistical test level was corrected to 0.05/2 = 0.025 for these two parameters. The secondary response variables in this study were: Energy intake, Se intake, serum Se, serum IsoP, serum T3:T4 ratio, Se digestibility, apparent CP digestibility, digestible Se intake (the absolute amount of digestible Se), digestible CP intake and urinary Se:CT ratio relative to the digestible Se intake. For these response variables a *p*-value of \<0.05 was used as the threshold for significance. The model contained a fixed effects structure of actual CP intake (g/kg BW^0.75), diet type and their interaction and a random structure of dog. If the interaction was not significant (*p*≥0.05), the model was reduced to CP intake and format as fixed effects and dog as random effect. Energy, Se, digestible Se and digestible CP intake were confounded with CP intake, and therefore CP intake was omitted from the model for these parameters. Distributional assumptions were checked by visual inspection of residuals. Results are reported as means (± SEM). To assess the impact of time between a dog’s last meal and urine collection on the relative urinary Se:CT ratio, a likelihood ratio test was carried out by including time into the fixed effects structure of the model. No significant differences were found when time was included as a parameter; consequently, time was omitted from the structure of the mixed effects model for relative urinary Se:CT ratio. # Results All biochemistry and haematology results were within the normal range for healthy dogs (data not shown). The average Se concentration of the kibble diets was 22.3 μg/MJ (range 13.9–30.3) and 40.6 μg/MJ (range 34.8–47.5) for canned diets. Energy intake did not differ between the diet types, but Se intake was lower on kibble than canned diets. Apparent CP digestibility showed a significant diet type by CP intake interaction (*p*\<0.001). Apparent CP digestibility increased with increasing CP intake, but only in canned diets. The digestible CP intake did not differ between diet types. Kibble diets had a higher Se digestibility on average than canned diets (*p*\<0.001) and Se digestibility did not significantly change with CP intake (*p* = 0.753,). On average, Se digestibility was 62.3% (± 1.5) in kibble and 44.9% (± 1.9) in canned diets. The average digestible Se intake was lower in kibble than canned diets. The GPx response was lower on average in kibble compared to canned diets (*p*\<0.001) and decreased with an increasing CP intake (*p*\<0.001). Serum Se concentrations did not change significantly with diet type (*p* = 0.158), but did with CP intake, whereby serum Se concentrations decreased with increasing CP intake (*p*\<0.001). No association of diet type (*p* = 0.069) or CP intake (*p* = 0.216) on IsoPs was found , nor with the T3:T4 ratio (*p* = 0.102 and 0.960, respectively). The relative urinary Se:CT ratio was higher on average in kibble diets than in the canned diets (*p*\<0.001). No association of CP intake with urinary Se:CT ratio was found (*p* = 0.059). When urinary Se excretion was expressed per absolute amount of digestible Se, dogs on kibble diets excreted more Se in their urine than dogs on canned diets (*p* = 0.001) and excretion decreased with increasing CP intake (*p* = 0.017, data not shown). # Discussion This study showed that the availability of Se through digestion and metabolism changed considerably with diet type. The higher apparent Se digestibility in kibble compared to canned diets suggests that the use of a single recommendation for Se inclusion levels in pet foods for both diet types needs reconsideration. Previous *in vitro* findings showed that Se in canned diets was more susceptible to a decrease in Se accessibility, i.e. the amount that is available for absorption in the gastro-intestinal tract, due to processing. The average Se accessibility was also lower in canned than kibble diets. In the *in vitro* study, canned diets had an average Se accessibility of 58% (± 4.01, n = 20), while the Se accessibility of kibble diets was 72% (± 3.95, n = 23). Accordingly, in the present study canned diets showed a lower Se digestibility than kibble diets. The difference in apparent Se digestibility observed between the diet types may be caused by the Se species used in these diets. The meat and meat-by-products that are used in canned diets, mainly contain organically bound Se (e.g. selenomethionine). Therefore, it may be that the dietary sulphur- containing amino acids (methionine and cysteine) compete for absorption with the organically bound Se. Sodium selenite, which is often used (as in the kibble study diets) to supplement pet foods that do not comply to the recommended allowance (mainly kibble diets), is absorbed through diffusion and thus does not have to compete for absorption with organically bound sulphur. Interestingly, bioactivity in this study measured as GPx, was higher in canned than kibble diets. This may be explained by a higher amount of digestible Se in canned diets. Even though the apparent Se digestibility of canned diets was much lower than kibble diets (44.9% vs. 62.3%, respectively), the absolute amount of digestible Se was still higher in canned diets (7.7 ± 0.4 vs. 6.6 ± 0.4 μg/kg BW^0.75, respectively). Both this study and the *in vitro* study were performed with commercially available diets and in both studies a higher dietary Se concentration was found in canned than kibble diets. In the *in vitro* work, canned diets contained on average 34.8 μg Se/MJ (± 6.25, n = 20) and kibble diets 22.5 μg Se/MJ (± 2.42, n = 23). In this study, similar contents were measured (canned diets: 40.6 μg Se/MJ, kibble diets: 22.3 μg/MJ). This indicates that the ingredients for canned diets tend to contain a higher amount of Se, which suggests this may be inherent to this diet type. In addition to the higher amount of digestible Se in canned diets, the urinary Se excretion relative to Se intake was lower in canned compared to kibble diets, even per absolute amount of digestible Se. This suggests that the percentage of retained Se per digestible Se is higher in canned than kibble diets. Consequently, a higher amount of digestible Se from the canned diets can be used for either non-specific incorporation into body proteins, or incorporation into the Se-dependent antioxidant GPx. In the latter situation, this results in a higher bioactivity of Se from canned diets, as demonstrated in this study. A difference in speciation is the most likely explanation for the differences between the diet types. Unfortunately, efforts to determine Se speciation were unsuccessful due to detection limit issues. However, it is known that the kibble diets in this study are supplemented with inorganic sodium selenite and that the Se in the canned diets in this study derives solely from the ingredients, which consist mainly of meat and meat by-products, and thus it is primarily organically bound. It should be acknowledged that apparent digestibility was calculated in this study. Some of the absorbed Se is excreted via the bile into the faeces, which may cause an underestimation of the Se digestibility. However, it is not known whether there was a difference in biliary Se excretion between the diet types. Given the higher dietary fat concentration in canned diets, it may be assumed that bile excretion is higher in canned compared to kibble diets, but this does not necessarily mean a higher Se excretion via bile. Gregus *et al*. suggest that Se excretion via bile is enhanced by binding to an organic acid. So, if there was a difference in the amount of Se excreted via the bile, this may also be due to a difference in Se speciation. Taking blood GPx concentration as a measure for bioactivity may suggest a superior antioxidant function of canned diets, but this was not associated with a higher concentration of the oxidative stress parameter, isoprostanes. It cannot be excluded that a difference between diet types may be found if other parameters of oxidative stress were measured or if IsoP’s were measured during a longer period. Like GPx, iodothyronine deiodinases are also Se-dependent hormones. Several studies have shown that a higher dietary Se concentration has a positive influence on the T3:T4 ratio in humans, kittens and dogs. In this study, no effect on T3:T4 ratio was found and they were shown to be within a clinically normal range. This may be because the difference in digestible Se was not large enough between diet types. Remarkably, no interactions were found for any of the parameters measured in this study. However, there was a negative association between CP intake and both GPx and serum Se, irrespective of diet type. It is unlikely that Se intake was responsible for these results as there was no negative, but actually a positive association between Se and CP intake. Se digestibility was also not significantly associated with CP intake, therefore, unlikely to be responsible for the negative association between bioactivity and CP intake. Furthermore, urinary Se excretion corrected for digestible Se decreased with increasing CP intake, which suggests a higher retention with increasing CP intake. The higher amount of retained Se was not incorporated into GPx, but left in the blood stream as serum Se or excreted via the urine. Therefore, taken together it is very likely that a part of the Se from canned diets was incorporated into body proteins as selenomethionine instead of methionine. To clarify whether the effects found in this study were attributed to the absolute amount of Se or the Se species, semi-purified diets could be used in the future. However, it is a challenge and maybe even impossible to manufacture diets with two different processing types, using exactly the same ingredients. Furthermore, such a study would not be relevant for the Se digestibility and bioactivity of commercially available diets. This implies that the effects of CP intake in this study include factors associated with the formulation of a diet to a certain CP concentration. Similarly, the effect of diet type is not just a result of processing, but includes the choice of ingredients for that particular diet type. In conclusion, the diet type, whether this is caused by processing, Se speciation, or otherwise, causes canned diets to have a lower apparent Se digestibility, but a higher amount of the digestible Se is retained in the body and thereby enhancing Se bioactivity. So the can can, what the kibble can't. This study suggests that the recommendation for Se inclusion levels in pet foods should take diet type into account. Further studies are warranted to quantify the exact origin of the difference between the diet types, but this requires valid and sensitive parameters to monitor Se adequacy. The impact of processing conditions and protein sources on Se digestibility and bioactivity is another topic that warrants more investigation. It should be noted that it is not sufficient to solely measure Se digestibility, because bioactivity and urinary excretion measurements may give a different picture. Therefore, the choice of parameters can be crucial for the conclusion when studying nutritional modulation of Se status in animals. The authors express their special thanks to Joachim Neri of the Department of Applied Analytical and Physical Chemistry at the Faculty of Bioscience Engineering of Ghent University for total Se analyses and to Daniel Vermeulen of the Division of Livestock, Nutrition and Quality at the Science, Engineering and Technology group of the Catholic University of Leuven for the thyroid hormone analyses. [^1]: KG, KB and LGA were employed by the WALTHAM^® Centre for Pet Nutrition at the time of the study. This does not alter the authors' adherence to PLOS ONE policies on sharing data and materials. None of the other authors have conflicts of interest to declare. [^2]: Conceived and designed the experiments: MvZ MH KG LGA GPJJ. Performed the experiments: MvZ. Analyzed the data: MvZ KB AC. Contributed reagents/materials/analysis tools: MvZ GDL. Wrote the paper: MvZ. Reviewing and editing the manuscript: MH KG KB AC LGA GDL GPJJ.

# Introduction Gene-environment interplay in immune phenotypes has been extensively studied using steady-state cellular immune profiles, functional immune responses, post- vaccination responses and V,D and J usage biases in naïve and memory T and B cell compartments. Some of these studies have identified genomic correlates associated with specific steady-state immune phenotypes and vaccine responses. However, there are conflicting findings regarding the relative importance of genetic versus environmental factors in regulation of immune phenotype, warranting further observational studies in humans and mechanistic studies in mice on regulation of individual immune phenotypes. Memory subsets in T and B lymphocytes are immune subsets that are generated in response to past immunogen exposure. Immunological memory, providing long-term persistence of antigen-experienced cells contributing to rapid and robust responses following re-exposure, is likely to have evolved in an ecosystem where environmental challenges including repeated infections would be the norm and the persistence of long-lasting antigen-specific cells generated during immune responses would confer survival advantage. However, it is possible that larger memory lymphocyte pool sizes may carry costs such as restriction of space for the more repertoire-diverse naive T cell compartment, attrition of pre-existing memory, and other bioenergetic costs. Such selection may well result in ‘optimum’ sizes of memory lymphocyte pools, and memory cells in such pools could show diversity in cellular life-spans, depending on the exposure rates, prevalence and virulence of pathogens in the ecosystem. Diversity in pool sizes of memory lymphocytes could thus be determined by a combination of genetic variability and diversity of environmental exposures. A number of mechanisms can be envisaged regulating the pool size of the memory T cell compartment, including cumulative life-time antigen-exposure and re- exposure, antigenic persistence, degree of expansion, cell survival, attrition and niche-space availability. Immune cells occupy a limited niche space in lymphoid organs or in the periphery. This niche size could be a function of size and/or structure of supporting lymphoid tissue architecture, along with intrinsic properties of cells occupying the niche. Similar determinants could affect steady state levels of transient cell subsets such as immediate-effector T cells and plasmablasts, but their steady state levels would be expected to fluctuate more with short-term environmental changes. On this background, we have conducted the present study to address the following questions: (1) to assess inter-individual and temporal variability in lymphocyte memory and effector compartments by peripheral blood leucocyte immunophenotyping in serial bleeds from healthy young adult human volunteers, (2) to assess the degree of similarity in lymphocyte memory phenotype among human siblings that could suggest regulation by hereditary or shared environmental factors and (3) to assess the cell-intrinsic regulation of memory lymphocyte pool sizes using bi-parental mixed bone marrow chimeras (with independent inbred mouse strains as parental strains) to test for parental genotype-driven inheritance of CD4 and CD8 T cell memory pool sizes. Together, our data from both humans and mice provide novel exploratory insights in the determination of memory lymphocyte pool size. # Materials and methods ## Ethics approval and consent to participate ### Human Informed written consent was obtained from all human volunteers who participated in the study. The study was approved by ethics committees of National Institute of Immunology (approval number IHEC/AKS/45/2013) and All India Institute of Medical Sciences (approval number IEC/NP-471/2013). The methodologies used in the study were in accordance with approved guidelines. All experimental protocols used in this study were approved by Institutional Ethics Committee (Human Research) of National Institute of Immunology and All India Institute of Medical Sciences. ### Animal Mice were maintained and used according to the relevant rules and regulations of Government of India and with the approval of Institutional Animal Ethics Committee, National Institute of Immunology (approval number 381/15). All mice experimental protocols were in accordance with approved guidelines and were approved by Institutional Animal Ethics Committee of National Institute of Immunology. ## Human subjects This study used samples from 2 separate cohorts of human volunteers. \(1\) To test longitudinal temporal changes in immune profile, we used a longitudinal observational study design where 45 healthy adult volunteers were recruited into the study and were bled 3-monthly for 12 months (total of 4 bleeds). Sample size calculation was done using 'pwr' package in R. The sample size needed to obtain a power of 0.8 at a significance level of 0.05 considering the number of groups being 4 (since we were looking for fluctuation in parameters across 4 time points) for modest effect sizes (Cohen's f = 0.25) was 44. We recruited 48 volunteers (additional 4 individuals) initially to account for attrition. Out of these, 45 individuals completed the study (all time points). Two samples were used for standardization and data analysis was done using a sample size of 43. Volunteers were recruited by posters in educational institution that detailed the nature of the study and interested volunteers were asked to contact the investigator. All volunteers were residents of National Capital Region (Delhi) during the study. The period of recruitment and sample collection was between 2013 to 2015. All volunteers were above 18 years of age and in good general health. Individuals with chronic illness, history of regular medication intake, recent vaccination (within 6 months of study or during the study) or pregnancy (in women) were excluded. Blood collection was avoided in volunteers who were suffering from infectious illness and during convalescence (2 weeks after full recovery). All volunteers were provided information about the study before recruitment and written and verbal consent was collected at each blood sampling. The study was approved by institutional ethics committee of National Institute of Immunology (NII). All protocols were in accordance with approved institutional guidelines. For testing correlations between the various memory and effector cell subsets, we also utilized the published raw data from a larger (n = 71) cross-sectional study of adult healthy volunteers who were also residents of NCR Delhi, who participated in a previous study. \(2\) To study the concordance in immune memory between siblings, we conducted a cross-sectional observational, family-based study in which 37 families (n = 80) were recruited, in which two or more siblings were willing for participation. All volunteers were more than 18 years, in good health and were residents of National Capital Region (Delhi) when a single blood sample was collected after informed written consent. Recruitment of volunteers was done using posters detailing the study. Duration of recruitment and sample collection was between 2014 to 2015. Since the degree of similarity between siblings in immune memory phenotype was not known at the start of the study, this was designed as an exploratory study and a sample size of 30 was considered. We collected 80 samples from 37 families out of which two samples were used for technical standardization and the rest (n = 78, 36 families) were used for final data analysis. This study was approved by institutional ethics committees of NII and All India Institute of Medical Sciences (AIIMS). Exclusion criteria for volunteers were similar to that mentioned above for the serial-bleed volunteers. ## Blood collection, sample processing and flow cytometry of human samples Ten ml heparinized peripheral blood was collected from healthy subjects after informed written consent. To avoid batch-to-batch variability due to potential diurnal variation in immune parameters in blood, sample collection for all batches were conducted in the morning. Blood samples were processed without delay (less than three hours). Total leukocyte counts and differential counts were obtained from standard hematologic methods. Peripheral blood mononuclear cells (PBMCs) were separated by density gradient centrifugation. PBMCs were washed, counted and divided into aliquots of about 1 million cells per ml per vial and cryopreserved in 10% DMSO in bovine serum until assays were performed. For flow cytometry, PBMCs were thawed, washed and incubated with the following antibodies organized into two cocktails: The T cell cocktails consisted of CD3 (UCHT1, eBioscience), CD4 (OKT4, eBioscience), CD8 (SK1, eBioscience), CD45RO (UCHL1, eBioscience), CCR7 (150503, BD) and the B cell cocktail consisted of CD19 (SJ25C1, BD), CD20 (2H7, eBioscience), CD38 (HIT2, eBioscience), CD43 (eBio84-3C1, eBioscience), CD27 (M-T271, eBioscience). A single control sample was run along with all the samples to ensure that gating is comparable across experiments. Samples were acquired in BD Verse and analysis was done using flowjo (Treestar). CD4 and CD8 memory cells were defined as the CD45RO+ fractions of CD8 and CD4 cell compartments and includes both central memory (CM) and effector memory (EM) subsets. Transient T effector-memory, RA+ve (TEMRA) cells were defined as CD45RO- CCR7- subsets of CD8 cells and of CD4 cells. Memory B cells were defined as CD27+CD43- fraction of the CD19+CD20+ B cell subset. Plasmablasts were defined as the CD38+CD20- subset of CD19+ B cells. ## Mice The following strains of mice were used for the study: C57Bl/6J, BALB/cJ, SJL/J, CBA/CaJ, B6.SJL-Ptprca Pepcb/BoyJ (B6.SJL), CB10-H2b/LilMcdJ (BALB/b), FVB/NJ, C3H/HeOuJ. Mice strains used in the study were obtained as breeder stock from the Jackson Laboratory (Bar Harbor, ME) and bred and maintained under specific- pathogen free conditions in the Small Animal Facility of the National Institute of Immunology. Mice were regularly tested for specific pathogens and were found to be negative. All mice were maintained and used according to the relevant rules and regulations of Government of India and with the approval of Institutional Animal Ethics Committee, National Institute of Immunology. All experiments used age (8 to 9 weeks) and gender matched mice. Age- and gender- matched littermates were used as controls. For making mixed bone marrow chimeras, bone marrow cells from donors were transferred intravenously by retroorbital injection. For retroorbital injection, mice were anesthetized using ketamine-xylazine given through intraperitoneal injection according to institutional guidelines. Mice were euthanized by cervical dislocation in all the experiments. ## Ex vivo cell preparations Spleen was dissected from mice euthanized by cervical dislocation and teased between a pair of frosted slides to obtain single cell suspension. Red blood cells in the suspension were lysed by osmotic shock using water, washed and then re-suspended in complete medium. Phenotyping of mouse spleen T cells was done using the following antibodies (all from eBioscience): CD4 (RM4-5), CD8 (53–6.7), CD44 (IM7), CD62L (MEL-14), CD25 (PC61.5), CD45.1 (A20) and CD45.2 (104). NK cells were phenotyped using the following antibodies (all from eBioscience): CD90 (53–2.1), B220 (RA3-6B2) and Ly49b (DX-5). Surface staining was done by incubating 1x106 cells in staining buffer (PBS containing 2% BSA and 0.05% NaN3) for 30 mins on ice. The cells were washed thrice with cold staining buffer and samples were acquired on BD Verse and analysis was done using flowjo (Treestar). Doublets were excluded from the population using height and area parameters of forward and side scatter. CD4 T effector memory (TEM) cells were defined as the CD4+ CD25- CD44hi CD62Llo fractions and CD8 T effector memory (TEM) cells were defined as the CD8+ CD44hi CD62Llo fractions. Central memory T cells (TCM) were defined as CD44hi CD62Lhi fractions. ## Mix bone marrow chimeras F1 hybrids were irradiated at 800rads in gamma chamber (BARC, Mumbai) with Co⁶⁰ as a source for gamma rays. Bone marrow cells from each parental strain were mixed in 1:2, 1:1 and 2:1 ratios and a total of 15 million cells transferred into irradiated F1 generation mice (n = 24 for B6.SJL—CBA/CaJ chimera and n = 19 for SJL/J—BALB/cJ chimera) and reconstitution allowed for two months (the duration for optimum reconstitution was ascertained by standardization experiments and previous literature). Based on the ratio of CD45.1 to CD45.2 in reconstituted chimera, biological outliers showing extreme ratios more than 10- fold from the expected were removed from further analysis (4 outliers in B6.SJL—CBA/CaJ chimera and 2 outliers in SJL/J—BALB/cJ chimera were removed). Phenotyping was done by using above mentioned CD markers. ## Statistical analysis All statistical analysis were done using R (V3.5.1) in Rstudio (V1.1.383). To test the equality of intra-individual variance and between-individual variance for any variable representing a compartment, we used a bootstrapping (resampling) method. We randomly sampled groups of individuals a large number of times and calculated their variance. The large resampled data resulted in the generation of distribution of differences of variances. The difference between within-individual variance and between-individual variance was calibrated against this distribution and p-values computed for each parameter. For comparison of sibling pairs vs. unrelated pairs, we used bootstrap resampling method to calculate the p-values. As the unrelated individuals are not naturally paired, we took random samples from the unrelated individuals for different parameters and created a distribution of differences for each parameter. The p-values were then computed with help of this randomly generated distribution. This procedure was followed for each variable representing a compartment. Since bootstrapping was done for each variable, the issue of multiple testing was not relevant. Correlations between cell subsets were estimated using Spearman’s correlation coefficient. The Bonferroni and FDR method was used to correct for multiple comparisons in testing the significance of elements of the pairwise correlation coefficient matrix. For mice data, frequency and cell count comparisons were done using a non-parametric test (Mann Whitney). Comparison of multiple inbred mouse strains was done using Analysis of Variance (ANOVA) with post-hoc testing for pair-wise comparison. ### Gene expression analysis Gene expression (microarray) data of sorted splenic ‘naïve’ CD4+ve CD62L-ve CD4 T cells of mice published previously was obtained from GEO database (accession: GSE60337) and analysed using GEO2R tool. Genes with p-value of less than 0.05 after multiple testing (Benjamini) was considered significantly differentially expressed. Gene Ontology enrichment analysis was done using web-based Gene Ontology enrichment analysis and visualization tool (GORILLA) and Database for Annotation, Visualization and Integrated Discovery (DAVID). To reduce false positive results in enrichment, in all enrichment analyses, the enrichment of candidate genes that were differentially expressed were analysed against a background of all the genes that were used in array experiment. # Results ## Human peripheral T and B memory cell levels show temporal stability Firstly, we examined if variation in the memory T and B cell subsets in the population were relatively stable over time, or were significantly affected by fluctuations possibly contributed by short-term environmental exposures. To test this, we characterized the peripheral blood leucocyte subset phenotype of a group of human volunteers \[n = 43, male: female ratio = 25:18, mean age = 25.8 years (SD = 1.8)\] every 3 months for a period of 1 year. Gating strategy for the memory subsets is indicated in. For cell subset frequencies the following parameters were quantified: Naive B cells, memory B cells and plasmablasts (all expressed as % of total B cells), naive CD4 and memory CD4 (expressed as % of CD4 T cells), and naive CD8, memory CD8 and CD8 TEMRA (expressed as % of CD8 T cells). TEMRA subset in CD4 T cells was poorly resolved and hence not quantified. Representative data of frequencies of each subset over 4 time points are shown in (for B cell subsets), (for CD4 subsets) and (for CD8 subsets). These are shown quantified with p-values in (quantification of cell subset frequencies) and (quantification of cell subset counts). Variations of memory B, memory CD4 and memory CD8 frequencies were significantly lower within individuals than between individuals. On the other hand, intra-individual variances in plasmablasts were no different from inter-individual variances. Naive subsets of B cells, CD4 T cells and CD8 T cells also showed lower within-individual variance compared to between-individual variance similar to the corresponding memory subsets. When absolute cell numbers per μl blood were calculated, CD4 and CD8 memory T cells showed low within-individual variance compared to between- individual variance and plasmablasts showed considerable within-individual variation (consistent with results of cell subset frequency). Curiously, CD8 TEMRA frequencies and counts tended to be similar to CD8 memory in that they showed lower intra-individual than inter-individual variation. It remained possible that inter-individual variability could be influenced by potential confounders including age and gender. In order to rule out this possibility, we looked for age and gender differences in the various immune parameters in our cohort. Age showed no correlation between the parameters examined, ruling out possibility of age as a confounder in our analysis. When we tested for gender differences, CD8 memory and CD8 naive frequencies and counts were seen to be different between males and females (with females showing lower CD8 memory frequencies compared to males). In order to adjust for gender, we re- analyzed the intra-individual versus inter-individual variance comparison in males and females separately. Differences in intra-individual variation versus inter-individual variation persisted even when adjusted for gender differences in this fashion, suggesting that gender differences alone cannot account for higher inter-individual variation seen for these subsets. Overall, these data suggested that short-term variations in memory B and T subsets within-individuals are relatively small compared to variability in these parameters that is seen between individuals. ## Correlations between CD4, CD8 and B cell memory Since memory cells can be long-lived and could have been generated by antigenic exposure in the relatively remote past, it remained possible that cumulative antigen exposures contribute to determining steady-state memory T and B cell levels. Microbial exposure commonly leads to generation of CD4 T-dependent B cell responses, since CD4 T cells are necessary for B cell germinal centre responses and hence vital to memory B cell generation. Thus, it would be expected that there would be coordinated accumulation of memory cells in the CD4 T and B cell compartments leading to a positive correlation between CD4 T memory and B memory cell frequencies, if cumulative antigenic exposures were to contribute substantially to determining memory T and B cell frequencies. However, there was no correlation between CD4 T cell memory frequencies and B cell memory frequencies in a given individual. Thus, cumulative antigen exposure may not be a major determinant of memory CD4 and B cell levels. Interestingly, CD4 and CD8 memory T cell subsets showed a strong correlation (correlation coefficient = 0.49, FDR p = 0.009), suggesting shared determinants of memory cell subsets within the T cell lineage. On the other hand, frequencies of CD4 TEMRA or CD8 TEMRA cells or plasmablasts did not show any correlations with memory T or B cell frequencies or with each other, except for the expected negative correlation seen in subsets that share a common parent gate. We also confirmed these results using another larger (n = 71) cohort of individuals from a previously published study and is shown in and). ## Siblings show concordance in T and B cell memory levels We tested whether T cell and B cell memory phenotypes show concordance in sibings by conducting a family based cross sectional study where peripheral blood leucocytes from 78 full siblings from 36 families were immunophenotyped \[male: female ratio = 38:40, mean age = 32.3 years (SD = 12.7)\]. Descriptive characterization of range of immune parameters in the cohort is shown in. We compared the differences in cell frequencies between siblings with the differences between random pairs of non-siblings for the 8 immune parameters previously mentioned. Unrelated controls were age-matched. The analysis showed that differences in cell frequencies between siblings were significantly less than the differences between random pairs of non-siblings for memory CD4, memory CD8 and memory B cells, but not for plasmablasts and CD8 TEMRA cell frequencies, suggesting that the memory and naive B, CD4 and CD8 subsets could be determined by shared early developmental environmental factors or hereditary factors. However, this analysis was intended as an exploratory study and more data from larger studies would be necessary to truly establish hereditary determinants. Raw data of immune cell frequencies and counts of all the parameters described in the study along with demographic characteristics of the 3 cohorts used in this study \[namely, serial bleed (n = 43), sibling (n = 78, 36 families) and previously published cohort from reference \#31) (n = 71)\] are provided as supplementary spreadsheets (and Tables). We also made a comparison of all the samples (n = 192) from these 3 cohorts with 2 other cohorts from North America (Stanford cohort) and Europe (Belgium) where published raw data was available. Since the age distribution of the 3 cohorts was different, we restricted our comparison to a narrow age interval (18–35 years) to avoid confounding due to age. We see some differences between cohorts for many subsets, but these comparisons are likely to be influenced by a variety of confounding factors, particularly flow cytometric technical factors, and carefully controlled planned multi-institutional studies would be needed to allow plausible interpretations of these differences. However, it is noteworthy that plasmablast frequencies in our cohort are substantially higher than in European cohort, suggesting environmental exposure-related differences. ## Mouse strains differ in their memory T cell frequencies and show cell-autonomous regulation of memory T cell phenotype To examine the possibility of cell-intrinsic regulation of memory phenotype more rigorously, we examined independent inbred strains of mice. Since specific markers for B cell memory are uncertain and since this compartment in mice may be smaller than in humans, we restricted these studies to T cell memory subsets alone. The gating strategy used for defining T effector memory (TEM) CD4 ( to) and CD8 T cells ( to) in a group of representative mice available in our laboratory are shown in. TEM populations in CD4 and CD8 subsets (CD44hi CD62Llo) were distinguished by unambiguous contours, whereas in some strains (eg. BALB/cJ) the T central memory (TCM) populations (CD44hi CD62Lhi) were difficult to gate out from the CD44 negative naïve pool using the conventional markers of memory (CD44 and CD62L). TCM phenotype CD4 T cells in the BALB/c strain have been previously shown to contain recent thymic emigrants with a CD44-high phenotype. TCM phenotype CD8 T cells have also been reported to contain antigen-inexperienced, non-memory (“virtual memory”) cell types as well as regulatory CD8 T cells, further confounding our analysis of memory subsets. Hence, we have avoided TCM phenotype T cells and focused on the unambiguously defined TEM populations for interpreting differences in memory between the genetically different strains. Our preliminary data showed that CBA/CaJ and C57BL/6J showed differences in both CD4 TEM and CD8 TEM compartments with C57BL/6J showing higher CD4 TEM and CD8 TEM frequencies and counts than CBA/CaJ ( to). On the other hand, BALB/cJ and SJL/J differed in CD4 TEM but not CD8 TEM compartment ( to), with SJL/J showing significantly higher CD4 TEM frequencies and counts than BALB/cJ. To test the source of the differences observed in CD4 TEM frequencies between the strains, we prepared bi-parental mixed bone marrow chimeras in which bone marrow cells from both parents, in various ratios, were transferred into irradiated F1 mice. We used heavy irradiation that is reported to successfully engraft parental donor cells in F1 recipients and ensured endogenous NK cell depletion post-irradiation. We also confirmed that chimerism is successful by comparing injected donor cell ratios with ratios of CD45.1 to CD45.2 lymphocytes in reconstituted F1 mice and found no statistically significant differences, ruling out the possibility of unequal reconstitution by donor strains. The mean frequency of endogenous cells was 10.1% in B6.SJL–CBA/CaJ chimera (SD-6.5) and 25.9% in BALB/cJ—SJL/J chimera (SD-12.3). After waiting two months for generation of memory cells from the donor genotypes, these mice were phenotyped for CD4 and CD8 TEM compartments. Memory CD4 and CD8 frequencies of the donor- derived cell population ( to) in these chimeras showed similar trends as that in parental strains. These results suggest that genetic factors contribute in cell intrinsic fashion to determining TEM CD4 and CD8 cell frequencies. ### CD4 gene expression in strains that differ in CD4 effector memory phenotype Since bone marrow chimera data suggested cell-autonomous regulation of memory phenotype, we hypothesized that intrinsic gene expression differences in naive CD4 T cells between the mouse strains could contribute to differences in CD4 TEM frequencies that are observed across the strains. We tested this using gene expression data (microarray) of splenic CD4 T cells from multiple mouse strains available from the immunological genome project. Immgen gene expression data comes from sorted CD4+ CD62L+ spleen cells ('non-EM' CD4 T cells), which could potentially include CD44-high CD4 TCM population in addition to genuine naïve CD4 T cells. Hence, we selected 2 mouse strains (CBA/CaJ and SJL/J) that showed maximum differences in CD4 TEM frequencies and counts, without showing differences in CD4 TCM levels (, p-values shown). CD4 TCM frequencies were very low in these 2 strains, making comparison of CD4+ CD62L+ gene expression between these strains interpretable. A comparison of gene expression profile of CD4+ CD62L+ve splenocytes (obtained from GEO accession: GSE60337) of CBA/CaJ and SJL/J identified 290 genes that were significantly different (p values \< 0.05 after multiple correction). Gene ontology enrichment analysis identified significantly enriched GO processes associated with the 290-gene list. As expected, MHC molecule related transcripts were differentially expressed and came up as significantly enriched, reassuring that the analysis we performed could actually pick up biologically relevant differences. In addition, there was significant enrichment for lipid biosynthesis (11 genes, p value: 1.78E-05, adjusted p value: 0.004) and lipid metabolism (16 genes, p value: 1.13E-04, adjusted p value: 0.02), implying that these pathways could be of biological relevance to regulation of CD4 T cell memory phenotype. When data from multiple strains were pooled together, there was a positive correlation between CD4 TEM and CD8 TEM frequencies, similar to what is observed for humans, suggesting similar mechanisms in regulation of CD4 and CD8 effector memory or dependence of one subset on the other. # Discussion There are a number of possible explanations for the observation of a substantial extent of variation in the relative frequencies of memory-phenotype B cells, CD4 T cells and CD8 T cells. Such variations in biological parameters are commonly interpreted as errors in technical or experimental factors, although in real populations, biological variance may well have meaningful interpretative value. We have used descriptive characterization of the human peripheral blood leucocyte phenotype in a small cohort to examine the possible determinants of variation in these leucocyte subsets. By comparison of intra-individual variance of memory and effector subsets of T and B cells, we find that immune subsets that respond to day-to-day fluctuations, such as the plasmablasts show similar intra-individual and inter- individual variance, while B cell memory, CD4 memory and CD8 memory cell frequencies are very much stable over a one-year period within individuals. This suggests that memory subsets are not determined by short-term environmental fluctuations, unlike plasmablasts. Within-individual variation can thus possibly be an explanatory factor in the overall variation seen among individuals for effector plasmablast subsets, but not for memory T and B cell subsets. A recent study that quantified technical and biological variation in human immune phenotype also found high intra-individual variation in CD4 TEMRA cells in comparison to other immune lineages. Since CD4 TEMRA subsets in our study was not discreetly defined by flow cytometry we count not quantify these subsets. Our data show that CD8 TEMRA cells did not behave like 'effector' cell subsets in that they do not show considerable intra-individual short-term fluctuation. This could be related to indications that CD8 TEMRA cells may be functionally different from CD4 TEMRA effector cells, despite being phenotypically similar. Our findings of a lack of correlation between B cell memory and CD4 memory also argue against the possibility of cumulative exposures determining both the memory subsets, although it remains possible that cumulative exposures regulate one but not the other subset. The positive correlation we find between CD4 memory and CD8 memory suggest that T cell memory levels could be regulated by similar mechanisms, distinct from how memory B cell levels are regulated. One limitation of our study is that we have quantified bulk memory cells rather than antigen-specific memory cells, considering the difficulty in technical quantification of rare antigen-specific populations. Also, antigen-specific responses would be subject to very many confounders such as when and how much natural immunisation happened. Since these factors are difficult to control, we have used characterization of bulk memory populations in B and T cells. We attempted to quantify the degree of similarity in immune phenotype between siblings to see if there was any indication that these memory phenotypes are heritable. Although our sibling study was not as powerful as previously published twin studies for understanding heritability, our sibling data support and extend the interpretations from our serial bleed data, although further large-scale studies are needed for confirmation. Thus, siblings did not show concordance in effector T and B subsets, suggesting that variation in those subsets may be environmentally driven. On the other hand, siblings were far similar to each other in memory subset frequencies than non-siblings were, suggesting a hereditary component or shared environmental factors in regulating memory lymphocyte frequencies. Any similarity between siblings could be attributed to early life influences, since it is well recognized that both intra-uterine and neonatal stress impacts immune system development, and a recent study has shown a prominent role for co-habitation in determining immune phenotypes. Our findings are consistent with two previous studies which report high degree of heritability for majority of immune subsets. Another study, too, reported high degree of heritability for CD4 memory subsets, although it found non-heritable factors to be important for most other immune phenotypes. Since our interpretations are predominantly based on associations, we complemented the study with experimental data from mice. Although laboratory animals show restricted genetic diversity compared to human populations, we exploited the fact that independent inbred strains are genetically distinct and are grown in a homogeneous environment in a controlled animal facility. Hence, it is worth examining if differences in leucocyte subset levels between strains are genetically driven. We immunophenotyped some common strains of mice and chose two pairs showing substantial differences, SJL/J and BALB/cJ for further experiments on CD4 memory and C57BL/6J and CBA/CaJ for CD8 and CD4 T cell memory phenotype. It is notable that not only did splenic CD4 and CD8 memory frequencies show differences, but that total numbers of these cells per organ also showed the same differences, indicating that subset frequencies are reasonable surrogates for the pool size of these subsets. We also have noted that animals maintained in our facility had relatively high frequencies of TEM phenotype T cells even at steady state. Even though mice were harbored in a specific-pathogen-free facility, it remains possible that there might be variations in microbial antigenic burden or gut microbiome composition between laboratories, although these factors are difficult to quantify. Long years of reproductive isolation because of inbreeding could also have contributed to these differences. Our mixed bone marrow chimera experiments allowed us to examine whether the CD4 and/or CD8 memory levels were determined in a cell-autonomous fashion. Our data suggest that donor genotype-specific cell-intrinsic factors strongly influence both CD4 and CD8 memory T cell pool size. Using gene expression data of mouse splenic CD4+ve CD62L+ve T cells available in the public domain we attempted to characterize genetic differences that could explain differences in CD4 memory phenotype. There are a number of limitations in our approach. Firstly, we do not evaluate gene expression differences in the same mice (or even mice from the same small animal facility) that we experimentally find phenotypic differences for, and differences in phenotype that we observe could be influenced by additional facility-specific environmental factors as well. Secondly, the array based gene expression data available in the public domain that we used contain only 2 replicates for each strain (except for C57BL/6J), limiting statistical power and increasing the chances of picking up false positive differences. Thirdly, the "naive" population as defined by Immgen is based on CD4+ve CD62L+ve gate, and does not exclude CD44+ cells, potentially including central memory cells into the population. Our analysis attempts to overcome this limitation by comparing those strains which differ only in CD4 TEM, and not CD4 TCM subsets. In spite of these caveats, our analysis is still useful as a preliminary exploration that can be extended in further experiments using multiple replicates and strains. Remarkably, lipid/fatty acid metabolism pathways that came up as significant in as limited an analysis as this, was recently suggested to play crucial roles in T cell activation. This exploratory analysis is thus likely to provide further clues to direct future mechanistic studies. Thus, our data suggest cell-intrinsic factors as potential determinants of the memory pool size of T cells as well as, probably, B cells. However, in a natural ecosystem where animals are exposed to substantial and continual antigenic exposures, the balance between cell-intrinsic genetic effects on memory and the effects of environmental factors is likely to be nuanced and quantitative. Further studies using varying antigenic burdens and experimental immunizations will be necessary to address these issues. Mechanistically, it will be interesting to explore whether genetic cell-intrinsic factors affect memory generation during immune response or memory attrition after the peak of immune response. Our data provide interesting insights and directions for future work in understanding the genesis and consequences of the regulation of niche size of lymphocyte memory. # Supporting information We acknowledge Department of Biotechnology, Government of India and Department of Science and Technology, Government of India for funding the study. EM Effector memory TEMRA T, effector memory, RA positive CD Cluster of differentiation [^1]: The authors have declared that no competing interests exist. [^2]: ‡ These authors also contributed equally to this work.

# Introduction A habit is an inclination or aptitude for some action acquired by frequent repetition, showing itself in increased facility to performance and reduced power of resistance. One of the commonest oral habits is sucking, a reflex present at birth, although oral contractions and other sucking reflexes have been observed before birth. Sucking habits could be nutritive (breast and bottle feeding) or non-nutritive. The commonest form of non-nutritive sucking (NNS) is digit sucking. Several studies have evaluated its etiological factors and suggest that fatigue, boredom, excitement, hunger, fear, physical and emotional stress, and insufficient satisfaction of sucking need in infancy are situations that could stimulate digit sucking habits. Sucking may provide happiness and a sense of security when a child faces difficult times. It may also give a feeling of warmth and contentment. Detrimental effects of digit sucking include disturbances in arch form, recurrent otitis media, the possibility of accidents, development of latex allergy, tooth decay, oral ulcers and sleep disorders. Others include wrinkled, chapped or blistered fingers, ulceration, corn formation, dishpan thumb as well as reduced peer acceptance. Digit sucking may also accompany behaviors like trichotillomania. The association between digit sucking and dental caries has been studied but results have been inconclusive. Yonezu and Yakushiji found that children with finger-sucking habits were more likely to be free of caries by age three years. Their finding was associated with increased inter-dental spacing which resulted from flaring of teeth due to digit sucking. On the contrary, a study conducted in Baghdad reported an increase in caries severity with NNS. NNS was associated with malocclusion, making tooth cleaning difficult and allowing accumulation of dental plaque. The aims of this study were to determine the prevalence of digit sucking, caries and oral hygiene status in the study population, and the association between digit sucking, caries and oral hygiene status of children from six months to 12 years, resident in Ife Central Local Government Area (LGA) of Osun State, Nigeria. # Methodology ## Study Design This was a cross sectional study that recruited participants from National Population Enumeration sites in Ife Central LGA. These geographical sites were selected because the participants there were familiar with the conduct of such surveys. ## Study Setting Ife Central LGA is a semi-urban area of Osun State, which hosts the Obafemi Awolowo University and its Teaching Hospital. The 1991 census put the population of the LGA at 96,580 while the estimated population for the year 2004 was 138,818. The child population for the LGA is about 14,000. ## Study Population Participants were included in the study if they were between the ages of six months and 12 years, living with their biological parents or legal guardians who consented to participate in the study. ## Sample Size Determination The sample size was calculated using Leslie Fischer's formula for study population \>10,000. Based on a prevalence of 34.1% of oral habits of children aged four to 15 years old, determined by Quashie-Williams *et al*, a sample size of 1,011 children was necessary to identify 345 children with oral habits, giving a non-response rate of about 10%. ## Sampling Technique The sampling procedure was a (three-level) multi-stage cluster sampling aimed at selecting eligible persons with known probability. Stage 1 involved the random selection of enumeration areas within the LGA.At the sites, every third household on each street was selected. Stage 2 involved listing eligible individuals within households. Stage 3 involved selection of actual respondents for interview. Only children present in the house at the time of study were eligible and one child per household was selected. Details of this sampling technique has been reported by Folayan *et al*. in an earlier publication from this same database. ## Data Collection Data were collected through face to face interviews using a structured questionnaire. Experienced field workers who had been engaged in past national surveys were recruited and trained on the study protocol. The interviewers collected all information from respondents and submitted to survey supervisors who reviewed the questionnaires. Mothers were requested to respond on behalf of children below eight years, based on evidence that responses of mothers on questionnaires have a higher correlation with children’s responses. However, where the mother was unavailable, fathers completed the questionnaires. Each child’s socio-demographic characteristics were obtained. ## Socio-economic Status Socio-economic status was determined using an adapted version of a socio- economic index described by Olusanya *et al*.. The tool has been tested and found valid and reliable in Nigeria. Data were collected on educational level and profession of respondent’s parents. The mother’s level of education was classified as ‘no formal education, Quranic or Primary school education’ and scored as 2; Secondary school education was scored as 1; and Tertiary education scored as 0. Father’s occupation was categorized into three: Civil servants or skilled professionals with tertiary level of education scored as 1; Civil servants or skilled professional with secondary level of education scored as 2; Unskilled, unemployed individuals, students, and civil servants or skilled professional with primary or Quaranic education were scored as 3. The scores for the mother and father were summed to give social classes I–V, where classes I and II represented the upper class, class III the middle class and IV and V, the lower socio-economic class. When a child had lost a parent, the socio-economic status was determined from the living parent. ## Digit Sucking Habit Questions were asked about type and number of digits sucked and the frequency of engaging in the habit. Options were, ‘irregularly’, ‘once a week’, ‘2–3 times a week’, ‘once a day’ and ‘several times a day’. Duration of sucking was explored and options ranged from ‘less than a minute’ to ‘1–5 minutes’, ‘5–10 minutes’, ‘10–20 minutes’, ‘20–30 minutes’, and ‘almost continuously’. The intensity of the sucking habit was explored by asking if or not an audible sucking or popping sound was heard while sucking. ## Dietary History Dietary history was obtained with a dietary chart which recorded all study participants meals and snacks, taken both at meal times and in between meals for three consecutive days, including two week days and a weekend or public holiday. ## Intra-oral Examination Intra-oral examination was conducted in the homes of all participants to determine the presence of caries and oral hygiene status. The participants were examined sitting, under natural light, using sterile dental mirrors and probes by trained dentists. The teeth were examined wet and debris removed with gauze where present. Radiographs were not used in this study. ## Caries Profile Caries diagnosis was based on the recommendation of the WHO Oral Health Survey methods. The decayed, missing and filled teeth index (dmft/DMFT) was used. The number of decayed, missing or filled teeth were summed together to give the dmft/ DMFT score for the primary/permanent dentition. For the purpose of analysis, caries status was further divided into caries present or absent. ## Oral Hygiene Status The Oral Hygiene Index–Simplified (OHI-S) described by Greene and Vermillion was used to determine the oral hygiene status. Its components, the Debris Index and Calculus Index, were obtained based on six numerical determinations representing the amount of debris or calculus found on the surfaces of index teeth 11, 16, 26, 31, 36, 46 and 51, 55, 65, 71, 75, 85 in the permanent and deciduous dentitions respectively. Debris and calculus scores were totaled and divided by the number of surfaces scored. Scores were graded as 0.0–1.2 = Good oral hygiene, 1.3–3.0 = Fair oral hygiene and \> 3.1 = Poor oral hygiene. ## Standardization of Examiners Clinical investigators were qualified dentists undergoing postgraduate residency training as Paedodontists or Orthodontists who were calibrated on the study protocol and the WHO criteria for caries diagnosis, including the use of the dmft/DMFT index and OHI-S. Training was followed by practice on patients. Each investigator examined and scored children for oral lesions as prescribed in the study protocol. Results were subjected to a Cohen’s weighted kappa score analysis to determine intra- and inter- examiner variability. The intra-examiner variability ranged between 0.89–0.94, while inter-examiner variability ranged between 0.82–0.90 for caries detection and OHI-S. ## Theoretical Model for Statistical Analysis A hierarchical theoretical model with the following four blocks was employed for the analysis of predictors for the presence of dental caries: 1) age of child. 2) socio-economic and demographic factors, 3) Oral Hygiene Index and 4) oral habit. Age was considered a potentially confounding factor, which informed the adjustment of the developed model for this variable. The second block included socio-economic and sex as distal factors in the theoretical model since they could influence oral hygiene status and the digit sucking habits. Oral hygiene practice may be a moderator variable of the association between digit sucking and caries, so it was included in the third block. In the fourth block, the digit sucking was recorded as the oral habit with the assumption that this variable may influence caries risk. A hierarchical theoretical model with the following three blocks was also employed for the analysis of predictors of poor oral hygiene: 1) Socio-economic and demographic factors, 2) caries status and 3) oral habit. The fundamentals that informed the developed model for predicting poor oral hygiene were the same for predicting the presence of caries. Socio-economic status, age and sex are potentially confounding factors for caries and oral hygiene status. In the second block, caries status was included as a moderator variable for oral hygiene. In the third block, digit sucking was recorded as the investigated oral habit that may influence the risk for poor oral hygiene.. ## Data Analysis The ages of the 10 children six to 11months were rounded off to 1 year for ease of data analysis. Descriptive analyses were conducted to determine the prevalence of digit sucking, caries and oral hygiene status. Bivariate analysis was conducted to test the association between dependent variables (presence of caries and oral hygiene status) and the independent variables (child’s age, gender, socio-economic status). Where appropriate, chi square tests were conducted. Logistic regression was used for inferential analysis. The hierarchical modeling started with the first block whose variables were adjusted simultaneously for each other. Only those variables whose p value were \<0.4 entered into the subsequent models. Variables of the second block were adjusted simultaneously for each other and for the variables whose p value were \<0.4 in the previous step. Digit sucking was included in both models irrespective of the p value in the previous steps because of the need to evaluate the effect of digit sucking on the presence of caries and poor oral hygiene. The significance of each variable was considered at the time of entry in the model (p value ≤0.05). All other blocks were then added in succession, following the same procedure. The estimated coefficients were expressed as odds ratios (ORs) and their 95% confidence intervals were also calculated. The Hosmer-Lemeshow goodness-of-fit test was done to confirm the consistency of the models. Where data were skewed, the dichotomized version was used. Statistical analysis was conducted with SPSS (version 17.0) for windows, while STATA software (version 10) was used for the logistic regression. Statistical significance was inferred at p ≤ 0.05. ## Ethical Consideration Ethical approval was obtained from the Ethics and Research Committee of the Obafemi Awolowo University Teaching Hospitals Complex Ile-Ife (ERC/2013/07/14). Approval to conduct the study was also obtained from the Local Government Authority. The study was conducted in full compliance with the study protocol. Written informed consent was obtained from the parents of study participantsafter duly explaining study objectives, risks and benefits, voluntary nature of participation and freedom to withdraw at any time. All children aged eight to 12 years also provided written assent. Efforts were made to minimize risks such as loss of confidentiality and discomfort, to participants. All data were collected without the identifier (names and addresses) of participants. Participants experienced no direct benefit and no compensation was paid. However they were given token gifts of stationery or a small tube of toothpaste containing 1450ppm of fluoride. The fluoride level of drinking water sources in most LGAs in the South West geopolitical zone of Nigeria where Ile-Ife is situated is between 0.00pm—0.30ppm highlighting the need to promote further use of topical fluoride. None of the gifts exceeded a value of \$0.50. # Results Only the data of 992 children recruited for the study were complete enough for analysis. This represents 90.1% of the proposed 1,011 study participants. None of the children recruited refused to participate. Participants included 508 (51.2%) boys and 484 girls (48.2%) with a mean age of 5.83 ± (3.15) years. There were 497 (50.1%) study participants in the 1 to 5 year age group and 495 (49.9%) in the 6 to 12 year age group, with mean ages of 3.15 ± (1.35) and 8.53 ± (1.90) years respectively. ## Digit Sucking Habit and Caries Profile Of Study Participants shows the socio-demographic and digit sucking profile of study participants. Seventy one (7.2%) of them had digit sucking habits. Fifty five (77.5%) engaged in thumb sucking habits while 16 (22.5%) sucked other digits. The majority of children with digit sucking habits (56.3%) fell into the 1 to 5 year age group. The prevalence of digit sucking was highest among the 2 year olds and least among the 7 and 12 year olds. There were no significant differences in the proportion of children aged 1 to 5 years and 6 to 12 years (p = 0.27), male and female participants (p = 0.37) and children in the different socio-economic classes (p = 0.40) who sucked their digits. highlights the caries profile of participants. One hundred and four children (10.5%) had dental caries. Significantly more females than males (61.5% vs 38.5%; p = 0.01), and more children in the 6 to 12 years than the 1 to 5 years age group had caries (71.2% vs 28.8%; p = 0.00). There was no significant difference in the proportion of children from each of the socio-economic strata who had caries (p = 0.13). The dmft score ranged from 0 to 8 with a mean score of 0.22 ± (0.80). There were 192 unrestored carious teeth, nine missing and three filled primary teeth. The DMFT score also ranged from 0 to 4 with a mean score of 0.04 ± (0.30). There were 29 unrestored carious teeth, three missing and two filled permanent teeth. Only eight (11.3%) of the 71 children with digit sucking habits had dental caries. There was no significant difference in the proportion of children with or without digit sucking habits who had caries (11.3% vs 10.4% p = 0.82). None of the children with digit sucking habits had missing or filled teeth. ## Digit Sucking Habit and Oral Hygiene Status of Study Participants The mean OHI-S score was 1.27 ± (0.73). Mean OHI-S score was significantly better in the 1 to 5 years age group compared with the 6 to 12 years age group (0.98 vs 1.56; p \< 0.001). No significant gender difference was observed in the mean OHI-S scores (1.28 vs 1.27; p = 0.73). There was a significant difference in mean OHI-S scores (1.07 vs 1.29; p = 0.02) of digit and non-digit suckers. highlights the association between oral hygiene status of study participants and dependent variables. The distribution of OHI-S scores was not significantly different across gender (P = 0.89) and socio-economic strata (P = 0.29). ## Digit Sucking Habit and Caries Status of Study Participants The association between the type of digit sucked, habit severity, caries and oral hygiene status was further explored. This revealed that there was no significant association between the type of digit sucked (p = 0.45) or severity of digit sucking (p = 0.53) and caries presence. Neither was there a significant association between the type of digit sucked (p = 0.32) or digit sucking severity (p = 0.79) and oral hygiene status. ## Predictors of Caries highlights the results of logistic regression to determine the predictors of caries. The Hosmer-Lemeshow test confirmed the goodness of fit of the model (p = 0.44). Age, gender and oral hygiene status were the significant predictor of caries for the study population. The odds of having caries increased with increasing age (OR: 1.09; 95% CI: 1.02–1.17; p = 0.01). Females had significantly increased odds of having caries when compared to males (OR: 1.79; 95% CI: 1.17–2.74; p = 0.01). Children with fair oral hygiene status also had significantly increased odds of having caries when compared with children with good oral hygiene (OR: 1.71; 95% CI: 1.09–2.68 p = 0.02). The odds of having caries also increased for children with middle (OR: 1.68; 95% CI: 0.95–2.94) and low socio-economic status (OR: 1.58; 95% CI: 0.87–2.85) when compared with those with high socio-economic status. Children with poor oral hygiene (OR: 2.51; 95% CI: 0.85–7.44) compared with those with good oral hygiene, and children with digit sucking habits (OR: 1.28; 95% CI: 0.58–2.81) compared with those without digit sucking habits also had increased odds of having caries. However, these findings did not reach statistical significance. ## Predictors of Poor Oral Hygiene highlights the results of the logistic regression to determine the predictors of poor oral hygiene. The Hosmer-Lemeshow goodness-of-fit test confirmed the consistency of fit of the model (p = 0.24). Age and presence of caries were the only significant predictors of poor oral hygiene. Children aged 1 to 5 years had reduced odds of having poor oral hygiene when compared with children aged 6 to 12 years (OR: 0.27; 95% CI: 0.21–0.36; P\<0.001)., Children with caries had increased odds of having poor oral hygiene when compared with children without caries (OR: 1.66; 95% CI: 1.07–2.59; P = 0.03). Female children (OR: 0.85; 95% CI: 0.65–1.11) and children who were digit suckers (OR: 0.58; 95% CI: 0.34–1.01) had decreased odds of having poor oral hygiene when compared with male children and non-digit suckers respectively. These findings were however, not significant. # Discussion This study is the first to determine the prevalence of digit sucking habit, caries and oral hygiene status of children at a population level in Ile-Ife. The prevalence of digit sucking, caries and poor oral hygiene among the study population was low. Digit sucking was not a predictor of caries and poor oral hygiene status. It however increased the odds of having caries and good oral hygiene in the study population, though these findings were not significant. Being a female and aged six to12 years were significant predictors of the presence of caries for the study population. Having caries and being between six and 12 years old were also significant risk factors for poor oral hygiene. The use of a household survey for study participants’ recruitment made the findings of the study generalizable to the study environment. This is because the recruitment method increased the probability of including children targeted for the study from all the socio-economic strata in the study population, irrespective of their ability to be enrolled in school or not. The robust analytical approach also reduced the chances of spurious inferences. However, the recruitment of study participants from enumeration sites familiar with research studies inherently introduced a bias into the study sample. Despite this limitation, the study was able to provide very useful information on oral health status and its relationship with digit sucking habits. One of the highlights of the study is the low prevalence of digit sucking in the study population when compared with prior reports. This supports previous suggestions of variability in the prevalence of NNS habits in different cultures. Proffit noted that the prevalence of oral habits is lower amongst less cosmopolitan communities like our study community, where children had ready access to their mothers’ breasts for a long period and rarely suck other objects. A second highlight is the insignificant difference observed in the proportion of children who were digit suckers across the different age groups, gender and socioeconomic strata. This is unlike a prior report of a decline in digit sucking habit with increasing age. More girls than boys were reported to continue digit sucking after beginning school. Adair also reported a higher prevalence of digit sucking habits in children with working class mothers. Children have to compete for their mother’s limited time and may resort to digit sucking for comfort. The low prevalence of digit sucking among the study population may have made it difficult to pick up age, gender and socioeconomic differences in the sub-analysis conducted. Third, the non-statistical association between digit sucking and caries observed in this study differed from prior findings of Yonezu and Yakushiji. Similar to the reports of Misbah, we observed that digit sucking increased the odds of having caries although our results did not reach statistical significance. We however did not observe a significant association between digit sucking and the risk of poor oral hygiene, a probability that Misbah alluded to. We suggest that digit sucking may have been protective due to increased salivary flow resulting from the habit. Future longitudinal studies may however help increase understanding on the association between digit sucking, caries and oral hygiene status. Fourth, the low caries prevalence in this study is similar to previous reports in many sub-Saharan African countries. We found age and gender to be predictive factors for caries. Age is a known risk factor for caries; as age increases, caries risk increases. The relationship between age and caries risk had been reported in prior studies in our study environment. However, the finding that females have increased risk for caries as shown by this study and other previous studies is still debatable. A few studies had highlighted an increased risk for caries in males. However, increasing evidences seem to show that females are truly at increased risk for caries. Differences in salivary composition, salivary flow rate, hormonal fluctuation, dietary habits and genetic variations increase the risk of females for dental caries. Despite the low caries prevalence, the range of the dmft and DMFT reported in this study implies that there is the need to actively identify children at risk for caries and manage them. Oziegbe and Esan reported higher ranges of dmft/DMFT for children in the same environment as ours. Their study however included children aged 4 to 16 years, which meant that they had more children whose teeth had been exposed to the oral environment for a longer duration, than the children in our study population. Increased duration of tooth exposure to the oral environment implies that teeth face greater risk of acidic assault and caries formation. The mean dmft score of our study population was also higher than the mean DMFT, a finding that has also been observed by others. Fifth, caries presence and being 6 to 12years old were predictive factors for poor oral hygiene. This has been earlier reported by Gopinath *et al*. Worse oral hygiene status in older children may be due to poor oral hygiene practices as children become independent of parental oral hygiene supervision. This underscores the importance of using various motivational methods to promote oral hygiene practices in adolescents. The risk of caries is heightened when the oral hygiene is poor.With the outcome of this study showing that caries presence increases the odds of having poor oral hygiene, the prospect of a cycle of caries and poor oral hygiene being set up throughout the life course of children in the study environment, if specific interventions are not instituted in adolescence to improve the oral hygiene, is high. In conclusion, though the findings of this study have increased public information about the associations between digit sucking, caries and oral hygiene status, it was not able to provide conclusive evidence on the association between the variables. The non-significant association between digit sucking, caries and oral hygiene status, and prior reports, on the detrimental effect of digit sucking on oral health makes it important to promote discontinuation of the habit as soon as feasible. We appreciate all the children, parents and field workers who participated in this study. We also appreciate Prof. F. J. Owotade for his assistance with the statistical analysis and Dr. Kimon Divaris for the critical review of the manuscript. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: KAK MOF EOO. Performed the experiments: HOA TAO NKO NMC OVO. Analyzed the data: KAK MOF. Contributed reagents/materials/analysis tools: KAK MOF HOA TAO EOO NKO NMC OVO. Wrote the paper: KAK MOF.

# Introduction Estuaries are biologically productive, supporting rich fisheries and diverse habitats, including oyster reefs, sea grass meadows, and vast expanses of fringing marshes. But their very richness coincides with human habitation, which has resulted in damage from pollution, overfishing and habitat destruction. Mid- Atlantic estuarine fisheries have severely declined from habitat alterations, pollution and overfishing. For example, the loss of a key species, the eastern oyster *Crassostrea virginica*, has had significant effects on estuarine ecosystems of eastern and Gulf Coast North America. Oysters and other estuarine bivalves affect estuarine water quality by removing particles , and influencing nitrogen cycling. Oyster reefs also create three-dimensional benthic habitat that enhances diversity of other suspension feeders and offers important refuge from predators. Oyster growth and disease rates vary substantially along the estuarine salinity transition between fresh and marine waters. Oysters on the Atlantic and Gulf Coasts exposed to marine salinity are readily infected by two diseases. The parasite *Haplosporidium nelsoni*, or MSX, caused 90–95% mortality in the eastern oyster *Crassostrea virginica* in Delaware Bay in the 1950s. MSX likely arrived in eastern North America from Japan, perhaps through an intermediate oyster host, and has spread to the eastern oyster from Florida to Nova Scotia. The infection period is seasonal and disease can be reduced by moving oysters to lower salinities where survival of MSX is poor. However, oyster growth and reproductive success decreases in lower salinities, and survival rates decrease below 5 psu. Oysters have faster growth rates in higher salinities, but MSX infections decrease survival, with a few exceptions of evolved resistance to the disease. MSX infections occur in the mesohaline and polyhaline zones of estuaries, but infection rates are much lower and often absent and oysters can grow in oligohaline zones of 6–12 psu. A similar tradeoff between growth and disease exists for the other major oyster disease, the alveolate protistan *Perkinsus marinus* known as Dermo. First discovered on the Gulf Coast of the United States, it has spread to the northeast and is a major source of mortality in marine waters. Increases of coastal sea surface temperature over the past few decades, especially in the form of winter warming, have facilitated the disease's northward spread. Like MSX, Dermo does not thrive in oligohaline salinities. In oligohaline waters, oysters grow slowly but have refuge from disease and from marine predators like whelks, oyster drills, flatworms, and starfish. In watersheds with controlled discharge, experiments have suggested that periods of increased river flow can temporarily reduce oyster disease, with enhanced growth during subsequent lower discharge periods. In natural estuaries, seasonal and interannual variability in river discharge leads to continuous variation in salinity. High discharge during freshets will lower salinity at a location, but droughts will increase salinity and potentially increase disease susceptibility. At the upper end of an estuary, increases in discharge may negatively impact oyster survival by reducing the frequency and duration of oligohaline conditions, making habitat that was formerly estuarine into a tidal freshwater river. Oyster restoration is a priority in many estuaries of eastern North America, and in particular the Hudson River estuary. The Hudson River estuary once supported among the richest oyster grounds in eastern North America, but signs of overfishing appeared early in the 19^th century, and urban pollution hastened the decline in the early 20^th century. Jamaica Bay supported thousands of oyster fishers through the 19^th century, but oyster populations are now negligible there due to pollution, habitat disturbance and the 1938 hurricane. The Tappan Zee-Haverstraw Bay (TZ-HB) region is a focus of restoration efforts in the Hudson due to historic oyster cultivation in that part of the estuary. In the 18^th and 19^th centuries Haverstraw Bay supported commercial oyster fisheries. In the 1950s, a time of below-average rainfall over the past century, the Flower and Sons Oyster Company moved their operation to the TZ-HB Bay region and raised juvenile oysters with high growth rates and survival. These results raised hopes that the broad shallow waters of TZ-HB with suitable bottom substrate and high benthic population densities would be well suited for oyster restoration. Oyster restoration objectives include not only reestablishment of fisheries, but also revitalization of a critical element of the estuarine ecosystem for increasing biodiversity and improving water quality. The current poor state of eastern oyster populations has led to skepticism for restoration potential, despite some successful efforts. Climate change is a one potential challenge for restoration. Increased sea surface temperature has facilitated the northward extension of Dermo and MSX, threatening oyster habitat in polyhaline and oceanic salinities. Regional shifts in timing and magnitude of precipitation with climate change will alter river discharges and estuarine salinities. Current climate models predict an increase in precipitation in the northeast U.S. of 5 to 8% in the next few decades and up to 30 percent by the end of the century with increases most likely in the winter and spring. Climate projections also suggest greater variability in streamflow with more frequent high and low discharge periods. The shifts in magnitude and timing of precipitation and discharge will affect the salinity distributions in estuaries and therefore the habitat, growth, and vulnerability of oyster populations and associated species. While estuarine oysters can tolerate freshwater during the winter, very low salinities cause high degrees of physiological stress under spring and summer temperature conditions. We are examining the Hudson River estuary, once a major oyster grounds and now a focus for restoration. We have combined regional studies of oyster performance (growth rate and survival) and estuarine modeling to predict physical conditions and potential impacts on oyster habitat under a regime of increased precipitation with climate change. # Results We investigated oyster performance in coastal and estuarine regions to evaluate tradeoffs between performance and disease occurrence. We compared growth, survival, reproduction, and disease occurrence at coastal sites from eastern Long Island, New York USA to western Raritan Bay, New Jersey USA and at sites in the TZ-HB section of the Hudson River estuary (river km 42–58). We quantified shell growth and disease prevalence (Dermo) of overwintered oysters that were transplanted from one hatchery (Fishers Island, New York USA) to replicate floating cages at 9 sites in 2008, and to 5 of these sites in 2009 (see “”). Shell growth showed a strong positive correlation with salinity. In contrast, Dermo was far less prevalent in the lower salinity sites of the estuary than in coastal sites. These results are consistent with the expected tradeoff between growth and disease with salinity, documenting the tradeoff more completely and at higher latitudes than previous work. The prevalence of MSX was low at most sites during 2008 and 2009. MSX was responsible for substantial mortality in 2008 at one site in the lower Hudson estuary (Pier 40, “P40” in, with a mean cumulative mortality of 43%). This elevated mortality due to MSX occurred near the mouth of the estuary, a location with higher salinities and greater salinity variability than the upper estuary sites in TZ-HB. Both 2008 and 2009 had higher than average precipitation (measured at Albany) and discharge in the Hudson River, but the timing of the high discharge period appears to be critical. At the TZ-HB sites salinities were in the range of 5–10 psu through July and August of 2008. We found generally high survival rates, albeit with low growth rates. During a discharge event in August 2008 salinity decreased below 2 psu at the site farthest up-estuary (Ossining, “OS”), corresponding with a mortality increase of ca. 30 percent. Precipitation and river discharge during the summer months of 2009 were greater than in 2008, with lower salinities at the estuarine stations and much greater mortality in TZ-HB. In contrast, mortality was minimal at the coastal sites in the study. In 2009, salinities in TZ-HB dropped to nearly 0 and remained around 3 psu for most of the summer. The populations farthest up-estuary (“OS”) died off completely and two other TZ-HB sites (“I” and “WI”) had significant mortality. Mortality at “I” and “WI” decreased in October as river discharge declined and salinity increased. A limited extension of the observations into 2010 offers additional evidence of the sensitivity of oyster growth and survival in TZ-HB to summer river discharge. Soft tissue growth, shell height growth, and survival were measured at the Washington Irving Boat Club in Tarrytown (“WI”) during the summers of 2008, 2009 and 2010. In 2009, summer precipitation and discharge were high: precipitation at Albany (averaged May 1 to September 1) was the highest in the 132 year record and average discharge in the Hudson ranked 13^th in the 93 year record. Average summer precipitation and discharge in 2008 and 2010 were significantly lower. Correspondingly, oyster growth at the Tarrytown site was much less and mortality was greater during the wet summer of 2009 than 2008 or 2010. To relate conditions during the observations to longer-term estuarine variability, we use a numerical model of the circulation and salinity in the Hudson River estuary that has previously been validated against observations. A hindcast of the salinity variability over the past ca. 90 years was made using the available discharge and tidal records. The model calculates the vertical salinity structure as well as the along-estuary distribution, and here we focus on salinities in the relatively shallow regions (depths less than about 3 m) on the east side of TZ-HB where leases for oyster culture were maintained in the 1950s and where restoration is most likely. Estuarine salinity depends inversely on discharge – as discharge increases, salt is pushed toward the mouth and salinity decreases. The model suggests that during high discharge periods, salinities in TZ-HB are frequently low enough to limit oyster growth and even survival. For example, averaging over the 90-year record, salinities in summer months in TZ-HB were in the range of 3 to 7 psu. In contrast, average salinities during the 5 years with the highest annual precipitation were on average 2 to 3 psu lower during the summer months. Model results in TZ-HB are also shown for 2008 and 2009. The increased precipitation during the late spring and summer of 2009 lead to decreased salinities during the summer, with salinities similar to the average conditions during historically high discharge years. While the annual precipitation and discharge were greater in 2008 than in 2009, the high discharge period in 2008 was during the typical spring freshet rather than during the summer months of oyster recruitment. Salinities from upper (river km 58) and lower (river km 42) TZ-HB over the full historical simulation demonstrate the inverse dependence between summer salinity (average July salinity shown here) and mean annual discharge. Conditions in the upper bay range from about 6 psu to essentially fresh, while the lower bay ranges from about 10 to less than 3 psu. The mean annual discharge and the mean discharge of the 5 years with greatest precipitation are indicated with markers on the abscissa for reference. The model results indicate that during high discharge years, only a very limited region of TZ-HB would retain sufficiently high salinities in summer to provide suitable oyster habitat. Mean annual discharge in the Hudson River depends primarily on regional precipitation. Most current climate models project increases in precipitation in the U.S. Northeast in the coming decades, with the greatest increases during winter and spring. Climate models predict a range of outcomes for summer precipitation. Oyster restoration prospects are sensitive to these projections, as summer is the season of oyster larval recruitment. The total projected increase in precipitation in the Northeast over the coming century is about 25 percent, similar to the difference between the average annual precipitation over the past 90 years and the average of the 5 highest precipitation years. If predictions of increased precipitation hold, particularly during the summer months, then decreases in salinity in TZ-HB may be detrimental to oyster survival and therefore restoration. In addition to the local salinity, the rate of salinity change can be a source of physiological stress and is a factor in oyster restoration. During high discharge, the estuarine salinity distribution compresses, and variability in salinity from tidal and meteorological forcing increases at any particular point due to the sharper salinity gradient. The temporal variability in salinity would be exacerbated by projected increases in the intensity of extreme precipitation events with climate change. Overall, the projected increase in precipitation and discharge would be expected to shift the location of suitable habitat for oyster growth. The bathymetry of the Hudson is such that appropriate water depths and appropriate substrate for oyster growth are sparse down-estuary from TZ-HB, so the total area with suitable water column and benthic conditions could be expected to decrease with a shift in the salinity distribution toward the mouth. Within the TZ-HB region, extensive suitable bottom areas exist that would support oyster growth and widespread larval recruitment has been observed there.An additional consideration is that increases in water temperature may exacerbate negative impacts of disease in oysters, particularly at coastal sites. # Discussion At present, the Tappan Zee – Haverstraw Bay region of the Hudson estuary provides suitable benthic habitat for oysters and a likely refuge from Dermo and MSX diseases. However, increased mortality in TZ-HB during the high discharge summer months of 2009 suggest that projected increases in precipitation with climate change may reduce salinities in this region below thresholds for oyster survival. Our modeling results suggest that discharges consistent with precipitation in future climate scenarios could decrease salinities in the region to levels below the threshold for oyster survival. The seasonal timing of precipitation and discharge remains a critical uncertainty in this assessment. While climate models generally agree that precipitation is likely to increase during winter and spring in the Northeast, uncertainty remains for the summer months that are important for oyster growth, spawning, and larval dispersal. Historically, high annual average precipitation correlates with lower salinities in July due to longer, higher volume freshets. Whether the trend continues depends on the future partitioning of precipitation between snow and rain and its effect on the timing of river discharge. Independent of the seasonal distribution, projections of increased variability in streamflow, are likely to be a stressor to oyster communities at the upstream margins of estuaries. Restoration of oyster populations in TZ-HB could have important implications for oysters throughout the Hudson-Raritan region. If populations could be restored, larvae from TZ-HB Bay might be exported to coastal sites in years when coastal populations with higher vulnerability to disease and predators fail to reproduce or survive. We found oysters recruiting to our cages in TZ-HB in the late summer of 2008, but could not determine if the larvae came from within the bay or from down estuary. Our observations in 2009 showing no recruitment in Jamaica Bay or the New York Harbor region suggests that the recruitment within TZ-H may have been indigenous. Thus the possibility for a metapopulation of interacting disease-prone, but high growth rate oysters on the coasts and low growth rate but disease-free oysters in the TZ-HB region could provide temporal reinforcement and promote overall survival of the regional oyster metapopulation. A model of connectivity has not yet been developed for this region, but restoration efforts would depend on maintaining a metapopulation of rapidly growing and disease-resistant local populations. In Chesapeake Bay, connections of similar distances have been shown to be feasible according to modeling studies. A broader assessment of effects of regional precipitation shifts on oyster populations in estuaries in eastern North America and the Gulf Coast could relate results to metapopulation design to maximize oyster recruitment and survival. Salinity structure in Chesapeake Bay, for example, is driven by variation of discharge in the major tributaries, particularly the Susquehanna. Anticipated increases in precipitation from climate change may cause major losses of oysters and estuarine habitat as salinity decreases, particularly in tributaries in the middle of the bay where isohalines may shift seaward by as much as 55 km. Delaware Bay has a small watershed and increased rainfall might have a salutary effect, driving low salinity waters and disease refuge into the shallow bay. Previous droughts were associated with expanded mortality from MSX as saline water moved into the upper reaches of Delaware and Chesapeake Bays. In general, the impacts of climate change on estuarine oyster populations will depend on how the modified salinity distribution corresponds to the location of suitable benthic habitat. The uncertainty of seasonal effects on changes in rainfall will strongly affect our predictions of potential for oyster restoration. The summer of 2009 was notable for increased precipitation and discharge during the late spring and summer, but climate predictions suggest increased precipitation may become more common in the future. In the Hudson, the shoals of Tappan Zee and Haverstraw Bay may evolve from a refuge from disease to an inhospitable habitat for oysters, eliminating a crucial component of a larger metapopulation. Even a decade of rainy years, such as the past decade in the Hudson, could hinder restoration efforts. Oyster restoration planning should take into consideration the response of the oligohaline transition between estuarine and fresh waters to potential shifts in forcing with climate change, in particular the magnitude and seasonal timing of discharge. The resilience of restored estuarine oysters may depend on the availability and proximity of suitable benthic substrate for colonization with shifts in the salinity regime. Significant uncertainty remains among predictions of climate change impacts on precipitation, as well as for other potential factors in oyster survival such as water temperature and sea level rise. Restoration efforts could address this uncertainty by focusing on estuarine regions that would allow for translation of the oysters in response to shifts in forcing and by continuously monitoring environmental conditions and oyster population response to better inform subsequent restoration efforts. Similar effects of climate change on the spread of disease have been widely noted and may portend major reorganization of natural communities in future decades. In the Hudson, the transitional zone of Tappan Zee-Haverstraw Bay and its vulnerability may provide lessons for estuaries throughout the world. The simultaneous effects of climate change on disease and physiological adaptations may give insight to the effect of regional climate change in other transitional environments. # Materials and Methods Eastern oysters, *Crassostrea virginica*, were placed in plastic mesh grow-out bags (14 mm mesh size) supported in wire cages suspended 1–2 meters below the surface at nine sites throughout the coastal New York, New Jersey, and Tappan Zee-Haverstraw Bay region. Two semi-rigid, rectangular shaped (dimensions of 94×43×7.6 cm) grow-out bags were placed in each wire cage. 300 oysters were placed in each grow-out bag, resulting in a starting density of 742 oysters m⁻². Oysters were purchased from the Fishers Island Oyster Farm and were spawned and settled in the summer of 2007 (data for) and overwintered before being transferred to the cages in June 2008. Oysters used in cages in 2009 were spawned and settled in the summer of 2008, overwintered and placed in cages in late May 2009 (data for). In coastal sites, three replicate cages (6 grow-out bags) were used, located about one meter apart. At Tappan Zee- Haverstraw Bay sites, two cages (4 grow-out bags) were each maintained one meter or more from the other. In both years, oyster height was measured with a random sample of 20 oysters from each sample bag without replacement every two weeks from June-November. We report oyster shell height for the October sampling. Since shell size was the same for all starting samples, the final mean shell height for a locality is a measure of shell growth. All live and dead oysters were counted to calculate survivorship. Cages and bags were cleaned of fouling organisms once every 2 weeks when measurements were taken. Temperature was monitored with in situ temperature loggers (TidbiT v2 temp loggers from Onset Corporation) attached to one cage at each of the 9 localities. Temperature was registered every 15 minutes. Salinity, temperature and dissolved oxygen were measured biweekly at cage depth using a YSI model 85 environmental TSO meter. Disease was assessed for occurrence and intensity of occurrence of MSX and Dermo in the laboratory. A sample of 30 oysters was tested once a year at each site in September. Oysters were dissected and biopsies of mantle and rectum tissues were incubated in Ray's fluid thioglycollate medium (RFTM) for the detection of *P. marinus*. Following incubation (1 week), biopsies were stained with Lugol's iodine and examined using a light microscope for the presence of enlarged, black stained parasite cells. Infection intensity was ranked (0–5) following a scale assessing the relative abundance of parasite cells in tissues (0: no infection, 5:heavy infection). MSX detection was performed using standard histopathology procedures. Briefly, a transverse slice of tissue roughly between 3 and 5 mm in thickness was made through the central region of the visceral mass to include digestive organs, gonads, as well as gill and mantle tissues. Tissue sections were placed in histo-cassettes and fixed in 10% buffered formalin. Following fixation, tissue samples were dehydrated and embedded in paraffin, sectioned (5 to 6 µm in thickness), and mounted on histology slides. MSX infection intensity was ranked as light, moderate or heavy based on the abundance of parasite cells in tissue sections and following general guidelines. The numerical model is an unsteady, quasi-2d solution for the along-estuary velocity and salinity distributions. The model has been previously applied to and validated for the Hudson River estuary based on comparisons with high resolution observations in a single year and against observations over several decades, corresponding with simulations presented here. The model was forced with river discharge upstream (USGS station \#01358000 from 1946 to present, \#01357500 from 1917, and \#01335754, from 1887) and with tidal water level downstream (NOAA stations \#8518750, \#8531680, and \#8534720). The model calculates the vertical structure of velocity and salinity at discrete points along the thalweg of the estuary (dx = 1 km). We extract model salinities at depths corresponding to the bed elevation on the shoals where the oyster sites were located. Precipitation observations were taken from Albany, NY (NCDC WBANID \#14735 and \#14796). Work on oysters was done with permission under permits to the New York-New Jersey Baykeeper Oyster Gardener Program (Raritan Bay and Jamaica Bay) and, for the other sites, under New York State Fish and Wildlife Scientific Collecting License number 1257 to Jeffrey Levinton. We thank Shauna Kuhn, Abigail Cahill, Pat Lyons, Megan Flennikan, and Sarah Winnicki for assistance in field work. We thank David Wethey for critical input. [^1]: Conceived and designed the experiments: JL MD DR AS BA. Performed the experiments: JL MD DR AS BA. Analyzed the data: JL MD DR AS BA. Contributed reagents/materials/analysis tools: JL MD DR BA. Wrote the paper: JL MD DR. Performed disease analysis: BA. Performed physical modeling: DR. [^2]: The authors have declared that no competing interests exist.

# 1 Introduction Worldwide, benefits from nature to society have been estimated to be worth more than the global gross domestic product. When ecosystems become degraded, the cost of restoration can be prohibitive, and businesses and communities who rely most directly on these services are typically the first to suffer the consequences. Neoclassical economics suggests that if property rights are clear and well defined (and if transaction costs are not too high), a social optimum can be attained via bargaining amongst ecosystem service providers and beneficiaries. This sets the basis for the market to theoretically protect and sustain those services. While this may work for some provisioning services over short time-horizons (e.g. food and fibre), markets often fail to reward those responsible for providing service (e.g. upstream farmers or forest managers whose work benefits those downstream) when benefits are hard to attribute a financial value to (e.g. mental health or spiritual benefits from nature) or when benefits mainly accrue to others in society (e.g. downstream flood protection) over longer time-horizons (e.g. climate change mitigation). As a result, many resource management decisions generate short-term private benefits to the owner or manager at the expense of longer-term public benefits, often leading to negative externalities (e.g. pollution or flooding). In response to this, governments commonly pay resource managers to adopt more sustainable practices and carry out other work that can protect or enhance public benefits from nature. Businesses may also pay for these public benefits for a variety of reasons, including the need to mitigate risks to their business (e.g. from climate change), reduce costs (e.g. by delivering cleaner water), secure social licence to operate or contribute towards corporate sustainability goals. This is increasingly being done via Payment for Ecosystem Service (PES) schemes, which offer monetary incentives to individuals or orgnanisations to adopt or alter behaviours, beyond what is legally mandated, to improve the provision of ecosystem services that would otherwise have been economically unviable to provide. However, there are a number practical challenges to the development and operation of PES schemes. Challenges that may deter buyers (such as food processors and water companies) and investors in ecosystem services (such as insurance companies and impact investors) include: the complexity of demonstrating the additionality and permanence of benefits (i.e. proving that they would not have happenned without investment and the benefits will be long- term), costs of monitoring and verifying benefits, coordination between investors to avoid non-paying beneficiaries piggybacking on investments (i.e. benefiting from the investment of competitors without contributing themselves) or benefits for one investor cancelling out benefits for others (for example, tree planting creating habitat for predators of a species being protected by a neighbouring scheme). There are also many potential barriers discouraging resource managers (for example, landowners, tenants and other businesses managing natural resources; the typical ‘suppliers’ whose actions shape ecosystem service delivery) from engaging in schemes. These include: poorly defined property rights, perceived (and real) risks of entering long-term contracts (including unknown impacts that managing for ecosystem services would have on land value), lack of clarity as to their eligibility for funding from public schemes after entering a privately funded (by private enterprise or investment) scheme, as well as more straightforward capacity issues relating to how they would implement and manage such schemes. There is also potential for private ecosystem markets to compete with publicly funded agri-environment schemes, which are becoming increasingly PES-like in their design. For example, the latest Rural Development Programmes under the EU’s Common Agricultural Policy pay more for environmental outcomes than ever before and post-Brexit agricultural policies in the UK are increasingly focusing on “public money for public goods”. Even if publicly funded schemes pay lower amounts over shorter time-horizons than privately funded schemes, they may still displace private funding if they are perceived to be simpler or more familiar, and hence lower risk to resource managers. The integration of different private ecosystem markets could address some of these issues by actively managing synergies and trade-offs. However, to date there have been few attempts to do this, and there is limited understanding of either the opportunities or barriers to integration of private markets. There is also limited analysis of interactions between public and private schemes, or how these might be better “blended”. While much has been written about international voluntary and compliance carbon markets in recent years, much less is known about the national and sub-national ecosystem markets that have proliferated in recent years, and how they operate or interact with each other. This paper therefore uses a comparative analysis of existing private ecosystem markets in operation or close to market at national and sub-national scales in the UK and elsewhere in Europe, to explore governance issues associated with integrating different types of ecosystem markets. Specifically, it aims to: - Develop a typology of ecosystem markets by comparing ecosystem markets currently in operation or close to market in the UK, Germany, Switzerland and the Netherlands; - Question some of the operating assumptions of ecosystem markets and offer insights that could enable the cost-effective operation of schemes that minimise trade-offs and integrate benefits across a wide range of land uses at landscape scales; and - Propose an approach that could be used to integrate multiple ecosystem markets, operating over multiple land uses and habitats, including the integration of private markets and the blending of public and private schemes designed to deliver public goods. The analysis includes all known private schemes operating or close to market in the UK, where the development of ecosystem markets has been a policy priority since the launch of the Woodland Carbon Code in 2011 and the 2011 Natural Environment White Paper (which included a Payment for Ecosystem Service Action Plan). It also includes all known privately funded schemes targeting peatland restoration in Europe, where innovative funding mechanisms have proliferated in recent years, providing insights into the operation of ecosystem markets internationally for this important habitat. # 2 Background There is a well-known and significant gap between the public funding currently available and the funds that are needed to address the twin challenges of climate change and biodiversity decline. In the UK alone, it has been estimated that it will cost £1.8M to meet Achai biodiversity targets, and the cost of reaching net zero GHG emissions by 2050 has been estimated at between £50–70 billion. However, there are significant challenges in delivering emission reductions in the land use sector, where it is estimated that it may cost £247 million to deliver net zero targets from peatlands, woodland and agriculture. This gap is likely to increase as Governments around the world respond to the economic impacts of the COVID pandemic of 2019–20. In the UK land use sector, this is compounded by post-Brexit agricultural policies, which will lead to an overall reduction in public funding for the sector by 2027 as support moves away from direct payments. The upfront costs of many nature-based solutions are prohibitive for owners and managers in the land use and marine sector, and it can be many years before monetizable benefits accrue, further exacerbating the funding gap. At the same time, members of the UK Investment Association managed £8.5 trillion in 2020 and the global bond market was worth \$21 trillion in 2019. Within this community is a small but growing group of impact investors who are willing to accept lower than market-rate returns on investment and higher levels of risk. There is also growing recognition in the corporate sector of increasing risks to business from the environment, with climate risks now commonly featuring on company risk registers. Although only 13 percent of US company directors ranked climate change as one of their top five risks for 2020, risk assessments over longer time horizons identify multiple risks from climate change, notably risks from extreme weather to infrastructure and supply chains, and “transition risk” as regulation and consumer preferences shift towards a low carbon economy, amplifying other more traditional risks e.g. being left behind by low carbon technology accelerations and resource scarcity. As a result, demand from the corporate sector is now growing rapidly for ecosystem markets, and there has been a recent proliferation of new schemes and markets, a number of which we review in this paper. These markets are being stimulated by Government investment, with the goal of using public funding to leverage private investment in natural capital. For example, in the UK, a Natural Environment Investment Readiness Fund was launched in 2021 to support the development of projects that can generate revenue from ecosystem services and attract repayable investment. The three-year £10 million programme is providing grants which project developers can use to build capacity and consortia to develop projects to an investible level. The UK will issue its first green government bond in 2021, setting an example to other governments on issuing green bonds in the year that the UK hosts the 26^th Conference of the Parties to the UN Framework Convention on Climate Change. The UK follows the example of Poland’s sovereign green bond (in 2016) and Germany’s inaugural green Bund (in 2020). # 3 Methods We conducted a comparative analysis of: 1) all known private ecosystem markets operating (or near to market) across dairy, arable, forestry and peatland systems in the UK; and 2) all four private peatland ecosystem markets known to be operating in Europe. For the purposes of our sample, we defined ecosystem markets as full developed platforms that could facilitate ongoing exchanges between multiple private buyers and sellers of ecosystem services in the UK and in European peatlands. As long as the scheme was designed primarily to facilitate private investment (and this was required in additionality criteria), schemes that also leveraged public investment were included in the sample. Schemes could facilitate investment directly through the purchase of ecosystem services or indirectly by providing credit supply and risk management, as long as the goal of the financial mechanism was to facilitate investment in ecosystem services. Schemes that were deemed out of scope included non-UK schemes (including international voluntary carbon markets), publicly funded schemes that did not require private finance as part of their operation, schemes at concept or early development stages, and single transaction bilateral arrangements that were not part of a longer-term scheme sourcing multiple projects for multiple buyers or investors. For this reason, voluntary and compliance carbon markets were not included in the analysis. Research was conducted in four phases, as shown in. Ethical approval was sought and granted from Newcastle University Ethics Committee in May 2018. Informed consent was gained from all participants, documented via signed consent forms with accompanying participant information sheets. For more detailed methods, see Gosal et al. and Olesen et al.. ## 3.1 Phase 1: Scoping A narrative literature review was conducted to identify all private ecosystem markets currently operating or near to market in the UK (in any agro-ecosystem) and all private ecosystem markets operating in peatlands in Europe. Unlike systematic reviews or meta-analyses, a narrative literature review is an expert- based “best-evidence synthesis” of key literature, which is better suited to reviews that aim to provide a broad overview via expert synthesis, where it is difficult to identify specific outcome measures. This literature also served to identify interview topics and questions for Phase 2. To ensure all relevant schemes were identified (including those close to market, which were not in the public domain) and refine the parameters of the analysis, scoping interviews were conducted for the UK comparative analysis and the two case studies, and a focus group was conducted with seven participants (including researchers, consultants and EU policy stakeholders) for the European comparative analysis of peatland schemes. During this phase, a number of new schemes were identified for analysis or removed from the study, on the basis of the inclusion and exclusion criteria in the previous paragraph. ## 3.2 Phase 2: Data collection Data was collected in 2020 via a review of documentation for each scheme and semi-structured interviews with scheme representatives and intermediaries, which covered governance and legal matters, economics and funding and the operation of each scheme. For the UK, 12 interviews were conducted with representatives the eleven markets: the Woodland Carbon Code (WCC), Landscape Enterprise Networks (LENs), Habitat Banking (HB), the proposed Natural Infrastructure Scheme (NIS), Nature-Climate Bond (NCB), Natural Capital Pioneer Fund (NCPF), Habitat Banking (HB) and the Blue Impact Fund (BIF). For the European peatland market analysis, a further 13 interviews were conducted with representatives of four private peatland ecosystem markets (the PCC in the UK, MoorFutures (MF) in Germany, max.moor (MM) in Switzerland and the Dutch Green Deal (GDNL). ## 3.3 Phase 3: Analysis Qualitative data from interview and focus group discussions were analysed thematically alongside documentation from each scheme. Interviews were recorded, transcribed and anonymised in line with ethical approval from the Newcastle University. The thematic analysis approach outlined by Braun and Clarke was used to undertake in-depth analysis of the interview and focus group transcripts in three stages: initial coding of ideas, views and concepts into minor themes; review and refinement of minor themes to identify major themes; evaluation of themes in relation to the objectives of the study to draw in relevant insights to the comparative analysis. Theoretical saturation was considered to be achieved when no new themes were identified from transcripts. ## 3.4 Phase 4: Triangulation Finally, preliminary findings from interviews and review of scheme documentation were triangulated via individual written feedback from interviewees (with those providing extensive inputs offered co-authorship), with the addition of a focus group for the UK schemes. In the focus group, findings from the interview phase were presented to participants for discussion in plenary, before breaking into two parallel groups to discuss options for integration between the three main private schemes in operation in the UK, and between public and private schemes. The focus group was attended by 12 participants including researchers, consultants, businesses, the third sector, an intermediary/broker and policy stakeholders from regulatory bodies and Government departments in Scotland and England. # 4 Results describes and then compares each of the schemes reviewed in terms of their approach to: validation and verification of outcomes; additionality and leakage; permanence; supply and demand issues; interaction with public funding; and scheme governance. These are discussed further in the Supplementary Material, Gosal et al. and Olesen et al.. The comparative analysis identified a number of points of commonality between the schemes that were reviewed (see and for more details). In summary, common characteristics and challenges across all schemes are: - Participation in all schemes was voluntary for both buyers and sellers. Clearing prices were reached between buyers and sellers occurred in a minority of schemes, mainly regional ecosystem markets. For the majority, prices were primarily determined by project costs, which were highly variable both within and between schemes. None based their prices on the price per tonne on the voluntary carbon market, which would typically have been too low to cover project costs. One of the ways that the four peatland schemes justified higher prices (compared to international carbon market prices) was by highlighting additional non-carbon benefits such as water quality benefits of restoration or biodiversity (more on bundling versus stacking of multiple services below). The marketing of co-benefits was common across all the schemes reviewed, but verification and quantification of co-benefits were limited in nearly all schemes (MF being the exception). - Most schemes used intermediaries to engage with project developers (e.g. landowners and tenants), or the scheme itself performed this function (e.g. BIF) and LENs used supply aggregators to aggregate sufficient density of supply within a single landscape. However, engagement with suppliers (typically landowners and managers) was a challenge for many schemes. In contrast, BIF had created a £90M project portfolio prior to entering its investment phase and did not foresee issues meeting demand from investors. - On the demand side, sensitivities around the willingness of businesses to share financial data were identified as a challenge to the establishment of co-procurement arrangements between buyers in schemes where this was possible. As well as this, additionality was an issue for buyers in some schemes where businesses were reluctant to pay for interventions that landowners/tenants should already be doing to comply with regulation and/or that could be paid for by public finance. - Across the schemes, consideration of the wider social distribution of ecosystem services was limited, although there was recognition of its importance for buyers with Corporate Social Responsibility goals. - Permanence was addressed primarily via contractual arrangements in the schemes reviewed, although Conservation Burdens (Scotland) and Covenants (England and Wales) were sometimes proposed by schemes as potential future options to provide additional assurances to buyers in some UK schemes, and BIF provided follow-on funding opportunities to enhance permanence. In addition to these common characteristics and challenges, the comparative analysis identified a number of important differences between the schemes that were reviewed, for example: - The four peatland schemes and WCC tended to validate and verify outcomes using site visits by independent certification bodies, HB was developing a third-party accreditation system and BIF accredited projects to relevant industry standards. However, validation mechanisms had not yet been developed for NCB and NCPF, and LENs and NIS provided validation in the form of evidence that interventions had been carried out, without requiring independent verification of ecosystem service outcomes. - Additionality was only assessed formally by the four peatland schemes, WCC and HB, typically via legal (e.g. projects go beyond what would be required by law), financial (e.g. projects would not be possible without carbon finance) and other additionality tests (e.g. application of biodiversity metrics in HB receptor sites). None of the other schemes applied formal additionality tests, relying instead on trusted intermediaries to manage additionality informally as part of the project design process (e.g. LENs) or identifying businesses that had been unable to fund sustainability initiatives via other means (e.g. BIF). - The peatland schemes and WCC tended to focus on selling (often fungible) climate mitigation benefits via market registries (e.g. the UK Land Carbon Registry run by IHS Markit). While co-benefits, such as biodiversity benefits were used to market these schemes as part of a bundle of services anchored on carbon, only MF quantified these benefits as part of an explicit package of multiple ecosystem services that were all being sold together. In contrast, other schemes were designed to sell or finance a wider range of (mainly non- fungible) ecosystem services, including water quality, soil function, biodiversity and animal welfare benefits, in addition to climate mitigation to buyers. None of the schemes stacked different fungible services for sale to different buyers. Additionality rules of fungible schemes meant that stacking of fungible services provided by the schemes reviewed would only be possible where the cost of delivering the service was too high to be financially feasible through the sale of a single service (e.g. the price per tonne of carbon would be prohibitive). However, where interventions deliver more than one service, and neither service could bring in sufficient funding to cover the cost of the intervention, stacking would in theory meet additionality tests in each scheme. For example, stacking could enable projects to meet “financial feasibility” tests where multiple sources of ecosystem service payments were necessary to fund the minimum threshold for private finance (15% in the case of PC and WCC). Alternatively, “economic alternative” tests could be met where the project would otherwise not be the most economically attractive option for that location, for example carbon finance alone is unlikely to outweigh the opportunity costs of replacing arable agriculture or horticulture with paludiculture or habitat restoration in lowland peat fenlands, but the addition of biodiversity finance might make habitat restoration economically attractive as an alternative to current land use. Finally, “barrier” tests could be met if it can be shown that the project would not be possible for any other reason without stacking payments for multiple ecosystem services. - Investments in the peatland schemes, WCC and HB tended not to be geographically linked to the locations or interests of buyers, who they sourced nationally, and some of these schemes ruled out international investment to avoid double-counting against national emission reduction targets. LENs, NIS, BIF, NCB and NCPF were able to accept funding from national and international buyers (e.g., overseas impact investors). However, LENs and NIS tended to focus on sourcing funding from regional stakeholders, on the basis that this is a scale at which synergies and benefit integration are easier to achieve. - Schemes relied to varying extents on public funding, both in terms of scheme operation and project financing. The peatland schemes and WCC were significantly more reliant on public funding for project financing and in many cases scheme operation than the other schemes reviewed. - Payment mechanisms varied significantly across schemes (and in some cases between interventions within schemes) with the use of different legal agreements, payment structures and investment aggregation platforms (ranging from intermediaries acting as demand aggregators to crowdfunding platforms). # 5 Discussion In this section, we will discuss some of the key differences between the schemes and markets included in the comparative analysis and explore the potential to integrate different types of ecosystem markets. In doing so, we explore the governance issues associated with private market integration and the blending of private and public funding for public goods. ## 5.1 Types of ecosystem market Based on the comparative analysis in and Supplementary Material, three different types of scheme emerged, based on their modes and geographical scales of operation, funding and outcomes: 1. **National carbon markets**, primarily sold verified, validated, additional and (often) fungible climate mitigation benefits (sometimes marketed as offsets), typically to national buyers within a single country at prices that reflect project costs more than they reflect carbon market rates, with permanence provided by legislation or long-term contracts and significant Government funding for projects and/or scheme operation. These differ from international voluntary carbon markets, which allow international buyers to purchase and trade carbon at market rates, and from compliance markets which are regulated by mandatory national, regional, or international regimes and only allow trading between regulated entities; 2. **Regional ecosystem markets** enabled buyers to manage environmental risks to their business by investing in a wider range of non-fungible ecosystem service outcomes (including water quality, soil function, biodiversity and animal welfare benefits), in addition to climate mitigation, typically to regional buyers, with varying levels of validation, verification, additionality and permanence and limited Government funding required for projects and/or scheme operation; and 3. **Green finance** provided risk-adjusted returns on investment for national and international investors (potentially including members of the public via crowdfunding) who were willing to accept lower than market rate returns, and financed the widest range of (sometimes fungible) ecosystem service outcomes, with verification of outcomes (and in one case additionality) using industry or Government agreed metrics and standards, permanence via long-term contracts or follow-on funding and limited Government funding required for projects and/or scheme operation. Although not included in our sample of UK-based schemes that are operational (or close to market) and peatland schemes in Europe, green finance mechanisms can also include loan-based schemes and insurance products. For example, Scottish Natural Heritage, Scottish Environmental Protection Agency, Scottish Wildlife Trust, RSPB, British Ecological Society and British Marine are developing a scheme based on loans with Lloyds Bank, in which commercial bank loans are be made to groups that can implement biosecurity measures to prevent the arrival or spread of invasive species or help eradicate them. Loans would be repaid from future savings on the costs of managing invasive species. For example, Willis Towers Watson have a Global Ecosystem Resilience Facility uses the prospect of reduced premiums to encourage investment in projects that reduce climate-related and other environmental risks to clients (e.g., coral reef protection and restoration to protect coastal businesses from storm surges), reducing risks and so making premiums more affordable. Corporate Social Responsibility (CSR) schemes are not included in the typology, as this is one of a range of motives for investing in ecosystem markets, and CSR can motivate investment in all three types of scheme identified above. The typology in provides an evidence-based distinction between the key types of ecosystem markets operating in the UK and Europe on the basis of their modes of operation, funding and outcomes. This may provide useful clarity for decision- makers in policy and practice who wish to expand the role of private investment (ranging from crowdfunding to green investment funds) in conservation and regenerative agriculture. For example, a practitioner may be able to use the typology to identify relevant existing schemes or develop new schemes that target the types of ecosystem services, project developers or investors they are most interested in. Alternatively, a policy-maker targeting climate change mitigation from the land use sector might prioritise the promotion of national carbon markets, whereas a Local Authority seeking to reduce flood risk might prioritise paying for or attracting private investment in natural flood management via LENs and/or investment in sustainable urban drainage systems via green bonds or other green finance mechanisms. A decision-maker interested in providing additional income streams for farmers might focus on a peatland scheme or LENs, and if they wanted to exclude overseas investment to ensure investments counted towards national net zero targets, they might focus on national carbon markets and regional ecosystem markets, rather than green bonds which tend to attract international impact investors. The typology also provides new academic insights into the operation of ecosystem markets in practice, which may challenge traditionally held notions of PES. In particular, regional ecosystem markets do not conform to a number of assumptions underpinning PES and financial markets, in which payments would normally be conditional on, or linked to, ecosystem service outcomes or returns on investment. For this reason, the next section considers the operation of this type of ecosystem market in greater depth. ## 5.2 Understanding the success of regional ecosystem markets The emergence and successful early operation of the regional ecosystem market model is particularly noteworthy, given how differently this model operates compared to national carbon markets and green finance. What constitutes a PES and how to define it is subject to much debate, but generally there is agreement on PES involving individuals or organisations (’buyers’) paying other individuals or organisations who manage natural resources (’sellers’) to deliver clearly defined benefits or “ecosystem services” from nature. While this definition of PES relaxes Wunder et al.’s original stipulation that transactions must be voluntary (they rarely are in publicly funded PES schemes), the conditionality of payments on the delivery of well-defined, agreed outcomes remains central to PES, and is widely assumed to be necessary to engender the necessary buyer confidence to underpin a functional ecosystem market. Therefore, the limited provisions for validation, verification and additionality in the regional ecosystem markets reviewed in this research may either be used to question whether these are indeed PES schemes, or to question how important conditionality is to the success of a PES scheme. Moreover, unlike green finance schemes, regional ecosystem markets are not designed to provide returns on investment. As such, it may at first glance seem surprising that the LENs scheme in particular had attracted significant levels of private sector investment and was proliferating across UK landscapes with new international LENs propositions being developed at the time of the analysis. Although prices across the schemes reviewed were dictated primarily by the costs of delivering projects and so varied from project to project, national carbon markets tended to calculate the cost of projects per tonne of carbon as a reference point to guide buyers. In contrast, LENs buyers had no way of knowing the likely climate benefits, let alone the price of these per tonne of carbon. Instead, they took a more risk- based approach to negotiating prices between buyers and sellers on the basis of risks to assets, supply chains or reputational risks, which could be reduced or avoided by paying for interventions in the landscape. In addition to providing a different reference point for buyers in the negotiation, the focus on risk often brought more senior executives responsible for risk management to the negotiation table with access to different budgets, compared to the sustainability and corporate responsibility officers typically involved in decisions around carbon offsetting. In addition, the metrics typically used to assess risk tended to be less precise than those used to assess offsetting, which may explain the willingness to work with trusted intermediaries to deliver risk reduction outcomes without the controls on verification, validation and additionality of projects that were seen in national carbon markets. This focus on risk management may also explain the broader range of interests captured by regional ecosystem markets, including for example, asset risks from flooding, supply chain risks arising from issues with water quality, soil function or animal welfare, reputational risks arising from threats to biodiversity, and the wider risks to assets, supply chains and reputation arising from climate change. This diversity of interests then drove demand for multiple ecosystem services, which had to be managed in space and time to avoid trade-offs where the delivery of one service (e.g. carbon via conifer plantation) compromised the delivery of another (e.g. biodiversity). However, this diversity of outcomes also increased the likelihood that companies who did not invest in a scheme may benefit from the investments of their competitors (the free-rider effect). The LENs scheme addressed the challenge of avoiding both trade-offs and free-riders by identifying multiple risks across landscapes used by a number of beneficiary organisations who could manage risks by working together at landscape scales. This increased the number of co-investors to reduce the free-rider effect whilst ensuring interventions worked together without generating trade-offs at the landscape scale through the identification of multiple interests across the investor community prior to constructing the landscape scale interventions to deliver against those interests. Aggregating demand for ecosystem services in this way also increased the overall amount of funding (by stacking payments for multiple benefits) and led to perceptions of long-term resilience in funding, as the risks of any individual investor withdrawing funding were reduced with an increased number and diversity of investors. This is consistent with the definition of a place-based PES scheme, which emphasises the multi-level governance of social, economic and biophysical attributes that shape a given place by bundling or layering the widest relevant range of ecosystem services in the same landscape. The successful aggregation of demand in LENs was in part due to the proactive role of trusted business-to- business brokers, compared to the national carbon markets, which tended to be managed by organisations with very different cultures and language (typically Non-Governmental Organisations, research institutes or Government agencies), who played a more passive role in engaging with investors. On the supply side, the limited requirements around verification, validation and additionality had the benefit of reducing red tape for land managers who wished to engage with regional ecosystem markets. Indeed, evidence from interviews with farmers working in the LENs scheme in Cumbria have shown widespread satisfaction with the scheme compared to the complexity of public agri-environment schemes. Although farmers still commented on the additional reporting burden, and had other criticisms of the scheme, engagement with the scheme was strong. The two most important drivers for farmer engagement with the scheme, according to a subsequent Delphi survey, were: i) additional, stable income for easily planned and reported, and flexible activities that were compatible with existing management; and ii) improving environmental outcomes and animal health. In addition to the relative simplicity of the regional ecosystem market model, trusted intermediaries were employed to actively recruit farmers, further reducing barriers to entry. These intermediaries aggregated suppliers of services, and so increased market potential (availability of saleable benefits) while reducing transaction costs (of contracting with multiple landowners/tenants). In contrast, national carbon markets were less proactive in recruiting land managers to develop projects. Neumann conducted a Social Network Analysis of PC and MF, showing little or no engagement with land management representatives in the two governance networks. Instead, decision-making was primarily taken by scheme co-ordinators in consultation with a small number of key researchers who acted as knowledge brokers, providing access to necessary evidence. There was limited active involvement from members of the policy community, although interviews showed that “weak ties” in the network to these more peripheral actors had played an important strategic role in gaining political support and funding for the two schemes. The role of the most engaged researchers in both networks was multifaceted, acting as trusted intermediaries to members of the policy community as well as providing access to evidence to inform scheme development and management. However, both networks were highly dependent on the knowledge, experience and trust that had been accumulated by a small number of scheme co-ordinators and managers, making the ongoing success of both schemes vulnerable to the impact of staff turnover (indeed, the Peatland Code Manager was replaced soon after the research was conducted). In the case of the Peatland Code, the Director had similarly strong networks, providing a degree of resilience to the management of the network. In the case of MF, despite stronger reliance on a single scheme co-ordinator and additional scheme coordinators in the other participating federal states, a larger number of researchers and practitioners played pivotal roles in the network, which may provide this scheme with more resilience to changes in staffing, compared to the Peatland Code. Despite the relatively informal governance arrangement of MF, compared to the two formal governance structures in the Peatland Code, the day-to-day operation of both schemes was similarly dependent on a small number of active members who regularly exchanged knowledge with others, and who were trusted by their network. More complex and formalised governance structures may be necessary to ensure accountability and transparency as new regional ecosystem markets develop and seek integration with national carbon markets. However, the successful operation of these schemes needs to mitigate the risk of losing key trusted individuals from the network, if these individuals provide access to expertise, political capital, funding and experience from across their networks. Equally, scheme resilience and delivery of outcomes may be strongly influenced by a small number of key players, which may limit the rate at which new schemes can successfully proliferate, based on their individual capacity. ## 5.3 Integrating private schemes The main reasons for integrating national and regional ecosystem markets that emerged from the stakeholder workshop (see phase 4 in Methods) were to increase levels of investment and drive more multifunctional outcomes from landscapes. National carbon markets have the potential to bring in new investors to regional ecosystem markets from beyond the region, and regional ecosystem markets have the potential to extend the range of habitats, land uses and interventions that can be funded, beyond those currently covered by national carbon markets. There is a danger that single habitat/service schemes, such as woodland carbon schemes may drive certain outcomes (e.g. climate change mitigation) at the expense of others (e.g. biodiversity), but by integrating national carbon markets and regional ecosystem markets, it may be possible to aggregate demand across multiple habitats and land uses for multiple ecosystem services, and so design schemes that reduce the likelihood of ecosystem service trade-offs. However, there are a number of governance and technical (e.g., additionality) challenges involved in integrating ecosystem markets. Integrating schemes could generate unwelcome levels of complexity, compared to retaining the status quo of separate schemes, given that these schemes are already operational without integration. There is also a danger that the “commercial force” of carbon markets (as one private sector stakeholder put it) might disrupt regional ecosystem markets that are not currently tapping into this market, leading to a significant refocussing of attention on a single ecosystem service. The need for schemes to deliver additional outcomes that would not otherwise have been delivered (or legally required) poses a more significant challenge to the integration of national carbon markets and regional ecosystem markets. As described in Section 3.2, regional ecosystem markets were less likely to include formal additionality tests, relying instead on quality assurance of work undertaken to deliver outcomes. However, if income streams for climate mitigation via a national carbon market are integrated with funding for a wider range of ecosystem services via a regional ecosystem market for a package of linked interventions on the same parcel of land, it may be difficult to ensure additionality tests are met. For example, if payments for water quality improvements are stacked on top of carbon payments for a peatland restoration project, it may be difficult to prove that the restoration would not have happened without the carbon finance. One solution to this is for national ecosystem markets to apply financial additionality tests (e.g. the Peatland Code and Woodland Carbon Code require a minimum of 15% carbon finance to be additional). In the case of the Woodland Carbon Code, projects can be de- registered if they integrate additional funding that was not declared prior to validation. Alternatively, although complex and currently untested, it may be possible to apportion credits to different budget contributions within a single project, limiting carbon credits to the proportion of the project funded by carbon finance. The simplest solution however, currently being pursued by UK schemes, would be to spatially separate the delivery of ecosystem services from different schemes, for example integrating peatland restoration and woodland creation in different locations upstream from farm-based projects managing soil carbon or planting hedgerows. The importance of intermediaries and brokers in achieving integration cannot be understated. In addition to working as supply and demand aggregators (see previous section), they also play an important role in identifying interventions and projects that could deliver monetizable benefits, demonstrating cash flows, evaluating risks and calculating potential return on investment, before presenting opportunities to investors, where relevant accrediting projects to standards (like those developed for national carbon markets) to increase investor confidence. Evidence from the comparative analysis suggested that communication and trust between scheme actors may be as important as the development of formal governance structures. For two of the peatland schemes analysed (PC and MF), there was evidence that researchers may play a more important role than has been previously appreciated, as trusted knowledge brokers and advocates rather than simply as providers of knowledge and evidence (c.f. Pielke). Financial brokers have the capacity to work across all three types of ecosystem market, and initiatives like the Broadway Initiative, Green Finance Institute and SRUC’s Thriving Natural Capital Challenge Centre in the UK are already connecting many private schemes and working with Government to create an enabling policy environment. Building this discussion, proposes three ways in which transactions between buyers and sellers could be managed to integrate both national carbon markets and regional ecosystem markets. In **Option 1**, a regional ecosystem market procures climate mitigation benefits from a scheme that is also supplying national carbon markets or green finance markets, either directly via a demand aggregator or intermediary (A), or with the demand aggregator procuring ecosystem services as part of a package of benefits arranged by a supply aggregator/intermediary (B). In **Option 2**, the carbon or green finance scheme acts as the supply aggregator, providing multiple functions from its own scheme with options (C) to source interventions from other supply side entities. The scheme may also supply additional climate mitigation benefits into national markets (D). In **Option 3**, the carbon or green finance scheme provides both demand and supply aggregation functions. Although this is the simplest integration option, it creates a conflict of interest because the same body is negotiating on behalf of both supply and demand side parties to the transaction. An important principle in integrating carbon, or any additional function into a multifunctional landscape trade, is that different income streams should be put together simultaneously to make a trading proposition, and that the full range of ecosystem services to be provided should be agreed prior to the proposition being agreed and implemented. Once implemented, there is typically little incentive for future buyers to pay for outcomes, since those outcomes are already being delivered, and the additionality tests of national carbon markets would not be met, since activities on the ground would demonstrably not be dependent on the additional payment. ## 5.4 Options for blending public and private finance for ecosystem services Finally, the research highlighted a number of potential areas of conflict between public funding for natural capital and privately funded ecosystem markets. These included the potential for public funds to outcompete private funds (e.g., where public schemes offer more attractive terms including shorter contract lengths and simpler or more familiar application processes), that would otherwise have enabled the market to deliver the public good. There was also considerable uncertainty over future public schemes as the UK develops and trials post-Brexit policy over a relatively long time-frame, which could freeze the market, with potential sellers withholding projects until they know whether they will get a better price or terms under existing private schemes versus future public schemes. To tackle these potential conflicts between public and private finance, three broad approaches may be considered: 1. Full public-private co-procurement of public goods, in which public and private finance are integrated into a single fund at a landscape scale designed to deliver multiple outcomes. An example of a private-led integration mechanism would be LENs, which could be adapted to integrate public funding as part of its demand aggregation process, co-ordinating landscape outcomes across multiple private and public investors. An example of a public-led integration mechanism like this would be Rural Land Use Partnerships, which are being piloted by Scottish Government to enable rural communities to shape natural capital investment priorities and provide benefits to these communities alongside land managers and other providers of ecosystem services. The advantages and disadvantages of this option are outlined in mechanism 1 in ; 2. Co-ordinated public-private funding of public goods, delineated in space or time, enabling the market to pay for as much as possible, while public payments focus on market failures and those who are not prepared to accept private finance. There are a range of mechanisms that could facilitate this, which are outlined in mechanisms 2–6 in ; and 3. Private funding pays for services that are not already being procured by public funding, with limited coordination. This is the “business as usual” scenario in most countries, and the primary mechanism that facilitates this are the additionality criteria in national carbon markets that enable public funding of schemes up to a certain amount (e.g. 85% in the Woodland Carbon Code and Peatland Code). The six options described in and show how public funding might be designed in future to incentivise participation in privately funded PES schemes, enabling the market to deliver significantly more public goods than it currently provides, while reserving public funding to address market failures and avoid distributional justice concerns about inequitable benefits arising from an entirely market-driven system. Several of these approaches may work best in combination. For example, funds delineation prioritises projects for the market that are able to deliver the most in-demand ecosystem services at the lowest price (often climate mitigation benefits), reserving public funds to pay for projects that are more expensive per tonne of carbon, but that offer other important ecosystem services that have a high value to society e.g. biodiversity or recreational benefits. A cost- benefit matrix or decision support tools such as the tool developed by Artz et al. for Scottish peatlands, could be used to identify sites most likely to deliver cost-effective GHG emission reductions based on the level and type of degradation, and factors likely to influence the cost of restoration, such as accessibility. At the same time, this tool could be used to delineate sites that would be more expensive to restore, but where there may also be important biodiversity and water quality benefits, reserving these sites for investors more interested in these outcomes, and prioritising public funding to sites and/or ecosystem services that the market fails to deliver. Where schemes do not allow overseas investment, the climate mitigation benefits of these privately funded projects count towards domestic net zero targets. However, where overseas investment is permitted, a balance has to be struck between the need to use public funding to meet net zero targets (and so designing public funding schemes to compete with private markets for the most cost-effective sites) versus prioritising sites where market failure is most likely to result in a lack of funding for public goods from nature. Funds delineation is relatively straightforward to implement (compared to many of the other options discussed below), but there is a danger that making room for private markets in this way doesn’t leverage additional private finance in some of the ways that other approaches can. Having said this, funds delineation is likely to stimulate some additional private investment by removing the option of public funding for the sites that are most attractive to the market, ensuring public schemes do not compete with private schemes, and increasing the number of projects that are therefore offered to the market. In contrast, carbon trigger funds and match funding provide a much stronger leveraging of private finance, directly stimulated by public investment. In the case of trigger funds, a proportion of public funding for projects is held back, and only released if a certain level of private investment can be secured within a particular time frame (typically after a project start date). The likelihood of securing private funding is then one of the selection criteria for public funding of these projects, ranging from signed contracts and letters of intent to the plans and track record of the project developer or an intermediary they are working with to bring in private investment. Carbon trigger funds are complex to administer, and a proportion of projects won’t ever actually get the private funding they seek. These projects won’t draw down the second instalment of their grant, leaving the public funder with projects that don’t reach their public good potential or private finance leveraging. However, if designed appropriately (e.g. restoring or planting up part of a site, rather than doing ground work across a whole site ready to investment that never materialises), projects may be able to deliver some benefits with public funding if they fail to get private investment. Because these are more likely to be in sites that deliver cost-effective climate benefits (to attract private investors), running a carbon trigger fund to stimulate private investment in the most attractive sites alongside funds delineation may provide governments with an attractive combination of leveraging power whilst considering issues of equity and distribution of benefits. One of the challenges of funds delineation is identifying which sites should be prioritised for public versus private funding, and carbon trigger funds are likely to prioritise projects on the basis of their ability to leverage private funding, rather than the efficiency with which they can deliver carbon and other benefits. The carbon guarantee is more likely to identify the most cost- effective sites for private buyers because it relies on a reverse auction to prioritise sites to be supported by the guarantee. It has the potential to mobilise private capital to finance restoration in the long term, replacing the year-to-year public grant system, whilst giving confidence to both project developers and investors. The mechanism has so far been tested through the Woodland Carbon Guarantee in England, but it has the potential to be replicated in other ecosystem markets especially in national carbon markets or where independent standards exist, such as the four peatland standards reviewed in. While the Woodland Carbon Guarantee typically relied on public funding to subsidise tree planting, future guarantee mechanisms could raise private capital to pay for capital works. For projects that require significant funding up front, for example to plant trees/hedgerows or do restoration works, there are currently two main options (which may be used in separately or combination). First, projects can use public grant funding, requiring the majority of project costs to be paid by the public and limiting the capacity for private funding to be leveraged. Second, some schemes allow projects to forward-sell “pending issuance units” once projects have been validated, prior to verification, at a significantly lower price than they would expect to achieve for verified units at a later date, to cover up-front project costs. However, this is a major disincentive for project developers who receive a fixed price for their carbon up front, which could have been worth significantly more had they been able to retain the units for future sale. However, the carbon guarantee mechanism opens an alternative funding mechanism for paying up-front project costs. If financing is agreed with impact investors with the terms made known to project developers prior to a reverse auction, these repayments (with interest) can be incorporated into bids, creating a floor price that enables projects to repay their finance, in addition to covering their own costs and profit. Investors may even provide commitments contingent on successfully accessing a guarantee. This then means that the carbon guarantee mechanism leverages both carbon finance and impact investment, giving the private sector confidence to invest in projects, knowing that project developers will be able to repay investment with a return by selling carbon units at the floor price via the guarantee if carbon markets are not able to sustain higher prices. If carbon markets are able to pay more than the floor price, then investors are repaid, and project developers retain any additional profits. There is a possibility under certain conditions that public funds (reserved in case the guarantee mechanism is triggered) are never used, and ecosystem services are delivered entirely via private funding, enabling public funding to be re-invested in future rounds. Conventionally, guarantees are offered at a project level, as shown in. Finally, it is possible to envisage the blending of public finance with multiple, co-ordinated private schemes. Funds delineation might reserve specific landscapes for private investment using a cost-benefit matrix, with carbon trigger funds, a match funding principle or guarantee mechanisms to leverage carbon finance, to pay for woodland creation, peatland and saltmarsh restoration or regenerative agriculture interventions that sequester and store soil carbon. Where interventions are too expensive to be paid via carbon finance alone, payments for biodiversity might be layered on top of the carbon finance to make projects financially feasible. If co-ordinated at a landscape scale by an entity such as LENs, it may be possible to aggregate demand for layered, fungible services such as carbon and biodiversity, with non-fungible services such as water quality and animal welfare (defined as a public good in UK law) for buyers seeking to reduce risks to their business from climate change or other drivers of change. At the same time, some land within the same landscape may be eligible for entry to agri-environment schemes to pay for interventions that deliver services not included in private schemes. # 6 Conclusions This paper has provided an empirical basis for a new typology of ecosystem markets based on schemes currently operating or under development in the UK and in European peatlands. Each have distinct operational scales of investment and delivery, modes of funding and governance models. Of particular interest are emerging regional ecosystem markets, which are stimulating and meeting demand for ecosystem services by framing demand in relation to business risks and aggregating both demand side interests and the supply of services, overcoming free-rider effects and minimising trade-offs between ecosystem services across a landscape. Contrary to assumptions underpinning traditional PES schemes, taking this approach may lead to strong and resilient demand for ecosystem services in the absence of tight coupling between payments and provision of benefits. However, integration of these regional ecosystem markets with national carbon markets and green finance mechanisms may provide an expanded range of investors and land uses from which a much wider range of services can be provided. The integration of private schemes may also make it possible to co-ordinate more effectively with public funding for ecosystem services, prioritising public funding towards landscapes and services not paid for by the market, and increasing the diversity and amount of funding for sustainable land management interventions. While the options for integrating private ecosystem markets proposed here are currently theoretical, there are now attempts to apply these integrative governance models in practice. Achieving integration between schemes is increasingly important as private ecosystem markets proliferate around the world. However, as separate schemes proliferate, so does the likelihood of competition and trade-offs between services provided by different schemes. The need to manage these trade-offs and ensure private investment contributes to multifunctional landscapes is therefore a key driver for considering how schemes can more effectively integrate with each other. As publicly funded schemes also become more PES-like in many countries, there is a risk of perverse outcomes if public funding pays for services that would otherwise have been provided by the market. However, by designing future public schemes to complement private ecosystem services, it may be possible to avoid these markets being crowded out, and even use public funding to leverage private investment, for example via carbon trigger funds. As Government budgets come under increasing pressure, stimulating ecosystem markets could help fill the funding gap, contributing to a green post-COVID economic recovery, and increasing the likelihood that ambitious climate and biodiversity targets are met. However, to unlock this private finance, mechanisms need to be developed to ensure public and private funding can be successfully blended in future nature- based projects, for example integrating funds delineation with carbon trigger funds or carbon guarantees (see discussion of). Robust standards (akin to those developed for peatland restoration in Europe) are needed to govern the development of new markets in a wider range of land uses and habitats, to provide investor confidence and ensure outcomes are delivered. Public funding may also be used to help these new markets develop investment pipelines with projects that are ready for investment with the associated staff and governance mechanisms to channel investment scale capital into nature-based solutions. In some contexts regulation may be considered, for example the integration of Net Biodiversity Gain in the planning system, requiring developers to make (typically offsite) provisions to compensate for biodiversity losses and provide additional biodiveristy gains. Government funding could also help unlock supply by employing facilitators to explain opportunities to owners and managers of land and marine assets, simplifying and democratising access to private finance. In conclusion, much still needs to be done to stimulate and integrate ecosystem markets, but with the right support and design, it may be possible to integrate multiple sources of private investment with public funding to start delivering the levels of funding needed to address the twin challenges of climate change and biodiversity loss. The typology developed in this paper was developed with reference to the analysis of private ecosystem markets in the UK and Europe, so care should be taken in applying this more widely. However, as the number of different markets and financial instruments facilitating the sale of ecosystem services grows, the conceptual clarity brought by this typology may aid those seeking to develop or access private finance in the context of green recovery and meeting net zero and other targets. # Supporting information There are no patents, products in development or marketed products associated with this research to declare. Thanks to Dr Gordon Mitchell (University of Leeds) for input to one of the reports that formed an early basis for this paper. Thanks to Julie Martin-Ortega (University of Leeds), Kerry Waylen (James Hutton Institute), Bryan Adkins (United Nations Environment Programme), Jo Pike (Scottish Wildlife Trust) and Rosmarie Katrin Neumann (Newcastle University) for useful comments on an earlier draft of this paper. 10.1371/journal.pone.0258334.r001 Decision Letter 0 Crossman Neville Academic Editor 2022 Neville Crossman This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 2 Mar 2021 PONE-D-20-40133 Integrating ecosystem markets to co-ordinate landscape-scale public benefits from nature PLOS ONE Dear Dr. Reed, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. Please submit your revised manuscript by Apr 16 2021 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <plosone@plos.org>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript: A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols> We look forward to receiving your revised manuscript. Kind regards, Neville Crossman, Ph.D. Academic Editor PLOS ONE Journal Requirements: Additional Editor Comments (if provided): I encourage the authors to respond very carefully to both reviews and put the effort into the major revision. When submitting your revision, we need you to address these additional requirements. 1\. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at <https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main _body.pdf> and [https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_titl e_authors_affiliations.pdf](about:blank) 2\. Thank you for stating the following in the Competing Interests section: "Mark Reed is Research Lead for IUCN UK Peatland Programme and sits on the Executive Board for the Peatland Code. Tom Curtis is a founding partner of 3Keel and helped develop Landscape Enterprise Networks. Matthew Hay is a Project Manager and Stephen Prior is co-founder of Forest Carbon Ltd. David Hill is founding owner of Environment Bank Ltd." Please confirm that this does not alter your adherence to all PLOS ONE policies on sharing data and materials, by including the following statement: "This does not alter our adherence to  PLOS ONE policies on sharing data and materials.” (as detailed online in our guide for authors [http://journals.plos.org/plosone/s/competing-interests](about:blank)).  If there are restrictions on sharing of data and/or materials, please state these. Please note that we cannot proceed with consideration of your article until this information has been declared. Please include your updated Competing Interests statement in your cover letter; we will change the online submission form on your behalf. Please know it is PLOS ONE policy for corresponding authors to declare, on behalf of all authors, all potential competing interests for the purposes of transparency. PLOS defines a competing interest as anything that interferes with, or could reasonably be perceived as interfering with, the full and objective presentation, peer review, editorial decision-making, or publication of research or non-research articles submitted to one of the journals. Competing interests can be financial or non-financial, professional, or personal. Competing interests can arise in relationship to an organization or another person. Please follow this link to our website for more details on competing interests: [http://journals.plos.org/plosone/s/competing-interests](about:blank) 3\. Please include captions for your Supporting Information files at the end of your manuscript, and update any in-text citations to match accordingly. Please see our Supporting Information guidelines for more information: [http://journals.plos.org/plosone/s/supporting-information](about:blank). \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Partly Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 2\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: N/A Reviewer \#2: N/A \*\*\*\*\*\*\*\*\*\* 3\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 4\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 5\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: This paper presents an attempt to identify the possibilities and impediments to integrated ecosystem service markets through an exploration of the emergence and framing of a small set of UK ‘ecosystem services’ markets and EU peatland markets. There are some clear lessons from the evaluation and the attempt to explore unification options has promise. Yet the markets explored are limited geographically and institutionally, and the exploration itself has some substantial weaknesses that I feel need to be addressed to support the conclusions drawn. For these reasons, expanded below, I have recommended reject. Major comments: It is not clear what is considered a ‘market’ and therefore considered in-scope or out of scope and why. Although there are several discussions in the paper it does not seem to me that there is clarity around supply and demand issues. At first glance, particularly from the discussion in the introduction, it appears only two-sided markets consisting of private sellers and buyers are within scope. Later in the paper it becomes clear that there are a range of different ‘sellers’, some of which could not be considered private, whilst some ‘markets’ may have large components of public funding (i.e. public buyers). The inclusion of investment funds and bonds is also complicating, on the one hand they appear to represent a private ‘demand’ in the market, however since in at least some instances the finance must be repaid at a future point in time they seem to be more about credit supply and risk management in environmental markets rather than actual market participants on either the demand or supply side. This confusion about what the paper covers is reflected in the commentary in section 2 ‘debate emerged over whether the analysis was considering schemes, markets, or stakeholder engagement frameworks’. In a nutshell the paper requires greater clarity about the ‘market’ analysed. This lack of clarity around what is considered a market means that it is unclear whether schemes such as The Countryside Stewardship Scheme should be included (publicly designed and funded agri-environment scheme), or some activities of not for profit nature conservation groups such as Wildlife Trusts (e.g. purchase and management of land for ecosystem services). More importantly, limiting the geographic scope of the paper to the UK and three markets focusing on peatland in Europe means that the potential for generalisation across landscape scale public good markets is limited. For example, the not-for-profit sector plays a much larger role in market formation and facilitation in the US (see for example The Nature Conservancy facilitates a number of landscape scale initiatives inclusive of markets) where there are a large number of different initiatives seeking landscape scale public good outcomes (at least in part) using markets. With respect to this point – the introduction should be altered to clarify that the exploration is limited to ‘ecosystem markets currently in operation or close to market in the UK, and peatland markets in Germany, Switzerland and the Netherlands’. Exploration of supply and demand issues (e.g. as specified in row in Table 1) only very briefly explores prices (which the discussion in text makes clear are not market clearing prices which would emerge were supply and demand to intersect) and does not address broader supply and demand issues (with the exception of BIF – where the discussion simply states ‘no problem expected’). A paper wishing to explore landscape scale outcomes should in my view credibly identify in the relevant case study markets whether there are constraints from both a supply and demand perspective. This step would help to identify overall potential feasibility, particularly given that some landscape scale outcomes can only be achieved with substantial participation – i.e. supply of the ecosystem service. Furthermore, the results conclude “… engagement with suppliers (typically landowners and managers) was a challenge for all schemes …” but this appears in conflict with the conclusion in the data summary in Table 1 where no such issues were identified. Indeed, it would be useful to identify whether these schemes (or an integrated version of them) are able to adequately address the landscape scale problem or the supply and demand is inconsistent with the solution scale required. The inclusion of national carbon markets into the discussion section is problematic from a language and practicality perspective. First, voluntary and compliance carbon markets were specifically excluded in section 2 yet seem to re-appear in Table 2 and in the discussion of types of ecosystem markets (because several case study schemes are categorized as such). Second, the basis for the conclusions in Table 2 and Section 4.1 are unclear. Early references to carbon markets (section 2) refer only to voluntary carbon markets. Noting the examples given in Table 2, the authors are indicating that there is a class of markets that are largely driven by carbon and operate anywhere within the boundaries of the relevant country. This should be made much more clear given the earlier language in the paper. Furthermore, there will likely be interaction with formal carbon markets which in effect set a floor price for this ecosystem service. The discussion of regional ecosystem markets would benefit from reference to the economic theory around club goods which these types of investment coalitions appear to be consistent with. This appears relevant at several points in sections 4 and 5. The paper presents a number of novel points which could be emphasised to further strengthen the paper: • There is a unfortunately a conflict between pure public goods represented by carbon and (at least partly) spatially delineated public goods such as flood mitigation – so integration of regional markets could either strengthen pathways to carbon markets and therefore the market OR weaken regional markets via the potential for institutional rules to prevent or complicate ecosystem service ‘stacking’. I suggest more attention could be paid to this point. • What are the challenges in moving from the facilitated, network style markets brokered by individuals towards a more anonymised trading approach? Is an anonymised trading approach incompatible with the market structures explored in the paper? (or are some of these markets essentially boutique?) • Regional ecosystem markets are essentially parallel markets under any scenario in the paper (because they are geographically focused). This means that they would always remain separate – but what can be learnt across them that facilitates integration in other ways across the overlapping national or ecosystem specific markets and green finance schemes? • Robust standards appear to make goods more ‘fungible’. Is there a role or a pathway that moves ecosystem goods along a continuum towards fungibility and market formation in the way that some of these markets are developing? It is also important to note that the ‘fungibility’ of carbon is due to agreed conversion of the green house gas forcing of a range of gases emitted to the atmosphere (e.g. methane particularly in a landscape scale setting). Reviewer \#2: Review of: Integrating ecosystem markets to co-ordinate landscape- scale public benefits from nature General: The manuscript is: 1) a review of several PES of varying types in UK and Europe, 2) an attempt to develop a typology of PES schemes, 3) an evaluation of advantages, disadvantages of different type of PES, and 4) evaluation of potential to further integrate ES markets. In general, I think that the topic is important in the context of a growing and diverse array of mechanisms to focus on ES service outcomes of investment and payments directly for ES outcomes and relatively little literature that looks at and evaluates of the diversity in this growing sector. I think that the methods applied interviews, focus groups and secondary research are appropriate for the task at hand. I also believe that a refined and improved version would provide valuable insight and be read, valued and cited by the relevant research and policy community. I do think that there are a number of writing and organisational issues in manuscript that would require improvement prior to the manuscript being published. I also feel that some very general conclusions seem pretty overstated. For example the manuscript ends by concluding that “with the right support and design, it may be possible to integrate multiple sources of private investment with public funding to deliver the levels of funding needed to address the twin challenges of climate change and biodiversity loss.” some additional specific comments are embedded in the attached commented original manuscript A more realistic conclusion that is backed by the evidence presented is that: we are currently seeing emergence of discussion and early small scale “niche market” implementation around the right support and design to integrate m.ultiple sources of private investment with public funding, but we remain a long way from being able to deliver the levels of funding needed to truly adequately address the twin challenges of climate change and biodiversity loss.” Specific Intro – the first part of discussion in section 4 – highlighted in yellow for the authors – is introductory material – not discussion in the sense that it is general and speaks to why the topic treated is an issue – its not discussion of the specific study findings. Move this text and using in re-write of intro. Methods- I’m happy with this section. It provides brief but complete explanation of what was done and I think that the appropriate and adequate methods are chosen and adequately explained. Results- The table in this section is a useful way to summarise a lot of relevant information and analysis. I read it carefully and referred to when reading discussion. The one thing that I did wonder about as the lumping together of what are two quite different public regulation driven markets – carbon and biodiversity offset. The former is easier to measure with better developed standardised protocols. The challenges of equivalence has made the latter more challenging to date. I was less pleased with the page and a half of results reporting that followed in two blocks of dot pointed text. I felt like that information could be better organised into a set of subheaded sections and that in some cases themes re- emerged under different dot points rather than being discussed all at once. Not sure what exactly the right sub-heading would be but could include: source of demand public versus private, verification/validation, intermediaries and supply/demand coordination, co-benefits, additionality. Discussion Mostly good discussion – I did wonder about extent to which conclusions were specific to the particular schemes investigated as opposed to the class of schemes generally. For example an advantage of local scheme described “successful aggregation of demand was in part due to the proactive role of trusted business-to-business brokers, compared to the national carbon markets, which tended to be managed by organisations with very different cultures and language (typically Non-Governmental Organisations, research institutes or Government agencies), who played a more passive role in engaging with investors.” I’m also aware that some green investors like Kilter Ltd investing for the Victorian Superannuation uses a similar model. Is this a chacteristic of local versus green investor? Conclusion We are still talking about a scale that is small and niche, this doesn’t really come across in conclusions. The conclusions can also be a bit better articulated regarding what recommendations for role of government. \*\*\*\*\*\*\*\*\*\* 6\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No Reviewer \#2: **Yes: **jeffery d connor \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <figures@plos.org>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0258334.r002 Author response to Decision Letter 0 21 May 2021 Reviewer \#1 Reviewer: This paper presents an attempt to identify the possibilities and impediments to integrated ecosystem service markets through an exploration of the emergence and framing of a small set of UK ‘ecosystem services’ markets and EU peatland markets. There are some clear lessons from the evaluation and the attempt to explore unification options has promise. Yet the markets explored are limited geographically and institutionally, and the exploration itself has some substantial weaknesses that I feel need to be addressed to support the conclusions drawn. For these reasons, expanded below, I have recommended reject. Major comments: It is not clear what is considered a ‘market’ and therefore considered in-scope or out of scope and why. Although there are several discussions in the paper it does not seem to me that there is clarity around supply and demand issues. At first glance, particularly from the discussion in the introduction, it appears only two-sided markets consisting of private sellers and buyers are within scope. Later in the paper it becomes clear that there are a range of different ‘sellers’, some of which could not be considered private, whilst some ‘markets’ may have large components of public funding (i.e. public buyers). The inclusion of investment funds and bonds is also complicating, on the one hand they appear to represent a private ‘demand’ in the market, however since in at least some instances the finance must be repaid at a future point in time they seem to be more about credit supply and risk management in environmental markets rather than actual market participants on either the demand or supply side. This confusion about what the paper covers is reflected in the commentary in section 2 ‘debate emerged over whether the analysis was considering schemes, markets, or stakeholder engagement frameworks’. In a nutshell the paper requires greater clarity about the ‘market’ analysed. Response: • We have now provided a definition for ecosystem markets in the first paragraph of the methods section: “For the purposes of our sample, we defined ecosystem markets as full developed platforms that could facilitate ongoing exchanges between multiple private buyers and sellers of ecosystem services in the UK and in European peatlands.” • We have listed the inclusion and exclusion criteria for our sample more explicitly in this paragraph too, addressing each of the points helpfully made by the reviewer: “As long as the scheme was designed primarily to facilitate private investment (and this was required in additionality criteria), schemes that also leveraged public investment were included in the sample. Schemes could facilitate investment directly through the purchase of ecosystem services or indirectly by providing credit supply and risk management, as long as the goal of the financial mechanism was to facilitate investment in ecosystem services.” • The point about including investment funds and bonds as “buyers” is correct – they are not technically buying ecosystem services, rather facilitating their production. Our point however, is that without that private finance, the services would not be supplied, so these are included in our definition of markets that “facilitate” buying and selling of ecosystem services. Green finance is a major player in the ecosystem market landscape, and to exclude this based on a narrow definition of markets would significantly constrain the utility of the paper for readers in policy and practice, where a growing number of finance mechanisms are making new ecosystem markets possible. • The final point about debate (not “confusion”) over which markets to include during the scoping phase has been clarified with reference to the clearer articulation of exclusion and inclusion criteria now included in the introduction to the methods section. Reviewer: This lack of clarity around what is considered a market means that it is unclear whether schemes such as The Countryside Stewardship Scheme should be included (publicly designed and funded agri-environment scheme), or some activities of not for profit nature conservation groups such as Wildlife Trusts (e.g. purchase and management of land for ecosystem services). More importantly, limiting the geographic scope of the paper to the UK and three markets focusing on peatland in Europe means that the potential for generalisation across landscape scale public good markets is limited. For example, the not-for-profit sector plays a much larger role in market formation and facilitation in the US (see for example The Nature Conservancy facilitates a number of landscape scale initiatives inclusive of markets) where there are a large number of different initiatives seeking landscape scale public good outcomes (at least in part) using markets. With respect to this point – the introduction should be altered to clarify that the exploration is limited to ‘ecosystem markets currently in operation or close to market in the UK, and peatland markets in Germany, Switzerland and the Netherlands’. Response: • Publicly funded schemes were excluded from the study on the basis that the scope was private ecosystem markets. The definition of markets and scope including more explicit inclusion and exclusion criteria has now been clarified, and we have explicitly stated in reference to our definition that publicly funded schemes that did not require private finance for their operation were excluded from the study • We agree that the potential for generalisation is limited by the geographical scope of our research. In the first sentence of the conclusion, we already made it clear that “This paper has provided an empirical basis for a new typology of ecosystem markets based on schemes currently operating or under development in the UK and in European peatlands”. However, we have made this more explicit now, at the end of the conclusion, in response to this helpful comment by the reviewer. The introduction already states the scope clearly when the aim of the paper is introduced: “This paper therefore uses a comparative analysis of existing private ecosystem markets in operation or close to market at national and sub-national scales in the UK and elsewhere in Europe”. Moreover, this is further clarified in the first objective of the paper under this broad aim, to Develop a typology of ecosystem markets by comparing ecosystem markets currently in operation or close to market in the UK, Germany, Switzerland and the Netherlands”. Reviewer: Exploration of supply and demand issues (e.g. as specified in row in Table 1) only very briefly explores prices (which the discussion in text makes clear are not market clearing prices which would emerge were supply and demand to intersect) and does not address broader supply and demand issues (with the exception of BIF – where the discussion simply states ‘no problem expected’). A paper wishing to explore landscape scale outcomes should in my view credibly identify in the relevant case study markets whether there are constraints from both a supply and demand perspective. This step would help to identify overall potential feasibility, particularly given that some landscape scale outcomes can only be achieved with substantial participation – i.e. supply of the ecosystem service. Furthermore, the results conclude “… engagement with suppliers (typically landowners and managers) was a challenge for all schemes …” but this appears in conflict with the conclusion in the data summary in Table 1 where no such issues were identified. Indeed, it would be useful to identify whether these schemes (or an integrated version of them) are able to adequately address the landscape scale problem or the supply and demand is inconsistent with the solution scale required. Response: • The reviewer refers specifically to Table 1, which provides only a high-level summary of supply and demand issues. However, there is an entire section on supply and demand issues in supplementary material. We wanted to discuss each row of the table in depth, whilst keeping the main body of the paper as concise as possible for readers, and so put the additional results and discussion in supplementary material. Given that this is easily missed, in addition to signposting this in the main text, we have also linked to supplementary material in the legend of Table 1, and we have signposted the supplementary material at each of the other points Table 1 is referenced in the main article. • Table 1 has now been extended to summarise issues with supply that are discussed in supplementary material to avoid the apparent contradiction between the main article and this table. The conclusion itself (in the first bullet list in section 3) has also been amended to better reflect the summary as it now stands in Table 1. • In addition to the extensive discussion of supply and demand issues in supplementary material, we have now discussed pricing more explicitly in the main article, considering which schemes reached clearing prices via negotiation between buyers and sellers versus schemes that based prices on the cost of interventions (first bullet list in section 3). We have also discussed stacking versus bundling of payments and how these payments could meet additionality criteria across multiple schemes (second bullet list in section 3). Reviewer: The inclusion of national carbon markets into the discussion section is problematic from a language and practicality perspective. First, voluntary and compliance carbon markets were specifically excluded in section 2 yet seem to re-appear in Table 2 and in the discussion of types of ecosystem markets (because several case study schemes are categorized as such). Second, the basis for the conclusions in Table 2 and Section 4.1 are unclear. Early references to carbon markets (section 2) refer only to voluntary carbon markets. Noting the examples given in Table 2, the authors are indicating that there is a class of markets that are largely driven by carbon and operate anywhere within the boundaries of the relevant country. This should be made much more clear given the earlier language in the paper. Furthermore, there will likely be interaction with formal carbon markets which in effect set a floor price for this ecosystem service. Response: • Although the exclusion of international voluntary carbon markets was already justified in the introduction and methods, this has been made more explicit by listing this as an example of a market that was excluded from the study in the revised introduction to the methods section • Table 2 did not and still does not mention voluntary or compliance markets at any point. We have however clarified our definition of national carbon markets and explained more clearly how these differ from voluntary and compliance markets now in the definition of carbon markets in the text (section 4.1). As part of this, we have also explained the very loose link to carbon market rates, given that prices tended to be based primarily on project costs • We have introduced the idea of carbon guarantees with set floor prices negotiated via revers auctions, which may be used to provide liquidity and confidence within national carbon markets in the discussion (see also the additional row in Table 3 and Figure 3) Reviewer: The discussion of regional ecosystem markets would benefit from reference to the economic theory around club goods which these types of investment coalitions appear to be consistent with. This appears relevant at several points in sections 4 and 5. Response: The paper now contains a proposal on how regional ecosystem markets can be interpreted in terms of “club goods”. The property of rivalry is interpreted as a form of “congestion” that generates disutility in investors (facilitating free-riding) if the number of investors is too high compared to the environmental benefit received. Although other investments may reduce the cost of participation in the schemes (especially for fixed cost schemes), they may induce free-riding if the marginal benefits perceived by the provision of ecosystem services occurs at a lower rate than cost reduction. A mechanism with fixed and variable cost is proposed, with government paying up front fixed cost of the investment, and private buyers negotiating prices of services with suppliers up to the point that the incremental change of participating to the markets equalise the benefits received. Due to space constraints, this has been integrated with supplementary material at the end of the paper. Reviewer: The paper presents a number of novel points which could be emphasised to further strengthen the paper: • There is a unfortunately a conflict between pure public goods represented by carbon and (at least partly) spatially delineated public goods such as flood mitigation – so integration of regional markets could either strengthen pathways to carbon markets and therefore the market OR weaken regional markets via the potential for institutional rules to prevent or complicate ecosystem service ‘stacking’. I suggest more attention could be paid to this point. • What are the challenges in moving from the facilitated, network style markets brokered by individuals towards a more anonymised trading approach? Is an anonymised trading approach incompatible with the market structures explored in the paper? (or are some of these markets essentially boutique?) • Regional ecosystem markets are essentially parallel markets under any scenario in the paper (because they are geographically focused). This means that they would always remain separate – but what can be learnt across them that facilitates integration in other ways across the overlapping national or ecosystem specific markets and green finance schemes? • Robust standards appear to make goods more ‘fungible’. Is there a role or a pathway that moves ecosystem goods along a continuum towards fungibility and market formation in the way that some of these markets are developing? It is also important to note that the ‘fungibility’ of carbon is due to agreed conversion of the green house gas forcing of a range of gases emitted to the atmosphere (e.g. methane particularly in a landscape scale setting). Response: • These are all interesting and pertinent questions, and we have done our best to answer these in the paper. However, we have already extended the paper significantly and to do these questions justice would take significantly more room than we have available given the word limit. As a result, some of these questions remain unanswered in this paper and we hope to explore them in greater depth in future work. • Nevertheless, we have integrated a more detailed discussion of bundling versus stacking of ecosystem services into section 3 now • Although technically separate, we explain how regional ecosystem markets and national carbon markets can interact in more or less integrative ways in the discussion Reviewer \#2 Reviewer: General: The manuscript is: 1) a review of several PES of varying types in UK and Europe, 2) an attempt to develop a typology of PES schemes, 3) an evaluation of advantages, disadvantages of different type of PES, and 4) evaluation of potential to further integrate ES markets. In general, I think that the topic is important in the context of a growing and diverse array of mechanisms to focus on ES service outcomes of investment and payments directly for ES outcomes and relatively little literature that looks at and evaluates of the diversity in this growing sector. I think that the methods applied interviews, focus groups and secondary research are appropriate for the task at hand. I also believe that a refined and improved version would provide valuable insight and be read, valued and cited by the relevant research and policy community. I do think that there are a number of writing and organisational issues in manuscript that would require improvement prior to the manuscript being published. I also feel that some very general conclusions seem pretty overstated. For example the manuscript ends by concluding that: “with the right support and design, it may be possible to integrate multiple sources of private investment with public funding to deliver the levels of funding needed to address the twin challenges of climate change and biodiversity loss.” … A more realistic conclusion that is backed by the evidence presented is that: we are currently seeing emergence of discussion and early small scale “niche market” implementation around the right support and design to integrate m.ultiple sources of private investment with public funding, but we remain a long way from being able to deliver the levels of funding needed to truly adequately address the twin challenges of climate change and biodiversity loss.” Response: This statement has been toned down and followed by a sentence unpacking some of the limitations of our work, cautioning over-generalisation. Reviewer: some additional specific comments are embedded in the attached commented original manuscript Response: These have all been considered and where relevant actioned Reviewer: Specific Intro – the first part of discussion in section 4 – highlighted in yellow for the authors – is introductory material – not discussion in the sense that it is general and speaks to why the topic treated is an issue – its not discussion of the specific study findings. Move this text and using in re-write of intro. Methods- I’m happy with this section. It provides brief but complete explanation of what was done and I think that the appropriate and adequate methods are chosen and adequately explained. Results- The table in this section is a useful way to summarise a lot of relevant information and analysis. I read it carefully and referred to when reading discussion. The one thing that I did wonder about as the lumping together of what are two quite different public regulation driven markets – carbon and biodiversity offset. The former is easier to measure with better developed standardised protocols. The challenges of equivalence has made the latter more challenging to date. I was less pleased with the page and a half of results reporting that followed in two blocks of dot pointed text. I felt like that information could be better organised into a set of subheaded sections and that in some cases themes re- emerged under different dot points rather than being discussed all at once. Not sure what exactly the right sub-heading would be but could include: source of demand public versus private, verification/validation, intermediaries and supply/demand coordination, co-benefits, additionality. Response: • The first part of the discussion has been moved to a new “background” section between the introduction and methods sections to avoid making the introduction too long and unfocussed • The results were originally organised under sub-headings which matched the rows of Table 1, but this was significantly longer than the available word limit, so the extended results were moved to supplementary material, where these sub-headings and a more in-depth description of our findings can be found. As a comparative analysis, we grouped our high-level summary of the findings into differences and commonalities between the schemes (the two bullet lists). In some cases, there are key differences and commonalities within the same theme, which we now realise may appear repetitive. We have therefore made it clearer that the two groups of points are only a summary, and that readers should refer to Table 1 and supplementary material for more detailed information. We have also extended the point about the sale of ecosystem services in the second list (differences between schemes), as pricing is already covered in the first list (commonalities between schemes), to make it clearer that the key differences are around their approach to bundling versus stacking of payments for different ecosystem services Reviewer: Mostly good discussion – I did wonder about extent to which conclusions were specific to the particular schemes investigated as opposed to the class of schemes generally. For example an advantage of local scheme described “successful aggregation of demand was in part due to the proactive role of trusted business-to-business brokers, compared to the national carbon markets, which tended to be managed by organisations with very different cultures and language (typically Non-Governmental Organisations, research institutes or Government agencies), who played a more passive role in engaging with investors.” I’m also aware that some green investors like Kilter Ltd investing for the Victorian Superannuation uses a similar model. Is this a chacteristic of local versus green investor? Response: We have checked this and other points made in the discussion, and they are clearly linked to the specific schemes in our results (in the case above, to LENs). However, as the scheme was only mentioned towards the top of the paragraph, we have mentioned it again later to make this clear and the quoted sentence now reads, “The successful aggregation of demand in LENs…” Reviewer: Conclusion We are still talking about a scale that is small and niche, this doesn’t really come across in conclusions. The conclusions can also be a bit better articulated regarding what recommendations for role of government. Response: Conclusions have been toned down and limitations added (see response above). We have added an extended discussion of the options for government to blend public and private finance, focusing on how three out of the six suggested mechanisms in Table 3 could be combined to use public funding to leverage private investment. These more detailed recommendations are also now highlighted in the conclusions. 10.1371/journal.pone.0258334.r003 Decision Letter 1 Crossman Neville Academic Editor 2022 Neville Crossman This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 27 Sep 2021 Integrating ecosystem markets to co-ordinate landscape-scale public benefits from nature PONE-D-20-40133R1 Dear Dr. Reed, We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements. Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication. An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at <http://www.editorialmanager.com/pone/>, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to- date. If you have any billing related questions, please contact our Author Billing department directly at <authorbilling@plos.org>. If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact <onepress@plos.org>. Kind regards, Neville Crossman, Ph.D. Academic Editor PLOS ONE Additional Editor Comments (optional): Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation. Reviewer \#2: All comments have been addressed \*\*\*\*\*\*\*\*\*\* 2\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 3\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#2: N/A \*\*\*\*\*\*\*\*\*\* 4\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 5\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 6\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#2: I'm happy with the revisions, the authors have addressed my initial concerns and those of the other reviewer well. This is a very good contribution to the literature done thoroughly with appropriate methods and well written - ready to publish in Plos Oen \*\*\*\*\*\*\*\*\*\* 7\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#2: **Yes: **jeffery d connor 10.1371/journal.pone.0258334.r004 Acceptance letter Crossman Neville Academic Editor 2022 Neville Crossman This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 10 Dec 2021 PONE-D-20-40133R1 Integrating ecosystem markets to co-ordinate landscape-scale public benefits from nature Dear Dr. Reed: I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department. If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact <onepress@plos.org>. If we can help with anything else, please email us at <plosone@plos.org>. Thank you for submitting your work to PLOS ONE and supporting open access. Kind regards, PLOS ONE Editorial Office Staff on behalf of Dr. Neville Crossman Academic Editor PLOS ONE [^1]: MR is Research Lead for IUCN UK Peatland Programme and sits on the Executive Board for the Peatland Code. TC is a founding partner of 3Keel and helped develop Landscape Enterprise Networks. MH is a Project Manager and SP is co-founder of Forest Carbon Ltd. DH is founding owner of Environment Bank Ltd. AG and SPA are employees of the Institute for European Environmental Policy. SA is an employee of Collingwood Investments Ltd and Project Maya CIC. AO is an employee of Forest Stewardship Council. RF is an employee of Finance Earth. Authors’ commercial affiliations do not alter our adherence to PLOS ONE policies on sharing data and materials. There are no patents, products in development or marketed products associated with this research to declare.

# Introduction Atrial fibrillation (AF) is the most common arrhythmia. In addition to deteriorating heart function, AF is associated with a high risk of stroke and systemic embolism and increased mortality, especially in patients with rheumatic mitral stenosis (MS). Thrombogenesis of the left atrium represents the underlying mechanism of AF-related embolism. Although left atrial appendage velocity is a predominant factor of thromboembolism in AF, inflammation is an independent predictor of thrombogenesis. A variety of inflammatory markers such as C-reactive protein (CRP), tumor necrosis factor-α (TNF-α), interleukin (IL)-2, and IL-6 have been linked to embolism of AF.The underlying mechanisms linking inflammation to thrombosis include production of tissue factor (TF) from mononuclear cells, increased platelet activation and over-expression of fibrinogen. In patients with AF, a pathological study reveals infiltration of macrophages and adjacent myocyte necrosis in atrium, which is absent in patients with sinus rhythm. However, the role of macrophages in thrombogenesis of AF is not clear. Macrophages play a critical role in non-specific defense (innate immunity), and also help initiate and regulate specific defense mechanisms (adaptive immunity) by recruiting other immune cells. Subpopulations of macrophages manifest different proinflammatory or anti-inflammatory responses. Proinflammatory M1 macrophages are recruited early during tissue damage to clear dead cells. By contrast, M2 macrophages, which have reparative functions and secrete proangiogenic or fibrotic mediators to promote wound healing, are recruited after M1 macrophages. No evidence is available supporting quantitative differences between subsets of macrophages or the underlying mechanisms in thrombogenesis of AF. In the present study, we analyzed the polarization of macrophages and related cytokines in the atrium of patients with rheumatic mitral stenosis with or without thrombus formation to identify the possible role of different subsets of macrophages in thrombogenesis of AF. # Methods ## Study population and tissue specimens Consecutive patients with rheumatic mitral stenosis undergoing valve replacement were enrolled between January 2014 and May 2015 at Nanjing Drum Tower Hospital affiliated to Nanjing University Medical School and Huaian First People’s Hospital. The patients were divided into three groups: AFThrombus group (n = 19), AF(+)Thrombus group (n = 57) and AF(+)Thrombus(+) group (n = 14). Two- dimensional echocardiography was evaluated in each patient pre- and post- operation. Patients with hyperthyreosis, sick sinus syndrome and renal disease were excluded. All medications prior to surgery were continued until the morning of surgery, except warfarin. The operations were as previously described. Left atrial appendage specimens were obtained prior to the establishment of extracorporeal circulation. Part of the tissue was fixed in 4% paraformaldehyde for immunohistochemistry and TUNEL assay. The remaining tissue was frozen in liquid nitrogen and stored at -80°C for Western blot. All procedures involving human tissue collection and analyses were approved by the Institutional Ethics Committee of Nanjing Drum Tower Hospital (Approval No. 201446) and Huaian First People’s Hospital (Approval No. 201431), and performed in accordance with the principles outlined in the Declaration of Helsinki. All participants enrolled in the present study provided written informed consent. ## Western blot Tissue samples were washed in phosphate buffered saline (PBS) and homogenized in RIPA solubilization buffer containing a 1% proteinase inhibitor cocktail (Roche) on an ice rotator. The samples were centrifuged at 12,000 rpm for 20 min at 4°C to precipitate the cell debris. The supernatant was transferred into fresh tubes and the protein concentrations were determined by BCA protein assay (Pierce). Equal protein mixtures were electrophoresed in 12% SDS-PAGE and transferred to polyvinylidene difluoride membranes. After blocking with 5% non-fat milk in Tris-buffered saline containing 0.1% Tween 20 (TBST) for 2h at room temperature, the membranes were incubated at 4°C overnight with the following primary antibodies: anti-NLRP3 (1:500, Cell Signaling technology) and anti-Caspase-1 (1:500, Cell Signaling technology). Anti-GAPDH (1:10000, Bioword) was used as the loading control. After washing with TBST, the membranes were incubated with horseradish peroxidase-conjugated goat anti-mouse (1:2000, Bioword) or anti- rabbit antibodies (1:5000, Bioword) at room temperature for 1h. The reactions were developed with enhanced chemiluminescence (Millipore), and images were obtained by film exposure. The bands were quantified using Image J software. All the quantifications of proteins were normalized against GAPDH, and the values of the other two groups normalized by the AFThrombus group. ## Immunohistochemistry All tissue specimens were fixed overnight in 4% paraformaldehyde and embedded in paraffin. Paraffin-embedded tissues were cut into 4-μm-thick sections and deparaffinized before endogenous peroxidase quenching and epitope retrieval. After blocking with 1% bovine serum albumin (BSA) in PBS for 30 min, slides were incubated with anti-HLA-DR (Abcam) and anti-CD 163 (Abcam) overnight at 4°C. After washing three times with PBS, slides were incubated with biotinylated secondary antibody for 20 min. Staining was visualized using the VectaStain ABC- AP kit. ## TUNEL assay To detect apoptosis, TUNEL assay was performed with paraffin-embedded sections using DeadEnd Fluorimetric TUNEL System (Promega), according to the manufacturer’s protocol. Cell nuclei were counterstained with DAPI (Sigma). The number of TUNEL-positive nuclei was manually counted. The total number of nuclei was determined by automatically counting the DAPI using Image J software (NIH, USA). ## Statistical analysis All continuous variables were described as mean and standard deviation. All categorical variables were described with absolute and relative frequency distributions. Comparisons of means between groups were performed by Kruskal- Wallis test. The χ² test or Fisher’s exact test was used for analysis of differences between groups. Statistical analyses were performed using SPSS version 13.0 software (SPSS Inc. Chicago, Illinois). Statistical significance was defined as *P* value\<0.05 (2-tailed). # Results ## Clinical characteristics We recruited 90 patients with mitral stenosis undergoing mitral valve replacement surgery. Age, gender, co-morbidities, NYHA classification, serum CRP levels, BNP, BUN, Cr and blood clotting time, were similar in the three groups. Although the incidence of stroke or TIA between groups was not significantly different, patients with AF showed a higher risk of embolism (5 in AF + Thrombus—group and 4 in AF + Thrombus + group, vs 0 in AF-Thrombus—group, *P* = 0.09). A slight elevation of INR was noticed in patients with AF (*P* = 0.02). All the patients with AF received warfarin routinely. After enrolling in the Department of Thoracic Surgery, warfarin was stopped and a bridge therapy with LMWH was initiated. ## Echocardiography parameters We compared echocardiography parameters between the three groups pre- and post- operation. In UCG pre-operation, the interventricular septum thickness, left ventricular end-systolic diameter, left ventricular end-diastolic diameter, postero-lateral wall thickness, and ejection fraction were similar in the three groups. Left atrium diameters in patients with AF were significantly larger than those without AF. After operation, left atrium diameters were reduced in all the three groups, compared with pre-operation values. In accordance with other studies, left atrium size is the predominant factor contributing to AF. ## M1 but not M2 macrophage markers overexpressed in MS patients with AF and thrombus formation To investigate whether specific macrophage subpopulations participate in thrombus formation in AF, we first assessed macrophage polarization in the atrium immunohistochemically. HLA-DR was used as a surface marker for M1 macrophages, while CD163 was a marker for M2 macrophages. As shown in, there was a clear predominance of M1 over M2 macrophages. HLA-DR was highly expressed in the atrium of MS patients with AF, when compared with MS patients without AF. HLA-DR was further up-regulated in specimens from AF patients with thrombus formation. It was noteworthy that several HLA-DR positive cells clustered under the atrial endothelium from AF(+)Thrombus(+) patients, which was not observed in tissue from AF(+)Thrombus and AFThrombus patients. On the other hand, the expression of CD163 was similar among the three groups. These results suggested that excessive M1 macrophage infiltration into atrium may play an important role in thrombogenesis in MS. ## NLRP3 inflammasomes over-activated in MS patients with AF and thrombus formation We found that excessive M1 macrophages infiltrated into atrium with thrombosis. IL-1β is an important cytokine from macrophages to mediate the inflammatory response. In response to stimuli, the expression of NLRP3 and IL-1β was up- regulated, followed by NLRP3 inflammasome complex with adaptor protein ASC, pro- caspase-1 and maturation of caspase-1. Consequently, mature caspase-1 cleaves pro-IL-1β into its active form IL-1β, mediating cell death. To elucidate whether IL-1β mediates thrombogenic effect of M1 macrophages, we analyzed the expression patterns of NLRP3 inflammasomes and IL-1β. NLRP3 expression was elevated in patients with AF compared with those without AF. In AF, NLRP3 expression was further increased in patients with thrombus formation. Although the levels of pro-caspase-1 between AFThrombus and AF(+)Thrombus group were similar, expression of pro-caspase-1 were significantly elevated in patient with thrombus. The levels of cleaved caspase-1, IL-1β and cleaved IL-1β in AF(+)Thrombus(+) patients were significantly higher than those with sinus rhythm, and were elevated when compared with AF(+)Thrombus patients. These results showed that in MS patients, NLRP3 inflammasome activation and elevated IL-1β were associated with thrombosis. ## Increased cardiomyocyte death associated with thrombogenesis in MS patients with AF Preceding results indicated that NLRP3 inflammasomes and IL-1β were involved in thrombogenesis in MS patients. The IL-1β is an important mediator of myocyte apoptosis, We used TUNEL assay to investigate apoptosis of additional myocytes in atrium was associated with thrombogenesis. TUNEL-positive cells were significantly increased in AF(+) patient’s atrium. When compared with AF(+)Thrombus patients, the number of TUNEL-positive cells were further increased in MS patients with AF(+)Thrombus(+) (left, middle and right panels). # Discussion In the present study, we demonstrated that proinflammatory M1 macrophages were predominant in atrium of MS patients with AF and thrombus. NLRP3 inflammasomes, which are primarily functional in macrophages, were activated in these patients. We also showed that increased cell death was associated with thrombogenesis in MS patients. These data indicate that macrophages play an important role in thrombogenesis of atrial fibrillation. The role of inflammatory cells in AF and AF-related thrombogenesis is not fully elucidated. Subsets of inflammatory cells show distinct pro- or anti- inflammatory functions. For example, M1 macrophages are pro-inflammatory and clear cell debris during early tissue damage. M2 macrophages have reparative function and secrete fibrotic factors, such as TGF-β, promote tissue healing. They are recruited after M1 macrophages during cell and tissue repair. However, little is known about the role of M1/M2 subsets in the pathogenesis and thrombogenesis of AF. A previous study reported infiltration of lymphomononuclear cells and concomitant necrosis of the adjacent myocytes in the atrium of AF patients, which were absent from the atrium of patients in sinus rhythm. Consistent with this study, we demonstrated massive infiltration of M1, but not M2, macrophages into the atrium of MS patients with thrombus. Additionally, several M1 macrophages were found under atrial endothelium. By contrast, in MS patients without thrombus, relatively fewer number of M1 macrophages were noticed inside the atrium reflecting a local inflammatory process in the atria of MS patients. Early M1 infiltration may represent the initial step of atrial inflammation and massive infiltration into atrial endothelium occurred with progressive inflammation highlighting the pathogenesis of AF-related thrombogenesis. Although a previous study observed infiltration of macrophage in atrium of AF patients, the underlying mechanisms of macrophages in thrombogenesis are still unknown. We postulated that macrophage-endothelial cell injury triggered thrombogenesis in MS patients. Enhanced endothelial expression of VCAM-1 via Toll-like receptor 4 signaling induced atrial thrombogenesis.Our TUNEL assay demonstrated that cellular apoptosis was increased in atrial endothelium of patients with thrombus. These data suggested that macrophages induced endothelial cell apoptosis, contributing to thrombogenesis in AF. IL-1β is an important proinflammatory cytokine that activates innate and adaptive immune responses. The processing of inactive pro-IL-1β depends on NLRP3 inflammasomes. Upon activation via TLRs on macrophages, the transcription factor NF-κB induces NLRP3 and IL-1β gene transcription. NLRP3 and pro-caspase-1 assemble into a complex inflammasome via apoptosis-associated speck-like protein containing a CARD (ASC). This complex activates caspase-1 itself, which further cleaves pro-IL-1β into its active form. This process represents post- translational modification of IL-1β. The serum level of IL-1β was altered in AF patients. It mediates myocyte apoptosis. However, its role in AF was not fully known\[–\]. Its role in endothelial cell apoptosis needs further exploration. Immunoblot assay in the present study showed that both pro- and cleaved IL-1β were increased in patients with AF and thrombus. NLRP3 and cleaved caspase-1 expression were significantly elevated in the atrium of MS patients with thrombus strongly indicating that for the first time, and to the best of our knowledge, that NLRP3 inflammasomes were activated and IL-1β was associated with thrombosis in MS patients. Immunoblot results demonstrated that cleaved caspase-1 and cleaved IL-1β were detected in MS patients with sinus rhythm.A predominance of M1 marcophages over M2 was seen in the atrium of all patients with rheumatic mitral stenosis. These data suggested that M1 macrophage infiltration and NLRP3 inflammasome activation mediated rheumatic mitral stenosis. In persistent MS, the M1-dependent inflammation was increased, contributing to thrombus formation. ## Limitations First, all experiments in the present study were performed in human atrial specimens. We could not use gain- or loss-of-function models to demonstrate a causal relationship between over-infiltration of M1 macrophages and NLRP3 inflammasome activation. Second, all patients enrolled were diagnosed with rheumatic mitral stenosis and were undergoing valve replacement. These patients might not represent a majority of AF patients. A further study of non-valvular AF patients is needed to test these hypotheses. Third, although endothelial impairment is a key step in thrombogenesis of AF, we did not counterstain TUNEL with CD31 to identify the endothelial cells. # Conclusions Infiltration of M1 macrophages and over-activation of NLRP3 inflammasomes may play a key role in atrial inflammation and thrombogenesis in patients with rheumatic mitral stenosis and AF. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: BX JX. Performed the experiments: GH WT. Analyzed the data: WT JC. Contributed reagents/materials/analysis tools: BW. Wrote the paper: JC GL. Assisted in western blotting and data analysis: SZ.

# Introduction Climate change studies can be categorized into three main communities. First, climate models (CMs) can be used to simulate climate variables using emission data of greenhouse gases (GHGs) and air pollutants (APs). Second, the impact, adaptation, and vulnerability (IAV) community assesses climate impacts using future climate information and socioeconomic conditions. Third, the integrated assessment model (IAM) community deals with future emissions of GHGs and APs and climate mitigation analyses. Emissions data are necessary for CM projections, and IAMs are used to generate future emissions scenarios that are integrated into CMs. Examples of such scenarios can be found in the Special Report on Emissions Scenarios (SRES) Nakicenovic and Swart and Representative Concentration Pathways (RCPs). These scenarios are ultimately used in climate model comparison projects such as the Coupled Model Intercomparison Project phase 3 and phase 5 (CMIP3, CMIP5: <http://cmip-pcmdi.llnl.gov/cmip5/>). In CM calculations for future scenario analyses, emissions data from IAMs must be disaggregated into a finer scale, or downscaled, for use in CMs. IAMs usually classify the world into 10–30 aggregated regions and simulate energy, land use, and relevant emissions. Regional aggregation provides sufficient detail to capture the characteristics of different parts of the world while avoiding unnecessary complexity. Well-known examples include the Asia Pacific Integrated Model (AIM), Global Change Assessment Model (GCAM), Integrated Model to Assess the Greenhouse Effect (IMAGE), Model for Energy Supply Strategy Alternatives and their General Environmental Impact (MESSAGE), and Regionalized Model of Investments and Technological Development (ReMIND). However, other research communities, such as the CM and IAV communities, use grid-based spatial resolutions for emissions. Downscaling methods have been comprehensively reviewed by van Vuuren *et al*., who classified downscaling methods into four categories: algorithmic downscaling, methods of intermediate complexity, fully elaborated models at relatively low levels of aggregation, and fully coupled models at national or grid scales. Downscaling is applicable to various indicators, but few studies have concerned global emissions. The global emissions studies of Hoehne and Ullrich and van Vuuren *et al*. used the algorithmic method. This method can be further classified into proportional and convergence downscaling methods, which have been applied by Hoehne and Ullrich and van Vuuren *et al*., respectively. The former consists of simple proportional downscaling. Meanwhile, the latter consists of allocating population first, and then allocating gross domestic product (GDP) and emissions. Then, the GDP per capita and emissions intensity (i.e., emissions per GDP) for each country are assumed to converge to values that correspond to regional aggregates. Finally, country-level indicators are downscaled to each grid. Recently, the interaction between IAMs and CMs has received great attention, and several studies have attempted to couple IAMs and CMs. van Vuuren *et al*. summarized and classified possible interactions between these types of models. They reviewed relevant research activities and suggested approaches for each research area. The need for such activities will become more urgent after the Shared Socioeconomic Pathways (SSPs) are developed as the next generation of SRES socioeconomic scenarios, because this will allow the climate study communities to interact more dynamically, as discussed by Ebi *et al*. and van Vuuren *et al*.. This research trend should not be limited to the IAM and CM communities, but should also involve the IAV community, as well as air pollution and health studies. Downscaling is a key facet of this broad-scale research activity. However, one question that arises is how downscaling methods affect projected climate variables, such as temperature and precipitation. Answering this question would provide a better understanding of the relationship between IAMs and CMs, and would be valuable to the IAV community. Motivated by this background, we had three objectives: document the downscaling method adopted in the AIM modeling framework, analyze the characteristics of the downscaled data for future scenarios, and input the downscaled data into the Model for Interdisciplinary Research on Climate (MIROC) to project climate variables with a focus on statistical significance. In the downscaling process, we assumed two extreme cases that sufficiently differentiated the downscaling patterns, thereby enabling the investigation into the effects of the downscaling method on climate projections. We used a downscaling method similar to that described by van Vuuren *et al*. due to its simplicity, clear description, and suitability for long-term scenario assessments. # Materials and Methods ## General Description The overall framework of this study (i.e., the model input and output) consisted of three steps. First, AIM/computable general equilibrium (CGE) was used to project emissions of GHGs, and APs were aggregated into 17 regions. Second, AIM/downscaling (DS) was used to downscale the emissions into a 0.5° spatial resolution. Third, MIROC was used to calculate the climate variables using the gridded emissions. We analyzed the results obtained from the AIM/DS and MIROC using two different factors (convergence and inertia) to downscale the emissions. The use of these two factors is explained in more detail in the discussion of the AIM/DS algorithm. We used sulfur emissions to directly compare the different implications of these two methods, as sulfur has a relatively large effect on climate compared to other APs. We differentiated the spatial allocation of sulfur emissions, and used the same information on spatial emissions for the other APs. Because AIM/CGE and MIROC have already been developed and had many other applications in which documentation is available on the model characteristics. Therefore, we focused predominantly on the AIM/DS methodology and its application. We analyzed the results from four perspectives. First, we examined the downscaled gridded emissions maps and compared the results for 2005 and 2100 obtained with the two methods. Second, we aggregated the downscaled emission information into national emissions and intensities, and examined the differences between the two methods to explain how the emissions intensity convergence assumption affected the emissions density. Third, we collected grid- based sulfur emission data and assessed the variability and distribution of sulfur emissions. These analyses were performed for each AIM/CGE aggregated region. We calculated the standard deviation (SD) and interquartile range (IQR; i.e., the difference between the upper and lower quartiles) for the gridded information as metrics of the variability of the predictions. Fourth, we applied *t*-tests and field significance tests to the climate model outputs to identify significant differences in the temperature and precipitation predictions. ## AIM/CGE We used an AIM/CGE model for this study, which is a recursive-type, dynamic, general equilibrium model that covers all regions of the world. Details of the model structure and mathematical formulas are provided by Fujimori *et al*.. The main inputs were population, GDP, food preferences, and assumptions of energy technology and air pollution controls. The model provided energy consumption, agricultural and land use indicators, and emissions of GHGs and APs. AIM/CGE considered carbon dioxide, methane, nitrous oxide, and fluorine gas to be GHGs, while black carbon, carbon monoxide, ammonia, non-methane volatile organic compounds, mono-nitrogen oxides (NO_x), organic carbon, and sulfur oxides (SO_x) were treated as APs. The regional, geographical, and industrial classifications are shown in, Fig A, and Table A in, respectively. With respect to assumptions of the future, population and GDP were based on the Shared Socio-Economic Pathway 2 (SSP2). Among the SSPs, SSP2 is characterized as a middle-of-the-road scenario. In addition, assumptions related to energy technology, food, and land were based on the SSP2. However, the original SSP2 sulfur emissions in the latter half of the 21^st century are low, owing to the assumption of continuous implementation of air pollution legislation, and results in minimal sulfur effects on climate. Because we aimed to explore the effects of downscaling methods on sulfur emissions and obtain meaningful insights on the effects of downscaling on climate, we changed the sulfur emissions scenario from the original SSP2. Instead of large emissions throughout the 21^st century, we simply and arbitrarily assumed an annual 1% decrease in emissions coefficients. The projected sulfur emissions by region are shown in. ## AIM/DS ### The AIM/DS algorithm AIM/DS is a tool used to downscale regionally aggregated emissions onto a grid- based map with a resolution of 0.5°. The algorithm is differentiated by sectors, and the sources of emissions are assigned to one of three groups. In group 1, GDP or population are the drivers of the emissions. This algorithm is adopted mainly for energy-related emissions, which we assumed to have a relationship with GDP or population. Group 2 is downscaled in proportion to the total regional emissions. The base year map information is scaled up or down according to the total regional emissions. The base year represents the AIM/CGE simulation starting year (i.e., 2005). Therefore, the spatial distribution pattern for future scenarios is the same as that of the base year. Group 3 is downscaled in proportion to total global emissions with the base year spatial map. The basic logic in the case of group 3 is the same as that of group 2, but the logic is applied to cross-border sectors. It should be noted that land use-related emissions were classified as group 2 in this study, and the emissions of this sector would ideally consider changes in land use. However, because energy- related sectors are the main sulfur emitters, we used proportional downscaling. In future studies, we will likely follow an approach that includes information on changes in land use. Here, we explain the methodology of group 1 (Eqs –), which is related to energy, solvent, and waste sector emissions, because the methods for group 2 and 3 are simple. We considered two factors, convergence and inertia. The basic logic was based on van Vuuren *et al*., who assumed that the intensity of emissions converged among countries that belonged to the same aggregated region. Inertia can be interpreted as the level of technology, level of legislation, or sector structures that exist in the base year. Eq represents these two factors. $${EMG}_{i,t} = {\sum_{r \in RI}{\alpha_{t} \bullet {EI}_{r,t} \bullet {DV}_{i,t}}} + \left( {1 - \alpha_{t}} \right) \bullet {EMG}_{i,t - 1}$$ where *EMG*_*i*,*t* is emissions in grid *i* and year *t*; *EI*_*r*,*t* is the emissions intensity in country *r* and year *t*; and *DV*_*r*,*t* is the driver in country *r* and year *t* (e.g., population and GDP). The factors α and (1- α) are the weighting coefficients between the convergence and inertia factors; α ranges from 0 to 1. If the factor is thought to be stronger than the inertia factor, the weight should be close to 1. A strong convergence factor indicates that the emissions intensity (e.g., emissions per GDP) of each grid cell converges to a certain value belonging to a region (e.g., the emissions intensity of a cell that belongs to Indonesia converges to the average of Southeast Asia). Conversely, a strong inertia factor indicates that the heterogeneity in the base year’s emissions intensity strongly remains (e.g., the emissions intensity of a cell that belongs to Indonesia remains the same as the base year intensity difference from the average of Southeast Asia). α is used to differentiate between the two methods. *RI* is a mapping from a set of country *r* to grid *i*. *DV*_*r*,*t* is an exogenous parameter, and *EMG*_*i*,*t-1* is the emissions in the previous year. The parameter *EI*_*r*,*t* is the product of two terms, as shown in Eq. The former is the country-specific emissions intensity associated with the average emissions intensity change ratio in the aggregated region, and the latter is the emissions intensity in the aggregated region. $${EI}_{r,t} = \left( {{EI}_{r,t0} \bullet \frac{\sum_{({r,rag}) \in RM}{{EIAG}_{rag,t}}^{}}{\sum_{({r,rag}) \in RM}{{EIAG}_{rag,t0}}^{}}} \right)^{\beta_{t}} \bullet {\sum_{({r,rag}) \in RM}{{EIAG}_{rag,t}}^{({1 - \beta_{t}})}}$$ where *EIAG*_*rag*,*t* is the average emissions intensity of the aggregated region *rag* (AIM/CGE; 17 regions) in year *t*, β is a parameter that represents the convergence of each region, *RM* is a mapping for the AIM/CGE-aggregated region to which the country belongs, and *t*0 is the base year (2005). If β = 0, all countries that belong to an aggregated region converge to the regional average intensity. We assumed that β was 1 in 2100 and 0 in the base year. The intermediate periods were connected linearly. Aggregating the emissions from each country obtained by the direct solution of Eq into the original 17 AIM/CGE regions yielded inconsistent numbers for the aggregated regional totals. Therefore, the *EMG*_*i*,*t* was updated by scaling using Eq. $${EMG}_{i,t}^{*} = {\sum_{r \in RI}{EM}_{r,t}} \bullet \frac{{EMG}_{i,t}}{\sum_{r \in RI}{EMG}_{ii,t}}$$ where *EMG*^*\**_*i*,*t* is the updated emissions in grid *i* and year *t*, and *EM*_*r*,*t* is the emissions in aggregated region *r* and year *t*. ### Tests of the two methods for determining the parameter α As shown in Eq, the weight coefficient α changes the pattern of emissions allocation. We examined two methods of determining α, M1-CONV and M2-INER. The basic strategy differentiating these two methods followed two extreme cases, as stated in the introduction. We used van Vuuren *et al*. as the starting point, as no other reports on downscaling methodologies for global-scale emissions were available. van Vuuren *et al*. assumed full convergence, which was an extreme case. The opposite extreme of this case would be strong inertia. If we could have found or created other extreme methods, we would have adopted them. However, to the best of our knowledge, these are the best available cases. Method M1-CONV assumed that the intensity of emissions eventually converged to the aggregated regional average. α was initially 0.0 in 2005 and increased linearly to 1.0 in 2100. In this study, the GDP and population of each country were provided exogenously, and the behavior of GDP per capita differed from that assumed by van Vuuren *et al*., who assumed that GDP per capita converged. The M2-INER method assumed that a mixture of the convergence and inertia factors determined the intensity of emissions, and that inertia was more important than convergence. α was assumed to be 0.1 in 2100. We selected 0.1 arbitrarily as a sufficiently small value to observe the impact of the assumption of α. Ideally, it would be allowed to vary as a function of future socioeconomic assumptions. For example, if a scenario represented an SSP4-like world, characterized by inequality, α would be low. By contrast, if a scenario represented an SSP1-like world, characterized by equality and sustainability, α would be high. ## Population and GDP spatial allocation We used national-level data for populations, urbanization rates, and GDP to generate the spatially gridded populations and GDPs. The base was the 2.5-arc- minute data from the Gridded Population of the World. We used 0.5-arc-minute population data to produce population distributions within the 2.5-arc-minute grid cells as the initial values. Initial population data and national urban population data were used in urban areas. We set thresholds for population density based on the initial populations in the 0.5-arc-minute data so as to match national urban populations. We treated the 0.5-arc-minute grid cells above the threshold as urban cells and those below the threshold as rural cells. For the 30-arc-minute grid cells, we used urban population/area ratios as the urban index. The Greenhouse Gas Initiative database provided by the International Institute for Applied Systems Analysis was the source of these data. We used the rank-size rule to estimate the populations of urban grid cells. It is an empirical law used to estimate previous city populations, but is also applicable to future populations. Then the GDP distributions were basically allocated based on the populations. Several geographical constraints were considered, including mountains, water bodies, and urban sprawl. Supporting Information 2 provides additional details on the methodology. ## Configurations of the atmosphere–ocean GCM experiments We used the coupled atmosphere–ocean global climate model MIROC5.2 in this study. MIROC5.2 is a minor upgrade of MIROC5.0, and is modified as follows. The horizontal resolution of the atmosphere component uses a T42 spectral truncation (\~2.56°). There are 40 vertical levels, and the top of the atmosphere is located at the 3-hPa pressure level, as in MIROC5.0. The simulation of solar radiation is improved in MIROC5.2 to deal with the infrared wavelength band 4–100 μm. In addition, the Minimal Advanced Treatments of Surface Interaction and Run-off (MATSIRO) land surface model, a parameterization for sub-grid scale snow cover distribution (SSNOWD) and representation of wetlands, which store some snow meltwater, is a new addition. MIROC5.2 uses COCO ver. 4.9 as the global ocean component, which includes several upgrades compared to MIROC5.0. A tri-polar grid is introduced in MIROC5.2 in place of the bi-polar generalized curvilinear coordinate used in MIROC5.0. The coordinate system is composed of a spherical coordinate portion south of 63.3°N and a bi-polar coordinate system in the Arctic region north of 63.3°N. The longitudinal grid spacing is 1° in the spherical portion. The latitudinal grid spacing is 1°cos*θ*, where *θ* is the latitude, while it is about 0.5° in the equatorial region and the Southern Ocean. The two coordinate singularities in the bipolar portion are placed at 63.3°N, 60°E in Greenland and 63.3°N, 120°W in Siberia. The zonal and meridional grid spacings are about 60 km and 33 km around the central Arctic Ocean, respectively. The number of vertical levels has been increased from 49 in MIROC5.0 to 62 in MIROC5.2. The profile of background vertical diffusivity has been changed in MIROC5.2. The empirical profile of Tsujino *et al*., used in MIROC5.0, is used in MIROC5.2 below a depth of 50 m. However, above 50 m, the vertical diffusivity is set to 1.0×10⁻⁶ m² s^-1 in the uppermost 29 m and gradually increases to 1.0×10⁻⁵ m² s^-1 at 50 m. The turbulent mixing process in the surface mixed layer is also updated in MIROC5.2. Areas where the surface is covered with sea ice are assumed to experience no surface wave breaking, resulting in near-surface mixing. A combination of the smaller background vertical diffusivity above 50 m and the suppression of turbulent mixing under sea ice cover better represents surface stratification in the Arctic Ocean. An online aerosol module of MIROC5.2, the Spectral Radiation-Transport Model for Aerosol Species (SPRINTARS), computes mass mixing ratios of the main tropospheric aerosols from emissions of aerosols and the aerosol precursors sulfate, carbonaceous aerosols (i.e., black carbon and organic matter), soil dust, and sea salt aerosols. The aerosol transport processes include emission, advection, diffusion, chemistry, wet deposition, dry deposition, and gravitational settling. SPRINTARS is coupled with radiation and cloud microphysics schemes to calculate the direct, semi-direct, and first and second indirect effects of aerosols. Because we compared the two experiments with different emissions of sulfur dioxide (SO₂), a precursor of sulfate aerosols, only sulfate aerosols are described. Sulfate aerosols are formed mainly by chemical reactions of SO₂ and dimethyl sulfide with oxides (hydroxide, ozone, and hydrogen peroxide), with monthly mean fields prescribed by the global chemical model, CHASER. SPRINTARS explicitly treats the chemical reactions related to sulfate aerosols and SO₂ dissolution into water. In addition, the SO₂ dissolution process is applied to in-cloud scavenging of SO₂. In the radiative calculation, the mode radii of the lognormal size distribution dependent on the relative humidity are assumed. The hygroscopic growth of sulfate aerosol particles is applied according to Tang and Munkelwitz. The assumed volume-weighted refractive indices are used for the internal mixtures between aerosol particles and water. External mixing is assumed for sulfate, soil dust, and sea salt. More detailed processes on sulfate aerosols and other aerosols can be found in Takemura *et al*.. We performed two experiments using MIROC5.2. Each experiment was driven by one of two SO₂ emission datasets provided by the AIM/CGE. The emissions data for SO₂ in the year 2100 were used in the experiments M1-CONV and M2-INER. For the other emissions data (e.g., black carbon and organic carbon) and forcing data (e.g., solar constant and volcanic aerosols), we applied pre-industrial conditions from 1850 in both experiments in which the anthropogenic emissions were zero. Both experiments were integrated for 40 years, and the last 30 years of data were analyzed to evaluate climatological differences between the M1-CONV and M2-INER methods that arose from difference in the SO₂ emissions data. It should be noted that the distributions and radiative effects of sulfate aerosol in these experiments might have differed from that of future projections with varying GHGs and aerosol emissions, because background concentrations of other pollutants and changes in climate could have affected sulfate aerosols. To test the statistical significance of the difference between these two experiments, we used a 250-year preindustrial control simulation of MIROC5.2 (piControl). # Results ## Overview of sulfur emissions Fig B, C, D and F in show the global and regional populations, GDPs, and primary energy supplies, the main drivers of sulfur emissions. shows the global sulfur emissions computed by AIM/CGE by region and sector. Total sulfur emissions gradually increased from 112 Mt/year in 2005, peaked in roughly 2050 (158 Mt/year), and then declined to 140 Mt/year in 2100. Because emissions were computed with driving forces (mainly energy consumption) and emissions coefficients, the decline only occurred in the latter period. These emissions were obviously higher than any of the RCPs shown by van Vuuren *et al*. and slightly higher than the maximum level in the Intergovernmental Panel on Climate Change Fifth Assessment Report (IPCC AR5) scenarios (Riahi *et al*.), implying that the scenario used in this study was near the maximum possible emissions. The main reason for these high emissions was the conservative air pollutant legislation assumption adopted in this study. A regional breakdown showed that China was the largest emitter in 2005, but its emissions became much lower by 2100. Meanwhile, India increased its total amount and proportional share of sulfur emissions. In addition, sulfur emissions increased in Africa. Both India and Africa were projected to experience steady increases in income and energy consumption over time (Figs B and D). A breakdown by sectors indicated that energy-related emissions, such as emissions from the industrial and energy sectors, accounted for most of the total sulfur emissions, and this characteristic remained throughout the 21^st century. Emissions from the energy sector decreased over time, industrial emissions remained more or less constant, and transportation-related emissions increased. As a result, the share of emissions from the energy sector declined from 54% in 2005 to 43% in 2100, whereas the industrial and transportation sectors contributions to sulfur emissions increased from 31% to 38% and from 3% to 8%, respectively, in the same timeframe. ## Spatial distribution of sulfur emissions shows the spatial distribution of sulfur emissions in 2005 and 2100 projected by M1-CONV (the 2050 maps are presented in Fig F and G). In 2005, there were high emissions from China, the east coast of the United States, and Europe. The distribution of emissions differed in 2100. Emissions from southern Asia, particularly the Himalayan region (i.e., Bhutan and Nepal), were remarkably high in 2100. Second, high emissions were observed in the west coast of Africa (e.g., Cote d’Ivoire) and eastern central Africa (e.g., Kenya). These trends reflected the projected increases in income and energy consumption in Africa and southern Asia. Third, emissions from China were projected to decline from 2005 to 2100, particularly along the east coast of China. illustrates the difference between M1-CONV and M2-INER computed by AIM/DS. The red shading indicates grid cells in which the M2-INER projections are smaller emissions than those of M1-CONV, while the blue shading indicates the opposite. Red-shaded regions include Nepal, western Russia, eastern United States, sub- Saharan Africa, North Africa, and the Middle East. Blue-shaded regions include southeastern Africa, Eastern Europe, the Korean Peninsula, the center of North America, and South Africa. These differences reflected the strength of the convergence effect. However, it was not straightforward to determine why the strength of the convergence effect differed. In principle, the reason was that the density of emissions (emissions per area) was initially higher in some regions than in other regions, resulting in remarkably large differences, although the ratio of the two methods was not particularly large for those regions. For example, in the Rest of Asia region (i.e., Bhutan and Nepal), M1-CONV projected higher emissions than M2-INER. illustrates the emissions intensities for the Rest of Asia region by country. The M1-CONV method did not converge because of the inclusion of non-energy sectors, which were not constrained to converge. The emissions converged more strongly in the latter half of the 21^st century for M1-CONV than for M2-INER. The intensities of emissions from Bhutan and Nepal were relatively low in 2005. In the M1-CONV method, the proportion of emissions from these countries increased, converging toward the average of this region, which was higher than the emissions from Bhutan and Nepal. In the M2-INER method, the emissions from Bhutan and Nepal remained unchanged from the base year. This revealed that these two countries did not always have a high intensity of emissions in the Rest of Asia region, and the level of the intensity did not entirely explain the spatial distribution in the maps. Instead, emissions per unit area were the primary driver. Another example was Egypt, which had a high emissions density in a very limited area, the northeastern part of the country, resulting in the dark red shading of Egypt in. As was the case for Bhutan and Nepal, Egypt did not always have the highest emissions intensity in North Africa, the region that included Egypt (Fig I). The other factor of the different geographical allocations was the difference in GDP growth across countries within aggregated regions, which was observed in the sub-Saharan region. In the GDP scenario assumption, Nigeria had much higher growth than South Africa and higher emissions were formed in Nigeria (Fig J). illustrates the difference between the two methods in the T42 spectral truncation (\~2.56°) resolution used in MIROC. Geographical features of the differences in emissions between M1-CONV and M2-INER in the original horizontal resolution were observed in the emissions differences in the MIROC5.2 resolution. Therefore, the geographical features of the emissions differences in the original horizontal resolution were appropriately implemented in the emissions data used in the MIROC5.2 experiments. In addition, by comparing 20-year averages of the mass column loading of sulfate aerosol in M1-CONV and M2-INER, significant differences in the mass column loading of sulfate aerosol were widely observed (Fig H). ## Frequency distribution of emissions density and intensity The frequency of distribution of emissions density and intensity can be investigated to better understand how spatial downscaling works. illustrates the frequency distribution of grids as a function of the emissions density (per area) and emissions intensity (per GDP) generated by the M1-CONV and M2-INER methods for the base year and 2100. It should be noted that the scale for the abscissa is logarithmic, and the ranges differ between regions. As the emissions intensity converged, the spatial allocation of emissions approached that of the GDP, and there was a large change in the emissions density distribution. The distributions for 2100 and 2005 simulated with M2-INER were similar. However, the M1-CONV distributions did not fully converge. There are two possible explanations for this behavior. First, the GDP density did not change substantially, and the emissions density could eventually become similar to the GDP density. Second, the scale of the abscissa is logarithmic, and the changes are difficult to discern. Meanwhile, similar figures made using emissions intensity based on emissions per GDP showed no apparent differences. The emissions intensities based on M1-CONV converged strongly, whereas those based on M2-INER did not. Next, we analyzed the distribution changes based on statistical indicators using the SD and IQR of both emissions density and intensity in the energy sector, which was the sector that accounted for the greatest share of sulfur emissions. As already noted, the emissions density differed little between 2005 and 2100, and the difference between the emissions density projected by the M1-CONV and M2-INER methods was not very obvious. By contrast, the statistics relevant to emissions intensity differed drastically between M1-CONV and M2-INER. In the M1-CONV projection, the emissions intensity converged to 0 in 2100, which was not observed in the M2-INER projection, although the emissions intensities calculated in M2-INER decreased for most countries. ## Climate variables The globally averaged 30-year mean temperature difference in the climate simulation between the M1-CONV and M2-INER methods (-0.0076 ± 0.053°C) was not significant. The confidence interval was estimated with a *t*-test at the 95% significance level. We assumed a degree of freedom of 18, which was estimated by considering the persistence in a first-order autoregression of global annual mean temperature time series of the piControl. The spatial distributions of the differences in the annual mean temperature and precipitation between the two methods are shown in (seasonal differences are shown in Figs K and L). The colored areas indicate differences that are statistically significant at the 95% level in *t*-tests with a degree of freedom of 58 (assuming 1 per year). Although there were local differences in the M1-CONV and M2-INER temperatures in 2100, the differences were small in most cases, and temperature differences significantly differed from 0 only in 9.0% of the total area. However, these significant differences may not represent clear evidence of sensitivity to the emission data, because 5% of the grid values are expected to be significant at the 95% significant levels by chance if the grids are independent of one another. Furthermore, in climate systems, considerable spatial correlations of the internal natural variability can lead to more or less than 5% of the total area with significant differences. To evaluate the possibility that these proportions of significant cases resulted from the internal natural variability alone, we applied a field significance test using the outputs of the piControl run. We computed the differences between two non-overlapped 30-year period averages of the 250-year piControl, and estimated the area fractions of grids with significant differences. We repeated these processes for all 28 combinations of two non-overlapped 30-year periods, and found that 96% (27/28) of all combinations had area fractions of significant difference larger than 9.0% when M1-CONV was subtracted from M2-INER. Therefore, there was no clear evidence that the two methods caused apparent differences in the projections of temperature change. Similarly, the differences in global mean precipitation were not significant (0.0022 ± 0.0054 mm day^-1; degrees of freedom: 44). The differences were statistically significant in only 4.8% of the areas. In the field significance test, 93% of all of the combinations of two 30-year piControl segments (26/28) had areas with significant differences larger than 4.8%. Therefore, the 4.8% of the global area that exhibited significant differences occurred by chance due to internal natural variability. # Discussion and Conclusions We described two downscaling methods of the AIM modeling framework and compared the two methods using the same aggregated sulfur emissions. One method is based on the convergence of emissions intensity, and the other is based on inertia. We performed climate model simulations using MIROC with spatial distributions of sulfur emissions based on the two methods, and then investigated the climate responses. The spatial allocation of emissions differed between the two methods. This difference was obvious in the case of emissions intensity, but was much less apparent in the case of emissions density. The projected temperature and precipitation in 9.0% and 4.8% of the total global area, respectively, differed significantly between the two methods. However, the field significance tests showed that these could have been produced by natural internal variability alone. Therefore, there was no clear evidence that the differences between the two downscaling methods led to additional significant uncertainties in climate projections. The conclusion derived from this analysis is that the differences in the spatial allocations investigated in this study did not have a significant systematic impact on global or regional climate. This study had several limitations. First, we only tested sulfur emissions; considering other gases could change the climate projections. Sulfur is a major gas that affects the pattern of cloud generation and radiative forcing, but black carbon might also be worth considering. We tested the downscaling method and determined the characteristics of the methods. Therefore, we limited the study to sulfur. However, it may be valuable to include other gases in future studies. Second, although the spatial allocation of GDP and population affect the geographical allocation of emissions, we used simple methods to allocate emissions. Using different spatial drivers could result in different emissions results, and the implications in terms of climate variability would then differ from those in this study. The M1-CONV method, for which the emissions intensity strongly converged, would be particularly sensitive to such changes. Third, this study focused entirely on climate responses, but it would have been beneficial to use the spatial data on emissions to study air pollution or its associated health effects. The conclusions may have been affected by the downscaling method, and more elaborate methodologies should be considered in future studies. Finally, we used a single climate model, MIROC, and conclusions of such analyses may depend on the climate model used. Thus, our results should be interpreted accordingly. # Supporting Information This study was supported by Global Environmental Research Fund grants 2–1402, S10-4, and S14-5 from the Ministry of Environment of Japan and the Program for Risk Information on Climate Change (SOUSEI program) from the Ministry of Education, Culture, Sports, Science and Technology (MEXT) of Japan. The authors are most grateful for the generous support provided by these funds. [^1]: Kazuhide Kushida, Kazutaka Oka and Minoru Yoshikawa are employed by Mizuho Information & Research Institute, Inc. There are no patents, products in development or marketed products to declare. This does not alter our adherence to all the PLOS ONE policies on sharing data and materials, as detailed online in the guide for authors. [^2]: **Conceptualization:** SF MA HS. **Formal analysis:** MA HS. **Funding acquisition:** TM KT. **Investigation:** SF MA HS. **Methodology:** SF MA HS TK HK. **Project administration:** SF. **Software:** KK KO MY SF MA HS HT. **Visualization:** MA SF. **Writing – original draft:** SF. **Writing – review & editing:** SF MA HS TK TH KT.

# Introduction Remarkable regularities are observed with respect to relations among the speed, mass, and energy expenditures of biological organisms. Most of these relations are allometric, i.e., they describe how a function or attribute of organisms changes with their scale (e.g., with mass). Although the accuracy of the relations tends to be limited to an order of magnitude, these relations hold over very large ranges of values and over multiple orders of magnitude. For example, one of the earliest such observations was Kleiber’s Law, which states that for the vast majority of animals–from tiny mouse to huge elephant–the organism’s metabolic rate is approximately proportional to the organism’s mass to the 3/4 power; the data for all biological organisms fall on the same curve. In a similar allometric fashion, it has been observed that for all biological organisms, the maximum speed *V* of animals increases proportionately to their mass *M* to the power of 1/6, i.e., *V \~ M*^*1/6*, even when the mass of organisms varies by many orders of magnitude. Another allometric law refers to the metabolic cost of transport (CoTm), which is the metabolic energy consumption that an organism requires to move a unit of mass of its body over a unit of distance. Equivalently, the amount of metabolic energy per unit time *Pm* required for the organism to move with speed *V* per unit of its mass M is CoTm = *Pm/(M·V)*. For all biological organisms, CoTm diminishes approximately with mass to the power of -1/3, i.e., *Pm/(M·V) \~ M*^*(-1/3)*. Metabolic energy is the energy that an organism obtains through its food, though typically measured as the organism’s consumption of oxygen. Unlike metabolic energy, the mechanical energy of an organism *P* is produced by the organism’s muscles in order to deliver forces by which it propels itself relative to the terrain and surrounding media, and moves its limbs with respect to the rest of its body. It has been observed that unlike the metabolic CoTm, the mechanical CoT = *P/(M·V)* remains approximately constant, or at least within the same order of magnitude, over a very wide range of *M*. Elaborating on the latter observation, Heglund proposed an empirical formula for estimating the mechanical power *P* expended by an organism of mass *M* in order to move itself at speed V: $$P = M \cdot \left( {0.478 \cdot V^{1.53} + 0.685 \cdot V + 0.072} \right)$$ Do similar regularities apply to artificial, human-made systems? If they do, how do they differ from those that apply to biological systems?. We are aware of one observation reported in this regard: the speed of ships increases with mass to the power of 1/6 along the same curve as fishes, and planes similarly follow the curve for birds. A potentially related observation was made by Marden and Allen who found strong similarities in mass-force relations of animals and human-made motors. Isalgue’s observation is intriguing and raises further questions to ask: Does such a relation apply to ground-mobile systems? Does the relation include energy expenditures? What is the extent of differences between biological and artificial relations of power, speed, and mass? Our paper is the first to explore these questions and provide initial answers. Our motivation in asking these questions combines foundational and applied interests. First, we seek fundamental insights into underlying mechanisms and limits, as well as opportunities to optimize the energy envelope of future artificial systems. Second, with growing interest in bio-inspired approaches to robotics, we seek to learn about the tradeoffs in speed, mass, and power of biological systems, including their behavioral and terrain navigation tradeoffs, as these tradeoffs may apply to robotic system design. For example, the CoT for even successful legged robots like BigDog and ASIMO is much higher than for animals. Advanced and challenging design approaches are required to achieve a CoT comparable to that of animals, e.g., the case of MIT’s Cheetah robot. If we were to know whether and to what extent the power-speed-mass relations of human-made systems exhibit a consistent relation to those of biological organisms, the designers of robotic systems would have additional guidance for specifying achievable performance targets. We illustrate this applied aspect later in the paper. This study makes the following contributions. First, we find that artificial ground-mobile systems of multiple types and over a great range of scale comply with the same power-mass-speed relation–the Heglund formula–proposed nearly 40 years ago for biological organisms. To our knowledge, this has never before been reported. Second, we show that the Heglund formula does not capture well the power-mass-speed relation for artificial ground-mobile systems at the extreme end of the mass scale, starting at approximately 35,000 kg and beyond. Third, we develop a modified formula in the spirit of Heglund’s pioneering proposal but that captures the behavior of both biological and artificial systems over a greater range of mass–more than 11 orders of magnitude. The paper is organized as follows. In the next section, we review the data we used in this study. The Results section describes our observations regarding the quantitative regularities we found in the data. After that, the Discussion section explores the meaning, implications, and limitations of our findings. The next section, An Application, offers an example of how our findings could apply to the development of future robotic vehicles. The paper ends with conclusions and recommendations for future work. ## Data In this paper, we focus on expenditures of mechanical energy and on corresponding data. Mechanical energy expenditures should not be confused with those of metabolic energy which considers consumption of chemical energy of the fuel (in vehicles) or food (in animals). For example, metabolic energy expenditure by an animal is typically derived from the rate of oxygen consumption, recorded during exercise, applying an energetic equivalent of 20.1 J to 1 ml of O₂ consumed, e.g., refs. Metabolic energy is outside the scope of this paper. In the case of vehicles, we take the mechanical power output of a vehicle’s engine *P*_*e* as approximately equal (for the purposes of our allometric study) to the mechanical power expended for the vehicle’s locomotion. In doing so we neglect the energy that a vehicle may expend while stationary, e.g., for operating sensors or air conditioning, as well as the expenditures of mechanical energy for maintaining the operation of the engine itself, e.g., tanks’ cooling fans that may take 10–15% of the engine output \[, p.258\]. In steady state locomotion, at constant altitude, the entire mechanical power output of the engine *P*_*e* is transformed eventually into heat and transferred to the environment of the vehicle (air and ground). This involves a complex chain of energy transformations ultimately ending in dissipative processes such as friction of sliding elements, gears, joints, and hysteresis in the rubber of tyres \[, p.228; \]. Similarly, in the case of animals, we take the net mechanical power output of an animal’s muscles *P*_*a* as approximately equal the mechanical power expended for the animal’s locomotion. In doing so we neglect the mechanical energy that an animal may expend while stationary. In steady state locomotion, at constant altitude, the entire mechanical power output of the muscles *P*_*a* is transformed eventually into heat and transferred to the environment of the animal (air and ground). This involves a complex chain of energy transformations ultimately ending in dissipative, heat-generating processes such as inelastic deformations of tendons and muscles. However, the conceptual similarity of *P*_*e* and *P*_*a* breaks down when we turn to the ways in which the mechanical power output is experimentally measured, and the corresponding availability of experimental data. In vehicles, the mechanical power output of an engine *P*_*e* can be measured directly in several relatively simple ways, typically involving measuring the torque and angular velocity at the end of the engine’s shaft, and for vehicles the data are usually available at least for the maximum “rated” power output approximately corresponding to the vehicle travelling at maximum rated speed and maximum load weight. In animals, on the other hand, measuring the rate at which muscles produce and output mechanical energy *P*_*a* is extremely difficult and the corresponding data are essentially unavailable. Instead of *P*_*a*, experimental biologists measure so called external work of an animal, the rate of which we designate here as *P*_*ext*. This quantity in locomotion of an animal is derived using a number of approaches. Most of them assess the changes in the body’s potential and kinetic energy. The movement of the body’s center of mass can be determined either using a recording of the movement of a marker placed near the center of mass, or most commonly from the forces that the body exerts against the ground (or force plates), i.e., the ground reaction forces, often using the procedure popularized in. Examples include refs. For the purposes of our research, the challenge is that *P*_*ext* is not directly comparable to *P*_*a*. One major difference between *P*_*ext* and *P*_*a* is that *P*_*ext* includes the elastic energy stored in muscles and tendons of an animal and as such can be significantly greater than *P*_*a*. We explore this matter in detail in the Discussion section, where we show that although *P*_*ext* and *P*_*a* are fundamentally different quantities, *P*_*ext* can serve as an order-of-magnitude approximation of *P*_*a*. In the next section of the paper, we use *P*_*ext* as a surrogate of *P*_*a*. We show that *P*_*ext* exhibits the same regularity, i.e., complies with the same formula, as *P*_*e*. Then, in the Discussion section we return to deriving the relation between *P*_*ext* and *P*_*a*, and show that *P*_*ext* approximates *P*_*a* well within an order of magnitude, appropriately for our allometric study. For biological organisms, we obtained the data on speed, mass, and power *P*_*ext* via a rigorous review of experimental literature, which we believe to be exhaustive, to the best of our knowledge. These included cockroaches, spiders, crabs, humans, quails, chipmunks, dogs, kangaroos, horses, kangaroo rats, ground squirrels, spring hares, wild turkeys, penguins, stump- tailed monkeys, lemurs, greater rheas, Asian elephants, sheep, frogs, and lizards. For artificial ground-mobile systems, our data on speed, mass, and power *P*_*e* represent diverse classes of defense-related systems (artillery systems, tanks, etc.) along with civilian transportation, construction, mining, and outdoor recreational vehicles. # Results Being interested in the relation among speed, power, and mass of systems, we focus on how artificial and biological systems compare in terms of the previously described the Heglund formula. Remarkably, the data suggest that the Heglund formula, originally developed for animals of up to 70 kg of mass, applies rather well to artificial systems, e.g., vehicles at least 2–3 orders of magnitude heavier. compares the actual system/organism power versus power predicted by the Heglund formula. (Here the term “actual” refers to experimentally measured or estimated values in the case of biological organisms, and experimental or design specification values in the case of artificial systems.) Visual inspection suggests that both biological and artificial systems generally follow the Heglund formula over a very wide range of masses and powers of systems. There is, however, a visually notable deviation of actual power from the predicted power, approximately above 300–400 hp, where the data points refer mainly to tanks and heavy trucks. Excluding, for the time being, all data points above the 35,000 kg limit, the R2 between predicted and actual power (on the log10 scale) for the entire set comprising both biological and artificial systems is 0.988. The R2 for biological organisms is 0.989, and R2 for artificial systems 0.945. To further assess whether the Heglund formula applies equally well to biological organisms and artificial systems below 35,000 kg, we performed an analysis of covariance (ANCOVA) to compare the slope and intercept of the regression line fitted to organism data to the slope and intercept of the regression line fitted to artificial system data. We found no statistically significant difference between the intercepts (0.0160; *p*-value = 0.985) and the slopes (0.0214; *p*-value = 0.357). To include systems with a mass greater than 35,000 kg, we developed a new formula. The formula takes inspiration from the Heglund formula, in the sense that it approximates the power expended by a system as a product of two functions–a function of the system’s mass and a function of the system’s speed. Unlike the Heglund formula, the proposed formula also takes into account the deviations at high mass values. Specifically, we assumed a piecewise-linear model and used multiple linear regression to identify the coefficients in the following: $$P = A \cdot M^{b} \cdot V^{c},$$ or $$log(P) = a + b \cdot log(M) + c \cdot log(V)$$ where *a* = 0.006, *b* = 0.986, *c* = 1.12 for *M*\<35,000kg, and otherwise, *a* = 2.99, *b* = 0.489, *c* = 0.485; here *P* is power in watts, *M* is mass in kilograms, and *V* is speed in meters per second. The confidence intervals (at 0.95) for the intercepts and coefficients are listed in. With this formula (which for the sake of brevity, we refer to as the KGP formula, from the initials of the authors), the R2 for the entire set of data (comprising both biological organisms and artificial systems) is 0.987 and the mean absolute percentage error (MAPE) is 0.0269 both on the log₁₀ basis. For *M*\<35,000kg, a simplified formula *P = 1*.*01·M·V* provides a close approximation, with fit only marginally worse than by using and coefficients of. As seen in, the data for lower values of power (and, generally, lower mass) refer mainly to biological organisms, while the data for higher values of power (and mass) refer to artificial systems. # Discussion Let’s begin the discussion by revisiting the question of relation between *P*_*a* and *P*_*ext* which we introduced earlier in the Data section. In the following, we build on the approach and data of Cavagna and co-workers. Per Eqs 5 and 6 of, neglecting the negative work of muscles (as does), the efficiency γ with which muscles transform chemical energy into positive work is *γ = P*_*a**/P*_*met* *= (P*_*ext* *+ P*_*int**−P*_*elast**)/P*_*met*, where *P*_*a* and *P*_*ext* were introduced earlier, *P*_*int* is the rate of work associated with movements of an animal’s limbs with respect to its center of gravity, *P*_*elast* is the power recycled via elastic storage within an animal’s limbs, and *P*_*met* is the chemical power input to muscles. Per, *γ* ranges from 0.2 to 0.3, and empirical values of *α = P*_*ext**/P*_*met* range from 0.15 to 0.75, for a number of species (rhea, turkey, spring hare, kangaroo, dog and monkey). Then, rearranging, the mechanical power output produced by muscles (comparable conceptually to the mechanical power output of an engine) *P*_*a* *= γ·P*_*met* *= P*_*ext* *+ P*_*int**−P*_*elast* *= (γ/α)·P*_*ext*. Taking the range of values of γ and α mentioned above, we find that the values of *P*_*a**/P*_*ext* *= γ/α* are in the range 0.27–2.0. Therefore, although *P*_*ext* and *P*_*a* are not directly comparable, *P*_*ext* can serve as an order-of-magnitude approximation of *P*_*a*. Although the range of values above suggest that *P*_*ext* could be twice smaller or nearly 4 times larger than *P*_*a*, this is acceptable for our allometric, order-of-magnitude study. For example, to become an outlier in (see the discussion of outliers later in this section), a data point would have to disagree with the predicted value of power by a factor of approximately 5. And although the values of *α* provided in cover only 6 species, illustrates that data for many other species fall into the same pattern, consistent with the premise that *P*_*ext* can serve as an order-of-magnitude approximation of *P*_*a*. With this, returning to the results of the preceding section, we see a strong similarity between biological and artificial systems in terms of the relation among speed, mass, and mechanical power of locomotion: both organisms and artificial systems agree closely with the Heglund formula and its extension, the KGP formula. This is remarkable for several reasons. First, the agreement covers an enormous range of mass and power values–over 11 orders of magnitude in mass, from a cockroach (less than 1 g) to a loaded freight train (nearly 100,000,000 kg) and 12 orders of magnitude in terms of mechanical power. Second, the Heglund formula was originally based on data limited to only a few animals with masses not exceeding 70 kg. There is no obvious reason why it should continue to be as valid when extrapolated to systems of dramatically greater mass–up to 3 orders of magnitude greater, about 35,000 kg. Third, the Heglund formula was developed originally for biological organisms only. It is notable that the formula is also valid for systems of entirely different morphology and functionality, such as trucks and tanks of up to 35,000 kg in mass. Turning to the KGP formula, let us first mention outliers. Following, we determined data to be an outlier if the standardized residual was greater than three scaled median absolute deviation (MAD) from the median standardized residual. Out of 316 points in the data set, only 15 are outliers. Most of them are bicycles and elephants, not surprisingly as both of these have been discussed in prior literature as remarkably efficient in terms of energy cost of locomotion. Bicycles had been mentioned as outliers in and elephants’ external mechanical CoT had been found in to be far lower than any other animal. Continuing to explore the KGP formula, and particularly its coefficients , we note that for organisms and systems below 35,000 kg, the exponent for mass *b* is close to 1.0, similar to the Heglund formula. The exponent for speed *c* is higher than 1.0, i.e., the power grows nonlinearly with speed, also similar to the Heglund formula. To put it differently, the mechanical CoT, *P/(M·V)*, increases with speed. This is a trend common for many mobile systems, as reflected in the Gabrielli–von Karman diagram. Then, as systems become heavier, approximately above 35,000 kg, the relation among power, mass, and speed enters a very different regime. As seen in, power depends on mass with an exponent significantly less than 1.0. Furthermore, the dependence of power on speed diminishes. To put it differently, the CoT (or the specific resistance in the Gabrielli–von Karman terminology) becomes less dependent on speed and diminishes with mass proportionately to *M*^*(b-1)*, where *b* is between 0.315 and 0.663. What could explain this behavior?. The literature offers numerous explanations for the diverse allometric relations among mass, body length, speed, and power expenditures of biological organisms. Typically, such explanations build on the organism’s need to minimize energy expenditures or the need to maintain acceptable level of stress in bones. In a similar spirit, we too offer an explanation of why, as we have shown in this paper, heavy mobile systems’ CoT diminishes with mass, proportionately to *M*^*(b-1)*, where *b* is between 0.315 and 0.663. For a system like a heavy tank, a key constraint on its speed is the sheer stress *S* it imposes on the ground, which cannot exceed a soil-dependent value of *S*_*max*. Such a system expends power *P \~ S·F·V*, where *F* is the system’s footprint–the area of contact with the ground. A heavy system attempting to reach its maximum speed is likely to be limited by *S*_*max*, a constant for a given soil type and its moisture content. Therefore, its power *P \~ S*_*max* *·F·V*. Then, *P/(M·V) \~ (S*_*max**·F)/M*. Because the footprint is proportional to the square of a system’s linear dimension, which in turn is proportional to *M*^*1/3*, we have *P/(M·V) \~ (S*_*max**·M* ^*2/3**)/M*, and therefore *P/(M·V) \~ S*_*max**·M*^*(-1/3)*. This is qualitatively consistent with our findings. Having offered an explanation for one of our key findings, we cannot however recommend giving it too much credence. We are reluctant to rely on explanations that are mono-casual in nature, e.g., based on a stress within the system or optimization of energy consumption, etc. To explore this point, let’s consider the remarkable multiplicity of factors governing the maximum speed that a vehicle of a given gross weight can develop using a given source of power (e.g., an internal combustion engine). For this purpose, we refer to the NATO Reference Mobility Model, extensively developed over several decades and experimentally validated on thousands of vehicle tests in multiple countries. For example, Vong et al. explore how a given vehicle’s speed can be limited by dozens of different constraints involving soil properties, terrain characteristics, drivers tolerance for shocks, vehicles geometry and features. The constraints that govern the maximum speed and power requirements of an artificial, engineered ground-mobile system are complex and multi-faceted, involving multiple heterogeneous factors and interacting among themselves in diverse ways. This may be even more so in case of intricate biological systems. As such, future work must look beyond mono-causal explanations for the allometric relations discussed in this paper. ## An application The findings of this paper are particularly relevant to designing robotic, ground-mobile systems. For one thing, robotics tends to have a significant theoretical and practical affinity to bio-inspired approaches and analogies. The growing interest in legged robots is one example. As such, the common regularity of biological and artificial systems is of interest. Furthermore, like biological organisms, robotic systems used for defense applications are more likely to operate on offroad terrain than a typical transportation vehicle, and our data collection strategy includes a broad range of data relevant to offroad locomotion. Determining the feasible yet ambitious targets for tradeoffs among the power, speed, and mass of future terrestrial robots, particularly for defense applications, is a difficult task. It is undesirable to base such targets on current experience, because military hardware is often developed and used for multiple years and even decades; therefore, the specifiers and designers of such hardware must base their targets–competitive yet achievable–on future technological opportunities not necessarily fully understood at the time of design. To be sure, much research literature discusses detailed models for the analysis and design optimization of power-efficient robots, e.g., refs. Here our intent is different: we explore how robotic applications can benefit from the KGP relation. Since the KGP formula allows the designer of a robotic system to estimate power requirements from the system’s desired mass and speed, it may help serve as a preliminary check of the design’s realism and potential performance bounds. Here, we consider a preliminary design concept–a quadrupedal robotic “mule” that we call Exploratory Design for Mule-like Equipment Carrier (EDMEC). EDMEC weighs 600 kg and travels at a speed of 2 m/s. EDMEC is envisioned to carry a payload up to 90 kg consisting of munitions, food, extra batteries, and communication equipment. This quadrupedal system will be able to travel in complex terrains (dense forests, rocky plains, and moderately difficult mountain trails) and over obstacles like building rubble. We estimated the total power required for EMDEC locomotion as 5 kW. However, the KGP formula predicts that a 600 kg vehicle moving at a speed of 2 m/s would need 1,200 W of power. This means that EMDEC consumes 4.1 times more power than predicted by the KGP formula. Is something wrong with EMDEC’s design or our power consumption model? Not necessarily. Consider, where we plotted a number of robotics systems in comparison with the data in. These systems include a number of commercially available or fielded systems (green asterisks), research systems (green squares), and the EMDEC concept (the purple star). Most of these data points are positioned well above the diagonal curve, indicating that the corresponding systems consume more power than the KGP formula predicts based on animals and conventional (tracked or wheeled) vehicles. Particularly notable are the legged systems that lie significantly above the curve: the University of Maryland Micro-robot, NASA Valkyrie, and Boston Dynamics Atlas. These systems consume 64, 52, and 60 times more power, respectively, than the curve predicts. Agility Robotics Digit is another outlier. Two of the most efficient walking robots, the Cornell Ranger and the MIT Cheetah, use 3.0 and 3.6 times the power predicted by the KGP curve, respectively. On the other hand, we should also note there are three data points that are close to, or even below, the curve and whose power is of the same order of magnitude as EMDEC. These are wheeled or tracked systems (Clearpath Robotics Jackal and Warthog, and iWarrior 710), which tend to be more efficient than their legged counterparts. The relative inefficiency of legged systems is a common topic of discussions in literature, e.g.. In fact, many ground vehicles lag behind their animal counterparts in movement efficiency when not moving on specially prepared surfaces like roads. Much of the legged robots’ inefficiency can be explained by the lack of passive elements like tendons in animals. Passive elements allow for energy to be stored and recycled back into the system, thus increasing efficiency. In fact, animal tendons can contribute up to 45% of the power required for locomotion in kangaroos and 40% in horses and camels. Another reason for why artificial systems perform less efficiently than animals is due to difficulties and inefficiencies in sensing and path planning. Although attractive for efficiency reasons, including passive elements in the design of a legged robotic system often comes at the cost of reducing the system’s versatility. Passive elements that recycle energy back into the system restrict the movement of appendages and often need to be empirically tuned to specific operating environments. Without passive elements, a legged robot sacrifices some efficiency but gains adaptability to variable environments, which helps it move in difficult environments inaccessible to more efficient wheeled and tracked vehicles. # Conclusions We found that artificial ground-mobile systems–as diverse as ground robots, small utility vehicles, trucks, and tanks–exhibit a consistent regularity of relation among mass, power, and speed. For the range of mass from 1 g to 35,000 kg, this regularity is similar to the Heglund formula, known since 1980s and applied to a range of ground-mobile animals. Therefore, a single formula describes organisms and systems, from a cockroach to a battle tank. These findings point to a fundamental similarity between biological and artificial locomotion that transcends great differences in morphology, mechanisms, materials, and behaviors. We also show that for very heavy vehicles, ranging approximately 35,000–100,000,000 kg, the relation among power, mass, and speed enters a different regime. The CoT (or the specific resistance in the Gabrielli–von Karman terminology) becomes less dependent on speed and also diminishes with mass, proportionately to M^(b-1), where b is on the order of 0.5. To cover the two different regimes–both below and above 35,000 kg–we propose a piecewise-linear regression formula that closely agrees with the available data. The agreement covers enormous ranges of mass and power values–over 11 orders of magnitude in mass, from a cockroach (less than 1 g) to a loaded freight train (nearly 100,000,000 kg), and over 12 orders of magnitude in terms of mechanical power. When the proposed formula is considered in relation specifically to an important class of future ground vehicles–legged robots–we note a consistent inefficiency in current designs: the power consumption is from 3 to 60 times greater than predicted by the formula. In other words, today’s legged robots are significantly less efficient than animals or tracked and wheeled vehicles of comparable speed and mass. This, however, may be the inevitable price to pay to allow them to adapt to diverse and difficult terrains. With regard to theoretical explanations of these empirical findings, we note that the constraints governing the maximum speed and power requirements of an artificial ground-mobile system are numerous, complex, and multi-faceted, involving multiple heterogeneous factors and interacting among themselves in diverse ways. This may be even more so in case of intricate biological systems. As such, we find it difficult to give too much credence to theories that are mono-casual in nature, e.g., based on a stress within the system or optimization of energy consumption. Future work must explore the theoretical space beyond mono-causal explanations for the allometric relations discussed in this paper. # Supporting information The views presented in this paper are those of the authors and not of their employer. Phyllis Mcgovern of CCDC Army Research Laboratory (ARL) searched for multiple books and articles that supplied the raw data for this research. Jody Priddy of the U.S. Army Engineer Research and Development Center assisted with obtaining data related to numerous ground vehicles. Carol Johnson, the ARL editor, improved the style and the grammar of the manuscript. 10.1371/journal.pone.0249066.r001 Decision Letter 0 Nyeem Hussain Md Abu Academic Editor 2021 Hussain Md Abu Nyeem This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 28 Aug 2020 PONE-D-20-21456 From Cockroaches to Tanks: the Same Power-Mass-Speed Relation Describes both Biological and Artificial Ground-Mobile Systems PLOS ONE Dear Dr. Kott, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. Please submit your revised manuscript by Oct 12 2020 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <plosone@plos.org>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript: A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols> We look forward to receiving your revised manuscript. Kind regards, Hussain Md Abu Nyeem, Ph.D. Academic Editor PLOS ONE Journal Requirements: When submitting your revision, we need you to address these additional requirements. 1\. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at <https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main _body.pdf> and <https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_titl e_authors_affiliations.pdf> 2\. We note that you have stated that you will provide repository information for your data at acceptance. Should your manuscript be accepted for publication, we will hold it until you provide the relevant accession numbers or DOIs necessary to access your data. If you wish to make changes to your Data Availability statement, please describe these changes in your cover letter and we will update your Data Availability statement to reflect the information you provide. \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Partly Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 2\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes Reviewer \#2: I Don't Know \*\*\*\*\*\*\*\*\*\* 3\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 4\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 5\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: The paper investigates allometric relationship of mechanical power expended per unit of momentum (cost of transportation, CoT) for a wide range of natural and artificial systems exhibiting terrestrial locomotion. Authors find that the previously known relationship proposed for the mass range of 10^-3-10^1 kg by Heglund \[13\] holds over a much wider range of masses, 10^-3-10^4 kg, in animals and engineered systems alike. They also analyze the data available for engineered systems in the range of 10^4-10^6 kg and find that a qualitatively different scaling law applies for that range. Two linear functions describing CoT in the sub-ranges are fitted to the data. The resulting piecewise-linear relation holds for the whole range of masses that was considered (10^-3-10^6 kg). The relationship is additionally verified by checking that it holds on the data on sauropods and elephants that was not used in fitting of the empirical law, and finding that the data agrees with the law. Some existing ground-based robotic systems are considered in the context of the considered allometric relations and, for the most part, are found to be outliers. Implications of this fact for design of such systems are discussed. All the claimed findings are, to the best of my knowledge, novel. While allometric studies that include both animals and engineered systems are not unprecedented (e.g. ref \[39, 17\] of the manuscript), no systematic studies of that sort were done before for terrestrial locomotion. Pre-existing work is properly cited. Data collection methods used in the study are mostly sound. the only issue I found was the method in which the elephants and sauropods data was obtained. Mechanical power expended by the animals is derived from literature estimates of metabolic power by multiplying it by a constant factor. The factor is taken from the study on horses (\[27\], masses around 5e2 kg) and used for animals with masses at least two order of magnitude larger. Meanwhile, in Introduction the authors mention that the metabolic and mechanical cost of transportation scales differently. That makes the proportionate estimate dubious at best. I ask that authors do one of the following: 1\) provide a justification for proportional estimation of mechanical cost of transport from the metabolic cost, for the relevant mass range, or 2\) provide a more careful estimation of mechanical cost of transport for elephants and sauropods and redo the related analysis and Figure 4, or 3\) drop the elephant and sauropod data analysis from the paper altogether. Another issue with this part of the study is that the data, while available in raw form in the primary sources, is processed to obtain the mechanical CoT estimates. It would be beneficial to make the processed data available. Also, Figure 4 is cluttered. Please only show the relevant range of masses and consider dropping the data for animals and artificial systems other than elephants and sauropods. This issue, however, is minor, as it the data is only used to provide additional support to a claim established with other data. Even if that part of the study is removed its primary claims can be sufficiently well supported. Data analysis methods are sound. There is a minor presentation issue with ANCOVA analysis at the bottom of page 4: a single value is given as "difference" between slopes, and the value (0.964) is too large to be the true absolute difference between slopes that are close to 1. Please provide both of the slopes with confidence intervals. Another minor issue encountered throughout the paper: values are reported to the third digit of the error margin. It is customary, at least in physical sciences, to round error margins down to one significant digit (two if the first one is 1) and also round the point estimate to the same decimal. For example, 0.964 \\\pm 0.0224 becomes 0.96 \\\pm 0.02. Similarly, for p-values usually only one significant digit is reported. Such conventions vary by the discipline, so I won't ask an adherence to this one, but in my opinion it could improve readability of the paper. There is also a similar minor issue with the power estimate for EDMEC given in the Supplementary. The estimate is valid, but crude and must be treated as such. In light of that, the power consumption estimate of 4920W should be truncated to a single significant digit and presented as 5kW. Language-wise, the manuscript is well-organized and accessible. There are a few places with typos and unclear language (listed below), but that is easily fixable and did not affect my ability to understand the intent of authors. I recommend that the paper is accepted for publication after these issues are addressed. ---------------- Abstract: "We show empirically not only..." - possible forgotten "that" Introduction: "metabolic rate scales approximately to the 3/4 power of the organism's mass" - awkward wording In the discussion of CoTm, a typo: "P/(M\*V)" -\> "Pm/(M\*V)" Results: p.10 last paragraph: "the slopes and intercepts of organism and system regression lines" - unclear Supplementary: The explanation of power expenditure section: "...is on the order of a magnitude or nearly constant magnitude of 10 body lengths per second..." - awkward Reviewer \#2: The authors test how predictable the relationship between mass and mechanical power is for biological and artificial systems based on a previously proposed formula. The authors extend the formula to consider systems of larger mass, where the previously proposed formula failed. The authors then use this metric to analyze the efficiency of current artificial legged walkers and find that for the most part they are several times more inefficient than predicted by the formula. They discuss the reasons for this departure and argue for more biologically inspired technologies as a direction forward. The manuscript is straightforward, well written, and technically sound. \*\*\*\*\*\*\*\*\*\* 6\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: **Yes: **Anton Bernatskiy Reviewer \#2: No \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <figures@plos.org>. Please note that Supporting Information files do not need this step. \<gdiv\>\</gdiv\> 10.1371/journal.pone.0249066.r002 Author response to Decision Letter 0 29 Sep 2020 All responses are in the file submitted as part of the revised submission. 10.1371/journal.pone.0249066.r003 Decision Letter 1 Nyeem Hussain Md Abu Academic Editor 2021 Hussain Md Abu Nyeem This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 9 Nov 2020 PONE-D-20-21456R1 From Cockroaches to Tanks: the Same Power-Mass-Speed Relation Describes both Biological and Artificial Ground-Mobile Systems PLOS ONE Dear Dr. Kott, Thank you for submitting your manuscript to PLOS ONE. All the comments of the editor and earlier reviewers have been addressed well. However, a few questions about the comparison measures and the implications of the given data need further clarification. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. Note PLOS ONE strives to facilitate timely review with our continued effort to improve the speed and quality of the review process. However, please understand that due to unavoidable circumstances of the academic editor, your manuscript experienced an unusual delay this time.  Please submit your revised manuscript by Dec 24 2020 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <plosone@plos.org>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript: A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions, see: <http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols> We look forward to receiving your revised manuscript. Kind regards, Hussain Md Abu Nyeem, Ph.D. Academic Editor PLOS ONE \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation. Reviewer \#1: All comments have been addressed Reviewer \#3: (No Response) \*\*\*\*\*\*\*\*\*\* 2\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Yes Reviewer \#3: Partly \*\*\*\*\*\*\*\*\*\* 3\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes Reviewer \#3: Yes \*\*\*\*\*\*\*\*\*\* 4\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes Reviewer \#3: No \*\*\*\*\*\*\*\*\*\* 5\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#3: Yes \*\*\*\*\*\*\*\*\*\* 6\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: All of the issues I pointed out in my review have been fully addressed. The paper now satisfies all the requirements of PLOS ONE: it reports a novel result and it is technically sound. The reviewer thanks the authors for a good read! Reviewer \#3: The biggest challenges to this paper is in understanding the different metrics being used for ‘system/organism power’, perceiving whether these different metrics are reasonably comparable, and in interpreting the implications of the presented relationships. Comparing metrics… It is reasonable to begin with the Heglund (Cavagna, Taylor, Fedak) studies of the 1980s for the broad-brush scaling data and exponent fits. Of the three I recall (metabolic, center of mass and center of mass + ‘internal’) the third is selected for the initial scaling equation (Heglund et al. 1982 IV). This is the energy put into the center of mass added to the energies of the limbs about the center of mass (with this metric, it is assumed there is no transfer between the two). However, it is the second that is used for the majority of the biological data survey. Why not use center of mass power for both (Heglund et al., 1982 III)? Power proportional to mass and velocity (so a constant mechanical cost of transport): this appears to provide a better fit to the data too ( Table 1: a=0; b=1; c=1?). A bigger issue is that neither form for mechanical power is (and is acknowledged to be) a very poor predictor for the metabolic or ‘biological engine’ power. But what is the ‘system’ power being described for the vehicles? It is presumably not the direct equivalent to the mechanical power demand of the animals – on flat level ground at constant speed this would be zero. Presumably it is the power supplied by the engine…? Under what condition? Maximum power or maximum speed? And then, what of the horse-drawn guns? Is the work calculated the ‘animal’ way or the ‘machine’ way? And so it is not immediately clear to me that the chosen animal and machine powers are reasonably comparable. Implications… If we believe there is some valid equivalency between the two metrics for power, the striking finding is the lack of evidence that a wheel improves matters. If I were to start off with a one-human-power runner (measured the animal way), and then put the human mass on a bicycle, I would hope one human-power to drive either a much more massive load or to travel much more quickly. That this is not evident across the data presented (compare 1000kg bull v.s truck) leads me to wonder whether 1) the two metrics for power are not comparable, or 2) trucks and tanks are horribly un-wheel-like. Where does a motorbike on a road sit on this line? And a small train on a railway track? It would be nice to see where a couple of ‘good’ (fast or economical) wheeled vehicles sit on this plot. Minor points To offer ‘data on request’ or ‘see pdf’ is not very 2020. The data are indeed available online <https://apps.dtic.mil/sti/pdfs/AD1098609.pdf> (Which may bring into question the issue of novelty – that is a policy decision outside my remit) but why not provide them as a spreadsheet so that the reader can easily start trying out their own statistical approaches? The scaling relationship for the maximal speed in running animals falls over a bit at the larger sizes (Garland etc.) Please double check the values taken from Heglund – there may be an issue with conversion to SI units. See Blickhan, R. & Full, R. J. 1987 Locomotion energetics of the ghost crab II. Mechanics of the center of mass during walking and running. J. Exp. Biol. 130, 155-174. It appears odd to resort to using metabolics to derive the elephant ‘animal- power’. Does not Genin, Willems, Cavagna, Lair and Heglund (J. Exp. Biol., 2010) provide a more appropriate starting point? It may be that the elephant data point would then fit rather poorly: note that Genin et al. report CoM energy fluctuations 1/3rd that of smaller mammals. And including sauropods as empirical points does feel a bit of a stretch. \*\*\*\*\*\*\*\*\*\* 7\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous, but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: **Yes: **Anton Bernatskiy Reviewer \#3: No \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <figures@plos.org>. Please note that Supporting Information files do not need this step. \<gdiv\>\</gdiv\> 10.1371/journal.pone.0249066.r004 Author response to Decision Letter 1 22 Dec 2020 The comments of the first 2 reviewers have been already answered to their satisfaction in previous revisions of the paper. This revision addresses the comments of Reviewer \#3, and a detailed Response to Reviewer 3 has been provided as a file included in this submission. 10.1371/journal.pone.0249066.r005 Decision Letter 2 Nyeem Hussain Md Abu Academic Editor 2021 Hussain Md Abu Nyeem This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 15 Jan 2021 PONE-D-20-21456R2 From Cockroaches to Tanks: the Same Power-Mass-Speed Relation Describes both Biological and Artificial Ground-Mobile Systems PLOS ONE Dear Dr. Kott, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. Please submit your revised manuscript by Mar 01 2021 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <plosone@plos.org>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript: A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions, see: <http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols> We look forward to receiving your revised manuscript. Kind regards, Hussain Md Abu Nyeem, Ph.D. Academic Editor PLOS ONE Additional Editor Comments (if provided): The paper has been significantly improved and addressed all the major questions of the previous reviewers. However, the academic editor is still interested in learning how the remaining questions of the current reviewer are addressed. Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation. Reviewer \#3: (No Response) \*\*\*\*\*\*\*\*\*\* 2\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#3: Partly \*\*\*\*\*\*\*\*\*\* 3\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#3: Yes \*\*\*\*\*\*\*\*\*\* 4\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#3: Yes \*\*\*\*\*\*\*\*\*\* 5\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#3: Yes \*\*\*\*\*\*\*\*\*\* 6\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#3: The major issue with this paper is that it assumes comparability between two potentially quite different forms of power. The vehicle values are for engine power, if possible, the ‘rated power’ (responses). This is the power (torque x angular velocity) at the end of the drive shaft. The rated power is useful in informing (once other losses are considered, and given suitable gearing) the capacity of a vehicle to accelerate, pull a load up an incline, or overcome drag. The animal values are the rate of mechanical work of the center of mass during steady state locomotion. Is it reasonable to assert an equivalency between ‘rated power’ and ‘center of mass power’? The difficulty is that the ‘center of mass power’ of a wheeled vehicle at any steady speed (even exceedingly high, and against drag) on level ground is zero. So ‘rated power’ cannot sensibly translate to ‘center of mass power’. But is the reverse likely? Does the center of mass power of an animal display something equivalent to its rated power? One can certainly imagine cases where this fails (consider cycling), but might it be reasonable for legged locomotion? For the purposes of this paper, it would appear sufficient to state it as an assumption that it is reasonable, with whatever justification can be thought of and while acknowledging that others have also made this assumption. At the moment, this is not sufficiently addressed. The term ‘rated power’ only appears in the responses – a description of the vehicle power close to the one I give above is required. Also in the responses is the justification: “We do believe that for all systems (biological or artificial) in our study we are looking at fundamentally the same metrics: the amount of mechanical energy expended to propel the system.” This is insufficient. Why is the center of mass power (that can be zero given wheels) ‘the amount of mechanical energy expended to propel the system’. Until an explicit work-around is given for this, any informed reader will view this study as a comparison of apples and oranges. I wonder whether the biological literature is as exhaustive as claimed. I would suggest doing a citation search on Cavagna, 1975 ‘force platforms as ergometers’. Off the top of my head, are the reported values for cats, tortoise, penguins all unusable? Any lizards? \*\*\*\*\*\*\*\*\*\* 7\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#3: No \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <figures@plos.org>. Please note that Supporting Information files do not need this step. \<gdiv\>\</gdiv\>\<gdiv\>\</gdiv\> 10.1371/journal.pone.0249066.r006 Author response to Decision Letter 2 21 Feb 2021 We accepted the reviewer's comments and provided revision accordingly. See the Response to Reviewers document provided with this submission. 10.1371/journal.pone.0249066.r007 Decision Letter 3 Nyeem Hussain Md Abu Academic Editor 2021 Hussain Md Abu Nyeem This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 11 Mar 2021 From Cockroaches to Tanks: the Same Power-Mass-Speed Relation Describes both Biological and Artificial Ground-Mobile Systems PONE-D-20-21456R3 Dear Dr. Kott, We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements. Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication. An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at <http://www.editorialmanager.com/pone/>, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to- date. If you have any billing related questions, please contact our Author Billing department directly at <authorbilling@plos.org>. If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact <onepress@plos.org>. Kind regards, Hussain Md Abu Nyeem, Ph.D. Academic Editor PLOS ONE Additional Editor Comments: The paper has been significantly improved and reasonably addressed the previous comments of the reviewers. Upon the 'ACCEPT' recommendations of the all the three reviewers, and seeing the improvements in the revised versions, the academic editor is now also convinced for its publication.  Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation. Reviewer \#3: All comments have been addressed \*\*\*\*\*\*\*\*\*\* 2\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#3: Yes \*\*\*\*\*\*\*\*\*\* 3\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#3: Yes \*\*\*\*\*\*\*\*\*\* 4\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#3: Yes \*\*\*\*\*\*\*\*\*\* 5\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#3: Yes \*\*\*\*\*\*\*\*\*\* 6\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#3: I am grateful for the consideration given to my previous comments. I shall give a couple of additional thoughts, but certainly do not require further responses. I would agree that the external mechanical power of a legged locomotor might be taken as a reasonable – at least order of magnitude – estimate of the minimum actuator work. Elastic mechanisms may play some role, and mean that it cannot be taken as an absolute minimum value, but the effect of these are likely – at the scales of interest here – to be negligible. However, the true actuator (muscle or engine) work may be much, much higher, and this may not necessarily be covered by an ‘order of magnitude’ argument. ‘Internal’ mechanical work demands may be significant; and external work could approximate zero. The Adaptive Suspension Vehicle (Waldron et al., 1984) carried its driver horizontally and steadily: external power approximately zero. However, the motor power is not. So: cases can be imagined where the ‘apples with apples’ might not be true. How important this is (after all, horses really do not look at all like the ASV) to interpreting the current study can, in my view, now be left to the reader. \*\*\*\*\*\*\*\*\*\* 7\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#3: No 10.1371/journal.pone.0249066.r008 Acceptance letter Nyeem Hussain Md Abu Academic Editor 2021 Hussain Md Abu Nyeem This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 15 Apr 2021 PONE-D-20-21456R3 From cockroaches to tanks: the same power-mass-speed relation describes both biological and artificial ground-mobile systems Dear Dr. Kott: I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department. If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact <onepress@plos.org>. If we can help with anything else, please email us at <plosone@plos.org>. Thank you for submitting your work to PLOS ONE and supporting open access. Kind regards, PLOS ONE Editorial Office Staff on behalf of Dr. Hussain Md Abu Nyeem Academic Editor PLOS ONE [^1]: The authors have declared that no competing interests exist.

# Introduction Up to 80% of patients experience acute preoperative anxiety prior to elective surgery, and can have both physiologic and psychological consequences. Previous studies have demonstrated that excessive preoperative anxiety increases the intraoperative anesthetic requirement and can prolong recovery. A rapid and objective assessment of preoperative anxiety might be useful for improving perioperative patient care and potentially enhancing the recovery process. The current “gold standard” for evaluating acute anxiety is the State-Trait Anxiety Inventory (STAI) which consists of two separate 20-item questionnaires. The Brief Symptom Inventory has 53 items which measures several affective states including anxiety. However, these anxiety measurement tools are time consuming which limit their usefulness in the preoperative setting. The numeric verbal rating scale (NVRS) has been utilized for assessing preoperative anxiety. However, this 11-point scale requires that a patient describe their acute anxiety level in numerical terms. Furthermore, there is a poor correlation (coefficient = 0.50) between the NVRS and the Spielberger State Anxiety Inventory scores. Ekman and colleagues have shown that facial muscle patterns can be used to reliably detect emotions and to distinguish between differing types of emotions (e.g., anger vs. fear). A facial affective scale has been previously validated as a reliable tool for assessing acute pain in children. Additionally, a visual analog scale (VAS) is a simple self-rating tool for rapidly assessing the level of state anxiety. Therefore, we speculated that a pictorial facial scale would be a suitable alternative to the simple NVRS and the VAS for assessing the patient’s level of state anxiety in the preoperative period. The objective of this preliminary investigation was to develop a user- friendly tool for the rapid assessment of state anxiety during the perioperative setting. The proposed visual facial anxiety scale (VFAS) was designed to allow surgical patients to identify their level of acute preoperative anxiety using simple facial depictions. # Methods The research was approved by IRB of Cedars Sinai Medical Center (No: Pro00041348). The participants provide their written informed consent to participate in this study and the ethics IRBs approved this consent procedure. A total of 265 healthcare providers \[anesthesiologists (n = 98), anesthesiology residents (n = 27) and perioperative nurses (n = 140)\] at Cedars-Sinai Medical Center in Los Angeles, CA were invited to participate in this pilot study. The VFAS was comprised of 11 similarly styled facial expressions labeled A0 through A10 (the number was ‘blinded’ to the participants).The NVRS were listed in order on one piece of paper, and the categorical anxiety levels, namely none, mild, mild-moderate, moderate, moderate-high and highest, were listed on a separate piece of paper. The participants were asked to: 1) match each separated face to a corresponding number, 0 (no anxiety) through 10 (highest anxiety), and 2) assign one face to each of six anxiety categorical anxiety variables: none, mild, mild-moderate, moderate, moderate-high and highest anxiety. The faces were purposely presented randomly to avoid any visual bias when assigning the faces to a number and a category. ## Statistical analysis The frequency that a particular face was assigned to a number on the numerical scale and to an anxiety level on anxiety category (none, mild, mild-moderate, moderate, moderate-high and highest was determined. The Spearmen correlations were calculated to investigate the relationships of the VFAS to the six categorical anxiety levels. The highest frequency of a face assigned to a level of the numerical anxiety scale resulted in a finalized order of faces corresponding to the 11-point numeric rating scale. # Results shows which face had the highest frequency for each of the NVRS anxiety scores (0–10). A total of 261(98.5%) participants chose the face A0 as representing a 0 anxiety score, and 250 (94.3%) participants chose the face A1 as representing a score of 1. Similarly, more than 96% of participants chose the faces A8, A9 and A10 to represent anxiety scores of 8, 9 and 10, respectively. However, the results demonstrated obvious disagreement from levels 3–5. For example, the faces A1, A2, A3, A4, and A5 were all selected to represent anxiety score of 4, but A4 was ultimately chosen to represent anxiety level 4 because it was chosen with the greatest frequency. Finally, the order of the 11 faces best corresponding to the NVRS scores of 0–10 was determined to be A0, A1, A2, A3, A4, A5, A7, A6, A8, A9 and A10, respectively. For the association between the VFAS and the six categorical anxiety levels, 260 (98.1%) participants considered the face A0 as representing ‘no’ anxiety, 250 (94.3%) participants picked the face A10 as representing ‘highest’ anxiety and 147 (55.5%) participants chose the face A8 as representing ‘moderate-high’ anxiety. Predictably, there was a less obvious pattern for the other anxiety categories. Only 92 (34.7%) participants categorized the face A1 as depicting a mild level of anxiety, 81 (30.6%) participants categorized the face A5 into the mild-moderate anxiety level, and 95 (35.9%) considered the face A7 as representing a moderate level of anxiety. Further analysis using Spearman correlation determined which face would represent these three anxiety categories. Spearman analysis shows a significant correlation between the faces A3 and A5, which had the two highest frequencies assigned to the mild-moderate anxiety category (r = 0.58). A5 was ultimately chosen to represent the mild- moderate category of anxiety due to its higher frequency (30.6% vs 24.9%). A significant correlation was also found between the faces A7 and A6 in the moderate category (r = 0.87). The face A7 was selected to represent the moderate anxiety level because of its higher frequency (35.9% vs22.6%). Finally, the faces A0, A1, A5, A7, A8 and A10 were assigned to the six categories, none, mild, mild-moderate, moderate, moderate-high and highest anxiety levels, respectively. # Discussion Numerous facial scales have been used to measure pain intensity, including the Color Analog Scale, Face Pain Scale and the Wong-Baker Face Pain Rating Scale. Facial pain scales are validated and reliable tools for measuring acute pain. A facial scale providing expression of emotions such as happiness, anger and anxiety is sensitive, reliable and easy to be understood. Excessive anxiety prior to surgery can produce harmful physiological consequences.The measurement of preoperative anxiety could be useful in administering appropriate preoperative medication (e.g., an anxiolytic drug) to facilitate the induction and recovery process. The proposed VFAS is a simple six-point scale consisting of six faces representing an increasing level of anxiety from none (a neutral facial expression) to highest (a facial expression displaying extreme fear). In contrast to a NVRS which requires patients to transform their subjective anxiety state into numbers on a continuous scale from 0 to 10, the new VFAS as an instrument which utilizes recognizable facial representations of various anxiety states, thereby strengthening its construct validity. The correlation between the subject’s responses to the State-Anxiety Inventory questionnaire and to a Faces scale was found to be 0.70, a level regarded as providing evidence of criterion validity by experts in scale development and validation. The facial scale is presented to patients as an alternative to a numeric linear analog scale (e.g., NVRS). Since it is often difficult for patients to quantify their level of anxiety using a simple numeric scale anchored by two extremes of anxiety (namely, no anxiety and highest level of anxiety), we proposed that a pictorial presentation of different facial expressions as a relatable measure of the patient’s levels of acute \[state\] anxiety. The healthcare provider merely has to ask the patient to choose which of the six facial expressions which most accurately reflects their level of anxiety at that precise moment in time. The new VFAS is quick and easy to administer, and therefore easy to incorporate into routine clinical practice. In our previous pilot study, the facial scale was created using different style faces, and it was utilized as an alternative to the NVRS for assessing the level of preoperative \[state\] anxiety. However, there were too many faces in this early scale and only a moderate correlation was found between the previous facial scale and NVRS. Furthermore, the inconsistent style of the 11 faces could result in confusion for the patient. The new six-facial VFAS was designed, and through analysis of frequency selection and Spearman correlation, the survey of the scale by anesthesiology and perioperative nursing personnel resulted in six faces being chosen to compose the new VFAS. In the testing of the order of the scale items, all participants independently placed the scale items in the order from the lowest to the highest anxiety level on an 11-point NVRS and then made associations with levels of acute anxiety on a six-point categorical rating scale. The highest frequency of a face assigned to a number on the NVRS resulted in a finalized order of faces. When the Spearman correlation between multiple faces and anxiety levels were similar, the one with the highest frequency was chosen to represent the specific anxiety level. This indicates that the new VFAS has the property of rank order. By narrowing down the number of facial expressions from 11 to 6, the scaled items are fairly equally spaced and not overlapping, and this support for the application of the VFAS as an interval scale measure. This was a preliminary study designed to validate a new instrument for assessing anxiety which utilized experienced perioperative healthcare workers rather than actual surgical patients. The clinical relevance is limited because we utilized healthcare professionals as ‘surrogate’ patients in this preliminary evaluation of the new six-facial VFAS. When healthcare professionals utilized this instrument for assessing the level of acute anxiety, it displayed evidence of interval scale properties of rank order and equality between the points on the scale. The healthcare provider can very quickly and easily ask the patient to choose a facial expression which corresponds to their current level of anxiety. One of the major limitations of this preliminary study is absence of clinically- anxious study population. Only healthcare providers (namely, staff anesthesiologists, anesthesiology residents and perioperative nurses) were invited to participate in this pilot study. However, since all of these individuals work in the preoperative area, they are familiar with evaluating perioperative anxiety in patients. Further studies are clearly needed to validate this new facial anxiety scale in a surgical patient population with varying levels of preoperative anxiety. These preliminary data demonstrate the potential clinical utility of the proposed new VFAS for assessing preoperative \[state\] anxiety. ## Conclusions The new six-facial VFAS appears to be a valid tool for assessing the severity of acute \[state\] anxiety, and could be easily implemented in routine clinical practice without adding significant additional work for the clinical staff providing caring to surgical patients. [^1]: The authors have declared that no competing interests exist. [^2]: **Formal analysis:** XZ. **Investigation:** XZC OLEL OD. **Methodology:** RY PW. **Project administration:** XZC OLEL JF OD. **Resources:** RY OD PW. **Supervision:** RY PW. **Validation:** RY PW. **Visualization:** XZC RY PW. **Writing – original draft:** XZC RY. **Writing – review & editing:** PW.

# Introduction Childhood cancer treatment is accompanied by multiple direct and late toxicities. Preferably, these toxicities are anticipated and ideally prevented already prior to and during cancer treatment. Impaired future fertility is a major concern for childhood cancer patients and their parents. In the past, awareness of infertility as unexpected sequelae was raised only after reaching adulthood. Currently, patients, parents, survivors and healthcare professionals acknowledge the importance of discussing the risk of premature ovarian insufficiency (POI) and consequent infertility due to gonadotoxic cancer treatment. This includes the need for preservation already at an early stage even before the start of cancer therapy. It is challenging however, to timely identify patients at risk, to triage and to inform patients of gonadal damage risk on an individual basis before cancer treatment, and to offer the possibility of further counseling for fertility preservation by a fertility expert, without delay of cancer treatment in full cohorts of pediatric patients. The clinical focus upon presentation with new oncologic disease is often on the diagnostic process and swift stratification towards the most effective cancer treatment rather than on preventing potential toxicities. Therefore, our dedicated oncofertility working group created a standardized and easy to apply oncofertility care plan including a gonadal damage risk stratification tool, as this was deemed indispensable to ensure adequate and timely oncofertility care. The aim of the care plan was to identify, inform and triage 100% of all new patients and to timely refer 100% of the girls at high risk (HR) of gonadal damage for expert fertility counseling. This observational retrospective PEARL (<u>P</u>resErving ov<u>AR</u>ian function through cryopr<u>e</u>servation and informing gir<u>L</u>s with cancer about infertility due to gonadotoxic treatment) study evaluated the use of this standard oncofertility care plan in the first year 2019 in a full national cohort after centralization of pediatric oncology care. # Methods Pediatric oncology care in the Netherlands, previously dispersed in 7 expertise centers, was merged into one national center, the Princess Máxima Center for pediatric oncology in Utrecht in May 2018. This centralization serves the mission to further improve cure rates, while also decreasing early and late toxicity. ## The female oncofertility care plan Since 2015 a multidisciplinary dedicated team with representatives of all departments (S1 Table), prepared an oncofertility care plan for newly diagnosed children respecting their right to receive personalized oncofertility care. The oncofertility care plan for newly diagnosed girls is based on international and national literature and professional experiences. The content of our care plan had been intensively discussed with and is approved of by the joined ethical committee of the UMC Utrecht and Princess Máxima Center. This plan includes five steps (S2 Table). ### 1) Identification of all new patients Identifying newly diagnosed girls with cancer in our center is pursued on a daily basis by a dedicated oncofertility nurse practitioner (coordinator) together with the involved pediatric oncologist. The daily refreshed financial administration, the tumor board lists and clinical ward rounds are used to timely identify new patients. ### 2) Triage gonadal damage risk Pursuing standardized triage is done using a gonadal damage risk stratification tool on risk of future gonadal damage. This clinically applicable and easy to use gonadal damage risk stratification tool covers the (most commonly) used European treatment protocols in the Netherlands. As pediatric oncology is a dynamic field, the risk stratification is subject to updates with every new treatment protocol introduced in our hospital and enhancements over time are possible with increasing knowledge. In 2019 the cyclophosphamide equivalent dose (CED) score, expected abdominal radiotherapy (with dose to the ovary), hematopoietic stem cell transplantation (HSCT), and (ovarian) surgery served as the basis premise for this gonadal damage risk stratification tool. In 2019 treatment protocols were classified as LR, IR or HR of gonadal damage with CED scores of 0-4000mg/m2, 4000-8000mg/m2 and more than 8000mg/m2 respectively including radiotherapy and surgery. Evidently, other factors such as very young age at diagnosis, prognosis, psychosocial and ethical issues are taken into account upon counseling of patients. The oncofertility plan was mirrored during a working visit to the oncofertility team of the Cincinnati Children’s Hospital, USA, in September 2018 before implementation. ### 3) Informing patients Timely (preferably before start of gonadotoxic treatment) all patients are informed on their specific gonadal damage risk by their pediatric oncologist and/or the oncofertility nurse practitioner (coordinator). ### 4) Counseling a subset of girls by a oncofertility specialist Timely (before therapy or not leading to delay of therapy) counseling by the oncofertility specialist (gynecologist) is available. In our institute this is available for patients at low risk (LR) and intermediate risk (IR) on request, but actively encouraged in the subset of patients who are at high risk (HR) of gonadal damage. ### 5) Fertility preservation Preserving fertility is an option for highly selected, counseled, eligible (HR) patients after shared decision making. Four different methods are available as standard care fertility preservation in our hospital. a) Ovariopexy (OP), which is only useful in girls in whom radiation to the ovary is expected to do significant (and chemotherapy minor or no) damage. b) Oocyte harvest and cryopreservation (OC), after hormonal ovarian stimulation, which is only feasible for oncology patients who are postmenarcheal and when treatment can safely be postponed at least 2 weeks. c) Ovarian tissue cryopreservation (OTC) by unilateral ovariectomy (OTC-o) or d) partial ovariectomy (OTC-p). The care plan offers OTC to all girls newly diagnosed with cancer with high risk of gonadal damage in our center who, in general, are aged 0–18 years at presentation. The DCOG amended-Edinburgh criteria (S3 Table in) are used to carefully confirm eligibility of girls for OTC in our center. The content of the counseling by the oncofertility gynecologist is described in the S2 Table in. ## Patients Only newly diagnosed girls with pediatric cancer between 1 January 2019 and 31 December 2019 were included. The evaluation was conducted in spring 2020. ## Evaluation of the full first year of oncofertility care implementation Application of the standard oncofertility care plan for newly diagnosed girls and the five step process was evaluated. Baseline data including age at diagnosis, type of malignancy, proposed treatment(-protocol and arm), triage result (gonadal damage risk estimation) and curative or palliative intention of treatment were collected. The date of diagnosis was defined as the date of communicating the cancer diagnosis including the explanation of the intended treatment by the pediatric oncologist with the family as recorded in the patient files. The date of the start of chemotherapy was defined as the starting date of the intended treatment protocol. Date of triage, information provision (pediatric oncologist or oncofertility nurse practitioner (coordinator)) and, if applicable, date and content of the fertility preservation counseling were retrieved from the medical records. Timely was defined as before starting cancer treatment, with the exception of acute lymphoblastic leukemia (ALL) and non- Hodgkin lymphoma (NHL). For these patients triage and informing of the family is generally postponed to the moment of reaching complete remission (CR) or treatment arm allocation. In ALL and NHL patients this moment of complete remission (CR) harbors an added benefit for eventual future use of preserved tissue, because of decreased risk of harboring minimal residual disease or leukemic infiltration. In addition, in renal tumor patients in our center the timely triage and information moment is defined as after surgery when definitive treatment stratification, including optional radiotherapy, is defined. ## Statistical analyses Descriptive statistics are reported including 95% confidence intervals (CI) for the main findings. ## Ethical approval Ethical consent was obtained from the METC Utrecht (Medical Ethical Committee Utrecht) and the need for informed consent was waived for this retrospective observational part of the PEARL study (METC research file NL72115.041.19 version 2, METC-protocol number 19/783, Netherlands trial register number NL8192). Information was retrieved from the medical records of all newly diagnosed girls in the Princess Máxima Center in 2019. # Results In 2019, 261 girls with a median age of 8.4 years (range: 0.0–18.1) were newly diagnosed with pediatric cancer in the Netherlands. Two died within days of presentation prior to treatment allocation. Of the remaining 259, 228 (88.0% (95%CI: 0.835–0.914)) patients were timely identified and triaged. Of the 31 patients who had not been timely identified and triaged, 28 were retrospectively classified as LR and 3 as HR. Characteristics of triaged patients seem representative (S4 Table in). Combining the triaged and the non-triaged patients the 259 patients were classified as LR, IR and HR of gonadal damage in 179 (69.1% (95%CI: 0.632–0.744)), 32 (12.4% (95%CI: 0.089–0.169)) and 48 (18.5% (95%CI: 0.143–0.237)) cases respectively. Risk of expected gonadal damage had been communicated timely with parents and child in 99/228 (43.4% (95%CI: 0.372–0.499)) patients out of the aimed 100% of patients to inform. Of these 42/151 (27.8% (95%CI: 0.213–0.354)) were LR, 28/32 (87.5% (95%CI: 0.719–0.950)) IR and 29/45 (64.4% (95% CI: 0.498–0.768)) HR. Reasons for not informing timely triaged patients were available for 125/129 cases, including 14 HR patients. Poor prognosis (n = 13), low risk of gonadal damage (ALL n = 19, renal tumors n = 11, resection only n = 28, other n = 20), psychosocial issues (n = 9), wait-and-scan regimens: no therapy (n = 14) and palliative setting at diagnosis (n = 3) were recurrent reasons for not informing these timely triaged patients. Two patients in the HR group were initially classified as LR with resection only. However, after treatment intensification their risk shifted to HR and subsequently these patients had unfortunately not been informed again. Of 42 LR timely informed patients, 5 families requested and received further counseling by a fertility expert. Fertility preservation was not advised nor pursued in these 5 patients. Of 45/228 (20%) triaged HR patients, 29 families had received personalized information. The reasons why 16 HR patients did not receive information after triage are depicted in. In all 29 informed HR cases counseling by a dedicated oncofertility gynecologist was strongly advised and 25 (86.2%) (of the aimed 100% counseled HR patients) appreciated such counseling. The 4/45 HR families that did not wish referral for counseling felt that they had been sufficiently informed (three were eligible for preservation, but did not wish preservation or counseling and one girl was very young (\<6 months)). In 16 out of the 25 HR girls counseling led to an attempt to preserve fertility. One patient underwent the combination of OC+OTC-o (n = 1). This patient initially had time to delay treatment for oocyte harvest. However, due to the limited number of harvested oocytes and insufficient time to perform another cycle, additionally OTC-o was performed. In five patients fertility preservation was performed prior to the start of cytotoxic treatment, eleven preserved gonadal tissue during treatment (S1 Fig). No intra-operative or post-operative complications, such as wound infections, bleeding requiring transfusion, reoperation, ICU admittance or mortality were reported. Reasons for not preserving ovarian tissue in HR patients mainly included uncertainty about the success of future auto-transplantation or use of the tissue. The option of OC in young mothers (\<40 years) of pediatric cancer patients for future oocyte donation was mentioned during counseling (S2 Table in). In 2019 none of the mothers pursued this option. # Discussion This retrospective study in a national cohort evaluated our first year of centralized oncofertility care in girls with cancer. It aims to search further improvement of quality of care. In the first full year this risk stratification seems to have led to timely identification and triage of 87.4% of all newly diagnosed girls with cancer using a protocol(-arm) based oncofertility gonadal risk stratification tool. Fertility preservation in 16 patients did not cause delay in the onset of oncological treatment. Of the 31 not timely identified and triaged patients, 28 were LR and three were HR patients. Hence, although improvement is necessary, identification of most HR patients seems feasible. To ensure timely provision of information on potential gonadal damage to all patients, we experienced the importance of daily and central coordination of identification and triage. For that purpose we built a logistic administrative system. We learned that appointing a dedicated oncofertility nurse practitioner who coordinates the navigation of all newly identified patients in the hospital is of utmost importance. Subsequently, patients can be navigated through the oncofertility triage system in due time, mostly before starting cancer treatment and in close communication with the multidisciplinary team (S2 Table). Triage can be a complicated effort, which lies just outside the main priorities and expertise of pediatric oncologists. Therefore, we developed a standardized gonadal damage risk stratification tool, based on international protocol(-arms), used at that time in our country. This was deemed instrumental. This tool improved over time and the most recent, currently used gonadal damage risk stratification tool is available in the S5 Table in. We identified and triaged 82.2% of all girls aged 13 years and older and timely informed 49.3% of them. Unfortunately, our retrospective study shows that 16 HR patients had not been informed after triage. Several reasons were reported such as very poor expected outcome, serious (co-)morbidity, psychosocial challenges and young age. The three clinical practice guidelines of the American Society of Clinical Oncology contain evidence-based recommendations for fertility preservation for patients with childhood cancer. A study of compliance with these guidelines reported that none of the 136 patients above the age of 13 had been counselled for fertility preservation. This illustrates how difficult the oncofertility care logistics process can be in real life clinical practice. Evidently, the IR and HR groups are the most relevant group to inform timely on their gonadal damage risk, as these patients might be eligible for fertility preservation. Nevertheless, providing information also to LR patients has been shown to be of value for survivors and family. Previous surveys have shown that many patients and families worry about future infertility at diagnosis already. Compared with available literature our percentage of informed patients is reasonably high, even though, we did not reach our aimed 100% informed patients. More effort is needed in the coming years to ensure that all patients with an acceptable cure rate are at least informed about their fertility. Preferably, HR patients will also be referred for counseling to a fertility expert (gynecologist) to explore the opportunities of fertility preservation. We learned that as treatment is sometimes intensified (n = 2), gonadal damage risk may increase and patients may need to be re-triaged, re-informed and that counseling may need to be reconsidered. For that purpose, presence of an oncofertility nurse practitioner (coordinator) at the multidisciplinary tumor board can enhance awareness in the oncology team. Why a large proportion (mainly LR) was not timely informed was not always documented. Hence, we learned it is important to facilitate a standard documentation process in the summary part of medical records stating whether oncofertility information is provided, and if not the reason why. This is now standard in our current practice. This is consistent with the recently published consensus of the international guideline harmonization group (IGHG), which stated that all childhood cancer patients and their families have the right to be informed regarding their gonadal damage risk. As indicated, adjustments have already been made in our current standard care oncofertility plan. We integrated recent recommendations from the IGHG guideline that classifies a CED score of 6000mg/m2 as high risk instead of the 8000mg/m2 which we used in 2019 in our gonadal damage risk stratification tool (S5 Table). Additionally, low risk does not mean no risk, as patients with a very low risk may experience infertility after cancer treatment. We should take into account that individual susceptibility and genetic variation may also influence individual risk of gonadal damage. The optimal moment to provide information regarding gonadal damage risk has been discussed in our oncofertility working group. Previous studies showed that for both patients and parents the preferred moment is at diagnosis, before cytotoxic medication is applied. This corresponds to our aim to inform all girls at diagnosis as intervention is still feasible then. However, we discovered that for female ALL and NHL patients the moment of triage and information provision can be postponed to the moment of CR or treatment arm allocation. We amended this early in the implementation phase of the oncofertility care plan. In addition, over time, for children with renal tumors (with the exception of full blown ruptured patients), we postponed the information process until after surgery (4–6 weeks), as the final gonadotoxic treatment stratification takes place based on histological stage and subtype. More recently, we learned that choosing the moment of ovarian preservation in patients with large abdominal tumors, such as neuroblastoma, 3–6 weeks into treatment may be beneficial for surgical and safety reasons, despite the adverse gonadotoxic influence of 1 or 2 courses of chemotherapy. Even though all patients may request additional counseling by experts, we learned this opportunity is not always utilized. Of 45 HR patients only 25 were counseled and of these 25 patients, only 16 chose to preserve gonadal tissue. OTC-o was the most common procedure. Oocyte cryopreservation before cancer treatment was no option for most girls in our cohort due to young age and/or lack of opportunity to delay oncologic treatment ( and S1 Fig). As no adverse events occurred, we consider OTC a safe procedure although our numbers are obviously limited. From previous reports, only limited information is available on the safety of OTC. In 2019, we chose not to not perform OTC-p to avoid previously reported bleeding risks and as the majority of our population was very young with small ovaries. Although age is no absolute contra-indication for OTC, we are hesitant to perform OTC in children under the age of 1 year in our center based on the published suggested potential higher anesthesia risk in infants. So far, we did not perform ovarian tissue cryopreservation in infants under the age of 1 year. Nevertheless, evidence for this anesthesia risk in laparoscopic procedures is not strong. The risk of gonadal damage will therefore always need to be weighed against the risk of direct toxicity for individual patients. Thus to infants who will, with no doubt, receive high dose HSCT (e.g. Juvenile myelomonocytic leukemia (JMML)) or high dose total-abdominal radiotherapy (e.g. after extensive rupture at presentation in renal tumor or neuroblastoma patients), counseling will be offered and OTC seriously considered. Decisions to perform OTC were always based on shared decision making. One of the main reasons for deciding against the OTC option was the communicated uncertainty of success of future auto-transplantation. Although studies in adults have shown promising results, ex-vivo maturation of ovarian material harvested during childhood is still not pursued. Future auto-transplantation of ovarian tissue from children is still considered experimental, in contrast to OTC, which is now considered standard care. This is explicitly explained to patient and parents during fertility counseling in our hospital (S2 Table). Future research on auto-transplantation of ovarian tissue harvested in prepubertal girls will shed more light on the effectiveness of, and may lead to more patients opting for, OTC in the future. Alternatively, OC after finalizing cytotoxic treatment is a feasible option for patients older than 16 years. This can be done starting 1 year after the end of treatment and with a sufficient ovarian reserve. However, there is substantial evidence that patients with excessive doses of alkylating agents, local irradiation, or following HSCT already have diminished ovarian reserve. It is conceivable that they may not benefit from such procedures. Thus, also for these patients preventive strategies at diagnosis, as used in our oncofertility plan, have a higher chance of creating fertility options in the future. Even though most young mothers are informed about the option of OC, none pursued this option in 2019. This may be influenced by the fact that the costs are not covered by insurance companies, as they qualify as “social freezing”. However, we did not investigate the reasons in this retrospective study. When this option is pursued the oocytes are stored under the mother’s name to prevent children to feel obliged to use oocytes of the mother. When age permits, mothers are advised to freeze their oocytes after the end of cancer treatment of their daughter and thus at a less stressful time. To improve our oncofertility care, the prospective part of the PEARL study currently evaluates the oncofertility care from a patient and parent point of view. It will explore in depth the reasons to preserve or not and the effect of ovarian tissue cryopreservation on the ovarian reserve. Patients’ and parents’ recall of fertility information in cancer survivors is known to be limited. We also aim to evaluate whether information provided by the pediatric oncologist or the dedicated nurse practitioner (coordinator) is deemed sufficient by LR and selected IR patients. The prospective study will further analyze whether the provided information is consumed, comprehended, and still remembered at the moment of discontinuation of therapy. Furthermore, we will analyze whether an information moment at the end of treatment would be a welcome addition to the quality of care of individual patients. In the prospective study the effectiveness of a protocol(-arm) based oncofertility gonadal risk stratification tool will also be evaluated. # Conclusion Our study suggests that it may be valuable and clinically feasible to timely identify and triage 87% of all newly diagnosed girls using a protocol(-arm) based oncofertility gonadal risk stratification tool. OTC seems a safe procedure and shared decision making led to a highly selected subgroup of patients for OTC. However, the safety of OTC needs to be confirmed in large prospective studies. In our center implementing oncofertility care did not cause delay in the onset of cancer treatment. We will continue to use the adjusted oncofertility care plan and evaluate this in the prospective PEARL study which started in 2020. We hope that our oncofertility care plan and risk stratification tool may be of use to other pediatric oncology institutes. # Supporting information We acknowledge J.I. Geller, pediatric oncologist and O. Frias, fertility patient navigator, from the Cincinnati Children’s Hospital, USA, for their lively discussions and their willingness to critically review our oncofertility care plan in 2018. [^1]: The authors have declared that no competing interests exist.

# Introduction The Affordable Medicines Facility- malaria (AMFm) was hailed as ‘one of the most important recent advances in fighting malaria’, designed to expand access to effective antimalarials in the public and private sectors. An estimated 3.3 billion people are at risk of malaria, with 80% of cases in sub-Saharan Africa. Artemisinin combination therapies (ACTs) are widely regarded as the best treatment for uncomplicated malaria. However use remains low, reflecting both unreliable public sector supplies and low availability and high prices of ACTs in the private sector, which has an increasingly important role in treatment seeking for malaria–. These factors lead patients to use older, less effective antimalarials such as sulfadoxine-pyrimethamine (SP) and amodiaquine. There is also concern about the use of oral artemisinin monotherapies, which may contribute to the development of artemisinin resistance. The AMFm, an ACT subsidy mechanism, was set up by the Global Fund to Fight AIDS, Tuberculosis and Malaria to address some of these barriers to ACT access. AMFm was launched in 2010 as eight national scale programmes in seven countries, comprising Ghana, Kenya, Madagascar, Niger, Nigeria, Tanzania and Uganda (mainland Tanzania and Zanzibar were considered as separate pilots). AMFm aimed to increase availability, market share and use of quality-assured ACTs (QAACTs) while reducing prices, thereby increasing coverage. This was also intended to result in ‘crowding out’ of other antimalarials from the market. AMFm consisted of three main elements: negotiations with QAACT manufacturers to reduce prices; a copayment made by the Global Fund to these manufacturers for every purchase made, representing 80–99% of the factory gate price; and supporting interventions to increase ACT awareness and appropriate use. AMFm could operate through the public, private for-profit and private not-for-profit sectors. Eligible importers, termed first line buyers, placed orders with approved manufacturers. Orders were then forwarded to the Global Fund AMFm Secretariat for approval. All copaid QAACTs had a green leaf logo on the packaging for identification as a subsidised good quality antimalarial. Despite these bold aims, AMFm has remained controversial, with concerns that the reduced price would still present access barriers for the very poor, that the subsidy would not be transmitted to the retail level and therefore not benefit the intended recipients, and that low levels of diagnostic coverage would lead to poor targeting and overtreatment of parasite negative individuals–. ## Malaria treatment in mainland Tanzania In Tanzania, about 90% of the population is at risk of malaria. The Tanzanian government introduced the ACT artemether-lumefantrine (ALu) to replace SP as the first line drug for treatment of uncomplicated malaria in 2006, with quinine as second line treatment. According to national guidelines, ACTs are provided free at public facilities for under fives, pregnant women, the elderly and those who cannot afford to pay, although these exemptions are sometimes not fully implemented. Oral artemisinin monotherapies have been banned since 2008, while non-oral artemisinin monotherapies are allowed for treatment of severe disease. ACTs are designated as prescription only medicines (POMs), while SP is over the counter (OTC). Pharmacies are allowed to stock POMs, while Duka la Dawa Baridi (DLDB, meaning ‘drug store’ in Kiswahili) are only allowed to stock OTC drugs. Since 2006, Tanzania has been in the process of upgrading DLDBs to Accredited Drug Dispensing Outlets (ADDOs), where dispensers undergo a 35 day training course and are allowed to dispense a limited range of POMs including ALu and quinine. A few general stores and kiosks also stock antimalarials, although this is not permitted. Diagnosis of fever cases in public health facilities was mainly based on symptoms alone until the Government changed the guidelines in 2010 to require parasitological confirmation for treatment of malaria. Malaria rapid diagnostic tests (mRDTs) were rolled out in public health facilities in Tanzania between 2010 and 2012. Drug retailers are not allowed to stock mRDTs. ## AMFm implementation The AMFm grant agreement between the Global Fund and the Tanzanian government was signed in August 2010. By December 2011 ten private sector first line buyers were registered in mainland Tanzania, of which five had placed orders with manufacturers. The first copaid drugs for the private for-profit sector arrived in October 2010. By the end of 2011, about 8 million ACT packs had arrived in country for this sector and by the end of 2012 an additional 16.6 million doses had been delivered. A maximum recommended retail price (RRP) of 1,000 Tanzanian Shillings (TSh) (\$0.64) was set for an adult dose in private for-profit outlets. The Medical Stores Department was registered as the first line buyer for the public sector. Public sector orders and deliveries were delayed, and only started arriving in July 2011. By December of that year 4.9 million doses had been delivered to the public sector, and by the end of 2012 a further 4.9 million doses. Additional public sector supplies were provided by the US President's Malaria Initiative, which donated 6.5 million ACT doses in 2011 and 4.7 million doses in 2012. Despite these additional supplies, high stockout levels in public health facilities were reported during AMFm implementation, although these were similar to stockout levels prior to AMFm. AMFm supporting interventions began in January 2011 in mainland Tanzania, and included use of national level mass media and community level communications, to raise awareness of the copaid drugs, the RRP and the green leaf logo. The mass media campaign began after the national launch in April 2011, and included TV and radio adverts and printed material. Community level communications, including mobile video units, road shows, clinic shows and school activities were implemented through community health workers and community based organisations, and restricted to two districts per region due to budget constraints. Training was the other major supporting intervention implemented in mainland Tanzania, focusing on the continued upgrading of drug stores to ADDOs. Upgrading had been occurring region by region since 2006. Prior to AMFm implementation, ADDOs had been introduced in eight of Tanzania's then 21 regions. Tanzania had 21 regions at the time of the study, though this has subsequently been increased to 25. An additional six regions were covered by the end of 2011, and a further six by the end of 2012, with progress slower than planned due to delays in disbursements from the Global Fund. In addition, a one day re-training programme covering malaria, Integrated Management of Childhood Illness and family planning was implemented in ADDOs in two regions in August to September 2011. Other smaller scale supporting interventions comprised strengthening of pharmacovigilance and monitoring and evaluation activities. This paper explores the impact of AMFm from both the supply and demand sides, specifically assessing changes between baseline and endline (pre and post AMFm implementation) in the following five areas: - Availability of QAACTs - Price of QAACTs - Market share of QAACTs - Choice of provider for treatment of fever - ACT use Supply side data are drawn from outlet surveys conducted in 2010 and 2011 in all regions of mainland Tanzania as part of the multi-country Independent Evaluation of AMFm. Demand side data are drawn from household surveys conducted in 2010 and 2012 in three regions of Tanzania with varying malaria transmission (Mwanza, Mbeya and Mtwara), as part of the IMPACT2 project ([www.actconsortium.org/IMPACT2](http://www.actconsortium.org/IMPACT2)). # Methods The study had a non-experimental, before and after design. As AMFm was implemented nationally there were no comparison areas. Outlet surveys were conducted using methods adapted from the ACTwatch project. Baseline outlet survey data collection took place from September to November 2010, and endline data collection from October 2011 to January 2012. 49 wards were randomly selected at baseline and endline, respectively, with probability proportional to population size, stratified by urban/rural location. One ward had to be dropped at baseline so data from 48 wards was analysed. Wards were designated as urban or rural using National Bureau of Statistics Census classifications, with mixed wards classified as urban if more than 70% of the ward was classified as urban. In each selected ward, every outlet with the potential to sell antimalarials was visited, including public and private health facilities, pharmacies, ADDOs, drug stores, general stores, kiosks and community health workers. Outlets were identified using official lists from district and national authorities, by consulting with district pharmacists and other local leaders, and by driving or walking down every street within a ward to locate all outlets. In large wards with a population over 30,000 people, wards were segmented, and one or more segments of the ward were randomly selected for the survey. As pharmacies were relatively rare but thought to be a major source of treatment, they were oversampled, whereby all pharmacies in the district in which the selected ward was located were visited. The sample size was calculated to detect a 20 percentage point change in outlets stocking a QAACT between baseline and endline, in urban and rural domains, with 80% power, 5% significance and at baseline assumed a design effect of 4 and 40% availability of QAACTs. Using these criteria, 305 outlets which stocked antimalarials were required in both urban and rural domains at baseline and endline. Estimates of the average numbers of outlets per ward were used to estimate the number of wards required to reach this number of outlets: 9 urban and 39 rural at baseline. This was then adjusted at endline to 20 urban and 29 rural wards, reflecting baseline estimates of availability and the higher degree of clustering observed at baseline in urban areas. Screening criteria were used to identify outlets with an antimalarial in stock at the time of visit or within the previous three months. Following verbal consent, a questionnaire was conducted with the most senior staff member present, in Kiswahili, with data collected using Personal Digital Assistants (PDAs). Questions about outlet characteristics were asked, and details about every antimalarial in stock at the time of visit were recorded, as well as the volumes of each product sold in the past seven days. Antimalarials were classified as quality-assured ACTs (QAACTs), non-quality- assured ACTs, artemisinin monotherapies (broken down by oral and non-oral forms), or non-artemisinin therapies such as SP, chloroquine, mefloquine and amodiaquine. QAACTs are ACTs that comply with the Global Fund's quality assurance policy. Price and market share data were calculated using Adult Equivalent Treatment Doses (AETDs), the amount of a drug needed to completely treat a 60 kg adult. For example, to calculate the price per AETD of a paediatric package of ALu with 6 standard tablets (20 mg artemether and 120 mg lumefantrine), the price would be multiplied by 4 to calculate what it would cost for an adult equivalent dose of 24 tablets. The household surveys took place in Mwanza, Mbeya and Mtwara regions, which have varying malaria transmission and epidemiology. Baseline household data collection took place from June to October 2010, and endline from May to September 2012. Mwanza is located next to Lake Victoria, Mtwara is in the south on the coast, and Mbeya is located in the southern highlands. In 2011–2012 malaria prevalence among children aged 6–59 months was 18.6% in Mwanza, 17.4% in Mtwara and 0.5% in Mbeya. Based on the national distribution of socio-economic status, Mbeya had the lowest percentage of people living in the lowest wealth quintile of the three regions, at 7.7%, compared to 20.8% in Mwanza and 35.5% in Mtwara. mRDT roll out in public facilities took place in all three regions between the baseline and endline household surveys, in early 2011 in Mwanza and Mbeya, and in mid-2012 in Mtwara. The ADDO programme had been implemented in Mbeya and Mtwara regions by baseline, and was implemented in Mwanza region after endline household data collection. In each region, 80 enumeration areas (EAs) were randomly selected using probability proportional to population size. In selected EAs, every household was visited and mapped using Global Positioning System. After obtaining a list of all the households within an EA, 32 households were randomly selected while in the field, 24 to be visited first, and the remaining eight to be included sequentially as replacements if any of the initial 24 households were unavailable or refused to be interviewed. At baseline, the maximum number of households visited per enumeration area was 32 in Mwanza, and 24 in Mbeya and Mtwara. At endline the maximum was 24 in all regions. The sample size was calculated to detect a 10 percentage point increase in children under five with a febrile illness who obtained ACT within 24 hours of fever onset. With 80% power, 5% significance and an assumed design effect of two, 480 children under five with fever were required in each region. The questionnaire was translated into Kiswahili, and data were collected using PDAs. Questions on household demographics were asked of the household head, and questions on history of fever were asked of all household members. All members who reported fever in the 14 days prior to interview were asked about treatment sought and drugs and blood tests obtained. The guardian was interviewed on behalf of children below 12 years old. In addition, a finger prick blood sample was taken from all consenting members, from which an mRDT was performed (ICT Diagnostics, Cape Town, South Africa). Written consent was obtained from everybody who was interviewed or had blood taken, or from their guardian. Data analysis for the outlet and household surveys was performed using Stata version 11 (College Station, Texas). Stata survey procedures were used to account for the survey design and stratification. Changes in availability and use were assessed using the design based F-test. R version 2.14.2 was used in outlet survey analysis for obtaining p-values for the change in retail prices using the Wilcoxon Rank Sum test. 2011 and 2012 prices were converted to 2010 prices using the Tanzania consumer price index. Prices were converted to US dollars using the average interbank rate for 2010. Socio-economic status quintiles were calculated using principal components analysis, based on the first principal component, and using standard Demographic and Health Survey variables. In addition, key informant interviews were conducted at national, regional and district level to capture information on the process of AMFm implementation and relevant contextual factors. Interviewees were drawn from government bodies such as the National Malaria Control Programme and the Tanzania Food and Drug Authority, non-governmental organisations such as the Clinton Health Access Initiative, organisations implementing supporting interventions, as well as regional and district medical officers and other local staff. We draw on the information gathered to inform the Discussion section of this paper. Ethical approval was obtained from the Institutional Review Board of Ifakara Health Institute and the London School of Hygiene and Tropical Medicine Research Ethics Committee for both the outlet and household surveys, including collection of blood samples during the household survey, and also from the Institutional Review Board of ICF International for the outlet survey. The investigator from the US Centers for Disease Control and Prevention provided technical assistance for the household survey but was not actively engaged in data collection for either survey. # Results ## Antimalarial availability, price and market share Of all outlets visited, 709 met the screening criteria at baseline and 799 at endline. 93.0% and 99.9% of outlets that met the screening criteria were interviewed at baseline and endline respectively. Outlets were classified as: public health facilities (dispensaries, health centres and hospitals); private health facilities (for-profit and not-for-profit dispensaries, clinics and hospitals); specialised drug sellers (SDSs) (pharmacies and drug stores, including DLDB and ADDOs); and general retailers (general stores and kiosks). Private not-for-profit and private-for-profit health facilities were pooled due to low numbers obtained. Only one community health worker stocked an antimalarial at baseline and none at endline, so results for this subgroup are not presented separately but are included in total estimates. In total, 15.8% and 14.0% of outlets visited stocked an antimalarial at baseline and endline, respectively (p = 0.31). Antimalarial availability was over 80% at baseline and endline in public health facilities, private health facilities and SDSs, while general retailers had the lowest availability of 4.3% at baseline and 0.9% at endline (p = 0.024) Of outlets stocking antimalarials, the percentage with at least one staff member with a health related qualification was above 98% in public and private health facilities at both baseline and endline, and increased from 89.7% to 97.1% in SDSs (p = 0.002). ### Availability of QAACTs Availability of specific antimalarial categories was calculated out of outlets stocking any antimalarials. QAACT availability increased from 25.5% among all outlets at baseline to 69.5% at endline (p\<0.001). This was mainly due to the substantial increase in SDSs, where availability increased from 12.8% to 69.6% (p\<0.001). Availability in general retailers increased from 4.3% to 20.6% (p\<0.001). No change was seen in QAACT availability in public health facilities, which was 80.1% at baseline and 81.4% at endline (p = 0.86). A decrease in the availability of non-artemisinin therapies in public health facilities led to a small but significant decrease overall (98.4% to 94.8%, p = 0.020). There was a small increase in availability of non-quality-assured ACTs (14.2% to 25.3%, p = 0.046), which was not significant in any one outlet type. The first line drug ALu accounted for 94.8% of QAACTs and 4.7% of non- quality-assured ACTs found in stock at baseline, and 94.4% of QAACTs and 9.0% of non-quality-assured ACTs at endline. ### Price of QAACTs The median price of QAACTs was zero in public and private facilities as they were supposed to be provided free in all public and most private not-for-profit facilities. Median price data for SDSs are shown in, for tablets only to enhance comparability across antimalarial types as QAACTs are only available in tablet form. The median price of QAACT tablets in SDSs fell from \$5.63 to \$0.94 (p\<0.001). Non-artemisinin therapy tablets had a median price in SDSs below \$1 at baseline and endline, and at endline the median price of QAACT tablets was equal to that of non-artemisinin therapy tablets. The median price of non- quality-assured ACT tablets also decreased in SDSs from \$7.92 to \$6.87 (p = 0.025) though the magnitude of the decrease was not as great as that for QAACTs. ### Market share of QAACTs The market share of QAACTs as a percentage of all reported antimalarial sales increased from 26.2% to 42.2% among all outlet types combined (p = 0.033), with a particularly large increase in SDSs from 2.2% to 34.0% (p\<0.001). No significant changes were seen in QAACT market share in public or private health facilities. The market share of non-quality-assured ACTs did not change significantly and the market share of artemisinin monotherapies was minimal, below 1.0% in all outlet types at baseline and endline. Therefore the increased QAACT market share was at the expense of non-artemisinin therapies, for which the market share decreased by 13 percentage points among all outlet types combined. ### Urban/rural variation in availability, price and market share When broken down by urban and rural areas, large increases were seen in QAACT availability in both urban and rural areas overall, and especially in SDSs. At baseline, there was higher availability of QAACTs in urban SDSs than rural SDSs (p = 0.028), but at endline there was no significant difference between the areas. Decreases in median QAACT prices were seen in both urban and rural areas, although greater reductions were seen in urban areas as there were few QAACTs in rural areas at baseline and the majority of these were very inexpensive. The increase in QAACT market share was almost completely due to changes in rural areas, where market share increased by 25.4 percentage points, compared with 1.8 percentage points in urban areas. However, significant increases were seen in market share in SDSs in both urban and rural areas. ## Treatment sought and obtained for malaria 5,423 and 5,511 households were interviewed at baseline and endline, respectively, involving 20,874 and 20,102 full interviews of household members or their guardian. Over four fifths of household heads worked in agriculture, and about half had completed primary school. Overall parasite prevalence among all age groups according to study mRDT results was 17.5% at baseline and 12.0% at endline (p\<0.001). ### Choice of provider for treatment of fever Overall 69.5% of people with fever sought care at baseline and 73.6% at endline (p = 0.074), with seeking care defined to include care sought outside the home and drugs obtained from home/a neighbour. Only 3.8% of people with fever at baseline and 2.7% at endline sought care in more than one place. There was a marked shift from baseline to endline away from public health facilities and general retailers and towards SDSs overall and in each age group, with an increase from 41.3% to 54.1% in people visiting SDSs (p\<0.001) and a decrease from 25.3% to 16.8% in people visiting a public health facility (p\<0.001) as the first source of care overall. The change in use of SDSs and public health facilities was considerable in both age groups, with the increase in SDS use being highest among under fives, from 28.6% to 48.3% (p\<0.001). ### ACT use Overall the percentage of people with fever who obtained an antimalarial was 40.6% at baseline and 35.9% at endline (p = 0.070), and did not change to an important degree in either age group. The percentage of people with fever who obtained an ACT was 20.7% at both baseline and endline (p = 1.00). There was weak evidence for an increase in the percentage of people who obtained an ACT, out of those who obtained an antimalarial between baseline (51.0%) and endline (57.7%) (p = 0.077), corresponding with the slight fall in all people with a febrile illness getting an antimalarial and no change in those getting an ACT. Other marked changes were seen in these three core household indicators when broken down by where the patient sought care. shows the data separately for visits to public health facilities and SDSs, the two sources where the majority of people sought care. In public health facilities the percentage of people obtaining an ACT decreased from 57.4% to 46.1% (p = 0.035), but there were no marked differences in the percentage of people obtaining an antimalarial, or an ACT out of those obtaining an antimalarial. The percentage of people visiting SDSs obtaining an antimalarial decreased from 60.5% to 51.0% (p = 0.026), while the proportion obtaining an ACT increased from 18.5% to 26.9% in these outlets (p = 0.015). The percentage obtaining an ACT out of those who obtained an antimalarial increased from 30.6% to 52.7% in SDSs (p\<0.001). When broken down by socio-economic status there was no substantial changes in any of the three use indicators in any quintile. ### Use of diagnostic tests The percentage of people obtaining a blood test was 17.6% at baseline and 21.4% at endline overall (p = 0.78), with no major change in any age group. In public health facilities the percentage of people who obtained a blood test increased from 28.7% to 46.6% (p\<0.001), while in SDSs there was weak evidence of a decrease (4.1% to 2.1%, p = 0.061) ### Price paid for ACTs Households were also asked about the price they had paid for antimalarials. The median price paid for ALu tablets from SDSs at baseline was \$1.41 (interquartile range (IQR) \$0.85 to \$1.88) per AETD, and at endline was \$1.29 (IQR \$0.86 to \$2.15). At endline, the median price of ALu with the green leaf logo from SDSs was \$1.08 (IQR \$0.81 to \$2.15) and without the logo was \$1.29 (IQR \$1.08 to \$2.15). At baseline 68% of ACTs from public health facilities were reported to have been obtained for free and 63% at endline (p = 0.53). Of drugs not obtained for free, the median cost was \$0.70 at baseline and \$1.08 at endline in this sector. ### Regional and urban/rural variation in provider choice and ACT use Somewhat different treatment seeking patterns were seen in rural and urban areas and across the three regions visited. In rural areas and in Mbeya and Mtwara regions a shift away from public health facilities towards SDSs occurred, but in Mwanza and in urban areas this was not significant, reflecting the high usage of SDSs at baseline. The proportion of people with febrile illness who obtained an antimalarial decreased significantly in Mtwara, and in rural areas, while the proportion of people obtaining ACT and the proportion obtaining ACT out of those obtaining an antimalarial did not change to a substantial degree in any setting. # Discussion This paper has presented results from large scale baseline and endline outlet and household surveys to explore the impact of AMFm from both a supply and demand perspective. The paper demonstrates how very different inferences can be drawn from looking at each side of the market, and therefore the importance of a holistic approach to evaluation. Before discussing the results further it should be noted that the surveys had several limitations. They were both based on reported data from providers and community members, and therefore potentially subject to recall bias. In addition, outlet staff may have biased their answers to make them more socially desirable or due to fear of regulatory consequences. For example, they may have concealed certain POMs, including ACTs, if they were concerned that their outlet was not authorised to sell them, especially at baseline before ACTs were widely promoted in the private sector. In addition, they may have stated lower retail prices than they actually charge, especially at endline when the RRP was widely publicised. Similarly, in the household survey, respondents may have overstated their use of ACTs if they knew this was the ‘correct’ answer, or stated that they went to a public health facility instead of another source. Another factor to consider is the different timings and geographical coverage of the surveys. The outlet surveys were conducted between September 2010 and January 2012, in all regions of mainland Tanzania, while the household surveys were conducted between June 2010 and September 2012, in three regions only. The endline outlet survey was conducted relatively early on during AMFm implementation: twelve months after subsidised drugs first arrived and five months after the communication campaigns commenced, while the household survey was conducted nine months later. It is unclear whether a longer time period before endline would be associated with greater impact as the programme became more established, or less impact as the effect of initial training and communications waned. Finally, the evaluation was based on a before and after study design. Lack of control areas means that it is challenging to assess what would have happened in the absence of AMFm and to what extent the changes seen are attributable to the programme. Below we draw on findings from key informant interviewers to address these issues where possible. Key informants indicated that there were no major delays in obtaining copaid drugs by private sector buyers and the supply chain from manufacturer to buyers generally operated smoothly, after an initial slow start in early 2011. The outlet surveys showed that in the private sector QAACT availability and market share increased while retail price decreased after AMFm implementation, with the household surveys showing large improvements in ACT use among SDS customers. In these outlets, although the proportion of people getting an antimalarial decreased, those obtaining an ACT increased, resulting in a substantial increase in the proportion of people getting an ACT out of those getting an antimalarial. Key informants did not identify any other projects or contextual issues that could have been responsible for changes of this scale in the retail sector, implying that they were very likely due to AMFm. By contrast, outlet survey data showed no significant change was seen in QAACT availability or market share in the public sector. Household survey data showed that the proportion of people obtaining an ACT at public facilities actually fell. This likely partly reflected the persistence of ACT stockouts in public facilities, and the significant increase in blood tests between baseline and endline due to mRDT roll out. Health facility surveys in the same three regions in 2010 and 2012 also reported a significant decrease in the percentage of people with fever obtaining ACTs from 39.9% to 21.3%, along with an increase in the percentage of people obtaining a blood test from 15.9% to 55.8%. Key informants reported that the supply chain in this sector proved more problematic than that for the private sector, with severe delays in procuring and obtaining copaid drugs through AMFm channels. The 4.9 million doses of copaid QAACTs delivered by the end of 2011 were estimated to be only two to four months worth of supplies, and the extra doses donated by the US President's Malaria Initiative were still insufficient to prevent stockouts due to the low stock levels before AMFm, leading to similar QAACT availability in public health facilities at baseline and endline. These delays in receiving public sector copaid drugs were said to have been partly caused by delays in ordering due to the confusion over the new grant mechanisms, delays in approvals being granted, and delays in delivery owing to manufacturers' inability to meet the high global demand for drugs at the time. Despite the improvements seen on the supply side in the private sector, household survey data showed that the overall proportion of people obtaining an ACT during a febrile illness had not changed, with only one fifth of patients obtaining an ACT before and after AMFm implementation. There are two possible explanations for this apparent paradox. First, while ACT use increased in the private sector, there was a reduction in the proportion of public sector patients obtaining ACT, reflecting increased use of mRDTs, as described above. Secondly, the proportion of people obtaining an ACT remained higher in the public sector than the private sector at endline, so the shift in treatment seeking towards the private sector therefore also had a negative effect on the overall proportion of people getting ACTs. These two effects therefore cancelled out the increase in ACT use in the private sector. To clarify further, at baseline, 25% of visits for fever/malaria treatment were to the public sector, of which 57% resulted in an ACT being obtained, compared to 41% and 19% respectively in SDSs; at endline 17% of visits were to the public sector and 46% of these obtained an ACT, while this was 54% and 27% respectively in SDSs. The result was that at baseline, of people with fever, 14% visited a public health facility and obtained an ACT, while 8% visited an SDS and obtained an ACT; at endline, 8% of people visited a public health facility and obtained an ACT, and 15% visited an SDS and obtained an ACT. Overall, the proportion of people who visited a public health facility or drug store and obtained an ACT therefore remained unchanged at 22%. The market share of QAACTs overall increased significantly while the use of ACTs did not change, which could be affected by certain factors such as the different timings and geographical areas of the outlet and household surveys. At baseline the QAACT market share could potentially have been underreported which would have resulted in an overestimate of the change in market share. The observed shift in treatment seeking behaviour may have reflected a number of factors. Qualitative data collected as part of the IMPACT2 project indicated that community members were very aware of public sector drug stockouts, and as a result tended to bypass public facilities and seek treatment elsewhere. Coupled with an increasingly abundant supply of good quality, affordable medicines in the retail sector due to AMFm this could have encouraged people to make the private sector their first source of care. A continuation of the ongoing general expansion of the private health sector over time could also have played a part, although the shift seen over one year is likely too large to be attributed to this trend alone. The upgrading of DLDB to ADDOs might also be expected to increase the number of outlets selling POMs and therefore ACTs, encouraging more people to use the private sector. However, the shift in treatment seeking was seen in both ADDO and non-ADDO regions, and DLDB were also known to widely stock POMs illegally. Household survey data showed no change in the price paid for ALu tablets, as the prices paid at baseline were very low, and much cheaper than the QAACT prices reported in the outlet survey at baseline. A possible explanation for this is that some ALu tablets in SDSs at baseline may have been leaked from public health facilities, and sold very cheaply to household members, but concealed from interviewers during the baseline outlet survey. It is also likely that relatively costly ALu tablets available in the private sector at baseline were too expensive for most community members, and therefore rarely purchased or recorded in the household survey. A study in relatively remote areas of two Tanzanian regions in 2011–12 showed similar outlet survey results, with availability of AMFm subsidised ACTs increasing from 25% to 88% in ADDOs in Mtwara, and from 3% to 62% in Rukwa between February 2011 and January 2012. This was accompanied by a decrease in median price from \$1.03 to \$0.81. However, household surveys in the same region found a significant increase in ACT use among suspected malaria cases from 54.6% to 67.8%, mainly reflecting increased ACT use in the retail sector. A shift in treatment seeking towards the retail sector was seen, but in contrast to our findings, this reflected a shift from those not seeking care rather than from the public sector. However, results from the Tanzania HIV/AIDS and Malaria Indicator Surveys in 2010 and 2012 were consistent with our study findings, showing no change in the proportion of children under five who obtained an ACT out of all those with fever, or out of those who obtained an antimalarial. In comparison with the other seven AMFm pilots, mainland Tanzania was fairly successful in terms of outlet survey outcomes, with the third highest increase in QAACT availability, and the third largest fall in median QAACT price, but only the sixth highest increase in overall QAACT market share. Baseline and endline household survey data were available for four of the other seven AMFm pilots. Similar to mainland Tanzania, results from Zanzibar also showed large improvements in QAACT availability, affordability and market share in the private for-profit sector but no change in overall ACT use. As above, this likely reflected a reduction in the relative importance of the public sector as a source of antimalarials, with the share of all antimalarials distributed through the public sector falling from 37% at baseline to 13% at endline. However, these contrasting household and outlet survey results were not found in Uganda, Nigeria or Madagascar. Uganda experienced a large and significant increase in ACT use in under fives (24.0 percentage points), despite weaker performance on outlet survey outcomes. Nigeria and Madagascar saw smaller increases in ACT use of 6.7 and 5.0 percentage points respectively. Household survey data were not available for Kenya and Ghana, which experienced the strongest results for outlet survey indicators. The Global Fund has decided to integrate AMFm into core Global Fund grant management, after a transition period in 2013. This means that eligible countries will be able to use their core Global Fund resources to buy copaid drugs under the AMFm system, but this use of grant funds for ACT subsidies will compete with all other interventions for malaria, HIV/AIDS and tuberculosis. The Tanzanian government has committed to continuing with ACT subsidies, though concerns have been raised about poor targeting of copaid ACTs to those with malaria in the private sector. For example, Briggs *et al.* showed that after AMFm implementation in Mwanza and Mtwara, only 20% of people purchasing ACTs from DLDBs or ADDOs were parasitaemic. Strategies to increase diagnostic use in the private sector have received considerable attention in national and international policy discussions, with some stakeholders keen to build on experience from Cambodia where subsidised mRDTs have been distributed through the private sector for over a decade. The Tanzanian government is currently piloting the introduction of low cost and subsidised mRDTs in ADDOs, with the intention of improving targeting of subsidised ACTs to those with confirmed parasitaemia. This research has highlighted that other key policy considerations should include how to combine improved access to quality, affordable ACTs in the private sector while maintaining use of the public sector, particularly through improvements in public sector ACT availability. Ultimately the goal should be to have affordable, high quality testing and treatment available in both the public and private sectors. # Supporting Information The authors would like to thank ACTWatch for assistance on study design of the outlet surveys. Many thanks to the field teams, regional and district health officers and study participants. Thanks to Adam Wolkon for training and advice on PDA programming. Rebecca Thomson, Katia Bruxvoort, Matthew Cairns, Immo Kleinschmidt, Sarah Tougher, Barbara Willey, Kara Hanson and Catherine Goodman are members of the LSHTM Malaria Centre. Mark Taylor now works for the Faculty of Health Care and Social Work, University of Trnava, Slovakia. [^1]: The authors have the following interests: Co-authors Yazoume Ye, Ruilin Ren and Fred Arnold are employed by ICF International. There are no patents, products in development or marketed products to declare. This does not alter the authors' adherence to all the PLOS ONE policies on sharing data and materials. [^2]: Conceived and designed the experiments: RT CF AK KB ST MT YY RR FA KH SPK CG. Performed the experiments: RT CF BJ AK KB HN MT. Analyzed the data: RT CF ST MC IK YY AM BW KH CG. Wrote the paper: RT. Provided input on manuscript: CF BJ AK KB HN ST MC MT IK YY AM RR BW FA KH SPK CG.

# Introduction Human T-lymphotropic virus (HTLV) was the first exogenous human retrovirus discovered, and although three additional types of HTLV have since been described, only HTLV type 1 is found throughout the world. HTLV originated by transmission of simian T-lymphotropic viruses from non-human primates to humans, but only HTLV-1 and HTLV-2 are so far known to be transmitted from human to human. Although HTLV-1 can be further divided into seven subtypes (A-G), only the cosmopolitan subtype A is found world-wide. There is, as yet, very little data concerning the distribution of HTLV-2, a virus found in several native populations in North and South America and in pygmy tribes of Central Africa. However, HTLV-2 infection of people who inject drugs (PWID) suggests that this distribution is now broadening globally. HTLV types 3 and 4 have only so far been found in a few individuals in Africa. HTLV-1 is the causative agent of adult T-cell leukemia/lymphoma (ATL), HTLV-1-associated myelopathy/tropical spastic paraparesis (HAM/TSP) and HTLV-1-associated uveitis (HAU). HTLV-1 is also thought to cause local and systemic inflammation which, in addition to HAM/TSP for which the most data is available, probably leads to several other diseases collectively referred to as HTLV-1 associated inflammatory diseases (HAID). Data concerning HTLV-2 infection and diseases remains limited, but HTLV-2 also seems to be associated with HAM/TSP and other neurological disorders although the association with increased mortality from cancer remains controversial. A higher incidence of bladder or kidney infection and arthritis has been observed for both HTLV-1 and HTLV-2 infection, and HTLV-2 infected individuals have been shown to have a higher incidence of acute bronchitis and pneumonia than HTLV-seronegative individuals. There are three well-studied routes of human-to-human transmission for both HTLV-1 and HTLV-2: sexual contact, breastfeeding and blood products containing HTLV infected lymphocytes. Despite the fact that these routes of transmission resemble those for HIV-1, the global distribution of HTLV-1 is strikingly different. It is estimated that world-wide, 5–10 million people are currently infected with HTLV-1, but these are mainly concentrated in just a few areas of high endemicity such as sub-Saharan Africa, South America and the Southwestern region of Japan. Furthermore, clusters exist in the Caribbean and the Middle East, but near these regions of high prevalence there are areas with low endemicity. Within Europe, Romania is a country of relatively high HTLV-1 prevalence, with 5.33 of 10,000 first time blood donors testing positive. The reasons for this unbalanced distribution are not yet fully understood, but founder effect and the persistence of virus transmission by one of the three major routes seems likely. For example, Japanese programs to prevent mother-to- child transmission via breast-feeding resulted in a marked reduction of HTLV transmission. The storage and shipping of samples for subsequent detection of antibodies in epidemiological surveys can be simplified if dried blood spots (DBS) or dried serum spots (DSS) on filter paper are used. Numerous studies investigating various pathogens have shown DBS to be reliable for antibody detection and comparable to venous blood samples. DBS were successfully evaluated soon after the start of the HIV/AIDS pandemic and the method has since proven to be very useful for HIV-specific antibody screening. DBS were shown to be reliable for detecting anti-HIV-1/HIV-2 antibodies and serological markers of other pathogens, using either in-house or commercial assays to analyze the eluates. The use of DBS in the detection of anti-HTLV antibodies has also been validated in several studies. One early study initially established the methodology using simulated DBS, (i.e. known HTLV-positive serum (or plasma) samples diluted in HTLV-negative human blood cells) and then screened 10,135 DBS from neonates using the HTLV-1 gelatin particle agglutination test, confirming by ELISA and Western blot. An ELISA followed by Western blot is still recommended for HTLV diagnosis by the HTLV European network (HERN). Using this procedure, a panel of 23 serum samples provided by HERN and subsequently 126,010 DBS samples from pregnant women in the United Kingdom were tested for the presence of HTLV antibodies. HTLV antibodies could be reliably detected even in dried blood spot eluates derived from patients with serum titres as low as 1:50 in a modified anti-HTLV gelatin particle agglutination assay. In addition, several HTLV seroprevalence studies have been performed using DBS/DSS but without method validation. According to the available sparse data, the prevalence of HTLV infection in Germany within the overall population is very low. The rate among 58,747 neonates delivered in Berlin between 1996 and 1997 and screened by heel prick was found to be only 0.7 in 10,000 DBS, the lowest in Europe, although a screening of 248,000 sera from blood donors in the early 90's found a prevalence of 2.1 in 10,000. However, in addition to all German studies so far being somewhat out of date, only one included people with a known risk for blood-borne diseases (although relatively few). In addition to 100,852 samples from blood donors, this Frankfurt study screened 117 from hemophiliacs and 63 from local PWID but failed to identify a single, confirmed case of HTLV-1 infection. The first aim of the study presented here was to validate, in our hands, the use of DBS for HTLV testing using a small set of sera from known HTLV-positive individuals. The second was to determine the prevalence of HTLV in a German cohort of people who inject drugs (and therefore at high risk for blood-borne infections) using 2,077 DBS from PWID in eight German cities that had been collected using respondent-driven sampling as part of a multicenter sero- behavioural survey for HIV and hepatitis B and C. # Materials and methods ## Serological testing of DBS As part of a recent study using respondent-driven sampling, 2,077 DBS from PWID in eight German cities had been collected, with the number of DBS in the respective cities ranging from 130 (Leipzig) to 337 (Berlin). The capillary blood samples, collected by a trained person, were spotted with \~30μl per spot on filter cards (Whatman 903, GE Healthcare Life Sciences, Freiburg, Germany). The filter cards were dried for at least three hours or overnight at room temperature, then placed in plastic bags and sent to our laboratory. Antibodies were eluted in PBS containing 0.05% Tween-20 and 3% FCS by incubation overnight at 4°C, resulting in a 1:15 dilution with respect to the original blood volume. Eluates were stored at -20°C until testing for the presence of HTLV-specific antibodies according to a modified algorithm proposed by the HTLV European Research Network. First, eluates were tested individually with the “Murex HTLV I+II ELISA” (Diasorin, Dietzenbach, Germany) according to the manufacturer's instructions. Positive samples were tested again in duplicate with the same assay. For repeatedly positive samples, the intention was to contact the donor for a follow-up venous blood sample which would be used to repeat the ELISA and confirm seropositivity using the Western blot assay “MP Diagnostics HTLV Blot 2.4” (MP Biomedicals SAS, Illkirch Cedex, France) according to the manufacturer's instructions. If the participant was not available for a follow- up sample then stored DBS material from this person could have been used to confirm HTLV infection via PCR as described elsewhere. ## Validation of HTLV testing To validate HTLV testing from DBS, 14 randomly selected sera from individuals previously tested positive for HTLV antibodies were used. All subjects were asymptomatic carriers (AC) to reflect the expected situation within a group of non-hospitalized PWID. Characterized HTLV-1 and HTLV-2 sera for the Dried Blood Spot validation were additionally provided by the Communicable Diseases Research Tissue Bank of Imperial College Healthcare NHS Trust, London, UK. The proviral loads of these HTLV infected subjects ranged between 0.01% and 35%. Thirty μl of the undiluted sera as well as of three serial 10-fold dilutions in negative human sera were spotted onto two replicate filter papers. These DSS were then processed in the same way as the DBS, eluting the antibodies from the filters to give an additional 1:15 dilution before testing in ELISA. A confirmation Western blot of the highest dilution of each sample still reactive in ELISA was carried out as described above. ## Determination of HTLV prevalence among PWID in Germany Details of the study protocol have been published elsewhere. Briefly, between 2011 and 2014 a sero-behavioural survey was conducted among PWID in eight German cities in cooperation with low-threshold drug services to generate HBV-, HCV- and HIV- seroprevalence- and associated behavioral data among current injectors (DRUCK-study). Respondent-driven sampling was used to reach PWID who were not in regular contact with the low-threshold drop-in facilities or drug consumption rooms but were part of the social network of other participants. Inclusion criteria for the survey were (i) being at least 16 years or older, (ii) self- reported injecting drug use within the past 12 months in the respective city, (iii) willing to take part in a questionnaire assisted-interview and to provide a capillary blood specimen for serological and molecular testing, (iv) giving informed consent, and (v) not having participated in the study previously. DBS samples from all 2,077 PWID enrolled in the study were investigated. ## Ethics statement The Federal Commissioner for Data Protection and Freedom of Information approved the study protocol on 29/11/2012 (III-401/008#0035). Ethics approval was received on 19/11/2012 from the ethics committee at the medical university of Charité, Berlin (EA4/036/11). All adult subjects provided informed consent. No persons under the age of 18 were recruited and included in the study (although the minimum age for inclusion was 16 years). The median age of the participants within each of the eight cities varied from 29–41. Oral and written information on the study was provided to all potential participants. In most cases, participants signed thereafter a written consent form. In a limited number of cases, written consent was refused (for the reason of staying anonymous), but oral consent was obtained and given to the study site manager. In this case the study site manager signed the consent form in the presence of the participants as approved by the IRB. # Results ## Validation of HTLV testing To validate the use of DBS for HTLV screening, we tested sera from six known HTLV-1 infected individuals and eight HTLV-2 infected individuals with the diagnostic ELISA routinely used in our laboratory. In addition to the negative control sera provided with the assay, one serum from a healthy donor was included as control. To account for the probable different magnitudes of reactivity in individuals progressing to disease and asymptomatic carriers, serial dilutions up to 1:1000 were also tested. All HTLV-1 and HTLV-2 positive sera tested positive at the 1:100 dilution. Three HTLV-1 positive sera (H1\_#3, H1\_#4, H1\_#6) and three HTLV-2 positive sera (H2\_#2, H2\_#3, H2\_#5) were still positive at the 1:1000 dilution. The same sera (and dilutions thereof) were then used to create DSS to simulate the conditions and processing used for the study samples. As expected, the inherent 1:15 dilution resulting from elution of the sera from the filters reduced detectability. All DSS made with undiluted sera from the 14 HTLV positive controls were reactive. Three samples (H1\_#1, H1\_#2, H1\_#5) from the HTLV-1 panel (but none from the HTLV-2 panel) were already negative at the 1:10 pre-dilution. Two HTLV-1 positive samples (H1\_#4, H1\_#6) and two HTLV-2 positive samples (H2\_#3, H2\_#5) remained strongly reactive even at the 1:100 pre-dilution. All sera were then subjected to diagnostic Western blot to test this assay's suitability for confirmation. According to the manufacturer, a confirmed diagnosis requires reactivity to the two envelope proteins GD21 and rgp46-I or rgp46-II plus reactivity to one or both of the core antigens p19 and p24, or reactivity to both core antigens plus one of the envelope proteins. All but one sera fulfilled the criterion for HTLV-1 or HTLV-2 positivity (reactivity to GD21, rgp46-I or rgp46-II and p19 or p24), and other bands (e.g. p26, p28) were also sometimes visible at varying intensities. Only one serum (H1\_#3) demonstrated reactivity to GD21, p19, p24 and other bands but not to rgp46-I or rgp46-II, although this could still be confirmed to be an HTLV-1 infection because the reactivity to p19 was stronger than that to p24. After spotting on filter paper and eluting, only two HTLV-1 positive sera (H1\_#4, H1\_#6) remained unambiguously positive. Both remained positive even at a pre-dilution of 1:10, and one (H1\_#6) was positive at 1:100. Five of the undiluted HTLV-2 positive sera (H2\_#3 –H2\_#6, H2\_#8) were unambiguously positive after spotting and eluting, and three (H2\_#3, H2\_#5, H2\_#6) remained positive at a pre-dilution of 1:10. Eluates of the undiluted HTLV-1 positive serum H1\_#1 and HTLV-2 positive serum H2\_#2, despite being positive in ELISA, tested negative by Western blot. The filter eluates of the three HTLV-1 positive sera H1\_#2, H1\_#3 and H1\_#5 showed reactivity to a few core antigens including p19 and the GD21, which, according to the manufacturer, would be interpreted as an indeterminate HTLV-1/-2 Western blot result, but not as a confirmed HTLV infection. The filter eluates of two HTLV-2 positive sera (H2\_#1, H2\_#7) also showed reactivity to only the core antigen p24 and the GD21 antigen, resulting, as described above, in an indeterminate HTLV-1/-2 Western blot result. In contrast, for some of the sera, even the 1:10 pre- dilution could by typed by Western blot (H1\_#4, H1\_#6, H2\_#3, H2\_#5) and in one HTLV-1 positive serum H1\_#6 the 1:100 pre-dilution could also be confirmed. Taken together, all DSS eluates of the validation panel were found to be positive in ELISA, but testing eluates of six DSS from HTLV-1 infected asymptomatic carriers resulted in only two cases that could be confirmed by Western blot. In contrast, testing eight DSS from HTLV-2 infected asymptomatic carriers resulted in five confirmed HTLV-2 infections. A summary of the validation, with details of the ELISA and Western blot results, is given in. In conclusion, these data indicate that even for HTLV-infected individuals with low antibody titers, eluates from DBS can still test positive by ELISA, although a venous blood sample appears in a few cases to be necessary for confirmation. ## Determination of HTLV prevalence among PWID in Germany Initial screening of the DBS eluates was performed by ELISA as described above. Of the 2,077 eluates from DBS tested, six were initially found to be reactive. Three were only weakly reactive with a sample/cut-off ratio of \<2, and three were stronger with ratios of 2.7, 5.0 and 5.8. The sample/cut-off ratios of one strong positive HTLV-1 serum and the internal positive control which was included on every assay plate were both \>12.5. The six samples were tested again in duplicate in a second ELISA assay and found to be negative. Thus, no HTLV infections were confirmed in our large study population of German PWID. # Discussion HIV and HTLV are the only known exogenous human retroviruses, and of the four types of HTLV, only HTLV-1 and HTLV-2 are of epidemiological relevance, both being distributed worldwide, albeit usually in different population groups. Both viruses can spread easily within people who inject drugs (PWID), with especially HTLV-2 being found at high rates in PWID within the United States and Europe. Here we report the results of screening for anti-HTLV-1/-2 antibodies in 2,077 samples from German PWID collected during 2011–2014 as part of a multicenter sero-behavioural survey using respondent-driven sampling. The frequent sharing of needles and equipment by PWID place them at particularly high-risk for blood- borne diseases such as HIV, hepatitis and also HTLV. Studies in other European countries found varying seroprevalences of HTLV-1 and -2 among PWID. Although the prevalences for hepatitis B and C and HIV in German PWID are comparable to those in other European countries and are much higher than in the general population, we found no case of HTLV-1 or -2 infection in our study population. This finding is in line with an older study carried out in a small sample of 63 German PWID that also failed to find a single case of HTLV infection and, in addition, reflects the absence or very low prevalence of HTLV infection in the general population. In German blood donors and pregnant women, the proportion of HTLV infection was found to be 0 and 0.7 per 10,000, respectively. Our study was performed with DBS material prepared from capillary blood. The use of DBS samples for HTLV seroprevalence studies is not without precedence. In one Spanish study, DBS from 484 PWID were analyzed and 27 cases of HTLV-2 infection were detected, using both Western blot and PCR for confirmation. It is not clear from the paper how many samples needed to be confirmed using PCR due to having titers too low for Western blot confirmation. There are also older data available addressing the use of DBS for the detection of anti-HTLV antibodies, one in which simulated HTLV-1 positive DBS prepared using sera from HAM/TSP or ATL patients were evaluated and another carried out in the Gauteng region of South Africa in which 2,582 DBS from neonates were screened and 10 cases of HTLV infection were found. Both studies used the same procedure for testing the filter eluates, i.e. HTLV-I gelatin particle agglutination for screening followed by confirmation by Western blot. However, in these studies antibodies were extracted into a smaller volume from only a portion of the DBS, and the wide range of antibody titers in HTLV-infected individuals makes it difficult to evaluate the impact of such variations in extraction methods on the subsequent Western blot result. A large prospective study of pregnant women in Europe, also used DBS samples from France, England and Germany, and it is worthwhile to note that of the 104 samples repeatedly reactive by particle agglutination test or ELISA, only 83 could be confirmed by Western blot or Line Immuno-Assay. Furthermore, a recent study comparing 692 DBS and their corresponding venous blood samples for the serological detection of several infections, found that both methods successfully detected several cases, including one HTLV infection, although no confirmation assay was used. Finally, one large study analyzed 55,293 DBS for HTLV seroprevalence, and although Western blot was not used for confirmation, it was possible to investigate the corresponding blood samples for confirmation. Despite the frequent use of DBS in earlier HTLV seroprevalence studies mentioned above, our validation results using serial dilutions of 14 positive sera indicate an increased risk of missing positive samples that have exceptionally low antibody levels if DBS are used in place of regular serum or plasma for ELISA screening. One study using simulated dried blood spots with sera or plasma from HAM/TSP or ATL patients found only two filter eluates samples giving indeterminate results in Western blot confirmatory testing, whereas 24 samples could be positively confirmed. However, here the authors also found the antibody titers in filter eluates measured with the HTLV-1 gelatin particle agglutination test to be drastically reduced compared to the titers of the original sera or plasma samples. The frequency of HTLV infected persons with such critical antibody levels is presumably low but needs to be determined far more precisely in order to reliably estimate the limitations imposed upon DBS usage for HTLV surveillance studies in the future. Furthermore, the results obtained show that a direct application of the DBS material obtained using a procedure such as ours or similar for Western blot confirmation of an HTLV infection is not reliable and should rather be performed with regular serum or plasma or by PCR. The prevalence of HTLV in PWID is quite different across European countries. In the Irish city of Dublin, a prevalence of 14.6% HTLV-2 seropositivity in a sample of 103 PWID infected with HIV-1 was reported, but no HTLV-1. In this study, sera were tested by ELISA and infections were confirmed by Western blot and/or PCR using the same commercial assays used in our study (Murex HTLV I+II ELISA; MP Diagnostics HTLV Blot 2.4). In contrast, in a sample of 583 Portuguese PWID infected with HIV-1 a low seroprevalence of 0.51% was observed, and again only HTLV-2. The sera were tested by ELISA and Western blot using the same assays as in our study, and one of the three ELISA positive samples remained indeterminate in Western blot. It is worthwhile to note that each of the infected individuals reported to have lived in Spain for long periods of time and that the phylogenetic analysis of the LTR sequences from their viruses suggested a Spanish origin. In Madrid and Barcelona, relatively high prevalences of HTLV-2, 17/256 (6.6%) and 10/228 (4.4%) respectively, have been reported in PWID recruited via respondent-driven sampling. It is notable that in this study no HTLV infections were detected in PWID recruited in a third Spanish city, Seville. This study also used DBS collected over a period of nearly three years and all cases were identified by ELISA and confirmed by Western blot (and/or PCR) using the same kits used in our study. Furthermore, according to one brief report, unpublished data suggest an even higher prevalence in Spanish PWID coinfected with HIV-1, with 67/695 (9.65%) being infected, although none of the 432 HIV-1 negative, HCV positive members of the study participants in the same hospital tested positive for HTLV. A large study conducted in 3,574 PWID in Italy from 1986 to 2005 revealed different prevalences of HTLV infection in HIV-1 coinfected (6.7%) and HIV-1 negative individuals (1.1%). Serum or plasma samples in this study were first screened by passive particle agglutination (Fujirebio, Tokyo, Japan), and confirmation (as well as HTLV-1/HTLV-2 classification) was performed using the same ELISA and Western blot kits used in our study. Interestingly, despite the PWID study participants being recruited from locations throughout Italy, all infections were classified as HTLV-2. In a more recent report using samples from group of Swedish PWIDs, the ARCHITECT ® rHTLV-I/II system was used for screening and the INNOLIA™ HTLVI/II score assay for confirmation, and an HTLV seroprevalence of 35/1079 (3.2%) was found. Again, the majority of HTLV infections were attributed to HTLV-2, with only two scoring positive for HTLV-1 and another five as indeterminate. Finally, a very recent study carried out in Tallinn, Estonia, with 345 PWID recruited by respondent- driven sampling found only one case of HTLV-2 infection. In this case PBMC DNA was directly analyzed by nested PCR and no serological assays were used. It is currently assumed that at least two different introductions of HTLV-2 from the US into PWID populations in Europe took place, based on the observation that the HTLV-2a subtype is found predominantly in Northern and Eastern Europe whereas the HTLV-2b subtype is mainly found in Western and Southern Europe. The highly variable HTLV prevalences in PWID from different European countries or even within different cities in one country, including the zero prevalence among German PWID reported here, is quite puzzling. Furthermore the prevalences of HTLV in PWID seem to be very stable: for example in Spanish PWID, HTLV-2 but no HTLV-1 was found despite the increased proportions of migrants from regions in which HTLV-1 is more prevalent. Furthermore, the most recent study of HTLV seroprevalence in Stockholm found it to be comparable to previous studies from the early 1990s. Moreover, the fact that HTLV-2 is predominately found in PWID, even in countries where HTLV-1 is found in the general population, suggests a distinct founder events for HTLV infection in the various PWID groups studied so far. In conclusion, this analysis of 2,077 DBS samples from German PWID recruited by respondent-driven sampling in eight German cities between 2011 and 2014 found no evidence for cases of HTLV infection. These results are in line with previous serological surveys carried out in Germany but are somewhat different from the situation in some other European countries where low HTLV seroprevalences have been reported, particularly in PWIDs. We would like to thank Nadine Weser, Anja Wolf and Nicole Norley for their excellent technical assistance. Furthermore the technical assistance of Sabrina Neumann, Katrin Arndt, Hanno von Spreckelsen, Joana Tziolis, Estelle Kenfack Guepi, and Guido Vogt is gratefully acknowledged. Finally, we would like to thank Prof. Graham Taylor of Imperial College London for providing sera from HTLV-1 and HTLV-2 infected persons for use in the validation experiments. [^1]: The authors have declared that no competing interests exist.

# Introduction Drug-induced dyskinesia is an important clinical challenge in both Parkinsonian patients treated with L-DOPA and/or dopamine agonists and patients receiving neuroleptics. Since both classes of drugs primarily act via dopamine receptors, it is generally accepted that modulation of downstream signaling of these molecules forms the primary event in the pathophysiology of such movement disorders. However, animal models for drug-induced dyskinesia to dissect involved signaling pathways downstream of the dopamine receptors are sparse. Cenci and colleagues characterized <span class="smallcaps">l</span>-DOPA-induced dyskinesia (LID) in rats that were first rendered hemiparkinsonian via unilateral midbrain injections of 6-hydroxydopamine and subsequently treated with rather high doses of <span class="smallcaps">l</span>-DOPA. Dopamine receptors are pharmacologically differentiated into the dopamine D1 (D1R) and the dopamine D2 receptor (D2R) families. While the former activates adenylyl cyclases (AC) via Gα_s and Gα_olf the latter inhibits AC acting via Gα_i (AC types I, V, VI) and Gα_o (AC type I). D2R agonists can provoke dyskinesia in clinical stages of dopamine deficiency, – and inhibition of D1R signaling prevented dyskinesia in hemiparkinsonian rats. Further evidence that excessive dopamine receptor signaling is involved in dyskinesia was provided by rodent and primate studies overexpressing GRK6. GRK6 is a G protein-coupled receptor kinase which controls desensitization of dopamine receptors. RGS9-deficient mice represent a genetic animal model for the phenotype of drug- induced dyskinesia. Lack of this accessory protein of GPCR signaling seems to predispose for <span class="smallcaps">l</span>-DOPA and neuroleptic-induced dyskinesias. Regulators of G-protein signaling (RGS) form a heterogeneous family of GTPase activating proteins (GAP) that in addition to accelerating G-protein turnover have multiple other functions. RGS proteins regulate a variety of Gα subunits, but do not interact with stimulatory Gα_s proteins. However, Gβ5 is an obligate binding partner to RGS9-2 and this dimeric complex has been recently shown to directly modulate adenylyl cyclase function. The striatum is the major target of dopaminergic pathways in the CNS and contains most of the postsynaptic dopamine receptors. It also bears the highest expression levels of the long splice variant of the ninth member of the RGS family (RGS9-2). Modulation of striatal dopaminergic transmission by RGS9-2 is most likely restricted to D2R signaling as D1R are mainly coupled to Gα_olf in striatum but also Gα_s in other tissues. Mice lacking RGS9 display increased abnormal involuntary movements following dopamine depletion and subsequent administration of dopamine receptor agonists or L-DOPA. Consistently, overexpression of RGS9 in dyskinetic non-human primates resulted in a reduction of such L-DOPA-induced dyskinesia. The functional interaction of D2R and RGS9-2 is supported by the increased fraction of high affinity D2R present in RGS9-deficient mice. Accordingly, in a rat model of schizophrenia with sensitization to amphetamine and in patients suffering from schizophrenia, reduced levels of RGS9 were detected. It is therefore very likely that RGS9-2 has important functional effects on D2R-mediated signaling of striatal medium spiny neurons (sMSN) most likely mediated by accelerating the GTPase activity of G proteins. Although an enhanced striatal D2R-dependent dopaminergic signal transduction and some deficits in working memory and motor coordination , RGS9-deficient mice show an almost normal motor phenotype under unchallenged conditions. However, in RGS9-deficient mice that are treated with the D2R-specific agonist quinpirole following pretreatment with reserpine exhibit pronounced dyskinesia that is absent in wild-type (wt) mice. Under the circumstance of reserpine-dependent depletion of dopaminergic transmission, striatal dysfunction becomes unmasked that is normally balanced by compensatory changes on the functional or gene regulation level. A similar situation may be present, when drug-induced dyskinesia develops either in patients with Parkinson’s disease or schizophrenia receiving long-term therapy with L-DOPA or neuroleptics. The analysis of compensatory gene regulation in functional dopaminergic dysbalance may thus help both understanding the pathophysiology of drug-induced dyskinesia and identifying novel targets for antidyskinetic therapy. Therefore, we screened for compensatory changes in striata of RGS9-deficient mice. Based on the function of RGS9 in D2R signaling we hypothesized that RGS9-deficient mice should display decreased cAMP signaling as a consequence of overactive D2R. We found no evidence for decreased cAMP signaling but rather detected molecular changes in Ca²⁺ signaling. Our data point towards a Ca²⁺-induced potentiation of dopamine receptor signaling that may contribute to drug-induced dyskinesia in RGS9-deficient mice. # Materials and Methods ## RGS9-deficient Mouse Strain and Mouse Genotyping The generation and initial characterization of the RGS9-deficient mice was reported previously. All animal experiments were conducted with mice on a C57Bl6 background (\>12 generations) and in accord with accepted standards of animal care (NIH guidelines) and approved by the respective regional government agency of Saxony (Regierungspräsidium Leipzig, TV 42/08). Genotyping was done by PCR using mouse tail DNA and three primers. The two reverse primers complementary annealed either to MC1neopA-cassette in the inactivated RGS9 gene (5′- GGCTATGACTGGGCACAACA -3′) or to the sequence that is substituted by MC1neopA-cassette in RGS9-deficient mice (5′- ACAGCGGAAGCCATAGAGGA -3′). The attribution of the genotype resulted from the different size of the PCR product with the forward-primer (5′- TTGGGCTCTTGCTCGTGTTA -3′) (94°C for 2 min; 35 cycles of 94°C for 30 sec, 61°C for 30 sec, 72°C for 90 sec; 72°C for 10 min). ## RNA Isolation and Microarray Expression Analysis Striata (dorsal and ventral part) of 3-month-old wt and RGS9-deficient male mice were isolated by microdissection and total RNA was isolated using TRIzol reagent (Life Technologies, Carlsbad, USA) according to the manufacturer’s instructions. The RNA was further purified using the SV Total RNA Isolation System (Promega, Mannheim, Germany) according to the manual. For microarray analysis, RNA integrity and concentration were quantified on an Agilent 2100 Bioanalyzer (Agilent Technologies, Palo Alto, USA) using the RNA 6.000 LabChip Kit (Agilent Technologies, Palo Alto, USA). For reverse transcription (SuperScript II, Life Technologies, Carlsbad, USA) of total striatal RNA (1 μg) from 5 wt and 5 RGS9-deficient mice, an oligo-dT primer containing a T7 RNA polymerase promoter site (Genset SA, Paris, France) was used. After purification of cDNA by phenol-chloroform extraction, *in vitro* transcription using ENZO BioArray RNA transcript labeling kit (Affymetrix, Santa Clara, USA) was performed. Unincorporated nucleotides were removed using the RNeasy kit. Fragmented cRNA was hybridized to GeneChip Mouse Genome Arrays 430 2.0 (Affymetrix, Santa Clara, USA) at the IZKF Leipzig microarray core facility. The expression data were normalized with the Microarray Suite 5 (MAS5) algorithm using the R software package (<http://www.r-project.org/>). The annotation of the probe sets was obtained from the Affymetrix homepage. The datasets were filtered for transcripts that were detected as present in at least three microarrays of one group (wt or RGS9-deficient mice). Statistical testing was done using a two-tailed Student’s t-test. For gene ontology profiling, MAS5 normalized data with P≤0.01 were subjected to the Onto Express algorithm. The settings were: hypergeometric distribution, correction for false discovery rate (fdr). The MAS5 processed data were also analyzed using the Gene Set Enrichment Analysis (GSEA) algorithm. Gene sets were obtained from the Kyoto Encyclopedia of Genes and Genomes database (KEGG). GSEA settings were: 500 phenotype permutations, datasets collapsed to gene symbols and weighted enrichment with signal-to-noise metric. ## Preparation of cDNA and Quantification by Real-time PCR For quantitative real-time PCR analysis (qPCR), Platinum SYBR Green qPCR Supermix (Life Technologies, Carlsbad, USA), cDNA from 30 ng total RNA, 0.6 μM forward and reverse primers and 100 nM ROX (5-carboxy-X-rhodamine, passive references dye) were used. Oligonucleotide primers (Table S1) were designed using the Primer3 software to flank intron sequences if possible. PCR was performed in an MX 3000P instrument (Stratagene, La Jolla, USA) using the following protocol: 2 min 50°C, 2 min 95°C and 50 cycles of 15 s 95°C, 30 s 60°C. To confirm the presence of a single amplicon, product melting curves were recorded. The correct size of the amplicons was asserted by agarose gel electrophoresis and the identity verified by restriction enzyme cleavage or sequencing for at least 10% of the amplicons. Threshold cycle (C_T) values were set within the exponential phase of the PCR. Data were normalized to β2-microglobulin and ΔC_T values were used to calculate the relative expression levels. Gene regulation was statistically evaluated by subjecting the ΔΔC_T values (ΔC_T RGS9-deficient minus ΔC_T wt) to a two-tailed Student’s t-test assuming equal variances. Gene regulation values are given as 2^−ΔΔC_T ± SEM. ## Western Blot Analysis Striata of wt and RGS9-deficient mice were homogenized by sonication in phosphate buffered saline (PBS) containing 1 mM Na₂EDTA and a protease inhibitor cocktail (Roche, Basel, Switzerland). The protein content was determined using Bradford reagent (Bio-Rad, Hercules, USA). 50 μg or 20 μg total protein were subjected to 10 or 12.5% SDS-PAGE and transferred onto nitrocellulose membrane (Hybond-C Extra, Amersham, Piscataway, USA). After blocking with 5% non-fat dry milk in Tris-buffered saline (50 mM Tris, pH 7.6, 150 mM NaCl) containing 0.1% Tween-20 (TBST) for 2 h at 4°C, the blots were incubated with primary antibody for 1 h at room temperature or over night at 4°C. The primary antibodies were: anti-RGS9 (Santa Cruz Biotechnology Inc., Dallas, USA), anti-phosphoDARPP32 (Thr34) and (Thr75) (Cell Signaling, Boston, USA), anti-DARPP32 (Cell Signaling, Boston, USA), anti-phospho-p44/42 MAPK (Cell Signaling, Boston, USA), anti-MAPK (Zymed Laboratories Inc., San Francisco, USA), anti-CaMKIIβ (ProteinTech Group Inc., Chicago, USA), anti-phospho-Thr286 CaMKII (PhosphoSolutions, Aurora, USA), anti-CaMKIIγ (Santa Cruz Biotechnology Inc., Dallas, USA), anti-GluR2 (Santa Cruz Biotechnology Inc., Dallas, USA) and anti-actin (Molecular Probes, Eugene, USA). The dilution of primary antibodies was 1∶1000, except for anti-CaMKIIγ (1∶500) and anti-GluR2 (1∶200). After extensive washing in TBST, membranes were incubated with horseradish peroxidase- linked anti-mouse, -goat or -rabbit secondary antibody (Dianova, Hamburg, Germany) diluted 1∶10,000 in TBST for 1 h at room temperature. After extensive washing with TBST, immunostaining was performed using ECL Western Blotting Substrate (PIERCE, Rockford, USA). Band densitometry was carried out using the Scion Image software (Scion, Frederick, USA). Values are given as mean ± SEM. Protein concentrations were statistically evaluated using a two-tailed Student’s t-test assuming equal variances. ## Cyclic AMP Measurements in Acute Striatal Samples Striatal samples of defined size were dissected and individually transferred into ice-cold stimulation buffer (0.1% BSA, 0.5 mM IBMX, 5 mM HEPES in HBSS, pH 7.4) containing forskolin at the given concentrations. To investigate the influence of D2R signaling on adenylyl cyclase activity, 10 μM forskolin and quinpirole between 0.01 and 100 μM were added. After stimulation at 37°C for 1 h, samples were transferred into 500 μl ice-cold 0.1 N HCl, homogenized by sonication for 15 sec at 4°C (1 cycle, 100% amplitude, UP-50 H, Hielscher Ultrasonics) and incubated on ice for 30 min. Homogenates were centrifuged for 10 min and 14000×g at 4°C. Supernatants were transferred into new reaction tubes and dried at 56°C over night. Pellets were dissolved in 100 μl lysis buffer (5 mM HEPES, pH 7.4, 0.1% BSA, 1 mM IBMX, 0.3% Tween-20) and the cAMP concentration was measured in an AlphaScreen-based assay (Perkin Elmer, Waltham, USA) according to the manufacturer’s instructions. ## Electrophysiological Recordings Coronal slices (300 μm thick) containing the striatum were prepared from the brains of RGS9-deficient mice and their wt littermates. Slices were superfused with an artificial cerebrospinal fluid (ACSF) containing (in mM): 125 NaCl, 2.5 KCl, 2 CaCl₂, 1 MgCl₂, 26 NaHCO₃, 1.25 NaH₂PO₄, and 20 glucose, bubbled with 95% O₂/5% CO₂. The pH was adjusted to 7.3–7.4 and the osmolarity to 305 mOsM. We did not suppress GABA_A receptor currents, since we were recording at a holding potential of –70 mV close to the chloride reversal potential. All recordings were performed at room temperature. sMSN were identified by their morphology and characteristic electrophysiological properties including negative resting membrane potentials and slow capacitance transients. To confirm the morphology typical for sMSN, we labeled some cells (n = 5) using 200 μM Oregon- green-BAPTA1 (Life Technologies, Carlsbad, USA) and imaged these cells using a confocal microscope (Olympus Fluoview FV1000). For analyses, cells were accepted if the leak current was \<100 pA and the resting membrane potential below –70 mV to ensure homogenous recordings. Glass electrodes (5–8 MΩ) were filled with a solution containing (in mM): 150 potassium gluconate, 10 NaCl, 3 Mg-ATP, 0.3 GTP, 10 HEPES and 0.05 EGTA, adjusted to pH 7.3 and an osmolarity of 305 mOsM. The liquid junction potential of 15 mV was corrected post hoc. Spontaneous excitatory postsynaptic potentials (sEPSPs) were recorded at –70 mV. We did not apply tetrodotoxin but applied CNQX in some of the recorded neurons to confirm that AMPA receptor activations were the source of spontaneous postsynaptic activity. Spontaneous postsynaptic potentials were analyzed using ClampFit 9 software and the event detection, template search mode (Axon Instruments, Foster City, USA). Templates were generated by averaging 10 characteristic events. Excitatory sMSN afferents were stimulated with a glass electrode filled with ACSF and placed between the recorded sMSN and the cortex, typically ∼50–100 μm from the cell body using current clamp mode. Stimulus intensity was adjusted to yield large EPSP amplitudes without eliciting an action potential or inducing direct stimulation. Typically, we used 100–200 μA for duration of 0.1 s. Continuous stimulation was performed at a frequency of 0.5 Hz. Following 20 min of baseline stimulation, we applied 3 bursts of 3 s duration and a frequency of 100 Hz separated by 30 s. LTD was then measured for 30 min. Data points were calculated every 30 s by averaging the last 15 EPSPs (0.5 Hz). All recordings were performed using a Multiclamp amplifier (Axon Instruments, Union City, USA), filtered at 2 kHz and digitized at 10 kHz. Acquisition and analysis were performed using custom Clampex 9 software (see above). LTD data were analyzed using a two-tailed Student’s t-test. Data are expressed as mean ± SEM. # Results ## Transcriptome-wide Striatal Gene Expression Analysis To identify adaptive gene regulation in response to RGS9 deficiency, striatal gene expression was assayed on a whole transcriptome-scale in 3-month-old male RGS9-deficient and wt mice. RNA was prepared from striatal tissue, reversely transcribed and subjected to Affymetrix microarray hybridization. To validate the microarray data, we first analyzed relative expression levels of the RGS9 transcript. Surprisingly, the RGS9 transcript was found significantly up-regulated in the striata of RGS9-deficient mice (Table S5). These data were confirmed by qPCR (Table S4). Besides the catalytic RGS domain, RGS9 contains GGL and DEP domains, the latter of which had been substituted in frame by an MC1neopA-cassette to generate the RGS9-deficient mice. Two transcript variants of RGS9 were detected in RGS9-deficient mice, one long including the MC1neopA- cassette and one truncated variant where the MC1neopA-cassette was missing. Elevated transcript concentration of RGS9 detected in striata of RGS9-deficient mice resulted from both transcript variants since RGS9 probe sets and qPCR primer bound to C-terminal regions that were available in both transcript variants. Our Western blot analysis confirmed absence of RGS9 protein in striatal tissue of RGS9-deficient mice (data not shown). This is because the resulting mRNAs did not encode for immunoreactive RGS9 proteins. One can speculate that the up-regulation of the RGS9 transcript variants may be directed towards balancing the lack of RGS9 protein. Indeed, a large fraction of differentially expressed genes belong to the GO category ‘DNA-dependent regulation of transcription’ (Table S2). Because the promoter of RGS9 and its regulation is not characterized yet, we did not follow this interesting finding. Of 45,101 probe sets assayed in the microarray hybridization experiments, 22,962 probe sets, corresponding to 12,714 genes, were detected as present (≥3 present calls in at least one group) using the MAS5 algorithm. To determine the extent of differences in gene expression, we subjected the MAS5-processed microarray data to statistical analysis. We found that 2.6% (365 probe sets, corresponding to 327 genes) were significantly different between wt and RGS9-deficient mice (P≤0.01) (the complete list of differentially expressed transcripts is given on request). We then used the OntoExpress software to identify functional systems that show global differences in gene expression between wt and RGS9-deficient mice (Table S2). Differentially expressed transcripts grouped into processes which included transcription, synaptic transmission, posttranslational protein modification like dephosphorylation and palmitoylation, and locomotor behavior (Table S2). In a complementary approach, full array data were subjected to gene set enrichment analysis (GSEA) using predefined gene sets obtained from the KEGG database. Genes that are involved in Ca²⁺ signaling and in long-term depression (LTD), a main form of synaptic plasticity, were differentially expressed between wt and RGS9-deficient mice (Table S3). ## Expression Analysis of Genes Involved in Ca²⁺ Signaling and Synaptic Plasticity To verify the differences in gene expression identified by microarray analysis, we performed qPCR for selected transcripts involved in D2R signaling, Ca²⁺ signaling and the two forms of synaptic plasticity, LTD and long-term potentiation (LTP). Verification of microarray data is strongly recommended since different probe set data for one gene can vary in array analyses (see Table S4) and microarray and qPCR data do not always show strong correlation. We therefore considered as differentially expressed only transcripts that revealed similar tendency in both methods and were significantly regulated at least in one of the methods. Concerning D2R signaling pathway, there was no alteration in D2R transcript concentration (Table S4) but the G-protein α_i subunits 1–3, the adenylyl cyclase 3 (AC3), the catalytic subunit of protein kinase A (PKA) and phosphodiesterase 4b (PDE4b) were significantly down-regulated in RGS9-deficient mice. Many genes that are involved in Ca²⁺ signaling where down-regulated in striata of RGS9-deficient mice among them the metabotropic glutamate receptor 5 (mGluR5), the ionotropic glutamate receptor subunit α2 (GluR2), Ca²⁺-dependent proteins like calmodulin (CaM1), Ca²⁺/CaM- dependent protein kinase II (CaMK II), phospholipase C (PLCβ1) and a catalytic subunit of protein kinase C (PKC). We also investigated expression levels of genes encoding endoplasmic membrane proteins that regulate the cytoplasmic Ca²⁺ concentration, the ryanodine- (*Ryr1*) and inositol 1,4,5-trisphosphate receptors (*Itpr1*) receptor genes and the most abundant striatal sarco−/endoplasmic reticular Ca²⁺ ATPase (SERCA) gene (*Atp2a2*). While there was no significant alteration in the IP3R1 expression level, Ryr1 and SERCA2 were significantly down-regulated (Table S4 in). The intracellular localization and functional interaction of proteins differentially expressed on the transcript level in striata of RGS9-deficient mice is illustrated in. ## Expression and Phosphorylation State of Striatal Signaling Proteins To assess the functional relevance of the striatal gene regulation pattern detected in mRNA expression experiments, we subjected striatal homogenates of wt and RGS9-deficient mice to Western blot analyses and quantified selected proteins involved in Ca²⁺ signaling and synaptic plasticity. The dopamine- and cAMP-regulated phosphoprotein of 32 kDa (DARPP32) is a key component in the integration of dopaminergic and glutamatergic signaling and is highly enriched in striatal tissue. To evaluate differences in RNA expression data from microarray (no regulation) and qPCR (down-regulation in RGS9-deficient mice) DARPP32 protein levels were determined in Western blot analysis revealing no significant difference between both mouse strains. DARPP32 activity is regulated by phosphorylation and possesses four functional relevant phosphorylation sites: Thr34, Thr75, Ser102 and Ser137. Using phosphorylation- specific antibodies, we quantified relative phosphorylation state levels of the most important phosphorylated forms of DARPP32, pDARPP32-Thr34 and pDARPP32-Thr75. Interestingly, phosphorylation of DARPP32 was significantly increased at Thr34, but not at Thr75. The extracellular signal-regulated kinases 1 and 2 (ERK1 and ERK2) are downstream effectors of dopamine signaling that regulate excitability and glutamatergic transmission in striatal sMSN. The ERK activities are tightly regulated by phosphorylation. Although ERK1 and ERK2 mRNA levels (Table S4) and protein concentrations were lower in RGS9-deficient mice, ERK1/2 phosphorylation was strongly increased indicating sustained activation. Control of glutamatergic transmission is also dependent on Ca²⁺ signaling. Thus, we investigated concentration of Ca²⁺-dependent proteins, e.g. CaMK IIβ and CaMK IIγ. Furthermore, protein concentration of GluR2, the most abundant striatal AMPA receptor subunit that accounts for Ca²⁺ impermeability of AMPA receptors, was determined. In agreement with gene expression data, protein concentration of CaMK IIγ and GluR2 were significantly decreased in RGS9-deficient mice. Western blot analysis also revealed a reduced concentration of CaMK IIβ and phospho-CaMK IIβ in RGS9-deficient mice but tendencies were not significant. ## Functional Analysis in Acute Striatal Samples The hyperphosphorylation of DARPP32-Thr34, that was observed in RGS9-deficient mice, implies enhanced basal PKA activity or an imbalance in the activities of the opposing D1R and D2R signaling. To dissect pathways involved hyperphosphorylation of DARPP32 we assessed basal and stimulated adenylyl cyclase activity in acute striatal samples of wt and RGS9-deficient mice by a label-free AlphaScreen-based cAMP accumulation assay. Under basal conditions and forskolin stimulation RGS9^−/− samples had a tendency toward higher adenylyl cyclase activity, however the tendency was not significant at the individual data point. No significant differences between wt and of RGS9^−/− striatal samples were seen in quinpirole-induced D2R signaling following forskolin stimulation although the D2R levels appear to be per se below functional detection limits of the assay. ## Analysis of Spontaneous Synaptic Events The expression data (Table S4 in) indicate that GluR2 may be primarily involved in conferring changes in signal transduction induced by the lack of RGS9. Ca²⁺ permeability of AMPA receptors is markedly influenced by subunit composition of these receptors with GluR2 decreasing Ca²⁺ permeability. In fact, overexpression of Ca²⁺-permeable receptors can markedly harm cellular integrity and lead to neurodegeneration. In addition, the ratio of GluR1 to GluR2 also correlates to the amplitude of spontaneous excitatory postsynaptic potentials (sEPSPs). The significant reduction of GluR2 expression detected in striata of RGS9-deficient mice suggests a change in the GluR1:GluR2 ratio and hence a change in the amplitude of sEPSPs. To test this we analyzed sEPSPs in sMSN derived from wt and RGS9-deficient mice. sEPSPs did not only exhibit larger amplitudes in RGS9-deficient animals (−10.08±0.77 vs −7.85±0.69 pA) but also occurred at a higher frequency (6.31±0.66 vs 3.21±0.44 Hz). All sEPSPs in these cells were completely blocked by CNQX which confirms that these are originated by AMPA receptor activation. We cannot completely exclude that these changes found are exclusively related to postsynaptic AMPA receptors since AMPA receptors are also found presynaptically. However, our expression data strongly suggest that changes in mRNA expression levels mainly derived from sMSN. ## Analysis of Long Term Depression (LTD) Ionotropic glutamate receptors are involved in synaptic plasticity and long term changes following tetanic stimuli. Using Mg²⁺-containing solutions and current clamp mode, LTD, a major type of synaptic plasticity, is well recognized in sMSN and mostly attributed to activation of AMPA receptors. NMDA receptors should not be involved due to voltage and magnesium blocks. LTD is dependent on the expression of GluR2 in cerebellar Purkinje cells and LTD in midbrain dopaminergic neurons depends on the rapid insertion of GluR2. As GluR2 transcript and protein concentration are significantly reduced in RGS9^−/− striata, LTD should be dampened in RGS9^−/− compared to wt striata. In fact, when we applied tetanic stimuli and recorded evoked EPSPs over at least 30 min, we observed LTD of ∼10% in RGS9-deficient animals and ∼35% in wt animals. # Discussion Previous studies have employed the RGS9-deficient mouse strain as a genetic model system for potentiated D2R-dependent striatal dopaminergic signal transduction. Indeed, our gene expression data are highly consistent with striatopallidal D2R hyperactivity in RGS9-deficient mice which is counteracted through changes in mRNA expression levels of specific downstream signaling components. Transcript levels of all G-protein α_i subunits, the downstream signaling components of G-protein βγ heterodimers, PLCβ1 and the diacylglycerol- (DAG) and Ca²⁺-dependent protein kinase C β1 (PKCβ1) were significantly down-regulated. These findings can be interpreted as compensatory regulation to counteract persistent overactivity of D2R-dependent signaling. Unchanged basal cAMP levels in acute striatum biopsies of wt and ko mice may reflect the effectiveness of these changes at the functional level and/or the minor importance of AC inhibition by striatal D2R signal transduction. Indeed, we did not detect adenylyl cyclase inhibition by the D2R-specific agonist quinpirole in both wt and RGS9-deficient mice. The obvious lack of quinpirole-induced cAMP reduction is in agreement with a recent study in rat striatal tissue homogenates suggesting that adenylyl cyclase inhibition is not a significant downstream effect of striatal D2R activation. Relevant D2R-signaling mechanisms in striatopallidal sMSN include activation of PLCβ isoforms, inhibition of Cav2 channels – and opening of potassium channels – through interaction with G-protein βγ heterodimers. While the former leads to intracellular Ca²⁺ mobilization, the last two mechanisms result in a dopamine-dependent reduction in neuronal excitability. There is evidence from our transcriptome data that components of the intracellular D1R/Gs protein/AC signaling cascade (AC 3 and 7, β subunit of PKA) are significantly down-regulated in RGS9-deficient mice. Together with the similar cAMP responses to forskolin this may point towards a compensation of increased stimulation via D1R. Secondly, key substrates of PKA that regulate striatal control of motor function include AMPA receptors and Thr34 of DARPP32. In agreement with PKA overactivity, significantly enhanced phosphorylation of DARPP32-Thr34 but not of Thr75 was detected in Western blot experiments. In DARPP32, Thr34- and Thr75 phosphorylations have mutually antagonistic effects with the phospho-Thr34 form acting as an endogenous inhibitor of protein phosphatase 1. Overactivity of these intracellular effectors and the most abundant striatal AMPA receptor subunit GluR2 is also reflected by down- regulation of the respective transcripts and proteins. PLCβ activation is one of the intracellular key events after activation of D2R in striatopallidal sMSN. Inositol trisphosphate release by PLCβ triggers the liberation of Ca²⁺ from the endoplasmic reticulum. The time course of the intracellular Ca²⁺ signal is further shaped by Ca²⁺-induced Ca²⁺ release from the endoplasmic reticulum, a process that involves ryanodine receptors. Both transcripts Plcb1 and Ryr1, as well as protein kinase C (PKC), a target of the other second messenger released by PLC, diacylglycerol (DAG), were significantly down-regulated in RGS9-deficient mice. In fact, down-regulation of the Ryr1 transcript by a factor of 0.19 was the strongest gene regulation effect observed in RGS9-deficient mice. This is likely to represent an adaptive response to substantially potentiated Ca²⁺ signaling in striatopallidal sMSN. A somewhat puzzling finding is that the ATP2a2 transcript that codes for the most prevalent neuronal sarco/endoplasmic reticulum Ca²⁺-ATPase (SERCA) isoenzyme is also significantly down-regulated. SERCA enzymes participate in the control of \[Ca²⁺\]_i by sequestering Ca²⁺ into the endoplasmic reticulum. However, in a recent study, profound loss of striatal SERCA activity was also found in an animal model of excitotoxicity that is associated with excessive Ca²⁺ influx. Further stimulation of PLCβ isoenzymes comes from G_q/11 protein-coupled receptors, particularly class I metabotropic glutamate receptors. Both mGluR5 and the G-protein α_q subunit were also significantly down-regulated in RGS9^−/− striata. Taken together, the described effects may result in exaggerated Ca²⁺ release in striatopallidal sMSN. This finding would also be consistent with the above mentioned circuitry-based increased glutamatergic stimulation of RGS9^−/− striata. The intracellular Ca²⁺ concentration (\[Ca²⁺\]_i) controls a plethora of cellular effects, an action that is in many cases mediated by the intracellular Ca²⁺ sensor calmodulin (CaM). In sMSN, this includes regulation of gene transcription, cellular excitability and long-term modulation of ionotropic glutamatergic transmission. Many processes that control synaptic transmission including AMPA receptor function and synaptic delivery are subject to regulation by the Ca²⁺/CaM-dependent kinase II (CaMK II). Both, the CaM and CaMK II (isoforms β and γ) transcripts were significantly down-regulated indicating sustained activity of this pathway. Enhanced synaptic AMPA receptor activity is also functionally reflected in the significantly increased amplitude and frequency of sEPSP in voltage-clamp experiments. AMPA receptors are heterotetrameric complexes in which the presence of the GluR2 subunit defines the endogenous impermeability to Ca²⁺. There is, however, increasing evidence that Ca²⁺-permeable forms of AMPA receptors may participate in striatal plasticity. Both qPCR and Western blot experiments demonstrated significantly reduced striatal expression of GluR2 in RGS9-deficient mice (Table1;). This favors the formation of Ca²⁺-permeable AMPA receptors that may contribute to excessive Ca²⁺ signaling and altered long-term synaptic plasticity in sMSN. LTD is the most prominent form of synaptic plasticity in the dorsal striatum. Studies in transgenic mice in which direct and indirect pathway sMSN were labeled by selective GFP expression revealed that high-frequency stimulation- induced LTD is a specific feature of D2R-expressing indirect pathway sMSN. There is evidence that the underlying mechanism depends on the postsynaptic release of endocannabinoids in a process that involves D2R stimulation, L-type Ca²⁺ channels and group I metabotropic glutamate receptors (mGluR1, mGluR5) in a coordinated way. The striking reduction in synaptic LTD in RGS9^−/− sMSN may be the result of mGluR5 down-regulation or attenuation of the intracellular D2R signaling cascade. The marked disturbances in working memory that have been previously described in RGS9-deficient animals may be a behavioral correlate to this finding. We also found an increased ERK1/2 phosphorylation. However, this does not result from enhanced Ca²⁺ signaling and CaMK II overactivity because Western blot analysis revealing no significant difference in phosphorylation state of CaMK IIβ in RGS9-deficient mice. A number of studies indicate that ERK1/2 phosphorylation underlies striatal regulation of gene expression and long-term synaptic plasticity. In a mouse model of drug-induced dyskinesia, sensitized cAMP/PKA signaling activated ERK1/2, a pathway that is confined to D1R-expressing sMSN. It remains unclear, how constitutively potentiated D2R activity in RGS9-deficient mice modulates D1R-dependent transmission and whether this crosstalk occurs in the same sMSN. However, a significant subset of MSN (20–50% depending on the striatal area) but not all coexpress both D1R and D2R. In a previous study that analyzed striatal dopaminergic signal transduction in mice deficient for the G-protein α_o splice variant, attenuation of D2R signaling resulted in concomitant down-regulation of D1R-signaling components and diminished behavioral sensitization in response to cocaine. Enhancement of D1R transmission in RGS9-deficient mice may thus be an analogous compensatory response on a functional level to D2R overactivity. We cannot answer whether these compensatory responses involve alterations of the activity of the basal ganglia loop based on the data presented. However, the basic model relating to direct and indirect signaling pathways in the basal ganglia would predict, that increased D2R signaling of sMSN lead to decreased activity of subthalamic nucleus and medial pallidal (GPi) output neurons and subsequent overactivity of cortical motor neurons, which in turn would overstimulate sMSN. Our present data provide at least some evidence that such glutamatergic overstimulation may share alterations of downstream signaling molecules such as DARPP32 and ERK. In summary, RGS9-deficient mice show significant changes in striatal function including enhanced DARPP32 and ERK phosphorylation that are characteristically found in L-DOPA-induced dyskinesia. In this study, we have shown that striata of RGS9-deficient mice present changes in transcript levels of Ca²⁺ signaling components that may counteract D2R overactivity. Thus, prolonged dopamine depletion that releases the pressure on gene regulation unmasks dyskinesia in these animals. By analyzing specific gene expression in RGS9-deficient mice, we have identified a number of critical signaling components that may represent novel targets in antidyskinetic therapy. # Supporting Information We thank C.K. Chen and M.I. Simon for generating and providing RGS9-deficient mice. We are indebted to B. Nürnberg and H.A. Lester for helpful discussions. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: KB RS TS JS AG KS. Performed the experiments: KB VD. Analyzed the data: KB VD RS. Contributed reagents/materials/analysis tools: FW TS JS AG. Wrote the paper: KB RS TS JS.

# Introduction The stoichiometric proportions of elements in plant and animal tissues differ substantially, which causes all herbivores to be confronted with a high “stoichiometric mismatch”, : the proportions of C:x (C indicates the content of carbon, x indicates the content of another element) in the consumer's body are much lower than in its food. To build up and maintain its body composition, a herbivore must develop a strategy enabling the selective consumption or assimilation of particular elements. The strategies for how to cross such a stoichiometric threshold remain poorly understood. Invertebrates feeding on dead wood comprise an extreme case. C, H and O comprise 99% of wood mass. How can an organism build its body based on a substrate consisting almost exclusively of these elements? These compositional differences cause an extreme stoichiometric mismatch between the most common organic tissue on Earth and all the organisms exploiting this resource. The problem of the stoichiometric mismatch between plant biomass and its potential non-microbial consumers in terrestrial ecosystems has not been fully investigated. Fagan et al. observed the discrepancy in nitrogen and phosphorus content between living autotrophs (plant foliage and algae) and typical herbivores; the C:N and C:P ratios are 10 to 20 times higher in autotrophs than in herbivores. Schneider et al. demonstrated a high stoichiometric mismatch in the P content between cave-dwelling invertebrates and the plant detritus they consume. Wood-boring (xylophagous) insects living in the xylem of dead trees constitute another prominent example but have not previously been studied. As estimated from the available data on the elemental content in wood – and insects, the stoichiometric discrepancy may reach two orders of magnitude for living wood and three orders of magnitude for dead wood. This discrepancy concerns not only nitrogen and phosphorus but also K and Mg. Many of the studies on the nutritional mismatch between terrestrial herbivores and their food have focused exclusively on two major nutrients: N and P; however, other elements are also essential for the growth and survival of consumers, and a low level of these elements in food plants may cause a nutritional imbalance. No data are available for the content of minor essential nutrients in dead wood; however, it may be assumed that their concentrations are also negligible. To date, only a few papers have tackled the stoichiometry of microelements, including studies of freshwater or marine pelagic systems –, of terrestrial saprobiotic microorganisms, and of prairie grasshoppers. However, the majority of those studies concerned the consumption of living plant tissues, which are relatively nutrient-rich. Bark beetles, feeding on living phloem, are the most unbalanced herbivores studied to date. Sterner and Elser suggested that termites may possibly be the most unbalanced consumers, with a discrepancy of two orders of magnitude between the nutrient content of their bodies and the nutrient content of their food. Stoichiometric mismatch is not the only problem faced by xylophagous organisms. The poor digestibility of cellulose, hemicelluloses and lignin limits the energy budgets of wood-eating invertebrates and slows their growth. The life strategy of xylophages includes an extremely long development time, which is possible due to the relatively low mortality risk of larvae living deep in the xylem and may compensate for the low digestibility of food. However, it remains an open question whether the fundamental effect on the life history of a xylophage is caused by carbon (energy source) or other nutrients (assembling molecules of crucial functional importance). Although the symbiotic interactions of numerous xylophages with microorganisms may ease the cellulose digestibility constraint, the supply of nutrients in wood cannot be increased this way; their deficit can only be ameliorated by an external supply. However, despite the stoichiometric mismatch, wood-boring beetle larvae are capable of extracting from their food all of the elements necessary for growth and for controlling their metabolic processes. Therefore, the wood consumed must be supplemented with some nutrient- rich material. The obvious candidates are saprobiotic fungi that invade dead wood, which are capable of transferring large quantities of elements to the food source (see for a review). The aim of this study was to determine how xylophagous insects can manage the drastic stoichiometric imbalance of major and minor nutrients. We tested the hypothesis that rather than offsetting nutritional constraints with a prolonged development period, wood-boring larvae may balance their nutritional demands by the import of nutritional elements from outside the system by the action of fungi. Thus, avoiding the stoichiometric mismatch may differentially shape the life histories of dimorphic sexes and various species exploiting the same resources. To address this hypothesis, we compared the elemental contents in the bodies of three species of wood-boring beetles inhabiting the same pine stumps, differing in body size and life histories, with the elemental content of wood (potential food during larval development) at various stages of decay. We examined the levels of essential macro- and micronutrients (C, N, P, K, Ca, Mg, Fe, Zn, Mn, Cu, and Na). In decaying wood, in addition to elemental concentrations, we also estimated the amount of fungal tissue using ergosterol content as a proxy. # Materials and Methods Three common species of pine-xylem-feeding beetles were used: *Stictoleptura rubra* Linnaeus 1758 ( =  *Corymbia rubra* Nakano and Obayashi 1957;  =  *Aredolpona rubra* Viliers 1974), *Arhopalus rusticus* Linnaeus 1758 ( =  *Criocephalus rusticus* Haldeman 1847; Coleoptera, Cerambycidae) and *Chalcophora mariana* Linnaeus 1758 ( =  *Buprestis mariana* Linnaeus 1758; Coleoptera, Buprestidae). The development times for these species reported in the literature are 3 years in the smallest beetle, *S. rubra*; 2–4 years in *A. rusticus*, which is of intermediate size; and 5–6 years in the largest of these beetles, *Ch. mariana*. Pine stumps potentially inhabited by larvae of these beetles were collected in the Niepołomice Forest, approximately 20 km east of Cracow (southern Poland, 50°05′N, 20°21′E, elevation 184–212 m a.s.l.) in spring, summer and autumn during 2010–2012. The stumps were collected from approximately 80-year-old pine stands, one to four years after felling. The stumps (diameter measured at the top was approximately 40 to 90 cm; height measured at the center was approximately 10 to 50 cm) were cut slightly below the ground and were hand-split to collect wood samples, pupae and larvae. The adult beetles leaving the stored stumps were also collected. In addition, *S. rubra* adults were captured in the forest. The field studies did not involve endangered or protected species. No specific permissions were required for collecting beetles and stumps in this location. The wood samples were classified by degree of decay (after): (1) undecayed wood – hard and healthy, without visible changes caused by microorganisms; (2) moderately decayed wood – considerably changed by microorganisms, colored (purple or dark brown), wet and softer than (1) but still difficult to tear apart with a knife; (3) highly decayed wood – many visible changes, ample layers of white or brown rotting fungi, wet and soft, easily torn apart by knife or even by hand; and (4) corridor wood – a thin layer of wood from the walls of corridors made by xylophagous larvae, together with content (‘frass’, i.e., wood fragments and feces). Prior to elemental content determination, the insects and wood samples were freeze-dried (using Christ Beta 1–8 LD plus). All the samples were dried using a two-stage pressure/temperature regime: main drying at 0.34 mbar/−31°C and final drying at 0.0010 mbar/−76°C. We dried insects for five days and wood samples for seven days. For CHNS analyzer, we dried, ground and homogenized the material (one insect or wood sample taken from one stump per measurement sample), and for the mineralization (samples for AAS and colorimeter) we dried intact insects and ground and homogenized wood (wood sample taken from one stump per measurement sample). C and N levels were determined using a Vario EL III automatic CHNS analyzer. K, Ca, Mg, Fe, Zn, Mn, Cu and Na levels were determined by atomic absorption spectrometry (Perkin-Elmer AAnalyst 200 and Perkin-Elmer AAnalyst 800), and P content was determined colorimetrically (MLE FIA flow injection analyzer). Prior to analysis, the samples were mineralized by acid digestion: beetles in HNO₃, pupae in a solution of HNO₃, HClO₄ and H₂SO₄, and wood samples in a solution of HNO₃ and HClO₄. Insect samples consisted of one or a few individuals (as described above), and pupae, adult males and adult females were considered separately. Sulfanilic acid was used as the reference material for C and N analyses, and Certified Reference Materials (bush – NCS DC 733348, chicken – NCS ZC73016 and pork muscle – NCS ZC 81001) were used for the other elements. Sample sizes differed for particular elemental analyses of beetles and variously decayed wood because of the uneven abundance of particular species, sexes and developmental stages and the necessity to pool small specimens and match the separate requirements for every individual element measured (for details, see). Ergosterol content was measured in wood samples using a GC/MS Clarus 600 chromatograph (Perkin-Elmer). We used 73 wood samples collected from the same site as the samples for the elemental content analysis, but from different stumps. The degree of decay of the stumps was categorized in the same way as for elemental content analysis, but the 4th category (corridors) was omitted. We express the degree of stoichiometric mismatch between the insects and their food for element x most simply as the ratio of the stoichiometric ratios in food and in the consumer's body (Trophic Stoichiometric Ratio  =  *TSR*):where *C* – carbon content and *X* – content of element *x*. This index does not depend on the units used for stoichiometric ratios C:x (molar or mass units). Values of *TSR_x\>1* indicate stoichiometric mismatch, with severe mismatch indicated by a *TSR* value substantially different from unity. Because the variation of the *TSR* values cannot be directly measured, we used bootstrapping to estimate the mean *TSRs* for specific elements and food categories with confidence limits. First, the *TSR* values were computed from raw data on a given elemental content in the food and consumer. The values were drawn 15 times, and a mean was calculated. The procedure was repeated 500 times, giving 500 means (bootstrap samples), from which the final averages with confidence intervals were calculated. Principal component analysis (PCA) was employed to compare the multi-elemental stoichiometric relations among species, sexes and developmental stages. The data were log-transformed, centered and standardized by species but not by sample; thus, PCA was performed on a correlation matrix. To check for differences between the indicated clusters, we computed ANOVA independently for the 1st and 2nd axis scores. The Mann-Whitney U test and Kruskal-Wallis test were used for significance testing (p\<0.05) of the differences between species and sexes in the elemental composition and body size (dry mass). Statistica 10 was used for all statistical analyses except for PCA (Canoco 4.5). # Results ## Body composition of beetles The adult beetles differed in body dry mass between species and showed sexual dimorphism of body size, with females being larger than males (; for details, see), but the difference in body size between sexes was significant only for *S. rubra* (Mann-Whitney U test, p\<0.05). The sexes of the pupae were not determined; their mass was intermediate between the adult masses of the two sexes or larger than the masses of either (for details, see). The complete data on elemental content are presented in. The carbon mean content in imagines ranged from 51.7 to 63.3% d.m. and did not differ significantly between species or sexes, except for *A. rusticus* females, which had a significantly higher carbon content than males and females of other species. *Ch. mariana* pupae had a significantly lower C content than *S. rubra* pupae. The mean nitrogen content ranged from 6% to 10.9% and differed significantly between species and sexes. The N content was significantly lower in females of all the species, and the pupae showed the lowest values. Males of all three species had similar P levels (av. 0.64%, SD 0.1;). Male and female *S. rubra* did not differ in P content, but their pupae had a significantly lower P content than adults. The P content differed significantly between the sexes in *A. rusticus*, and the pupae of *Ch. mariana* had a significantly higher P content than *S. rubra* pupae. The levels of the eight other elements did not differ significantly within species, with only three exceptions: Ca (lower in male than in female *S. rubra*, higher in male than in female *A. rusticus*, Mann-Whitney U test, p\<0.05;), Mg (lower in male than in female *S. rubra*;), and Fe (higher in male than in female *A. rusticus*;). The interspecific differences appear to reflect taxonomic relationships at the level of the subfamilies Buprestidae and Cerambycidae: *Ch. mariana* pupae had higher K and Mg contents and lower Fe, Zn and Cu contents than *S. rubra* pupae. *Ch. mariana* males had higher Na and Mg contents and lower Fe, Zn, ad Cu contents than *S. rubra* males. *A. rusticus* males had higher Ca content than *S. rubra* males. PCA allowed for a simultaneous comparison of multi-elemental stoichiometries in all sex, age and species categories of the beetles studied. On the plane determined by the first two axes (54.4% of the total variance), the beetles tended to group according to taxonomy (species and subfamily) and sex. The 1st component was loaded mostly by the variance of Fe, Cu, N, and Mg, and the 2nd component was loaded by K, Mn and P levels. The clusters for both sexes of *S. rubra* greatly overlapped but differed from the pupae of this species. The other two species tended to differ between themselves and to maintain the body multi- elemental stoichiometry across developmental stages and across sexes; no overlap occurred between clusters for *Ch. mariana* pupae, *Ch. mariana* males, and the clusters for the representatives of Cerambycidae. These tendencies were partly confirmed as statistically significant by the ANOVA computed independently for the 1st and 2nd axis scores. ## Elemental content and ergosterol content in wood The relative concentrations of all elements except carbon tended to increase during the decay process. In the material from corridors, the elemental concentrations were similar to those in highly decayed wood, with the exception of Na, Zn and Mg (concentrations highest in corridors;, see for complete data set). The relative increments of the increase in concentration during wood decay (from category 1 to 3) were highest for nitrogen (23-fold), phosphorus (14-fold), copper (6.3-fold) and potassium (4-fold); the other elemental concentrations increased by 18% (Mn) to 136% (Fe). Ergosterol content in dead wood significantly increased along the decay gradient; each wood category significantly differed from the two others (Kruskal-Wallis test, p\<0.05). The respective median values were 39.6 µg/g (dry mass) for the undecayed wood, 169 µg/g for the moderately decayed wood and 385.7 µg/g for the highly decayed wood (see for more details). ## Stoichiometric mismatch expressed as the Trophic Stoichiometric Ratio (TSR values) The *TSR* values were highest for the undecayed wood for all the nutrients. After pooling the data for the three species and both sexes, the stoichiometric mismatch of undecayed wood as food, represented by the *TSR* values, reached three orders of magnitude for N, two orders of magnitude for P, one order of magnitude for K, Na, Mg, Zn and Cu, and less than one order of magnitude for Fe. No stoichiometric mismatch was shown for Ca and Mn: the *TSR* values for these nutrients were lower than one. The stoichiometric mismatch was lower in moderately decayed wood (two orders of magnitude for N and P, one order of magnitude for K, Na, Mg, Zn and Cu) and even lower for highly decayed wood (one order of magnitude for N, P, K, Na, Zn and Cu). The *TSR* values calculated for corridors were between those for moderately decayed wood and highly decayed wood for most of the elements but were the lowest for Na, Mg and Zn. # Discussion Our results show that dead wood tissue as the sole source for xylophagous beetles cannot provide sufficient nutrients for growing larvae to compose their bodies. In addition to N and P, we found that K, Na, Mg, Fe, Zn and Cu are also limiting nutrients. The larval diet is apparently supplemented by fungal tissues gradually infecting the decaying wood and transporting nutritional elements into the stump. Thus, “xylophages” (also called “wood eaters”) are in fact “fungivores”, as they feed on fungi. The nutritional demands differ among species belonging to different families, as determined by the content of several elements in the adult bodies: Fe, Zn, Cu, Mg, Na, and N. Within a family, the species do not differ significantly in elemental composition, but the sexes do differ in their elemental composition. The females require more time to develop because they are larger. The pine stumps in this study were inhabited mostly by *S. rubra* and *Ch. mariana*, which belong to different families and differ significantly in their body stoichiometries and life histories. However, the specimens of *A. rusticus*, a species quite similar to *S. rubra* in stoichiometry and in the length of development time, only rarely co-occurred in the stumps. ## Stoichiometry of xylophages For the three major body components (C, N, P), the body composition of adult xylophagous beetles falls within the broad range of values found in the few other coleopteran taxa studied to date. It has been suggested that besides a taxonomic idiosyncrasy or a presumed body mass allometry, N and P content may reflect the feeding strategy of invertebrates, with predators having higher concentrations of N and P than herbivores and detritivores. The xylophagous beetles studied here do not confirm this generalization: their N content is close to the high values reported for carnivores, and their P content is intermediate between herbivores and carnivores. However, our results are consistent with species-specific data provided by Fagan et al., who found high N levels in several cerambycid and buprestid beetles. Females of all three species studied here have significantly lower N concentrations than males of the same species, possibly associated with the higher fat content in the females. Because the sex of pupae could not be determined, and the samples most likely contain individuals of both sexes, their average body composition should be intermediate between males and females. The position of pupae of the two species studied on the PCA plot suggests that they have relatively high C levels and low N and P levels. This result may be attributed to the higher C:N ratio of the chitinous exuvium left behind at eclosion and the fat reserves exhausted during the pupal stage. ## Stoichiometry of wood decay The concentrations of the nutrients (N, P, K, Ca, Mg) measured in the sapwood and heartwood of living gymnosperm trees are an order of magnitude greater than in the undecayed wood of the dead pine stumps that we studied. The stumps, cut one to four years before the wood samples were taken for analysis, were remnants of trees cut when they were at least 80 years old. Palviainen et al., reported similarly low nutrient concentrations measured in pine stumps during the first five years of decay after cutting. Meerts suggested that mineral nutrients in a living tree may be recycled from senescing sapwood. After cutting, the stumps are exposed to weathering and may be further depleted of nutrients until the fungal mycelium growing into the wood enriches it with nutrients, particularly N and P, imported from outside the system –. Even so, there is still a significant difference in the stoichiometries between partly decayed wood and the insects feeding on it. The nutritional composition of corridor material is similar to that of the wood from which the samples were taken (moderately and highly decayed), except for Na, Zn and Mg, which are present in the corridors at higher concentrations. Most of the corridors that we measured originated from moderately decayed wood and less commonly from highly decayed wood. It is not clear whether the chemical composition of the wood is nonuniform and the larvae bore selectively in the more nutritious areas or whether the chemical composition of the corridors changed due to the more intense penetration of fungal mycelium following larval activity. ## The dynamics of nutrient content in decaying wood Two mechanisms contribute to the decrease of C:x ratios during decay, that is, to the enrichment of the wood with elements other than C, H and O: (i) the liberation of C as CO₂ due to microbial and animal respiration, and (ii) the import of nutrients from outside the system by fungal tissue (mycelium) growing into the decaying wood. To determine whether elements are imported in significant amounts during wood decay, we assessed the relative contributions of the two mechanisms from the stoichiometric proportions of specific elements in the wood at various stages of decay. The initial (*SRI*) and final (*SRF*) stoichiometric ratios in decaying wood for a given element x may be expressed as:where *pC_I* and *pX_I* are initial relative concentrations of carbon and of element *x*, respectively; *C_I*, *C_F*, *X_I* and *X_F* are the initial and final absolute amounts of carbon and of element x, respectively; α represents the proportion of carbon not released as CO₂ during decay; and *β* represents the coefficient of enrichment of element *x*. From), it follows that A value of *β\>\>1* would indicate an increase of the absolute amount of element *x* in the wood during decay. The coefficient *α* is defined as:where *M_I* and *M_F* are initial and final masses of decaying wood, respectively. The concentration of carbon decreases from 55% to 48.9% during decay; thus, *pC_F/pC_I  = 0.489/0.550 = 0.889*. The unknown proportion *M_F/M_I* can approximated from literature data concerning the rates of decay of coarse woody debris , including those of conifer stumps. The proportion of the remaining mass of wood *d* after *t* years of decomposition is usually described with exponential model. The experimentally evaluated constant *k* may range between 0.020 and 0.1101, depending on the tree taxon, environmental conditions and the methods used. Samples of wood at advanced stages of decay were taken from the stumps of trees cut 3 or 4 years before sampling. The model solved for 3 years of decay using these figures yields the proportion of remaining mass between 0.72 and 0.94 and solved for 4 years 0.64 to 0.92. Based on these estimates, the possible values of the coefficient α (eq. 5) may range from 0.57 to 0.84. The range of *β* values can be estimated for each element *X* using *β_x = α×SRI_x/SRF_x*, where *SRI_x* and *SRF_x* are the measured initial and final *C/X* ratios. The results show that independently of the assumed value of α, the amounts of Na, Ca, Mg, Zn, Mn and Fe do not increase during decay (coefficients of enrichment do not deviate substantially from 1.0, ranging from 0.7 to 1.4 for *α* = 0.57 and from 1.0 to 2.1 for *α* = 0.84), but the contents of N, P, Cu and K increase several fold, independent of the assumed value of *α*. Thus, the change in the relative concentrations of these elements is not only the result of the carbon escape from decaying wood but also the result of the net import of these elements from outside of the system. Although part of the additional N may be supplied by nitrogen-fixing bacteria, which are known to occur in decayed plant matter, fungi appear to be the only agents translocating other nutrients from outside of the system to the decaying wood. In fact, the ergosterol content (the fungal tissue proxy) increased along the decay gradient and differed significantly between dead wood categories. The measured ergosterol contents cannot be simply recalculated into fungal biomass because the conversion factors are strongly dependent on the fungal species. However, the contents of N, P. K and Cu in the mycelium of wood decaying fungi, and in the fruiting bodies of mushrooms, are two to three orders of magnitude higher than in undecayed wood (this study), while the contents of the other elements studied here differ less than one order of magnitude. Thus, the increase in nutrient concentration in decaying wood may be attributed to the action of fungi. An experiment in which xylophagous beetle larvae were fed with fungi instead of wood showed that fungi can be an adequate food source for these insects. Fungi inhabiting dead wood have been described as nutrient immobilizers, and our data support the view that fungi may serve as nutrient deliverers. The translocation of elements by fungi in the forest floor is a well-known phenomenon. ## Nutritional imbalance in wood-boring insects The concentrations of elements in the imagines and pupae were one or more orders of magnitude greater than in the potential food of the larvae, except for Ca, Mn and Fe, which showed a less pronounced discrepancy. The mismatch, expressed as the *TSR* value, is most striking for undecayed wood (three orders of magnitude for nitrogen) and diminishes as wood decay proceeds, but the differences remain large (; for full data set, see). The nutrient content of wood from corridors is close to that of highly decayed wood. The most extreme differences are for N and P, which are, respectively, 1500–2000 and 500–900 times less concentrated in undecayed wood than in the beetles. The concentration differences for Cu, K and Na are also significant. Cu is approximately 86 times higher in beetles than in undecayed wood. The K and Na concentrations are 54 and 50 times higher, respectively, in the beetles than in undecayed wood. Only Ca and Mn are available in excess. The other nutrients (K, Na, Mg, Fe, Zn, Cu) are much more scarce in dead wood, such that they may constrain the development of wood-eating larvae. All the elemental concentrations tend to increase as wood decay proceeds, although the *TSR* values remain quite high for N and P. Considering wood as the exclusive source of nutrients, xylophagous beetles seem to be faced with the most unbalanced diet of all organisms studied to date. Even termites live on a less unbalanced diet. The N and P stoichiometric mismatches were similar in all three studied species. We found differences concerning other nutrients: the Na mismatch for *Ch. mariana* is almost twice as high as that of the Cerambicidae beetles, whereas the Fe and Cu deficiencies for *Ch. mariana* are two to one order of magnitude lower. Thus, according to this study, different taxa of xylophagous beetles occupying the same nutritional niche may be faced with different stoichiometric mismatches concerning elements other than N and P. ## Nutritional limitations of xylophage life history and stoichiometric compensation by fungi Walczyńska, using experimental measurements of consumption, assimilation and growth efficiences of larvae feeding on pine wood (with stoichiometric conditions identical to the ones used in the present study), demonstrated that the low digestibility of wood may affect the life history of *S. rubra* by prolonging the development time. This result poses the following question: is the development time long enough to concentrate essential nutrients to the levels required in the body? We calculated the minimum growth period needed to collect sufficient essential element *x* (GP_x, years) aswhere *B* – average mass of a larva, *A_X* – concentration of element *x* in the body of an imago or pupa, *K* – daily food consumption rate averaged through the development period (after Walczyńska), *F_X* – concentration of element *x* in food. The conversion efficiency of limiting nutrients was assumed at 100%, and seasonal changes in the food consumption rate were incorporated (nothing was consumed for half the year). Eating an exclusive diet of undecayed wood would lengthen the time needed to form body tissues to an implausible 40 years for males and 85 years for females, and the overall assimilation efficiency would drop to improbably low values. Based on field observations, the maximum development time for *S. rubra* was estimated at three years. Only highly decayed pine stumps or the material from corridors could provide the beetles with enough of the most limiting nutrients. Larvae found in highly decayed stumps were quite well developed at the age of at least 1.5 years. The stumps were likely not highly decayed during early larval development. The great majority of larvae collected from stumps were found in moderately decayed wood, which is relatively poor in nutrients. The higher content of nutritive elements in material from corridors may suggest that larvae are capable of selecting areas of wood more heavily infected with fungi and thus providing adequate amounts of nutrients or that the activity of the larvae facilitates fungal infection through the corridors. Nonetheless, the enriched chemical composition of material from corridors permits the beetles to complete their life cycle within their maximal lifetime in males. The females must further prolong their life history to be able to assimilate the amount of nutrients sufficient to build up their bodies. Wood-boring insects, enhancing the decomposition process of dead wood by mechanical grinding and fragmentation of the solid stumps, depend in turn on the fungal supplementation of their food with nutrients. Thus, elemental transport by fungi plays a pivotal role in the function of forest ecosystems: matching the stoichiometric balance between the trophic links. # Conclusions 1\. During larval development, xylophagous beetles are confronted with a severe nutritional imbalance caused by poor digestibility of food and its stoichiometric mismatch with the beetles' bodies. 2\. These nutritional constraints are partly offset by the adjusted life histories of xylophages, with a development time extended to a couple of years. The life histories of dimorphic sexes and various species exploiting the same resources may differ, but computational simulations show that the prolongation of the development time is not sufficient to accumulate nutrients in adequate amounts. 3\. The nutritional balance of growing xylophagous larvae can be maintained due to the substantial enrichment of dead wood with nutrients imported from the outside by the mycelium of saprotrophic fungi. 4\. The fungal transfer of essential nutrients from the soil into the wood of dead trees is of fundamental importance for maintaining the detrital food web in forest ecosystems. # Supporting Information The authors are indebted to Ola Walczyńska, Filip Kapustka, Justyna Kierat, Łukasz Sobczyk, Ulf Bauchinger, January Weiner III and anonymous reviewers for constructive critical comments. We also thank Maciej Choczyński, Paweł Dudzik and Patrycja Gibas for assistance during the analyses. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: JW MF. Performed the experiments: MF. Analyzed the data: MF JW. Contributed reagents/materials/analysis tools: JW MF. Wrote the paper: MF JW.

# Introduction Diagnosis-related group (DRG)-based prospective payment schemes are commonly adopted in several national healthcare systems. The literature agrees that DRGs are effective in controlling the prices of medical services and that they incentivize cost containment goals. Hospital revenues depend on DRG prices that in turn, directly affect the financial risk borne by the hospital. The cost depends on many variables that rely both on patients’ conditions and the size, ownership (public, private), and organization (teaching or research hospital) of the hospital. With DRG, patients with similar diagnoses and homogenous resource consumption are grouped together and reimbursed in a similar manner. Diagnosis- related group reimbursement schemes are considered effective if the price is fairly defined and accurately reflects the costs incurred by the hospital for patient treatment and stay. Length of stay (LOS) is generally considered to be one of the most relevant indicators of hospital activity and resource consumption. As a result, a performant DRG system should be able to predict resource consumption for patients belonging to a specific DRG, mainly by considering the expected LOS. If the DRG reimbursement was not correctly defined, favorable conditions for opportunistic behavior may occur. Attention should be placed on the variation of patient severity and costs within a group. According to, the risk lies in the adoption of one of the following strategies: "creaming, the over-provision of services to low cost patients; skimping, the under-provision of services to high cost patients; dumping, the explicit avoidance of high cost patients". Considering neonatal diagnosis, DRGs seem to fail in “matching” the actual resource utilization. This is mainly due to the high LOS variability between groups, which directly correlates to the gestational age and weight at birth. It is estimated that costs for preterm babies are 25 times greater than for “in- terms” and those with no complications. In particular, the cost of Very Low Birth Weight (VLBW) babies, that is, infants with complications at birth and birth weights lower than 1,500 grams, can vary between €38,660 and €116,180 based on analysis conducted in the US, England, Finland, and Sweden. Respiratory Distress Syndrome (RDS) is one of the most frequent complications in preterm infants and is attributable to a deficiency of pulmonary surfactant in premature neonates. The risk of contracting RDS decreases significantly with gestational age. The management of RDS involves prolonged artificial respiratory support and the use of expensive therapy. For example, surfactant replacement helps maximize the surface area available for gas exchange in the lungs while a lack thereof causes breathing difficulties. Hence, surfactant therapy is commonly used for babies with RDS. In the Italian health-care system, the hospitalization of babies born with RDS fall under DRG 385, “neonates, deceased or transferred to another acute care facility,” and DRG 386, “extreme immaturity or respiratory distress syndrome, neonate". These codes match the US Medicare DRG codes of 789 and 790, respectively. Infants who do not survive after birth or are transferred to another hospital for the continuation of care, independent of disease type, are included in DRG 385. Diagnosis-related group 386 includes infants with a birth weight less than 1,000 grams and RDS patients regardless of whether they are preterm. Hence, a patient’s discharge status becomes an important discriminant affecting DRG assignment. In Italy, regions have been autonomous in the organization and management of healthcare, including the definition of the DRG tariffs, since the late 90s. Therefore, local regional governments can opt to adopt the national DRG tariffs or set new tariffs. In the region of Liguria, DRGs 385 and 386 show the same national tariffs. However, it should be noted that, with reference to Gaslini Children’s Hospital, a 20% increase is assigned: DRGs 385 and 386 are therefore reimbursed €6,522 and €36,866, respectively. Moreover, the Italian healthcare system provides a threshold value for each DRG, which indicates the maximum number of hospitalization days, beyond which an additional daily reimbursement is assigned to the hospital. The threshold value for DRGs 386 and 385 is equal to 135 days and 4 days, respectively. The hospital receives an additional reimbursement on a daily basis equal to € 354 and € 84, respectively, per day of hospitalization exceeding this threshold. However, this marginal cost sharing does not even fully cover per day hospitalization costs. For this reason, we do not believe that any incentive to strategic behavior could be provided by this additional (marginal) cost reimbursement. The present study aims at estimating the actual costs incurred for the hospitalization of extremely immature babies or infants affected by RDS. For this purpose, DRGs 385 and 386 patients, hospitalized at Gaslini Children’s Hospital in 2016, were analyzed using clinical and administrative records. # Methods The study is carried out as part of routine checks conducted in the Gaslini Children’s Hospital, and thus, ethical approval is not required. As with all studies conducted in the hospital environment, hospital management approved the study’s protocols. Management was responsible for ensuring the ethical aspects of all hospital activities as the entire study was organized in conjunction with Gaslini teams. Upon entering the hospital, all patients (or parents) sign an informed consent form regarding hospital treatment and its terms and conditions. The research was carried out in full accordance with the Italian law on privacy and all data were appropriately pseudonymized by Gaslini Children’s Hospital by associating each patient with a numerical code. Gaslini Children’s Hospital is located in Genoa, a city in Northwest Italy. It is a referral hospital for many disciplines such as general practice, neonatal, oncology, and orthopedic surgery. Since 1959, it has been formally recognized as a scientific institute for research, hospitalization, and healthcare by the Italian Ministry of Health, with about 30,000 hospitalizations and 550,000 outpatient services per year. During 2016, the hospital accepted more than 70% of regional hospitalizations with the 769 ICD-9-CM code. Furthermore, about 80% of the overall regional discharges with DRG 386 were made by Gaslini Children’s Hospital’s Neonatal Intensive Care Unit (NICU). The research data is based on 243 infants hospitalized in the hospital’s NICU from 1 January to 31 December 2016. Using the International Classification of Disease, patients with the principal or secondary diagnosis code 769, “respiratory distress syndrome in new-born,” were included in the study sample. This diagnosis code is annotated in the “Hospital Discharge Form.” In particular, infants discharged with DRG 386 were tagged with the 769 code as the principal or secondary diagnosis. A subset of 219 cases were discharged with DRG 386, while 24 cases died or were transferred to another hospital and discharged with DRG 385. The operating and management costs used in the present analysis were provided by the NICU. These included personnel costs (physicians and staff); utilities (electricity, water, gas, telephone); medical supplies and drugs costs; treatment tools; management costs (administration, maintenance, depreciation cost of medical equipment); and department overhead costs (cleaning, mortgages, kitchen and laundry, other). These costs were obtained from the department’s annual balance sheet and distributed among patients on a daily basis. Examinations, radiology, laboratory services, and other patient-specific direct costs were analytically imputed to each patient. Hence, assessing the total cost on a daily basis and taking into account care unit capacity, the mean hospital per day cost was estimated. Moreover, LOS, number of examinations, discharge status, and the neonatal characteristics of the infants were also considered in the analysis. By jointly considering the mean hospital cost, patients’ LOS, and regional DRG tariffs, the inpatient reimbursement adequacy was investigated. The K-means clustering method was implemented to create a homogeneous group of patients. Given that this method requires pre-determining the number of clusters, a dendrogram and the Within Group Sum of Square (WSS) error plot were used to identify the appropriate number of groups. The sample was studied before and after clustering. The variables included in the Cluster Analysis relate to the DRG type, birth weight, gestational age, number of examinations, gender, type of discharge (home, died, or transferred to another hospital), and place of birth (“inborn” if the infant was born at Gaslini Children’s hospital or “outborn” if born elsewhere). It should be noted that birth weight is a variable that cannot be affected by biased recovery policies. For each cluster, the mean cost was estimated, taking into account the assistance and examination costs for each patient included in the group. The results obtained were used to define new fees strictly related to the actual costs incurred by the hospital in treating these patients. Tobit regression models were used to capture the relationships between the costs and determinants of a newborn’s recovery. The LOS was used as a proxy for costs in order to avoid discretionary measures. The Tobit model is used, provided that LOS (the dependent variable in the regression) is left censored at zero. By comparing two Tobit models, we examined the influence of neonatal and clinical characteristics on LOS. Both models use the same set of infant characteristics such as gestational age, number of examinations, discharge status, gender, and place of birth to predict the LOS; however, the latter also considers the VLBW variable to investigate the economic impact of babies with birth weights lower than 1,500 grams on LOS. The models take the following general form: $$\left. y_{i}^{*} = \beta_{0} + {\sum_{k = 1}^{K}\beta_{k}}x_{i} + \varepsilon_{i}\mspace{720mu}\varepsilon_{i} \right.\sim N\left( 0,\sigma^{2} \right)$$ $$y_{i} = \left\{ \begin{matrix} {y_{i}^{*},\mspace{720mu} if\mspace{720mu} y_{i}^{*} > 0} \\ {0,\mspace{720mu} if\mspace{720mu} y_{i}^{*} \leq 0} \\ \end{matrix} \right.\mspace{720mu} i = 1,2\ldots.N$$ where $y_{i}^{*}$ and *y*_*i* represent patient *i’s* unobservable and observed LOS, respectively. If the observed LOS is 0 (under 24 hours), the observation is censored. *x*_*i* is the regressor, which will be discussed in detail in Eqs and; *β*_*k* (k = 1,2…N) represents the estimable coefficients of the model, *β*_*o* is the constant, and *ε*_*i* denotes the standard random error term. The first and the second model can be represented by Eqs and , respectively: $${LOS}_{i} = \beta_{0} + \beta_{1}{GA}_{i} + \beta_{2}D_{i} + \beta_{3}B_{i} + \beta_{4}G_{i} + \beta_{5}N_{i} + \varepsilon_{i}$$ $${LOS}_{i} = \beta_{0} + \beta_{1}{GA}_{i} + \beta_{2}D_{i} + \beta_{3}B_{i} + \beta_{4}G_{i} + \beta_{5}N_{i} + \beta_{6}{VLBW}_{i} + \varepsilon_{i}$$ where “GA” is the gestational age, “N” is the number of examinations, while “D,” “B,” and “G” denote the dummy variables of the model. “D” represents the type of discharge; it is 1 if the patient died, and 0 otherwise. “G” is the gender of the baby and is 1 if male and 0 if female. “B” is the place of birth, which is 1 if the infant is outborn and 0 if inborn. The second model differs from the first for the “VLBW” variable, which is a dummy that takes a value of 1 if the baby’s birth weight is lower than 1,500 grams and 0 otherwise. # Results The mean per-day hospital cost in the NICU is estimated to be €820. This value is obtained by the analysis of total cost reassignment on a daily basis and only considers the total number of beds in the care unit. Considering the different DRG types and their relative resource utilization, we estimate a mean hospitalization cost of €2,168 and €39,419 for DRGs 385 and 386, respectively. Compared to the reimbursement, this results in a mean gain of €4,354 for each patient discharged with DRG 385 and a mean loss of €2,553 for each patient discharged with DRG 386. highlights the great sample variability in the LOS, number of examinations, birth weight, and gestational age. The LOS ranges from 0 (under 24 hours) to 262 days, number of examinations from 0 to 328, and birth weight from 435 grams to 4,140 grams with a mean of 1,781. The gestational age distribution is symmetric with the mean value (32) matching the median and the whiskers of the boxplot have a similar length. The LOS positively correlates with the number of examinations, and negatively with birth weight and gestational age. The number of examinations tends to increase if the infant has a low birth weight or low gestational age. Birth weight strongly positively correlates with gestational age. Conversely, no correlation occurs between the number of examinations and the birth weight and gestational age. reports the cluster analysis output. The green dots represent the observations while the components along the axis reflect the dimensions with the largest variance, which explain 61.25% of the total variability. These components are generated by the Principal Component Analysis, which reduces dimensional data to plot the clusters in a two-dimensional space. The value of the first component changes continuously in each cluster, while the second component’s value has discontinuous jumps, particularly for the red and blue clusters. Dots at the top of the plot represent patients discharged with DRG 385; the ones located higher died while the ones located at the average height were transferred to another hospital. Dots at the bottom refer to infants discharged with DRG 386. The amplitude of the circles indicates how similar patients are between groups. Cluster 2, in green, is the smallest and does not include deceased patients. Its dots are very close to one another, highlighting that this cluster has the lowest variability. On the contrary, Cluster 4, in red, is the biggest and, in fact, presents a greater variability within its group. It is comprised of patients with low birth weights and infants discharged with DRG 385 who died after birth. Therefore, it is quite a homogenous group from a neonatal characteristics’ perspective, but heterogeneous when looking at the status of discharge. summarizes the characteristics and descriptive statistics of each cluster. From the table, we can observe that Clusters 1 and 2 do not include VLBW newborns, but include at term infants, contrary to Clusters 3 and 4. In particular, Cluster 4 includes only VLBW patients with birth weights lower than 1,170, which accounts for 78% of all deceased babies in the sample discharged with DRG 385. It excludes patients transferred to other hospitals after birth. Moreover, 75% of the patients in this cluster have a gestational age less than 28 weeks. The mean hospitalization cost is €73,286, ranging from €268 to €265,669. Cluster 3 includes only preterm infants with a birth weight lower than 1,885 grams and a mean gestational age of 31 weeks. The VLBWs in this cluster have birth weights greater than 1,190 grams. The estimated mean hospitalization cost is €32,364 and ranges from €562 to €126,997. Cluster 2 includes only surviving babies with a birth weight between 1,920 grams and 2,800 grams and a mean gestational age of 34 weeks. Patients in this cluster have the lowest number of examinations. The mean hospitalization cost varies between €1,407 and €44,842, with an estimated mean value of €12,177. Finally, Cluster 1 mainly includes infants born at-term; its mean gestational age is 39 weeks and birth weight ranges between 2,870 grams and 4,140 grams. The mean hospitalization cost is €10,899 and ranges between €516 and €35,374. As shown in, birth weight is the main cluster discriminant. In fact, there are no overlaps amongst the groups. The black bold horizontal line in the boxplots emphasizes that infants with the lowest birth weights are characterized, on average, by the highest LOS and number of examinations. Nonetheless, the height of the red box indicates that low birth weight infants present the greatest variability. The patients with a greater birth weight/gestational age have less variability in their LOS and the number of examinations as reported in the green and blue box plots. Results from the Tobit regression are reported for each model in, with the corresponding 95% confidence interval plots. In both cases, LOS is the dependent variable. In order to run the regression correctly, we deleted observations with an LOS less than 1 (i.e., those infants who died within 24 hours) from the sample, resulting in 242 observations. In the first model, all variables are significant at the 1% significance level except the place of birth (significant at 10%). The type of discharge and gestational age has a great impact on LOS. A new-born who dies or who was born at-term has a significantly lower LOS than a surviving infant or a late preterm. Males have lower LOS than females. The number of examinations increases with LOS. In the second model, we notice that the VLBW variable has a significant (1% level) impact on LOS. Additionally, the impact of the patient’s death slightly increases, while the impact of gestational age and gender slightly decreases. The value of the constant is reduced by 44 days and the other variables maintain the same approximate value. The goodness of fit, represented by the pseudo-*R*, is 0.1336 and 0.1394 for the first and second models, respectively. # Discussion The effectiveness of the DRG system is widely discussed in the literature. Investigating the factors that determine the variations in the resource use is one of the main goals from a price redefinition perspective. Our study shows a low DRG performance in terms of predicting the resource use for babies affected by RDS; this is explained by the great variability between groups. The main drivers of this variability are the characteristics at birth and the type of discharge. Gestational age and preterm death have a significant impact on LOS. The analysis of the VLBW variable allows measurement of the significance of the impact of birth weight on the LOS. For a child with a birth weight of less than 1,500 grams, the hospital stay increases by almost 15 days. The second model fits the data better than the first, as reflected by an increase in the pseudo-*R*^*2*. The significance of the VLBW variable can also be seen in the cluster analysis results. Infants who are VLBWs with a birth weight lower than 1,020 grams should be included in a single group, while those ranging between 1,190 and 1,500 grams represent 50% of the points in a different cluster (often configuring outliers). The VLBW group with a birth weight lower than 1,020 grams has the highest variability in the LOS and in the number of examinations. The reason for this great variability is explained by the fact that some infants who belong to this group died after birth. Death before discharge occurs more frequently in the VLBW group. Therefore, the similarity within this group is more attributable to the birth rather than clinical characteristics. In particular, the group that includes infants with the lowest birth weight is characterized by a mean LOS value of 75 days, while the cluster represented for the most part by term infants has the lowest mean LOS value of 12 days. The DRG 386 includes both neonates with RDS and extremely immature infants. This implies that both preterm newborns and term infants with RDS should have a similar expected LOS and an equivalent resource consumption. However, from our results, it is evident that the admission cost for very preterm babies differs from that of term babies, even if they are discharged with the same DRG. Clusters 1 and 2, where extremely immature infants are not considered, have a similar mean cost; they mainly include infants discharged with DRG 386, for which the reimbursement is almost thrice the effective cost. On the contrary, the mean cost for surviving patients in Cluster 4 is approximately twice the refund rate. Cluster 3 exhibits a mean cost that is quite similar to the reimbursement; however, if we consider only the VLBW included in this cluster, it results in a mean cost of €39,166, which makes the refund rate inappropriate again. Additionally, considering a mean hospital day cost equal to €820, it follows that for the survival of VLBW babies, the reimbursement does not completely cover the inpatient cost, while the hospitalization for term or near term infants with RDS is more than rewarded by the reimbursement. Diagnosis-related group 385 does not take into consideration birth weight or gestational age and does not distinguish between infants deceased a few hours after the birth or infants that died after several days in the NICU. However, from the analysis, it is clear that for patients with RDS who die before 8 days, the reimbursement fully covers the costs. Therefore, our study clearly shows that the strong patient differences in LOS, number of examinations, birth weight, and gestational age indicate that DRG 386 needs to be improved and that a reclassification of neonatal admission with RDS may be a suitable solution. The actual reimbursement level for patients discharged with this DRG is underestimated on average. In fact, the related costs incurred by hospitals who accept a significant number of VLBW patients annually are much higher than their expected reimbursement. This might have undesirable effects on the Third Level NICUs which have to manage at least 50 VLBWs annually, in accordance with the State-Regions Agreement of the 16^th of December, 2010. Currently, we estimate a loss of €36,420 for every surviving baby with a birth weight lower than 1,170 grams. In this situation, the reimbursement benefits hospitals that mainly treat term infants with RDS, for which the study shows a gain greater than €20,000 per patient. In addition, the profitable reimbursement rate for those cases discharged after a few days could lead to the “up-coding” phenomenon, whereby physicians could inappropriately designate RDS as the main or secondary diagnosis in order to obtain larger revenues. According to our findings, four different tariffs might be set in order to reduce inefficiency and the undesirable effects that may occur given the high variability in preterm infants affected by RDS. The new tariff should equal the mean cost of hospitalization belonging to each cluster. Practically, the discriminant and observable variable that could be used by the policy maker to separate patients (and associate the correct reimbursement) among groups could be the "birth weight" variable, which is the only one that does not show any overlapping among groups. A new grouping based on the birth weight variable would also allow for the dichotomy between DRGs 385 and 386 to be overcome. In doing so, inefficiency in resource allocation and risk of strategic behavior by hospitals may be sensibly reduced. Notwithstanding, it should be noted that the risk of strategic behavior is very remote with reference to public hospitals (such as Gaslini Children’s Hospital) that do not pursue the profit maximization goal. Our study presents several limitations. Firstly, although the Gaslini Hospital represents almost the totality of infants affected by the pathology in a region, the analysis takes into account patients hospitalized in only one NICU, thereby limiting the number of cases. Secondly, by considering only Italian costs, the research lacks comparison with others European NICUs. Thirdly, the study does not consider the lodging cost for the caregivers during the infants’ hospital stay and the possible deterioration in their quality of life, thereby underestimating the average cost from this perspective. The efficacy and robustness of this study could be improved by including a greater number of cases. Moreover, the cluster analysis approach may also be applied to other episodes of care, contributing to a revision of the current DRG classification. Despite the limitations of this study, our results may support the regulatory authority to improve DRG ability and explain resource use among complicated infants by further “birth weight adjustment,” thereby discouraging unfair hospital behaviors. # Supporting information The authors would like to thank Gaslini Children’s Hospital for their invaluable cooperation in providing the data. [^1]: The authors have declared that no competing interests exist.

# Introduction The packaging of DNA into chromatin plays an important role in setting up specific gene expression patterns during early embryonic development. The level of packaging is regulated by specific histone modifications leading to a more open or closed chromatin structure, depending on the factors that bind to the specific modifications. Recently, several genome wide mapping studies have revealed specific distribution patterns of particular modifications. For example, H3 Lys-4 (i.e. H3K4) is often found methylated at cis-regulatory elements such as enhancers and promoters, and a combinaton of di-/trimethylated K4 at the promoter and K36 in the transcribed gene body, respectively, indicates reliably active transcription units. In contrast, methylations at H3K9 and H3K27 correlate with repressed domains. Under specific circumstances, chromatin fragments have been identified that simultanously carry active as well as repressive modifications, specifically the H3K4me3 and H3K27me3 marks. This chromatin state has been termed bivalent to indicate its ambiguous nature. Bivalent domains first have been described in embryonic stem (ES) cells, where they may keep developmental regulatory genes in a state poised for subsequent activation during cellular differentiation. However, bivalent domains are not restricted to pluripotent cells and most likely have functions beyond priming genes for activation. Altogether, these genome wide studies indicate covalent histone modifications as important regulatory components of development and cellular differentiation. A growing body of data suggests that histone modifications convey important information to early developmental programs in mammals. Histone modification patterns in gametes become established during germ cell differentiation, and nucleosomes, which are retained in low numbers in human sperm, carry histone modifications on particular developmental loci. This finding extends to zebrafish, who packages its sperm genome entirely into nucleosomes, rather than protamines. After fertilisation, widespread epigenetic reprogramming, occuring in cooperation with stochastic gene expression of lineage-determining transcription factors, helps prepare the earliest lineage decisions that generate trophectoderm, primitive endoderm and inner cell mass (reviewed in detail in refs.). Our knowledge about the epigenetic changes, which accompany the determination of definitive somatic cell lineages in the mammalian embryo, are derived largely from in vitro differentiation studies on pluripotent ES cells. Generally, ES cells are characterized by a derepressed chromatin structure that is highly permissive to the transcriptional machinery. The chromatin structure and function of ES cells changes dramatically during differentiation. For instance, the number and size of heterochromatic foci, visualized by immunostaining for heterochromatin protein 1 (HP1), increase upon differentiation. FRAP (fluorescent recovery after photo-bleaching)-analysis revealed that HP1 as well as core and linker histone proteins are hyper-dynamically bound to chromatin in the undifferentiated ES cell, but are more stably associated with chromatin upon differentiation. These changes extend also to covalent histone modifications. Examples are the differentiation-dependent increase in the repressive mark tri- methylated Lys-9 of histone H3 (H3K9me3), and the observed decrease in the global levels of acetylated histones H3 and H4. ES cells, differentiated under retinoic acid administration, undergo a two-fold reduction of H3K4me2 levels, while H3K9me1 and H3K79me2 levels increased moderately, describing nearly specific histone modification patterns for undifferentiated ES cells and their differentiated progeny. Striking changes in histone methylation and acetylation during human ES cell differentiation have also been observed by mass spectrometry for the aminoterminal tail domain of histone H4. Overall, this data shows that the fraction of nucleosomes with repressive histone marks increases during ES cell differentiation, reflecting the lineage-specific use of the genome in somatic cells. Interestingly, direct reprogramming of somatic cells to an induced pluripotent state reverses these changes largely, although not completely. During reprogramming, histones become hypermethylated at H3K4, bivalent domains are re-established, while repressive marks are diminished, indicating a general shift towards the transcriptionally permissive chromatin of pluripotent ES cells. Thus, transitions between pluripotent and comitted somatic cell fates during development or reprogramming involve global alterations in histone-modifications. Pluripotent ES cell lines are established and maintained under stringent culture conditions. Although being derived from endogenous pluripotent cell populations, they have no strict counterpart *in vivo*. Indeed, derivation of ES cells from the blastocyst inner cell mass causes a significant skewing of histone modification patterns, including H3K4 and H3K27 methyl marks, compared to the inner cell mass. The many cell types and tissues of the mammalian embryo arise from the epiblast after implantation, and very little is known about histone modifications at this time of development. Results from carrier ChIP analysis on microdissected tissues from post-implantation mouse embryos suggest that genes known to be bivalently marked in ES cells tend also to be enriched for these modifications in embryonic tissues. However, the bivalent status of these genes has not been rigorously established by sequential ChIP analysis and, thus, the opposing marks could be distributed between cells in a non-overlapping manner. Other vertebrate model organisms such as *Xenopus* and zebrafish have recently provided key information on epigenetic changes in early development. Histone variants and modifications from various *X. laevis* cell types including oocytes, sperm and somatic cells, have been characterized by immunoblotting and mass spectrometry, indicating unique histone modification signatures for each cell type. While these studies did not analyse chromatin from normal embryos, genome-wide RNA-Seq and ChIP-Seq technologies were combined to investigate histone modifications in the course of *X. tropicalis* development. This revealed a hierarchical acquisition of H3K4me3 and H3K27me3 marks, following the onset of zygotic transcription at midblastula. Specifically, spatial differences in the deposition of the repressive Polycomb mark H3K27me3 was predicitve of localized gene expression patterns. In contrast to mammalian ES cells, bivalent chromatin domains are practically absent from Xenopus embryos. Lysine trimethylation of H3K4 and H3K27 appears in the zebrafish epigenome only after the maternal-zygotic transition and in the same sequence as in frogs. However, sequential chromatin immunoprecipitation revealed the existence of bivalent chromatin domains in the zebrafish. While it is currently unclear, whether the observed differences for bivalent domains between frog and fish has a biological basis or reflects technological differences, these studies have provided evidence of genome-wide transitions in histone modifications during normal vertebrate development. The information from ChIP-based studies is invariably dictated by many technical parameters, most notably the antibody quality (discussed). We have sought to obtain quantitative information on different histone modifications by an antibody-independent approach. Here we report the results from mass spectrometry analysis of histone modifications present in bulk embryonic chromatin through *X. laevis* development. This analysis has revealed major quantitative shifts for several histone modifications known to be involved in gene regulation, and has also identified specific differences between pluripotent cells from frog embryos and murine ES cells. The quantitative shifts, which we observed during development, cluster into stage-specific histone modification profiles accompanying and potentially regulating the transition from pluripotent to determined cell states. # Results ## Experimental design In order to investigate histone post-translational modifications (PTM) during vertebrate embryogenesis, we purified core histone proteins from unmanipulated *X. laevis* embryos of four different developmental stages. The stages included blastula (Nieuwkoop Faber stage NF9), gastrula (NF12), neurula (NF18) and tadpole (NF37) embryos, representing key steps in vertebrate development. At late blastula, shortly after the onset of zygotic transcription, embryos consist mostly of uncomitted, pluripotent cells (refs., ; Nicetto and Rupp, unpublished data). By the gastrula stage, the germ layers have been induced, and embryonic patterning increases the cellular diversity of the embryos during neurulation. After hatching, tadpoles are composed largely of differentiated, although premetamorphic, somatic cells. To evaluate the results from the embryonic samples, we also prepared histones from Xenopus A6 and XTC-2 cell lines, as well as from murine germline- transmission competent GS-1 ES cells, and primary embryonic fibroblasts (MEF). The identity of the histone molecules was determined by mass spectrometry. Alignment of the primary sequences of *Xenopus* histones with core histones of other species such as human, mouse and fruitfly revealed high interspecies conservation of the H3 and H4 histones. Overall, we obtained a sequence coverage of 68.9% for H3 and 87.3% for H4 ( and data not shown), covering most of the known histone modification sites on these histones. Therefore, we focused our further investigations on canonical, replication-dependent histone H3 and on histone H4. ## Histone Modification Analysis and Quantification using LC-MS/MS For the identification of histone post-translational modifications we preferentially used high mass accuracy LC-MS/MS mass spectrometry. To obtain peptides of suitable masses, histones H3 and H4 were separated by SDS-PAGE, covalently modified by propionic anhydride and subsequently digested by the endoprotease trypsin. The resulting peptides were then separated by C18 reversed phase chromatography, subjected to tandem mass spectrometry and quantified by integrating the area under the extracted ion chromatogram (XIC) of the corresponding doubly and triply charged ions. When we analysed the chromatographic behaviour of differentially modified peptides, we detected a specific shift of retention times depending on the type of modification. Specifically, acetylation of the H4 peptide amino acids 4-17 leads to an earlier elution by approximately 0.7 minutes. From these findings we concluded that in addition to the mass of a histone peptide and its fragmentation spectrum, the elution time is also a unique parameter to characterise the modification of a peptide. In the following analysis we therefore used the retention time as a third parameter to determine the identity of a given peptide. The small mass and specific biochemical characteristics of two peptides, i.e. histones H3 peptide 3–8 and H4 peptide 20–23, precluded analysis by LC-MS/MS, and, therefore, were analysed by conventional MALDI-TOF mass spectrometry to investigate their modification states. Overall, we monitored 18 known modification sites (17 lysine residues, one serine residue) that are present on five peptides of H3 and four peptides of H4, respectively. Quite often, the type or extent of modification on a particular site is known to be associated with different biological functions. For instance, the Lysine residue at position 27 of H3 can be unmodified or acetylated, which is compatible with transcriptional activity; alternatively, it can be mono-, di- or trimethylated, indicating an increasing activity by Polycomb, which leads to transcriptional repression. Furthermore, the fragmentation of propionylated histones generates in some cases peptides, which contain more than one modification site (e.g. peptide 4–17 on H4; see,), allowing us to detect some combinatorial modification states that coexist on single peptide molecules. Based on the potential biological implications we view each unambigously identified mass shift as a distinct modification state. Based on this definition, we have characterized in toto 59 different modification states by either LC-MS/MS (52 states;) or by MALDI/TOF (7 states). For each developmental stage we quantified spectra from two (tandem LC-MS/MS) or three (MALDI-TOF) biological replicates, with an average correlation coefficient of 0.9962 between biological replicates, indicating high robustness for the reported epigenetic states. ## Active marks As described above, we detected multiple acetylation states at the aminoterminal tail of histone H4. Similar to what has been described in other organisms, the mono-acetylated peptides are predominantly modified at Lysine 16. In blastulae, 38% of the total H4 4–17 peptides carry a single acetyl group. The level of mono-acetylation rises in later developmental stages to a steady state of more than 40% abundance (see also). The percentage of mono-acetylation is slightly lower in tissue culture cells (38% in A6 and 34% in XTC-2, respectively;, panel A), which instead contain more unmodified peptide (54% in A6 and 58% in XTC-2). Of the many potential di-acetylated forms we detected only two states, i.e. simultanous acetylation of either K5 and K12, or K8 and K16. K5/K12 di- acetylation is quite abundant at the blastula stage (27%). Since this modification state is known as a hallmark of predeposition histones, this pool probably reflects maternally stored H4 molecules. This modification state was neither detected on histones from later stages nor on histones isolated from frog cell lines. From gastrula stage onwards, we detected a wave of K8/K16 diacetylated H4 in the embryo, increasing from 11% at NF12 to 17% at NF18, before decreasing again to 11% at NF37. The frog cell lines contain somewhat lower levels of K8/K16 di-acetylated H4 (appr. 7%). The tri- as well as the tetra-acetylated isoforms are less abundant in all embryonic stages (\<10%), even though they were 5-fold higher than in tissue culture lines (\<1.5%; compare). In case of the tri-acetylated isoform, all LC-MS/MS spectra point to a specific modification pattern of acetylation marks at lysines 5, 12 and 16, which argues against a zipper model of lysine acetylation that had been suggested previously for mammalian histones. On the H3 N-terminus, we find acetylation of K9 and K14. At low abundance we detected di-acetylated peptides (K9ac + K14ac) in all stages, however, we were not able to find acetylation of K9 without the acetylation of K14. This suggests that K9 acetylation either requires previous acetylation of K14 or occurs simultanously. The level of mono-acetylation of this peptide decreases significantly as development proceeds from 45% in blastulae to 31% in gastrulae, after which it increases again gradually up to 37% in tadpoles. As it is the case for H4 mono-acetylation, the tissue culture cells contain less acetylated H3 ( panel B). Di-acetylated H3 K9/K14 molecules are much less abundant and their levels do not change significantly during development. However, the levels are slightly lower again in the two frog cell lines, compared to blastulae, which contain the lowest value of the four embryonic stages (1.17%, see). In addition to the K9 and K14 residues, the nearby lysines 18 and 23 on the H3 N-terminal tail can also be acetylated. We find a similar pattern for both mono- acetylated states with a significant shift between NF9 and the three other embryonic stages. Like for the K9/K14 pair, we find K18 acetylation only in combination with acetylated K23. These levels are between 4% in NF12 and 9% in NF37. Monoacetylated K23 ranges from 57% in Blastula to 45% in the other embryonic stages. In the two *Xenopus* cell lines, the level of K23 mono- acetylation is around 20% and the di-acetylation at 2%. Thus, also for these H3 acetylation marks, the four embryonic samples are very different from the cultured frog cell lines. Further inwards of the N-terminal tail, we also detected acetylation of K27. This Lysine can also be methylated by PRC2 complexes (see below). Embryonic samples contain H3K27ac at relatively constant levels of 1–2%. This modification was less abundant in all other samples (≤0.54% (, panel I), and among these, ES cells had the lowest K27ac levels (0.06%;). These differences, in particular between frog embryos and murine ES cells, are remarkable, given that acetylation of K27 is known to antagonize Polycomb silencing in flies, and to mark active enhancers. Besides N-terminal acetylation marks, we also detected acetylation in the H3 histone-fold domain at Lysine 56 (H3K56ac;). This modification has been suggested to represent a predeposition mark, and it was shown to be linked to the core transcriptional network of human ES cells. In mammalian cells increased H3K56ac levels correlate with tumorigenicity and the undifferentiated nature of cells. In frog embryos, H3K56ac levels are always low (0,11%–0,32% abundance) and comparable to the levels in the two frog cell lines. We have also found Lysine-56 to be methylated (predominantly me1) in both embryos and cultured cell lines at low abundance. Very little is known about this PTM, which has been detected so far in human and *Tetrahymena* chromatin. Interestingly, both K56me1 and K56ac levels reach a minimum at gastrula stages, i.e. when germ layers become specified. MEFs and murine ES cells contain somewhat higher levels of K56me1. Since this mark has also been observed in *Drosphila* embryos (A.I., unpublished), the ability to methylate H3K56 appears to be a conserved feature among metazoa. The most prominent histone modification associated with transcriptionally active promotors is di- and tri-methylated lysine 4 of H3 (H3K4me2/me3). In order to analyse the tryptic peptide 3-8 that contains Lys-4, we used a different de- salting method that can be applied to very small hydrophilic peptides. This new method improves the signal to noise ratio significantly, thereby allowing the quantitation of the differentially modified isoforms. We found that the amount of the unmodified peptide increases significantly during embryonic development from 27% in NF9 to 44% in NF37. This is largely due to a decrease in tri- methylation from 28% in NF9 to 16% in NF37. The levels of the mono- and di- methylated isoforms remain relatively constant. In comparison, the two cultured frog cell lines have much higher levels of unmodified (67% in A6 and 62% in XTC-2), and much lower levels of di- and tri-methylated H3K4 (≤10% each;, panel F). This data indicates an unusual and transcriptionally permissive state of the uncomitted blastula epigenome, in which half of the histone H3 N-terminal tails appear to be di- or tri-methylated at Lysine 4. ## Repressive marks One of the best characterized modifications that is associated with heterochromatin and transcriptionally repressed genic regions is di-/tri- methylation of lysine 9 on the H3 tail. In contrast, K9 mono-methylation has been associated with regulatory DNA elements and maintenance of activation potential during differentiation. In all our samples, we find K9 in mono-, di- and tri-methylated form, although at very different levels. The different modification states of K9 fluctuate less than two-fold during frog development, except for K9me1, which shows a small but significant transient peak at gastrula, when it accounts for 20% of the total H3 9–17 peptides. A similar distribution of mono-, di- and trimethylated K9 is found also in the two cultured *Xenopus* cell lines. Combined, the K9me2/me3 marks make up for less than 4% of the total H3 tails in the different *Xenopus* samples, suggesting a rather low density of this mark in both constitutive and facultative heterochromatin. The tryptic peptide 20-23 from propionylated histone H4 with the sequence KVLR has one lysine residue at the position 20, which is known to be methylated. Its different methylation states have been linked to diverse epigenetic functions. The H4K20me2/me3 marks are correlated with the formation of pericentromeric heterochromatin, maintenance of genome integrity and transcriptional repression, while the role of H4K20me1 is not fully understood (see also review by). Using MALDI-TOF, we found in all samples peptides corresponding to unmodified, mono- methylated and di-methylated states. The amount of the unmodified peptide decreases slightly from blastula to tadpole stage, and is even lower in the A6 (20%) and XTC-2 (17%) cell lines. It is known that the specific properties of the tri-methylated H4 20-23 peptide make it very difficult to quantify this PTM by mass spec, and we have not reproducibly detected it in our samples. To determine the relative abundance of H4K20me3 in frog embryos, we performed Western blot analysis with antibodies against methylated H4K20 as well as pan- histone H3. This confirmed the consistent presence of the H4K20me3 mark during embryonic development (; see for cultured cells), and revealed also that its abundance increases appr. five-fold from blastula to tadpole stages. ## Opposing marks on the histone H3 27-40 peptide Methylation of lysines 27 and 36 within the histone H3 tail has been mapped to mutually exclusive regions within the genome of most eukaryotes. The H3K27me2/me3 marks are almost exclusively found at inactive genomic regions, whereas the H3K36me2/me3 marks are localized predominantly in actively transcribed genes. Both modifications reside on the same tryptic peptide (H3 residues 27–40), which allows us to determine combinations of modification states on these two lysines by mass spectrometry. However, as peptides carrying a single trimethyl group at either K27 or K36, or combinations of a monomethyl and a dimethyl group at these two lysines, are isobaric (i.e. they have the same mass), they cannot be distinguished solely by this parameter (see). Although tandem MS/MS analysis can facilitate the assignment of the modifications, the relative quantitation based on extracted ion chromatograms is only possible when the differentially modified isobaric peptides are physically separated. Indeed, tandem MS/MS spectra revealed that most peptides that have a higher methylation degree at position 27 elute earlier from a C18 reversed phase column than the corresponding ones that carry the same number of methyl groups on K36. With the exception of the isobaric mono-methylated peptides (i.e. K27me1 <u>or</u> K36me1), which were not separated on the C18 micro-column, this observation enabled us to identify 12 additional, distinct methylation states for the K27 and K36 positions on single H3 27–40 peptides. In genome wide studies, tri-methylation of H3K36 is found at strongly expressed genes, which are also preferentially enriched for the replication-independent histone variant H3.3. On canonical histone H3 (i.e. replication-dependent H3.2; ref.) that is highly abundant in early embryos, we have detected H3K36me3 by its fragmentation spectrum only in frog tadpoles at very low abundance. Its relative absence from embryos could be explained, if frogs deposit the K36me3 mark selectively on the replication-independent H3.3 variant, as has been reported for murine ES cells. The ratio of H3.3 to H3.2, as measured by comparing all spectra derived from the peptides unique for each isoform, decreases during development from 20% in blastula to approximately 7% in tadpoles. However, in all cases we find a very similar distribution of modifications within these peptides, and therefore focused our analysis on the peptide derived from the more abundant, canonical H3 molecule. Overall, this data indicates that H3K36me3 levels are low during frog development. Consequently, these findings imply that practically all the single tri- methylated 27–40 peptides of H3 represent K27me3, a conclusion that is supported by the fragmentation spectra. We found that H3K27me3 was readily identified in many, but not all samples. Notably, it was not detected in blastula and gastrula, and appeared only at very low levels in neurula (0.05%) and tadpole stages (0.1%;, C and D). However, in Xenopus cell lines (panel I), as well as in mammalian cells (see below), H3K27me3 levels were at least 100-fold higher than in bulk neurula chromatin. We conclude that the repressive K27me3 modification is present in our histone extracts, but extremely underrepresented in the early frog embryo compared to cultured frog and mammalian cell lines. Overall, the abundance of methylated H3K27 rises during development. This trend became particularly obvious, when we compared the XIC profiles of the di- methylated peptide species. While the K36me2 peak stayed more or less constant, the K27me2 peak increased gradually from 1.6% in the blastula to 14% in the tadpole stage. During this time, the ratio of H3K27me2 to H3K36me2 changed from 0.14 in the blastula to 1.6 in the neurula stage, i.e. by more than 10-fold. In comparison, the cultured cell lines A6 and XTC-2 showed the highest percentage of H3K27me2 peptides (33% and 22% respectively), and contained more than 10% of H3K27me3 (panel I). The abundance of peptides with combinatorial methylation on both K27 and K36 residues rises also siginificantly during development. In the frog blastula, less than 2% of H3 tails are simultanously modified on these sites, while one third is combinatorially modified in tadpoles. Among these, we found also low levels of unexpected combinations such as K27me3 paired with K36me1/me2 in late embryonic and tissue culture cell samples. In general, the level of modifications within this peptide shows a major increase during development. This trend may be connected functionally to the establishment of stable gene expression profiles as cells differentiate. ## Methylation states of pluripotent frog blastulae and murine ES cells Most cells in the blastula stage of *Xenopus* embryos are uncomitted and capable to differentiate into derivatives of all three germ layers. This raised the issue, whether the histone modification profiles we observed in the frog blastula represent common features of pluripotent cells. To address this point, we continued to analyse the histone PTMs of murine ES cells and of primary embryonic fibroblasts (MEFs), which were used as feeder cells in the ES cell culture and represent a murine somatic cell type for comparison. When we initially compared MEFs with ES cells, we observed different degrees of acetylation on the H3 18–26 and H4 4–17 peptides, indicating that we separated successfully ES cells from MEF feeders. In depth analysis confirmed then that the two murine cell types contained very specific profiles of histone modifications, and that the histone modifications of pluripotent cells from Xenopus blastulae were related to, but also clearly distinguishable from the profiles of pluripotent murine ES cells. Bivalently marked chromatin fragments, which simultanously carry the active H3K4me3 and the repressive H3K27me3 marks, are considered a hallmark of pluripotency. With our technical approach we cannot determine, whether histone H3 tails are modified on both Lysine 4 and 27, since these residues are located on different tryptic peptides. Nevertheless, we compared the overall abundance of modifications at these sites. For H3K4, this comparison revealed a high similarity between Xenopus blastula and murine ES cells. Of all samples, these two contained the lowest amount of unmodified K4 (below 30%), but maximal amounts of H3K4me2/me3 (\>50% together). In contrast, H3K4me1 levels were not only similar between blastula, ES cells and MEFs, but almost constant in all samples (3E). In contrast to H3K4, the methylation profiles of the H3 peptide 27–40, including Lysines 27 and 36, were very different between blastulae and ES cells. As described earlier, the majority of the di-methyl marks reside on K36 in frog blastulae. In ES cells, however, they were found predominantly on K27. MEFs had a similar dimethyl-distribution as ES cells, but contained significantly more K36me2 than these. A dichotomy was also found for the higher modification states of this peptide, including three to five methyl groups. Whereas in frog blastulae H3K27me3 is below the detection limit, mouse ES cells had with 8.3% the third-highest K27me3 abundance of all samples tested, and twice as much as MEFs. This huge quantitative difference is also seen in western blots. Thus, both mass spec and western blots indicate independently that the epigenome of ES cells is more than 100-fold enriched for H3K27me3 compared to frog blastulae. Combinatorial tri-methylated states, such as K27me1/K36me2 and K27me2/K36me1 were also significantly more abundant in ES cells than in frog blastulae, and comparable to the levels found in MEFs. Notably, the highest levels of tetra- (K27me2+K36me2) and penta-methylated (K27me3+K36me2) states of the H3 27–40 petides were also found in ES cells and to a less extent in MEFs, but were not detected in frog blastulae. Except for mono-methyl (K27 <u>or</u> K36) and di- methyl (K36) states, frog blastulae contain the lowest methylation levels for the H3 27–40 peptide of all the samples tested. Overall, these results indicate that frog blastulae and murine ES cells share high H3K4me2/me3 levels, but the former are dramatically undermethylated at the H3 27–40 peptide. In fact, uncommitted frog cells have a 15-fold higher level of the unmodified state of this peptide than pluripotent GS-1 ES cells. Based on this, the polycomb repression complex 2 (PRC2) activity seems to have a very different impact on the epigenomes of frog embryos and *in vitro* cultured murine ES cells. ## Stage specific histone modification profiles We have observed significant quantitative changes in a relatively large number of histone modification states during frog embryogenesis. In order to investigate, whether particular developmental stages could be described by characteristic groups of histone modification states, we performed hierarchical clustering analysis to identify potential epigenetic patterns, and we then generated a Heatmap of the obtained clusters to visualize the results (see for details). The dendrogram on the left axis of the Heatmap shows four well-defined clusters between the second and third level. The most obvious features of the heatmap are the four red-colored areas, each representing a cluster of histone modifications at their individual maximal abundance, which describe the four developmental stages that we analysed. The separation is very clear between blastula and tadpole clusters, where the red groups generally indicate a global shift from active to repressive histone marks as the embryonic cells undergo differentiation. It is less clear between gastrula and neurula stages, in particular at the bottom third of the heatmap, where several histone modification states persist at similar, intermediate abundances until the tadpole stage. As gastrula and neurula stages are biologically characterized by ongoing cellular diversification, these modifications may be associated with transitory features of germ-layer comitted precursor cell populations. Importantly, the apparent clustering provides unique evidence for a global, stepwise maturation of the embryonic epigenome in vertebrate embryos. # Discussion To our knowledge, this study represents the first description of bulk histone modifications and their steady state flux as it occurs *in vivo* during vertebrate embryogenesis. Using high accuracy LC-MS/MS and MALDI-TOF mass spectrometry, we have detected 59 different modification states on 18 sites - all Lysine residues, with the exception of histone H3 Serine 57 - derived from canonical H3 and H4 histones. These results provide key insight into global alterations of histone PTMs during development from uncomitted cells of the late blastula to the differentiated tissues of hatched, premetamorphic tadpoles. Mass spectrometry is a very quantitative technique, as shown by the proportionate increase in XICs upon loading of different amounts of synthetic peptides (both unmodified and modified) over several orders of magnitude. In addition, we have detected the full range of possible modifications for many peptides. The modifications that we have not found are known to be either of low abundance (Arginine methylation, Lysine ubiqitination) or unstable (phosphorylation; see. Finally, Western blots for the H3K27me3 levels of frog embryos and ES cells confirm the quantitative differences obtained by mass spec. Altogether, this indicates that the reported relative abundances of modification states are robust, and that it is very unlikely that they could be inflated due to non-detected modifications. At the time of germ layer induction, frog embryonic cells are characterized by a low point of repressive histone modifications, including H3K27me3. Our findings are in agreement with recent ChIP-seq results from Xenopus tropicalis and suggest that the formation of somatic cell lineages in Xenopus species occurs from an epigenetic state practically devoid of bivalent domains. Future studies in Xenopus will have to address, whether the derepressed state at blastula exists also in early pre-MBT stages such as the morula (see also below). Global changes in histone modifications occur in mouse embryos between fertilisation and the blastocyst stage (reviewed in refs.. However, histone PTMs have not been investigated from post-implantation embryos, except for one pioneering study using carrier ChIP. Here, both K4me3 and K27me3 marks were found on several developmental loci in epiblast tissue, but the bivalent status of these loci was not rigorously established and the opposing marks may derive from distinct cell populations. Nevertheless, this study demonstrates that K27me3 exists in the mouse embryo at the time of germ layer formation, unlike in Xenopus embryos. By hierarchical clustering analysis, we found that a large fraction of the observed histone modifications fall into profiles that apparently characterize each of the analysed embryonic stages. This finding extends the hypothesis of individualized epigenomes for different cell types to whole embryos. This surprising result does not necessarily imply that all cells of an embryo at a given stage share the same histone PTM profile, but rather reveals the existence of stage-specific constraints that shape the histone modifications according to global cellular transitions, such as the development from pluripotency to germ layer precursor to committed and differentiated cell states. In line with this hypothesis, the clearest differences are observed between histones from post- MBT, uncomitted cells of blastulae (NF9) and histones from the largely differentiated, somatic cells of tadpoles (NF37). Overall, the four clusters describe a general decrease in the abundance of active histone PTMs with a concomittant increase of repressive histone modifications. For future experiments, these clusters provide a blueprint of stage-specific chromatin states that can be used to investigate the impact and interrelationship of individual histone modifying enzymes. We noted that some modification types - e.g. K9me1, K14ac, K36me2 on histone H3, or K16ac on histone H4 - remained remarkably constant (\<two-fold changes) over this time course. Even though the local distribution of the PTMs at these histone residues will vary between cell types, this finding suggests relatively static or basic functions for them. For other sites, the steady state of PTMs can change very strongly within few hours, most notably between the late blastula and midgastrula stages. Here, the largest differences were observed on histone H3 for K27me1/K36me1 and K27me1/K36me2 (18-fold and 9.8-fold increase, respectively), as well as for acetylation and mono-methylation of H3K56 (11.5-fold and 18-fold decrease, respectively). Another example is represented by double-acetylation of Lysines 8 and 16 on H4, which rises from below detection limit to 11% abundance within this short time window. The disproportionate change in the steady state of these specific PTMs suggests that they are regulated by some developmental mechanism, possibly involving the signalling cascades that drive embryonic induction and cell differentiation. Since Lysines 27 and 36 are present on the same tryptic peptide, it was possible to distinguish a total of 12 different modification states. Specifically, almost half of the unmodified fraction of this peptide becomes modified in steady state between blastula and gastrula. This dramatic increase in K27 methylation is consistent with the results of recent ChIP-studies in *X. tropicalis* and *D. rerio*, which reported a hierarchical acquisition of H3K4 and K27 methyl marks. In extension to these studies, however, we have found that the K27me3 mark is at least 100-fold less abundant in *Xenopus* early embryos than in cultured frog or mammalian cell lines, including murine ES-cells. As a consequence, the K27me3 density in chromatin may be rather low throughout the embryo. Alternatively, it could be high within a subfraction of either cells or genomic regions, and consequently being depleted from the rest. Whatever may be the case, this finding is very unexpected, given current models on the epigenetic nature of pluripotency. Functional tests are required to clarify, whether at this low abundance (\<0.1%) H3K27me3 represses transcription in Xenopus blastulae as reported for human and murine ES cells, in which H3K27me3 is much more abundant (refs., ; and our data here). It is not known, why the chromatin of the frog blastula is so devoid of the H3K27me2/me3 marks. Recently, an antagonistic relationship between methylation and acetylation of Lysine 27 has been reported that reflects a competition between PcG and trithorax/MLL complexes, which recruit H3K27 methyltransferase or acetyltransferase activities, respectively. Since frog embryos contain the highest relative K27ac levels in our data set, one might speculate that some regulatory mechanism in blastula embryos favors MLL-mediated recruitment of HATs. However, a direct competition between trx and PcG proteins on target genes is unlikely to explain entirely the initial depletion in H3K27 methylation, because K27ac levels remain high during subsequent development despite the dramatic increase in K27me2/me3 from gastrula stages onwards. Nicklay and colleagues have incubated sperm nuclei with Xenopus egg extract *in vitro* to generate chromatin of “early embryo equivalent” that may represent the chromatin state before or at MBT. This sample contains quite abundant K27me3 levels (estimated 10–50%), confirming the maternal expression of biologically active PRC2 components in early embryos. However, neither we (this study) nor others have found evidence for this in NF9 blastulae. If chromatin of the “early embryo equivalent” corresponds to a physiological state, one would have to postulate that practically all K27me3 marks become erased before midblastula. The presence of maternal and potentially active PRC2 complexes in turn suggests for the unmanipulated frog embryo that PRC2 activity is either negatively controlled or efficiently antagonized by demethylation around the onset of zygotic transcription. The sudden increase of K27me2 at gastrula suggests a regulatory switch that quite rapidly brings a significant portion of the epigenome under the influence of the *polycomb* system. This switch may include the upregulation of Suz12 mRNA levels that has been described during gastrulation. During the amphibian blastula-gastrula transition, pluripotent embryonic cells are programmed by induction to germ layer related cell fates. This phase corresponds in mammalian embryos to the specification of the epiblast at implantation, from when on pluripotency is gradually lost. Although being low abundant, acetylation of H3K56 is specifically enriched at promoters bound by the core pluripotency factors Oct4/Sox2/Nanog. When ES cells differentiate, K56ac relocates to developmental regulatory genes. We have found that K56ac levels drop more than ten-fold between blastula and gastrula. It is possible that the transient depletion of K56ac from gastrula chromatin reflects the reported translocation of this mark, when pluripotency is lost in ES cells, and possibly in mammalian embryos. At the same time, the embryonic K9me1 level rises to a transient maximum of 20% abundance. H3K9me1 has been recognized as a stable mark of cis-regulatory elements for both active and inactive genes. Together, the observed changes in the H3K56ac and H3K9me1 profiles may thus be important for the programming of cell fates. The extraordinary abundance of H3K4me3 (almost 30%) in frog blastulae represents another puzzling observation. In mammals, the H3K4me3 mark is found on gene promoters and enhancers. Given that frog and human genomes are comparable in size and gene numbers, this level exceeds the expected genome proportion represented by these transcriptional regulatory elements. Notably, murine ES cells have similarly high levels, and even in somatic cells (A6, XTC and MEFs) its abundance is around 10% (see, and). Due to the unusual biophysical properties of this peptide, we could not record MS2 events, which would have allowed us to distinguish acetylated and trimethylated states. Although H3K4 acetylation levels are generally low, it is possible that some fraction of the H3K4me3 peak represents acetylated peptides. Nevertheless, one may speculate that H3K4me3 could be deposited differently in embryonic cells compared to somatic cells. In addition to the typical promoter-proximal peaks, which have been observed in recent genome-wide ChIP experiments both in Xenopus and zebrafish, this mark could exist at higher base-levels throughout the embryonic epigenome, and/or be enriched significantly in still unannotated areas. The high abundance of H3 K4me3 in embryos certainly merits further investigation. In general, the HMP of the frog blastula describes a highly permissive chromatin environment at the time of embryonic induction that is remarkably devoid of repressive methyl marks on H3K9, H3K27 and H4K20. This transcriptionally permissive epigenome may mechanistically explain, why some region or cell-type specific genes are initially activated in apparently ectopic manners in frog embryos. Subsequent refinement of gene expression patterns during gastrulation requires *de novo* synthesis of repressors, possibly including epigenetic repressors such as the PRC2 component Suz12. Based on our data, and on ChIP-Seq derived genome wide maps for H3K4 and H3K27 methyl marks, the epigenetic state of pluripotent Xenopus embryonic cells can thus be defined as “active” and “derepressed”. In this sense, they differ significantly from pluripotent mammalian ES cells and early post-implantation mouse embryos, which contain represssive marks such as H3K27me3 (our data on GS1-ES cells;). Our findings with Xenopus embryos underscore the necessity to further explore the epigenetic ground state of pluripotency in primary cells. # Materials and Methods ## Ethics Statement Animal work has been conducted in acordance with Deutsches Tierschutzgesetz; experimental use of Xenopus embryos has been licenced by the Government of Oberbayern (Projekt/AZ ROB: 211-2531.6-11/2001). ## Embryos and cell lines Embryos were generated by in vitro fertilization with adult X.laevis animals (Nasco, *Xenopus* Express) as described. Stable cell lines A6 (derived from adult frog kidney tissue) and XTC-2 (derived from Xenopus tadpole carcass)) were grown in 75% DMEM supplemented with 10% FCS, 100 u/ml penicillin/ 100 ng/ml streptomycin at 26°C in a humidified atmosphere with 5% CO₂. Murine embryonic fibroblasts were isolated from ICR mice at day 13 of pregnancy and subsequently grown in DMEM, 10% FCS, 2 mM L-glutamine, 50 U/ml penicillin, 50 µg/ml streptomycin. For chromatin analysis, non inactivated MEFs at passage 2 were used. Murine GS-1 ES cells were grown on inactivated MEFs (3 hrs. 8 µg/ml mitomycin C) in high glucose Dulbecco's modified Eagle medium supplemented with 10% heat-inactivated ES-qualified fetal calf serum, 2 mM L-glutamine, 50 U/ml penicillin, 50 µg/ml streptomycin, 1x nonessential amino acids (all reagents GIBCO BRL, Germany) and 0.1 mM ß-mercaptoethanol (Sigma, Germany). For chromatin analysis, the ES cells were transferred to feeder free conditions on gelatinated plates for three passages to avoid contamination of probe material with MEF derived chromatin. During feeder-free culture, ES cells were kept undifferentiated by adding 1000 units/ml purified recombinant mouse leukemia inhibitory factor (ESGRO; Life Technologies, Inc., Grand Island, N.Y.) to the medium. Monolayers were passaged by trypsinization at 70–80% confluency and the cells were maintained at 37°C in a humidified atmosphere of 5% CO₂/95% air. ## Histone Extraction Around 50 to 100 embryos developed to Blastula (NF9, 7 hpf), Gastrula (NF12, 13 hpf), Neurula (NF18, 20 hpf), and Tadpole (NF37, 54 hpf) stages were harvested and washed with 110 mM KCl, 50 mM Tris/HCl (pH 7.4 at 23°C), 5 mM MgCl₂, 0.1 mM spermine, 0.1 mM EDTA, 2 mM DTT, 0.4 mM PMSF, 10 mM Na- butyrate. Nuclei of the embryos were prepared by centrifugation with 2600xg, 10 min (3–18, Sigma) after homogenization by a 5 ml glass-glass douncer (Braun, Melsungen). For cell lines, 10⁷ cells were collected and washed twice with 1 ml PBS, 0.1 mM spermine, 0.1 mM EDTA, 2 mM DTT, 0.4 mM PMSF, 10 mM Na- butyrate and then resuspended in 1 ml PBS buffer containing 0.3% Triton. The cells were put on a rotating wheel at 4°C for 20 min and then centrifuged with an Eppendorf Centrifuge 5417C with top speed. The supernatent containing the cytoplasmatic fraction was discarded. The nuclear pellets of both *Xenopus* embryos and cell lines were resuspended in 1 ml 0.4 M HCl, incubated on a rotating wheel over night and dialysed against 3l of 0.1 M acetic acid/1 mM DTT. The dialysed histone solution was vacuum-dried in a Concentrator Plus (Eppendorf) and stored at −20°C. ## Mass Spectrometry Histones were separated by sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE). Coomassie blue-stained bands were excised and propionylated as described before. Digestions were carried out overnight with 200 ng sequencing grade trypsin (Promega); Before analysis, the peptide solution was desalted using Zip-Tips (Millipore) or porous carbon material (TipTop Carbon, Glygen). Bound pepties were eluted in matrix solution (saturated -cyanohydroxycinnamic- acid \[Sigma\] dissolved in 50% acetonitril \[vol/vol\], 0.3% trifluoroacetic acid \[vol/vol\]) and spotted onto a target plate. The target plate was loaded in a Voyager DE STR spectrometer (Applied Biosystems). The resulting spectra were analyzed using the Data Explorer and in-house- developed software Manuelito (open source: <http://sourceforge.net/projects/manuelito>). For quantification of the different post-translational modifications of the various peptides, the relative intensities of each were measured. The areas of all modifications of one peptide were summarized and thus the percentage of each modification was calculated. The statistical significance of quantitative changes during development or between samples was determined by Student's T-test. For LC-MS/MS analysis the digested histones were loaded onto a reversed-phase HPLC in an analytical C18 column (75 µm i.d packed with C18 PepMap™, 3 µm, 100Å by LC Packings) and eluted using a 80 min gradient from 5 to 60% ACN in 0.1% FA. Spectra were recorded by a LTQ-Orbitrap mass spectrometer operating in a top six mode where the six most intense peptide ions with charge states between 2 and 4 were sequentially isolated and fragmented in the linear ion trap by collision- induced dissociation (CID). All fragment ion spectra were recorded in the LTQ part of the instrument. For all measurements with the orbitrap detector, 3 lock- mass ions from ambient air (m/z = 371.10123, 445.12002, 519.13882) were used for internal calibration as described. Typical mass spectrometric conditions were: spray voltage, 1.5 kV; no sheath and auxiliary gas flow; heated capillary temperature 200°C. For the analysis of different histone modifications the resulting RAW files were converted into DTA and searched against the NCBI non- redundant database using the SEQUEST search algorithm to identify the corresponding histones using the BIOWORKS3.3 software package (Thermo). In order to determine the posttranslational modifications on the histone peptides, the data files were searched against a targeted histone database containing *Xenopus laevis* histone molecules from the NCBI non-redundant database by SEQUEST search algorithm with the following modification settings (static modification on lysine: 56.0262; variable modifications: me1: 14.01565; me2: -27.99490; me3: -13.97925; ac: -14.10565 on K and phos: 79.96633 on S). The resulting SEQUEST files were filtered for a peptide probability score of 0.0005 and Xcorrelation values of 1.5, 2.0, 2.5 and 3 for charge states 1, 2, 3 and 4 accordingly. A similar search against a reversed and scrambled histone decoy database did not lead to an identification of peptides that matched the filter criteria. Therefore we estimated the false discovery rate to be less than 1%. For quantification, only modifications with a probability score of lower than 0.0005 and top hits were used. For all peptides, which were quantified, an extracted ion chromatogram (XIC) was obtained from the raw file using the Xcalibur 2.0.7 software using the default ICIS peak detection algorithm (Thermo), extracting doubly and triply charged ions with a user defined mass tolerance of 5–10 ppm and a mass precision of 4 decimals to be able to distinguish between nominally isobaric modifications such as acteylation or trimethylation. The retention time for each specific peptide on the chromatography was highly specific when we used the same LC MS/MS method. We therefore only used these XIC areas for quantification, which eluted at time points where we observed at least two MS/MS spectra that unambiguously identified the identity of the peptide and the position of the modification within the peptide. ## Quantitative Western blotting Western blot analysis for H4 K20me3 was carried out according to standard protocols. Nuclei of 0.5–2 embryo equivalents were used, depending on developmental stage. After SDS-PAGE, proteins were blotted with 1.5 mA/cm2 for 1 h to nitrocellulose membranes (porcello NCL, Macherey-Nagel), which were blocked for 1 h using PBS, 3% BSA. Primary antibodies detecting H3K27me3 (1∶3 000; ref. ), H4K20me3 (1∶500; ref.) or core region-H3 (Abcam, 1∶25.000) were incubated using PBS, 3% BSA. Secondary fluorophore linked antibody (Li-cor, IRDye 700DX goat anti-rabbit, 1∶5 000) was incubated in PBS, 5% Milk powder, 0.1% Tween 20. Fluorescent signals were recorded using Li-cor Odysse wet membrane technique for 700 nm absorbance. ## Heatmap Generation We generated the Heatmap shown in using R (<http://www.r-project.org>) and the *gplots* package (<http://CRAN.R-project.org/package=gplots>). All functions were called using default parameters if not indicated otherwise. We first calculated the mean and the standard deviation of the quantifications obtained by Mass Spectometry across the four developmental cell stages (each stage containing two biological replicates) for each of the 57 modifications. Histones with small mean values contain low Mass Spectometry quantifications that could obscure the clustering analysis that we observe in the heatmap. Thus, as a preprocessing step of the data, we discarded modifications with mean values less than 0.5% (22 modifications). In order to compare appropriately the contribution of each modification to the specific cell stages, we standardized the mass spectometry quantifications individually using Z-scores. Using these Z-scores we then produced the Heatmap using the Ward's minimum variance method to perform the hierarchical clustering analysis. # Supporting Information We thank the following colleagues for their generous support: Gunnar Schotta for antibodies, Barabara Hölscher and Christiane Groß for cultured cell production, A. Villar-Garea, L. Israel and I. Forné for support and advice with Mass Spec, and Gunnar Schotta, Gustav Klobeck, and Sandra Hacke for comments on the manuscript. [^1]: Conceived and designed the experiments: TDS AI RAWR. Performed the experiments: TDS EM RD DN. Analyzed the data: TDS AI JMA-S RAWR. Contributed reagents/materials/analysis tools: TDS EM AI. Wrote the paper: TDS JMA-S RAWR. [^2]: The authors have declared that no competing interests exist.

# Introduction The Nuclear Factor kappa B (NF-κB) family of dimeric transcription factors regulates gene expression involved in multiple biological processes including immune and inflammatory responses and control of cell proliferation and death. In the absence of stimuli, a NF-κB dimer is kept inactive in the cytoplasm of most cells via association with a member of the inhibitor of NF-κB (IκB) family, which includes IκBα. An important regulator of canonical NF-κB signaling is the IκB kinase (IKK) complex which consists of two catalytic subunits, IKKα and IKKβ, and a regulatory subunit, NEMO (NF-κB essential modulator, IKKγ). Once activated by incoming signals, the IKK complex phosphorylates IκB to stimulate polyubiquitination and proteasome-mediated degradation to release NF-κB. The liberated NF-κB translocates to the nucleus and regulates its target gene expression. Because the IKK complex represents the convergence point in activating canonical NF-κB signaling, a considerable amount of research has been conducted to understand the mechanism of activation and regulation of the IKK complex. In particular, the role of the non-catalytic subunit NEMO in IKK complex regulation has been studied intensively (reviewed in). These studies have highlighted the role of NEMO as a ubiquitin binding protein to promote IKK activation. The recruitment of NEMO to polyubiquitin scaffolds assembled by the upstream signaling events permits the recruitment of the catalytic IKK subunits to the upstream kinase, TAK1 (TGFβ activated kinase 1), which is also recruited to the ubiquitin scaffolds via its ubiquitin binding subunits, TAB2/3, to be phosphorylated and activated. In addition to its well accepted role as an IKK regulatory subunit, NEMO also performs an additional upstream role to permit communication between the nuclear DNA damage activated kinase ATM (ataxia telangiectasia mutated) and the cytoplasmic IKK complex to induce NF-κB signaling in response to genotoxic agents (reviewed in). To investigate the distinct functions of NEMO, a variety of mutant forms of NEMO have been previously analyzed. However, since overexpression of NEMO can result in inhibition of NF-κB signaling, it is important to control the amount of NEMO expressed in different cell systems in order to define the various functions of NEMO without being confounded by artifacts associated with high NEMO expression levels. Even so, it is generally undefined what the physiological levels of NEMO are, i.e., how many NEMO molecules are expressed in different cell systems, and what the impact of different expression levels of NEMO is on NF-κB activation by different stimuli, including genotoxic agents. To answer these questions, purified recombinant full-length NEMO is needed to provide standards for quantification of cellular NEMO levels. Since purification of soluble, recombinant full-length NEMO from *E*. *coli* in high concentrations is technically challenging, we first optimized a NEMO purification protocol from *E*. *coli*. We then quantified the average number of NEMO molecules per cell using 1.3E2 NEMO-deficient murine pre-B cell clone stably reconstituted with full-length human NEMO (C5) previously characterized by our lab. Next, we generated 1.3E2 clones stably expressing different amounts of NEMO to determine the impact of known numbers of NEMO per cell on NF-κB activation by the DNA damaging agent etoposide, which creates DNA double strand breaks via inhibition of topoisomerase II. Our study demonstrates that a 10-fold range of NEMO levels has surprisingly little impact on NF-κB activation by etoposide and suggests that mutant NEMO phenotypes can be analyzed accurately as long as the level of NEMO remains within this range. We also used the C5 cell line to determine the average number of NEMO molecules per cell in several human cell lines often employed in NF-κB signaling studies to highlight the utility of the C5 cell line as a standard to determine the cellular content of NEMO proteins in different cell systems. # Materials and Methods ## Reagents The following antibodies were used: antibody against NEMO (FL-419, sc-8330, Santa Cruz Biotechnology), Flag-HRP (A8592, Sigma), Tubulin (CP06, EMD Millipore), Actin (sc-1616, Santa Cruz Biotechnology), and IKKα/β (sc-7607, Santa Cruz Biotechnology). Anti-rabbit and anti-mouse antibodies conjugated to horseradish peroxidase were obtained from GE Healthcare. Etoposide (E1383, Sigma) and bacterial LPS (L2880, Sigma) were purchased from Sigma-Aldrich. Glutathione-agarose beads and Halt Protease Inhibitor Cocktail was obtained from Thermo Scientific. ## Purification of recombinant full-length human NEMO protein To purify recombinant full-length human NEMO, NEMO/pGEX6p-1 plasmid was transformed into Rosetta2 BL21 cells and a transformant was grown in 20 ml of LB medium supplemented with 100 μg/ml ampicillin and 34 μg/ml chloramphenicol at 37°C for overnight. The overnight seed culture was diluted in 2 L of LB medium containing 100 μg/ml ampicillin and 34 μg/ml chloramphenicol and grown at 37°C until the absorbance at 600 nm reached 0.6–0.7. The bacterial culture was kept at 4°C for 30 minutes while the temperature of the shaker was cooled down to 28°C. 0.5 mM isopropyl β-D-thiogalactoside (IPTG) was then added to the culture, and the culture was incubated at 28°C for 3 hours to induce the protein. After harvesting the cells by centrifugation at 4000 rpm for 20 minutes at 4°C, the bacterial pellet was washed with 1X PBS and stored at −20°C for later purification or resuspended in 30 ml of lysis buffer (20 mM HEPES-KOH/7.6, 300 mM potassium acetate, 20% glycerol, 10 mM magnesium chloride, freshly added 1X Halt Protease inhibitor cocktail, freshly added 5 mM β-mercaptoethanol) per 1 L bacterial culture along with 125 μg/ml lysozyme. The culture was then incubated for 20 minutes on ice. The lysate was sonicated with 6 short burst of 20 sec followed by intervals of 40 sec for cooling and centrifuged at 10000 rpm for 20 minutes at 4°C to remove cell debris. The lysate was centrifuged again at 23000 rpm for 60 minutes at 4°C to remove the any leftover cell debris. The lysate was diluted up to 150 ml of lysis buffer. To equilibrate glutathione agarose beads (Thermo Pierce), 1 ml of the beads (wet bed volume) were washed with lysis buffer twice and carefully poured into a closed column. The beads were then allowed to settle without flow. Once the beads were gravity packed, 100 ml of lysis buffer was run through by gravity flow. The diluted lysate prepared above was loaded onto the equilibrated column and ran at a rate of 30 ml/hr at 4°C. The flow-through was reloaded onto the column and the binding step was repeated twice with a 50 ml/hr flow rate at 4°C. The beads were washed with 150 ml of lysis buffer with 30 ml/hr flow rate at 4°C. For elution of GST-NEMO, 300 μl of freshly prepared elution buffer (10 mM reduced glutathione, 50 mM Tris-HCl/8.0, freshly added 1X Halt Protease inhibitor cocktail, freshly added 5mM β-mercaptoethanol) was applied using gravity flow and collected in a 1.5 ml microcentrifuge tube. For cleavage of GST by GST- human rhinovirus 3C protease, the beads containing GST-NEMO were resuspended in 9 ml of lysis buffer, transferred to a new 14 ml conical tube, and washed with 10 ml of lysis buffer twice. 100 μl aliquot of the beads (wet bed volume) was transferred to an Eppendorf tube and twice washed with dialysis buffer (50 mM HEPES-KOH/7.6, 100 mM potassium acetate, 20% glycerol, freshly added 5 mM β-mercaptoethanol). 400 μl of dialysis buffer containing 4 μg of GST- human rhinovirus 3C protease was added to the wet beads and the beads were rotated at 4°C overnight. The beads were then centrifuged at 1500 rpm for 20 sec at 4°C to remove GST- human rhinovirus 3C protease and to recover the GST cleaved NEMO protein in the supernatant. To determine the concentration of the purified recombinant NEMO, proteins were analyzed on SDS-PAGE gel together with a serial dilution of BSA at known concentrations. The concentration of GST cleaved NEMO protein was confirmed by absorbance at 280 nm using Nanodrop (Thermo Fisher Scientific Inc.) using the extinction coefficient of 14440 M⁻¹ cm⁻¹ which was calculated from the NEMO sequence. ## Preparation of GST- human rhinovirus 3C protease GST- human rhinovirus 3C protease production was induced in BL21 strain *E*. *coli* with 0.5 mM IPTG for 4 hours at 25°C. Cells were collected by centrifugation, resuspended in lysis buffer (1X PBS, 250 mM NaCl, 0.1% Tween-20, 10 mM β-mercaptoethanol, 1 mM PMSF, 1 mM benzamidine), and lysed by sonication with 1 mg/ml lysozyme. Lysate was clarified by centrifugation for 1 hour at 45,000xg and incubated with glutathione agarose beads at 4°C for 1 hour. The beads were washed extensively with lysis buffer. Proteins were eluted with reduced glutathione added to the wash buffer, dialyzed into cleavage buffer (50 mM Tris-Cl pH 8.0, 75 mM KCl, 50% glycerol), and maintained at −20°C. ## Validation of purified recombinant NEMO by MS/MS analysis A 48 kD protein band from Coomassie blue stained sodium dodecyl sulfate (SDS)-PAGE gel was sliced out, cut into smaller pieces into an Eppendorf tube, and destained by washing three times for 10 min in 200 μl of 50% (v/v) acetonitrile solution containing 50 mM NH₄HCO₃. The gel pieces were then reduced by 50 μl of 20 mM dithiothreitol, 100 mM NH₄HCO₃ for 30 minutes at 55°C, alkylated in 50 μl of 20 mM iodoacetamide solution, washed with 100% acetonitrile, and dried at room temperature. 10 μl of modified trypsin (Promega, sequencing grade) (100 μg/ml) in 100 mM NH₄HCO₃ was added on the dried gel pieces in the Eppendorf tube and incubated at 37°C overnight. The tryptic peptides were injected onto the C18 column (Waters, 100um x 10cm) and eluted with a linear gradient for 60 minutes from 1 to 45% acetonitrile (in water with 0.1% FA) at a flow rate of 500 nl/min. The effluent was electro-sprayed into the LTQ (Thermo Scientific) mass spectrometer. Database search was carried out using both Sequest and X! Tandem algorithms, and identification confidence was calculated by a published method. Human protein sequence database (Uniprot) was used. ## *In vitro* translation and GST-pull down and co-immunoprecipitation assay Flag-IKKβ/pcDNA3.1(+) was *in vitro* translated with TNT Coupled Reticulocyte Lysate Systems (Promega) as described by the manufacturer using 1 μg of the plasmid and 50 μl final reaction volume. 1/50 of the reaction was separated by 10% SDS-PAGE and verified by immunoblotting with Flag antibody. For GST pull- down assay, 1 μg of GST or GST-NEMO protein was incubated with 1/5 of the translation reaction containing mock or IKKβ in 800 μl of IP buffer (50 mM Tris- HCl, pH7.5, 150 mM NaCl, 1 mM EDTA, 0.25% Triton X-100, 1 mM DTT, 1X Halt Protease inhibitor cocktail). The mixture was rotated for 5 hours at 4°C. 20 μl of glutathione beads was added to the mixture, and the binding mixture was rotated for another 2 hours at 4°C. After washing the glutathione beads 4 times with wash buffer (50 mM Tris-HCl, pH7.5, 300 mM NaCl, 1 mM EDTA, 0.5% Triton X-100, 1 mM DTT, 1X Halt Protease inhibitor cocktail), the GST-NEMO bound IKKβ was analyzed by 10% SDS-PAGE and immunoblotted with Flag antibody. For co- immunoprecipitation analysis, 1 μg of GST or GST-cleaved NEMO protein was incubated with 1/5 of the translation reaction containing mock or IKKβ in 800 μl of the IP buffer containing 2 μg of anti-NEMO antibody. The mixture was then rotated for 5 hours at 4°C and then 20 μl of protein G-sepharose was added to the mixture. The binding mixture was rotated for another 2 hours at 4°C. After 4 times of washing the beads with the wash buffer, the immunoprecipitated proteins were analyzed by immunoblotting as above. ## Cell culture 1.3E2 murine pre-B cells (obtained from Israel’s group) stably expressing 6x Myc-tagged human NEMO (C5) and 70Z/3 cells (TIB-158, ATCC) were cultured in RPMI1640 medium (GE healthcare HyClone) supplemented with 10% (v/v) fetal bovine serum (GE healthcare HyClone), 50 μM β-mercaptoethanol, and antibiotics (100 IU/ml penicillin and 100 μg/mlstreptomycin) as described previously. Human Jurkat T cells (TIB-152, ATCC) were cultured in RPMI1640 medium supplemented with 10% fetal bovine serum and antibiotics as above. Human retinal pigment epithelial (RPE) cells (obtained from Burkard’s group) were cultured in DMEM/F12 medium (GE healthcare HyClone) supplemented with 10% fetal bovine serum and antibiotics as above. Human embryonic kidney (HEK293) cells (CRL-1573, ATCC) and HeLa cells (CCL-2, ATCC) were cultured in DMEM medium (Corning Cellgro) supplemented with 10% fetal bovine serum and antibiotics as described above. All cells were grown at 37°C in 5% CO₂ humidified incubator (Thermo Scientific) except HEK293 cells, which were grown at 37°C in a 10% CO₂ humidified incubator on 0.1% (w/v) gelatin-coated dishes. All cells were passaged at least twice weekly before reaching a cell density of 1 x 10⁶ per ml for suspension cells or reaching 90% confluence for adherent cells. Immunofluorescence analysis was done as in except for anti-NEMO antibody (FL-419) used for detection of NEMO. ## Generation of NEMO reconstituted 1.3E2 cell clones 1.3E2 cells were reconstituted with wild-type human NEMO as described previously. Briefly, 5 x 10⁶ 1.3E2 cells were electroporated with 40 μg of human NEMO/pcDNA3.1(+), immediately split into three 10 cm dishes to ensure isolation of independent clones, and incubated for 24 hours in growth media. G418 (Mediatech) at 1 mg/ml concentration was then added to select for resistant pools over 4–5 days. Dead cells were removed by using lymphocyte separation medium (Mediatech). The G418-selected clones were isolated (∼40 clones from each of the three pools) and individually placed in each well of a 96-well plate with 0.2 mg/ml G418 for ∼2 weeks before expanding each clone (∼12–20 clones for each pool) for the analysis of NEMO expression by Western blotting with NEMO antibody, using the C5 clone as the standard in each gel. Generally, 2–4 clones from each pool had equivalent NEMO expression as C5. Protein extraction was done by 2x SDS Laemmli sample buffer or using Totex buffer (20 mM HEPES, pH7.9, 350 mM NaCl, 1 mM MgCl₂, 0.5 mM EDTA, 0.1 mM EGTA, 20% glycerol, 1% NP-40, 0.5 mM DTT, 1X Halt Protease inhibitor cocktail). ## Quantification of cellular NEMO numbers 10⁷ C5 cells were pelleted at 2000 rpm for 2 minutes and resuspended with 10 ml of 1X PBS resulting in a concentration of 10⁶ C5 cells/ml. The resuspended cells were aliquoted to make a series of diluted cells: 1 x 10⁴, 2.5 x 10⁴, 5 x 10⁴, 1 x 10⁵, 2.5 x 10⁵, 5 x 10⁵ and 1 x 10⁶ C5 cells. Purified recombinant NEMO protein (GST cleaved) was also diluted with 1X PBS to make 0.1, 0.5, 1, 5, 10, 50, 100 ng diluted protein series. 2x SDS Laemmli sample buffer was added to the diluted cells or recombinant NEMO protein. The samples were boiled for 5 minutes and separated on 10% SDS-PAGE gels. The separated proteins were transferred onto polyvinylidene difluoride (PVDF) membranes and visualized by Western blotting with anti-NEMO antibody. To calculate the number of NEMO molecules per cell on average, the intensity of visualized NEMO bands were measured by densitometry analysis. The intensity of bands at the linear range, corresponding for 1, 5, and 10 ng of the purified recombinant NEMO protein and 5 x 10⁴, 1 x 10⁵, 2.5 x 10⁵ C5 cells, were normalized to the intensity of band corresponding to 1 ng of the purified recombinant NEMO protein. A graph and equation were achieved from the normalized intensity to estimate the number of NEMO molecules per cell on average. Based on this equation, the intensity for 1 x 10⁵ C5 cells was equivalent to the intensity of band corresponding to 4.55 ng of the purified recombinant NEMO protein. Using molecular weight of 6x myc tag-human NEMO (62.6 kDa), 4.55ng equals to 72.7 fmoles or approximately 4 X10¹⁰ NEMO molecules, indicating approximately 4 x 10⁵ molecules of human NEMO protein is expressed in a single C5 cell on average. Three independent analyses were performed by three investigators on different days with the same outcomes obtained (with R² value ranging from 0.92 to 0.99). To quantify the number of NEMO molecules in different human cell lines, 3 x 10⁶ cells each of Jurkat, HEK293, HeLa, or RPE, and C5 were pelleted at 13,000 x g for 10 s in a table-top microcentrifuge at room temperature, rinsed once with phosphate-buffered saline (PBS), resuspended in 50 μL PBS, and lysed by the addition of 250 μL 2x SDS Laemmli sample buffer. Samples were then immediately boiled for 10 minutes, loaded by cell number in a titration from 10⁶ to 10⁴ cells, and separated on 10% SDS-PAGE gels. The separated proteins were visualized by Western blotting with anti-NEMO antibody as described above. Densitometry analysis using ImageJ software was used to measure the intensity of the visualized NEMO bands for each blot, and the signals were then normalized to either 2.5 x 10⁵ or 5 x 10⁵ of corresponding C5 cells for each blot. Using the normalized intensities, a linear graph and equation were generated for each cell type with R² value \> 0.94. The average number of NEMO molecules per cell was then determined by calculating the ratio of the slopes between each cell type Jurkat, HEK293, HeLa, or RPE, and C5 and multiplying this ratio by the estimated average number of NEMO molecules per C5 cell. ## Statistical analysis Values are presented as means ± S.E.M. with the indicated number of independent experiments. The statistical significance of differences between groups was determined by the Student's t test. P values less than 0.05 were considered statistically significant. # Results ## Purification of full-length NEMO To purify sufficient amounts of full-length recombinant human NEMO protein, a human NEMO construct was transformed into Rosetta2 BL21 cells and the NEMO protein was induced. Regardless of different induction temperatures (19–30°C), at least 60% of the induced NEMO protein was found in the insoluble fraction (lane 6). To reduce the aggregation of the NEMO protein and maximize the yield of NEMO protein in the soluble fraction, the bacterial lysate was diluted with a large volume of lysis buffer (150 ml per 1 L of the bacterial culture). The diluted lysate was slowly loaded onto a glutathione column with three repeated applications to further reduce the potential of NEMO aggregation. In addition, β-mercaptoethanol freshly added to all buffers increased the yield of the soluble NEMO protein, possibly by further preventing protein aggregation. Despite extensive washing with lysis buffer, the GST-NEMO fraction, which was eluted with elution buffer containing 10 mM reduced glutathione, contained nonspecifically bound bacterial proteins (lane 4). However, direct cleavage of GST-NEMO from GSH-beads with GST-tagged human rhinovirus 3C protease resulted in the recovery of cleaved NEMO protein fractions that contained significantly reduced protein contaminants compared to the GST tagged NEMO fraction (lane 5 and 6). Even so, the NEMO protein was extremely susceptible to aggregation after the GST tag was removed. Thus, it was important to maintain at least two bead volumes of cleavage buffer to prevent NEMO protein aggregation. Approximately 0.3 mg/ml of GST-cleaved full length NEMO and 1 mg/ml of GST-NEMO in a soluble state could be obtained. The nature of the purified NEMO protein was confirmed by excising ∼48 kD protein band from Coomassie blue-stained SDS-PAGE gel followed by MS/MS analysis. Sequence coverage with high confident peptides was 55.8%. Moreover, both purified GST-NEMO and GST-cleaved NEMO bound to *in vitro* translated flag-IKKβ by GST-pull down assay and co-immunoprecipitation assay, respectively. These results confirmed the purity and identity of the purified recombinant NEMO protein, which was used for subsequent cellular NEMO quantification studies. ## Determination of the average number of NEMO molecules in the C5 clone To estimate the average number of NEMO molecules present within a cell, we employed a 1.3E2 clone (C5) which stably expresses 6x Myc-tagged NEMO at approximately the same levels as endogenous NEMO levels detected in 70Z3 cells, the parental cell line of 1.3E2, using FL419 NEMO antibody. C5 and 70Z/3 cells display similar levels of NF-κB activation in response to etoposide (VP16) and lipopolysaccharides (LPS; membrane signaling control). Different amounts of purified recombinant NEMO protein along with cell extracts representing different C5 cell numbers were analyzed by Western blotting using anti-NEMO antibody. Based on quantification of the intensity of the NEMO bands through densitometric analysis, approximately 4 x 10⁵ molecules of human NEMO protein is expressed in a single C5 cell on average. ## Generation of 1.3E2 clones expressing different amounts of NEMO protein To determine the impact of different NEMO expression levels on NF-κB activation by genotoxic agents, we stably reconstituted NEMO in 1.3E2 cells with tag-less human NEMO as depicted in. An example of the Western blot analysis of 1.3E2 cell clones stably expressing different amounts of NEMO is shown in. Based on the expression of NEMO relative to the C5 clone, multiple independent stable clones were further expanded in order to determine the average number of NEMO expression per cell and their respective NF-κB activation potentials in response to etoposide treatment. When different clones were analyzed by immunofluorescence analysis using anti-NEMO antibody, the expression of NEMO detected in individual cells was relatively uniform. We next treated multiple 1.3E2 clones, stably expressing different levels of NEMO, with etoposide for 1 hour and analyzed NF-κB activation by EMSA. Since NEMO is prone to precipitate *in vitro*, the amount of NEMO in the clones was measured using two different extraction methods: 2x SDS Laemmli sample buffer and Totex buffer, the latter of which was used to prepare cell extracts for EMSA assay as NF-κB proteins are efficiently extracted in the native state. Amounts of extracted NEMO were similar in both methods for tag-less NEMO. To estimate the number of NEMO molecules in the NEMO reconstituted clones shown in, the equation derived in comparison of the NEMO bands between the clones and C5 were used. Based on our calculations, we estimated that clones 6E, 6B, 5D, 5E and 4C express 0.6, 2.7, 2.7, 6 and 7 x 10⁵ molecules of NEMO per cell. All of these NEMO clones showed almost no detectable basal NF-κB activity. In response to etoposide treatment, a trend of increasing NF-κB activation was observed as the amount of NEMO increased (peaking with clone 6B), which became modestly reduced in clones which expressed even higher amounts of NEMO. In contrast, the NF-κB activation in response to LPS was not affected by the amount of NEMO in this range. More careful analysis using both a time course and dose response showed similar trends with clone 6B showing the highest NF-κB activation and lower or higher amounts of NEMO than clone 6B showing reduced NF- κB activation. However, quantification of EMSA data from multiple biological replicates did not show statistically significant impacts (lower graphs). Thus, NF-κB activation by etoposide is surprisingly invariant across a 10-fold range of NEMO expression in the 1.3E2 cell system. ## Determination of the average number of NEMO molecules per cell in multiple human cell lines Various human cell lines are often employed to study NF-κB signaling by different stimuli, including genotoxic agents; however, the number of NEMO molecules expressed per cell in any human cell line remains undefined. Using the C5 clone as a standard, we sought to estimate the average number of NEMO molecules in the following human cell lines: Jurkat T lymphocytic, HEK293 embryonic kidney, HeLa cervical, and RPE retinal pigment epithelial cells. Cell extracts of each cell line, as well as those of the C5 clone, were prepared using 2x SDS Laemmli buffer and analyzed by Western blotting using anti-NEMO antibody. Quantification of the NEMO band intensities yielded the following estimates for the average number of NEMO molecules per cell for each cell line: Jurkat—1.3 x 10⁶; HEK293–1.0 x 10⁶; HeLa—1.0 x 10⁶; RPE—2.0 x 10⁶. All of these human cell lines are considerably larger than 1.3E2 pre-B cells, which may account for the relatively larger average numbers of NEMO molecules per cell. # Discussion In previous studies, investigation of the functional significance of NEMO and its mutants in multiple NF-κB signaling pathways utilized reconstitution of NEMO mutants in NEMO-deficient 1.3E2 murine pre-B cells, R5 rat fibroblast, and Jurkat human T cells, or NEMO-null mouse embryo fibroblasts. The interpretation of the functional significance of mutant NEMO proteins is confounded by the potential artifacts associated with their overexpression as overexpression of even the wild-type NEMO protein can significantly inhibit NF-κB signaling. Indeed, while NEMO was originally cloned through functional restoration of NF-κB signaling by its stable reconstitution in NEMO-deficient R5 and 1.3E2 cells, it was also independently cloned by another group as IKK/NF-κB inhibitor protein upon its overexpression. One approach to control the expression of NEMO mutants is to generate inducible cell systems (e.g., doxycycline-inducible system) or use of knock-in technologies (e.g., CRISPR). Alternatively, co-expression of a marker protein (e.g., CD2 or GFP) as a NEMO fusion or expression from an internal ribosome entry sequence within a single expression plasmid may be used as a surrogate marker of expression quantified and isolated by FACS sorting. Even with these approaches, it may be necessary to ensure the expression level of NEMO to be "physiological" by comparing the expression of mutant NEMO proteins to the endogenous levels of the parental cell systems as comparison to wild-type NEMO protein may be insufficient to draw accurate conclusions if the wild-type NEMO expression levels are highly overexpressed compared to the endogenous levels. We have previously employed stable 1.3E2 cell clones to study NEMO mutant phenotypes in NF-κB signaling induced by genotoxic agents. The expression of NEMO in stable 1.3E2 clones remains generally stable over many passages and cryopreservation steps, and individual cells within a given clonal population generally express similar levels as assessed by immunofluorescence analysis. In contrast, expression of NEMO in individual cells of a stable pool population is typically highly variable, making it a generally unsuitable system to study NEMO mutant phenotypes as the NF-kB activation in those individual cells with very low or high NEMO expression will likely show different phenotypes as those with more physiological NEMO expression levels within a stable pool. However, it is technically challenging to match the expression levels of NEMO in stable clones across different mutant NEMO versions. Our current study demonstrates that the levels of NEMO do not have to be precisely matched to draw functionally relevant conclusions in the context of 1.3E2 reconstitution system. Based on our quantification analysis, using the C5 stable clone that expresses approximately 4 x 10⁵ molecules of human NEMO protein per cell as a standard, we found that a range of 0.6–6 x 10⁵ NEMO molecules per cell (a 10-fold range) is functionally equivalent with statistically insignificant impact on etoposide-induced NF-κB activation as measured by EMSA analysis. Since our previous studies suggested that only “IKK-free” NEMO, which is not bound to IKKα/β, mediates nuclear signaling in response to genotoxic agents, we originally considered the possibility that the increased amounts of NEMO beyond some threshold level (perhaps set by the levels of IKKα/β subunits) might increase the level of NF-κB activation in response to genotoxic agents by increasing free NEMO. Indeed, there was a trend of modest increase in NF-κB activation by etoposide as the level of NEMO increased from clone 6E to 6B/5D. As the level of NEMO further increased, etoposide-induced NF-κB activation was modestly reduced and a greater decrease in NF-κB activation was found in some clones that expressed an amount of NEMO beyond this range (data not shown). Despite the varied NEMO levels from undetectable to that which is much higher than the C5 clone, the amount of IKKα/β expression was invariant, suggesting that the levels of IKK catalytic subunits are not regulated by NEMO protein in this cell type. Thus, estimation of the number of IKKα/β molecules per 1.3E2 cell and the definition of the stoichiometry of NEMO and IKKα/β molecules would be informative for further elucidation of the mechanisms involved in NF-κB activation by genotoxic agents. It is also possible that "IKK-free” NEMO is liberated from NEMO-IKK complexes upon stimulation with genotoxic agents and therefore the basal stoichiometry of NEMO and IKKα/β molecules may be insufficiently informative. As such, additional studies aiming at defining the dynamics of NEMO-IKKα/β complexes during cell stimulation may also be critical to improve our understanding of NF-κB signaling by DNA damaging agents. The level of NEMO expression in the C5 clone was stable over multiple cryopreservation cycles and over multiple independent analyses; therefore, it may be useful as a standard cell line to estimate the number of NEMO molecules in other human cell types as we have done so for several commonly used human cell lines. However, we cannot rule out the possibility of some posttranslational modifications of NEMO altering the degree of epitope recognition by anti-NEMO antibody. The cell lines derived from non-human species may also need to be recalibrated by the use of recombinant full-length NEMO protein standards generated from the species of interest as different anti-NEMO antibodies likely recognize NEMO proteins from different species with varying efficiencies depending on the conservation of the epitopes being detected. As part of the NEMO quantification studies, we generated recombinant full-length human NEMO protein to be used as a molecular standard. Even though multiple groups have previously purified recombinant NEMO protein for biochemical study, the recombinant NEMO proteins were generally purified as fusion proteins and we were unable to find a study that reports a purification protocol for untagged full-length recombinant wild-type NEMO protein. Therefore, we sought to optimize a NEMO purification protocol to obtain relatively pure and concentrated amounts of the soluble, full-length, and untagged NEMO protein. A well-appreciated obstacle in the purification of recombinant full-length NEMO protein is protein aggregation and precipitation. While the NEMO protein was more soluble as GST- fusion protein, it became more susceptible to aggregation when the GST was removed from the protein, suggesting GST-induced dimerization of NEMO potentially reduces NEMO aggregation at high concentrations. Agou et al. suggested the use of neutral detergents such as dodecyl maltoside to prevent this aggregation, however this increased the co-purification of bacterial DnaK protein with NEMO protein and thus decreased overall purity. Whitty and colleagues have recently generated an N-terminal seven cysteines-to-alanines substitution mutant of NEMO to increase the purification of soluble NEMO protein by preventing intermolecular disulfide bond formation. Another possible way to increase the solubility of NEMO protein is to co-purify NEMO with other proteins that interact with NEMO to prevent NEMO aggregation, such as the IKK catalytic subunits. We found that keeping the bacterial lysate dilute in large volumes and adding a fresh reducing regent to all the buffers increased yield of soluble NEMO protein. Moreover, direct cleavage of GST-NEMO proteins from the GSH-beads by GST-tagged human rhinovirus 3C protease in large reaction volumes also resulted in isolation of soluble NEMO with high purity and at a maximal concentration of 0.3 mg/ml. Further concentration of NEMO resulted in aggregation and precipitation under our current buffer conditions. The identity of purified NEMO protein was validated by MS/MS analysis as well as by IKKβ interaction *in vitro* and association with appropriate ubiquitin chains (data not shown). The NEMO purification protocol described here thus may be useful for purification of non-human NEMO proteins as well as future *in vitro* studies of NEMO in NF-κB signaling. # Supporting Information We thank Miyamoto lab members and Dr. Richard R. Burgess for helpful discussions. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: BH KM AJ AA SM. Performed the experiments: BH FPP EYC JY SM. Analyzed the data: BH FPP JY SM. Contributed reagents/materials/analysis tools: BH FPP KM EYC JY AJ AA SM. Wrote the paper: BH SM.

# Introduction Laminitis is one of the most painful and debilitating diseases in equine medicine, with significant implications in terms of horse welfare. It is characterized by deterioration of the lamellar tissue that connects the hoof wall to the underlying third phalanx in the equine hoof, with associated inflammation, heat, digital pulses, pain and lameness, and, in some cases, dorsopalmar rotation of the third phalanx. It can be classified as either acute or chronic, and it is not usually readily reversible. Laminitis can lead to permanent disability or even death. The most common etiology of laminitis cases involves gastrointestinal or metabolic disease. For a number of reasons, laminitis is considered a metabolism-related pathology, resembling the human diabetic foot, since hormonal influences, weight and pressure are involved. Evidence suggests that diabetic microvascular complications, including neuropathy, are caused by the activation of mechanisms related to the polyol pathway: advanced glycation end-product (AGE) formation with subsequent activation of the receptor for AGEs (RAGE), and lead to the activation of other factors such as nuclear factor kB, protein kinase C (PKC) isoforms, and the hexosamine pathway as a result of the overproduction of superoxide by mitochondria. Gastrointestinal disturbances resulting from ingestion of excess carbohydrates play a pivotal role in sepsis-related laminitis. Feeds rich in starch are used as highly digestible energy sources in horses. During digestion in the small bowel, starch is primarily broken down by amylase enzymes, leading to glucose liberation. Extremely large quantities of starch in the diet can result in starch overload; which means that undigested starch passes from the small to the large intestine, which can lead to a decrease in colon and caecal pH, often followed by colic laminitis. In this scenario, release of toxins from the hindgut is suspected to occur, which in theory induces degradation of the lamellar basement membrane and a loss in epithelial cell adhesion, with the subsequent activation of matrix metalloproteinases (MMPs) and leukocyte infiltration into the lamellae. According to some authors, AGEs accumulate in significant amounts in the hoof lamellar tissue in the acute phase of experimentally induced laminitis using the hyperinsulinemic model. Moreover, Valle et al. revealed increased plasma levels of pentosidine, a glycoxidative marker, in ponies with clinical equine metabolic syndrome. Methylglyoxal (MG) causes the formations of AGEs, leading to carbonyl stress. AGES derived from glucose and intermediates like MG maybe the major source of intracellular and plasma AGEs. Since MG is a highly reactive intermediate, it is converted into D-lactate, to prevent the formation of AGEs. Specifically, starch overload in the gut can lead to rapid, devastating changes that result in the elaboration of toxins and other substances: Gram-negative and -positive bacteria undergo carbonyl stress, leading to methylglyoxal biosynthesis—a highly reactive α-oxoaldehyde, mainly derived from the triose phosphates D-glyceraldehyde-3-phosphate and dihydroxyacetone phosphate, which would be detoxified into D-lactate under healthy physiological conditions. In equines, feeding large amounts of fermentable carbohydrates with subsequent acidosis increases the plasma levels of D- lactate up to 25 mMol/L. According to this mechanism, D-lactate can be considered a clinical marker of the development of laminitis, because it is more specific than its L-isoform and more stable than methylglyoxal. Matrix metalloproteinases (MMPs) have been long investigated for their role in the cleavage of the extracellular matrix, which precedes and enables remodeling of tissues and the selective degradation of membrane proteins and collagen IV and VII. Previous studies using *ex vivo* hoof explant models have shown that under specific conditions, basal epidermal cells can separate from their basement membrane as occurs *in vivo*, making them an attractive model for the study of laminitis. In these investigations, silymarin or LPS were used to stimulate a tissue reaction analogous to laminitis in hoof explants. The aim of the present study was to investigate the effects of MG on equine lamellar explants in an *ex vivo* model by evaluating macroscopic and histological structures, tissue integrity by means of the separation force test, and by quantifying MMP expression at different exposure time points. # Materials and methods ## Animals Nine male draft horses (Breton horses), reared for meat production, with an age ranging between 18 and 24 months, were enrolled in the study. No animals were specifically killed for the purposes of this study. Horses were slaughtered in a commercial abattoir in Volpiano (Turin, Italy), and front limbs were collected with the slaughterhouse owner’s consent under the control of a supervisor from the Italian National Health Service. ## Sample collection A complete physical examination of all animals was performed, with particular attention to symptoms and signs of laminitis (i.e.: lameness, typical stance, increased digital pulse, and increased hoof capsular warmth). Following slaughter, the right fore distal limb was disarticulated at the carpal joint, and the limbs were transferred onto ice within 45 minutes and brought to the dissection room of the Department of Veterinary Sciences of the University of Turin (Italy). ## Hoof explant preparation Hoof explant (HE) collection and treatment procedures are summarized in. Hooves were carefully cleaned using a hoof-knife and scrubbed with brushes and chlorhexidine solution (4%). All instruments were cleaned with chlorhexidine and sterile saline. Ice was used to keep all instruments and samples cool. Explants were collected according to the method reported in Mungall and Pollitt with slight modifications. Hooves were trimmed, removing the lateral and medial walls and cutting the digits into 4–5 sagittal slices. In this way, it was possible to obtain hoof wall strips measuring 0.8x1.5 cm in thickness containing 10–12 lamellae, consisting of the inner part of the hoof wall epidermal lamellae, dermal lamellae and the bone. Prior to incubation, HEs were washed three times with Phosphate Buffered Saline (PBS) under sterile conditions. ## *Ex vivo* tissue survival conditions–preliminary assay Before the execution of the present project, a preliminary assay was performed to identify the best culture medium for preserving HEs: the tests were run to evaluate the use of Dulbecco’s Modified Eagle Medium (DMEM) alone (with 4.5mg/L glucose, but without L-glutamine) *vs* DMEM plus 2% of antibiotic/antimycotic solution, 2% of L-glutamine, and 10% Fetal Bovine Serum (FBS) (DMEM+). All reagents were purchased by Sigma Aldrich (Milan, Italy). The HEs were incubated in 5 mL of their assigned medium in 6 well-plates. Plates were incubated at 37°C (5% CO₂ and 95% O₂ atmosphere) for 24 or 48 hours. The medium in the plates assigned to the 48 hour incubation condition was changed after 24 hours. ## *Ex vivo* tissue survival conditions Sample survival was evaluated considering macroscopic features (changing of colour, smell, and disruption) with histology as confirmation. In accordance with the results obtained from the preliminary assay (see details in “Results” section), the DMEM+ medium was used for the rest of the study. ## Experimental design HEs collected from nine different horses were incubated in triplicate with DMEM+ medium in six-well plates. MG was added to the HEs at the following concentrations: 50, 100, 150 and 200 mM, and incubated at 37°C (5% CO₂ and 95% O₂ atmosphere) for 24 and 48 hours. HEs cultured in the absence of MG were used as controls (K). In order to allow the execution of the entire experimental design, 45 samples were incubated. ## Separation force test The separation force test was used to evaluate structural integrity of the samples. All tests were performed in a blinded manner. All HEs were evaluated using a strain gauge as follows: one end of the HE was fixed to a tong and the other was attached to a force transducer and exposed to a maximum force of 4000 x *g*. Separation force was averaged over nine experiments for each time point (24 and 48 h) by the same operator. Only HEs that separated at the level of epidermal lamellae or dermal lamellae were considered as valid measurements. The HEs that were not broken at these points were considered virtually impossible to separate. The procedure was coordinated in order to immediately proceed with RNA extraction (details in section “RNA extraction and cDNA synthesis”). ## Histology One sample from each time point and from each MG concentration was fixed in 4% buffered formalin. These samples were then embedded in paraffin, and 3-μm slices were mounted on silanized glass slides. Samples were air-dried and stained histochemically with haematoxylin and eosin (H&E) and periodic acid-Schiff (PAS) (Dako, Milan, Italy). Slides were examined using a Nikon microscope, and images captured and stored with Image pro-plus software (Media Cybernetics, MD, USA). The same expert operator performed all evaluations, blinded to treatment group. The laminitic lesions were scored using the histopathological grading system proposed by Pollitt. Briefly, lesions of epidermal lamellae attributable to laminitis induced by the experimental design were graded using the following evaluation method: normal (grade N), mild (Grade 1), moderate (Grade 2), severe or extensive (Grade 3). The grading system was based on the degree of change observed in the lamellar basement membrane. Samples were then evaluated for lesions of the secondary epidermal lamellae. ## RNA extraction and cDNA synthesis Immediately after the separation force test, hoof explants (HEs) were placed in tubes containing RNAlater solution (Ambion), and the *stratum lamellatum* was disrupted using a TissueLyser II (Qiagen) with stainless steel beads in 1 mL TRI Reagent (Sigma Aldrich). Total RNA from tissue samples was extracted and purified from any residual genomic DNA using a DNAse I Recombinant RNAse free kit (Roche, Mannheim, Germany). RNA concentration was measured using spectrophotometry (BioPhotometer, Eppendorf, Germany). The ratio of the optical densities measured at 260 and 280 nm was greater than 1.9 in all RNA samples. cDNA was synthesized from 1 μg of total RNA using a RT high Capacity cDNA Reverse Transcription kit (Applied Biosystems, Foster City, CA, USA) according to the manufacturer’s protocol. The cDNA was subsequently diluted in nuclease- free water and stored at -20°C. ## Quantitative assessment of MMP gene expression Sufficient cDNA was prepared in a single run to perform the real-time quantitative PCR (q-PCR) experiments for all the selected genes. To determine the relative amount of specific MMP-2, MMP-14, and TIMP-2 transcripts, qPCR was performed using the CFX Connect Real-time System (Bio-Rad, Hercules, CA, USA). Primers for target and reference genes were designed for *Equus caballus* GenBank mRNA sequences using Primer 3 Software (version 4.0). In order to minimize the amplification of contaminant genomic DNA, oligonucleotides were designed using exon/exon boundaries analyzed using the IDT tool (available at <http://www.idtdna.com/scitools/scitools.aspx>) for hairpin structure and dimer formation. Primer specificity was verified using BLAST analysis, comparing results against the genomic NCBI database. reports primer sequences, gene accession numbers, and amplicon sizes. To establish primer efficiency, we used the dilution method: CFX Manager software (version 3.0, Bio-Rad, Milan, Italy) was used to calculate primer efficiency using the linear regression slope of the dilution series. Each primer set efficiency was between 95% and 100%. Multiple housekeeping genes were selected as potential internal control mRNAs: ß-2 microglobulin (B2M), ß-actin (ACTB), and glyceraldehyde 3-phosphate dehydrogenase (GAPDH). Based on the stable cycles quantification (Cq) values among the different experimental conditions, GAPDH was selected as the most suitable internal control. The q-PCR parameters were as follows: 95°C for 3 min, 95°C for 15 sec, and 60°C for 30 sec, for a total of 44 cycles. To evaluate mRNA expression, data obtained as Cq values of technical replicates were averaged and used to determine ΔCq values (ΔCq = Cq of the target gene–Cq of the reference gene). The ΔΔCq method was used to analyze the data, and results were expressed as fold changes compared with control samples. Assays were run in triplicate, and template control was included using water instead of cDNA. ## Statistical analysis Results were organized using Excel software (Microsoft Corporation, CA, USA) and data were analyzed with Prism 9.0 software (Graph Pad, CA, USA). The results of the separation force test and qPCR were analyzed using one-way Anova and Tukey’s multiple comparison test. Histology score data were recorded, and their distributions were analyzed using two-way Anova, setting time and concentration as variables. The normality of q-PCR data was checked using D’Agostino and Pearson’s omnibus normality test. Results are presented as means ± standard deviation (SD). Statistical significance was accepted at *p* values ≤ 0.05. # Results ## Preliminary assay Samples maintained in DMEM underwent a change of colour (from a white-light pink colour to green) and smell, indicating that autolytic processes were underway. Histological examination was performed for all samples in order to confirm the macroscopic evaluation. The separation force test was only performed in a few samples because the majority of samples already showed clear visual evidence of disrupted lamellae. On the contrary, samples maintained in DMEM+ medium showed no macroscopical changes, and histological evaluation confirmed the integrity of all structures; all samples were considered eligible for the separation force test. These details are shown in. ## Macroscopic evaluation Considering the results obtained in the preliminary assay, samples were kept in DMEM+ for all subsequent experiments. During the incubation period, all the samples were checked daily. No alterations in colour or other macroscopic changes were detected. ## Separation force test The separation force test was performed for all samples incubated with the different concentrations of MG at predetermined time points. The data reported in show that after 24 h of incubation in the presence of varying concentrations of MG, separation of the HE lamellar structures could be provoked in 55% of samples (25/45). After 48 h, the percentage increased to 77% (35/45). The representation in shows the averaged weights (kg ± SD) required to provoke separation. No statistically significant effect was observed according to the different MG concentrations, probably due to the high variability of samples. Statistically significant difference can be observed comparing the two time points (24h and 48h). Moreover, a trend for reduced tissue integrity in samples treated with 50, 100 and 150 mM MG for 48 h can be observed. ## Histology Histological analysis permitted the ranking of samples according to a scoring system, as described above in the Materials and Methods section; the distributions of scores per experimental condition are shown in. Hoof sections incubated for 48 h in the presence of high concentrations of MG scored the highest due to the alteration of all layers and were statistically different from samples incubated for 24h. In these samples (classified as laminitic stage 2), the following histological changes were observed under low magnification: wavy primary epidermal lamellae; the absence of connective tissue between secondary epidermal lamellae; and rounded basal nuclei. Under high magnification, the round shape of the nuclei was confirmed; basal cells showed a pale, cloudy and vacuolated cytoplasm; and the normal organization of palmar anatomy was no longer recognizable: the definition near the tip of the secondary epidermal lamellae was lost. ## Expression of MMP-2, MMP-14 and TIMP-2 mRNA in digital lamellar specimens The expression of genes encoding MMPs associated with inflammation, such as MMP-2 and MMP-14, was not substantially increased in digital lamellar samples following 24 h incubation with MG compared with control samples. Concerning MMP-2, higher levels of expression were observed in samples incubated for 48 h compared with controls, with differences also evident for the different MG concentrations. Nevertheless, these differences did not reach the level of statistical significance. This suggests that MG may have a concentration- dependent effect following 48 h incubation. That said, no significant differences between samples, or between the two experimental time points, were detected. Significant differences in TIMP2 expression levels were detected between the MG-treated samples and controls at 24 h at any concentration. At the 48 h time point, the expression levels of TIMP2 were inversely related to MMP-2 and MMP-14 levels. Results are presented in Figs. # Discussion Regarding the primary aim of this study, our results support the use of *ex vivo* equine lamellar hoof wall and dermis explants cultured in DMEM with the appropriate addition of FBS, antibiotic and antimycotic solution, and L-glutamine as a suitable methodology for obtaining a suitable viable model system for investigating the pathological mechanisms of laminitis. Here, we focused on the possible role of MG in laminitis. To this end, macroscopic and histological evaluations, the separation force test, and MMP-2, MMP-14 and TIMP-2 gene expression level assessments were performed. The preliminary assay was a milestone for the entire experiment: glucose supply was shown to be essential for the maintenance of an integral hoof structure in the explant model, corroborating the results of previous studies that evaluated the integrity of explants in the presence or absence of glucose. In accordance with the method published by Reisinger et al., all explants were cultivated in medium containing high glucose concentrations (4.5 g/L) and antibiotics, but in accordance with our results, FBS and L-glutamine are necessary for tissue preservation. The macroscopic evaluation of explants cultivated in DMEM+ showed the absence of any colour or smell changes throughout the entire experimental period, supporting the viability of the samples for further investigations into the pathogenesis of laminitis. Other papers have evaluated the possibility of using hoof explant samples but focus their attention on different metabolic goals (eg.: the role of insulin) or with the final aim to obtain cell culture. To the authors knowledge this is the first paper focusing specifically on the role of MG. The results obtained in the separation force test support the hypothesis that MG can lead to modification of the hoof lamellae, rendering them less robust and unable to resist separation following the application of a load that untreated samples are able to bear and remain intact. This evaluation confirmed that higher concentrations of MG and a longer incubation period (48 h) can cause irreversible damage that weakens hoof tissues. These results were confirmed by histological examination of HEs, which revealed major alterations in the basal lamella, the severity of which correlated with increasing concentrations of MG. The results of gene expression of MMPs and TIMP-2 in HEs are in accordance with those obtained by other groups. Increased levels of MMPs seem to be correlated to increasing degrees of extracellular matrix remodeling, more severe histological features, and resulting in structural weakness as assessed by the separation force test. On the contrary, it was expected that TIMP2 levels would decrease in a proportional way as previously shown. MMPs are important for maintaining the integrity of healthy hoof tissue. Virtually the entire lamellar region is non proliferative, and remodeling of the various cells of the secondary and primary lamellae occurs via controlled MMP activity. The results of the present work seem to be aligned with those obtained by de Laat et al., who showed MMP-2 to be secreted by basal cells and to be present in the lamellar tissue of healthy horses; however, the study by de Laat and colleagues did not show any increase in MMP gene expression, protein concentration, or activation in lamellar tissue taken from horses with insulin-induced laminitis, suggesting that it does not play a significant role in the development of this form of the disease. However, in the case of laminitis induced by starch overload or other causes, increases in MMP-2 have been detected and may be accompanied by the activation of other MMPS, such as MMP-9. MMP-14 may also play a key role in the maintenance of lamellar homeostasis under physiological conditions. One possibility is that MMP-14 and MMP-2 act simultaneously to remodel the non-proliferative basal cells of the inner hoof wall. MMP-14 has been shown to be regulated in breast cancer, where it is reported to activate MMP2 that is, in turn, associated with increased risk of malignancy and metastasis. We cannot exclude the possibility that a similar mechanism occurs during laminitis, where MMP-14 activation may cause the subsequent activation of MMP-2 and thereby trigger the entire laminitic process. The triggering factors able to induce laminitis have yet to be clearly identified. Although cellular and molecular events have been studied in the past, yielding important contributions to our current knowledge, much remains unknown about the early events of the disease process. One of the main known risk factors entails a metabolism-related pathology. The accumulation of triosephosphate intermediates induced by an increase in reactive carbonyl species is caused by an enhanced metabolic flux. Methylglyoxal belongs to a class of reactive carbonyl species known as the alpha-oxoaldehydes or dicarbonyls. It is derived from the spontaneous degradation of triosephosphate intermediates and acts as a highly potent glycating agent. Dicarbonyls, such as methylglyoxal, can modify DNA and react with the amino groups of intracellular and extracellular proteins to form AGEs. Increased levels of AGEs induce alterations in physiological cell cycles, leading to extreme stress conditions and even cell death. The interaction of AGE-modified proteins through cell surface receptors, such as RAGE, can lead to increased cellular activation and sustained inflammatory responses. Significantly higher levels of the aforementioned factors have been clinically demonstrated in various pathologies; for example, erythrocytes from diabetic subjects show higher levels of triosephosphate intermediates and methylglyoxal compared with those of healthy subjects. In addition to the metabolic causes of laminitis, many other theories have been proposed, of which the production of bacterial exotoxins seems to be one of the most important. Indeed, a great deal of research has been conducted in the attempt to elucidate a possible bacterial role. However, a study by Reisinger et al. demonstrated that lipopolysaccharide (LPS) extracted from bacteria *Escherichia coli* alone is not sufficient to induce laminitis *in vitro*. The majority of horses in the study by Tóth et al. developed laminitis after the administration of a low dose of oligofructose in the presence of endotoxins: this strongly suggests that endotoxins are unlikely to induce laminitis on their own, more likely contributing to the disease’s development and progression. One drawback of *ex vivo* studies is that tissues are isolated from the whole organism, thus the complex processes occurring on the systemic level are missing. For example, in the case of hoof explants, the lack of cytokines might explain why in some experiments that used LPS, high concentrations were necessary to induce lamellar separation. Moreover, the diffusion of nutrients and oxygen can vary throughout the explant and could depend on the explant dimensions (i.e., an excessive thickness could limit the permeability of the nutrients contained in the medium). To minimize all these effects, the present study employed a complex approach with several evaluations. In addition, in this study we have used male draft horses (Breton horses) that were relatively young; it would be interesting to apply the same *ex vivo* model for laminitis to even older horses of different breeds, such as pony breeds, to assess any differences in the studied parameters. # Conclusions The results of the present study provide evidence showing that it is possible to maintain HEs for at least 48 h following their harvest, and that sections could be used for *ex vivo* studies. The results of histological examination and the separation force test showed that increasing concentrations of MG lead to the deterioration of HE tissue integrity. Moreover, the effect of MG seems to be time -dependent because samples incubated for 48 h showed higher levels of deterioration compared with samples incubated for 24 h. RNA quantification of MMP-2 and -14 and TIMP-2 expression supported the occurrence of extracellular matrix remodeling induced by MG in all samples. Further studies are needed to advance our understanding of the role of glucose metabolism in pathogenesis of laminitis. # Supporting information The preliminary study results were presented in the Equine Colic Research Symposium of 2014 (Dublin, Ireland) and in Sisvet Congress 2017 (Naples, Italy). 10.1371/journal.pone.0253840.r001 Decision Letter 0 Males Jamie Staff Editor 2021 Jamie Males This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 9 Apr 2021 PONE-D-20-06143 Effect of sugar metabolite methylglyoxal on horse keratinocytes: an ex vivo model for laminitis PLOS ONE Dear Dr. Vercelli, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. Reviewer 1 has identified a number of questions that need to be addressed regarding aspects of the methods and analyses included in your study. Please ensure that you provide detailed responses to each of these questions and concerns, as well as addressing the presentational issues raised. Please submit your revised manuscript by May 21 2021 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <plosone@plos.org>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript: A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols>. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at <https://plos.org/protocols?utm_medium=editorial- email&utm_source=authorletters&utm_campaign=protocols>. We look forward to receiving your revised manuscript. Kind regards, Jamie Males Senior Editor PLOS ONE Journal Requirements: When submitting your revision, we need you to address these additional requirements. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at <https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main _body.pdf> and [https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_titl e_authors_affiliations.pdf](about:blank) Thank you for stating in your Funding Statement: The present study was funded with ex 60% fund of the University of Turin. Please provide an amended statement that declares \*all\* the funding or sources of support (whether external or internal to your organization) received during this study, as detailed online in our guide for authors at [http://journals.plos.org/plosone/s/submit- now. ](http://journals.plos.org/plosone/s/submit-now. ) Please also include the statement “There was no additional external funding received for this study.” in your updated Funding Statement. Please include your amended Funding Statement within your cover letter. We will change the online submission form on your behalf. 3\. We noticed you have some minor occurrence of overlapping text with the following previous publications, which needs to be addressed: \- <https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3605361/> \- <https://onlinelibrary.wiley.com/doi/full/10.1002/dmrr.2811> \- <https://beva.onlinelibrary.wiley.com/doi/abs/10.2746/042516406778400565> \- <https://www.sciencedirect.com/science/article/abs/pii/S0165242711000201?via% 3Dihub> The text that needs to be addressed involves: \- Lines 297-302 \- Lines 303-310 \- Lines 315-326 In your revision ensure you cite all your sources (including your own works), and quote or rephrase any duplicated text outside the methods section. Further consideration is dependent on these concerns being addressed. \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Partly Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 2\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 3\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 4\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 5\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: The authors have submitted a manuscript describing an ex vivo model of laminitis that they have created using a glucose metabolite (methylglyoxal) applied to lamellar explants. Such a model, if it could be validated, would be extremely useful for research in this area and has, in fact, been sought for some years now by researchers in the field. This could be a valuable contribution to those efforts - I have some questions and comments for the authors about their work, listed below (by line number in the manuscript): Line 1 - 'Effect of the sugar metabolite...'; ex vivo should be italicized Abstract: Line 15 - '...synthesized during the digestion process in horses...' Line 16 - '...excessive levels could lead to alterations in the hoof lamellar structure.' Line 18 - '...were applied to hoof explants (HE), which...' Line 19 - 'Microscopic' Line 20 - '..post-MG application.' Line 21 - '...(MMP)-2 and -14, and tissue inhibitor of metalloproteinases (TIMP)-2' Line 22 - '...at each time point for all...' Line 23 - '...mimicking laminitis. The separation force test revealed...' Line 26 - '...significant weakness, and samples...'; 'In the same samples, high levels of MMP-2 and -14 and low levels of TIMP-2 were present. Line 27 - 'All results support that high levels of MG could induce irreversible...' Introduction: Line 33 - 'It is characterized by damage/deterioration of the lamellar tissue...' Line 37 - '...acute or chronic, and it is not usually readily reversible.' Lines 38 - 39 - 'The most common etiology of laminitis cases involves gastrointestinal or metabolic disease.' Line 41 - '...hormonal influences and weight/pressure are both involved..' Line 46 - would add comma after 'isoforms' Line 48 - '...disturbances resulting from ingestion...' Line 49 - '...role in sepsis-related laminitis...' Line 50 - '...used as highly digestible...' Line 52 - '...glucose liberation...' Line 54 - '...followed by colic and laminitis.' (founder is redundant here, I believe) Line 56 - 'In this scenario, release of toxins from the hindgut is suspected to occur, which in theory induces degradation of the lamellar basement membrane...' Line 58 - '...infiltration into the lamellae.' Line 59 - '...can lead to rapid, devastating changes that result in the elaboration of toxins and other substances:...' Line 60 - 'Gram-negative and -positive...' Line 61 - '...stress, leading to...'; has this substance been shown to appear in equine serum/plasma following induction of a carbohydrate overload model of laminitis? That would seem to be central to the premise here, given the proposed route of exposure of the lamellae to this substance in vivo. Lines 64-65 - Would include some context-specific criteria where lactate may serve as a biomarker of laminitis risk (not laminitis itself) - certain situations are associated with hyperlactatemia but not laminitis risk (such as intense aerobic exercise), so would restrict this in some way to situations in which this is more likely to be the case (e.g., sepsis). Line 67 - '...which precedes and enables remodeling...' Line 70 - '...have shown that under specific...' Line 72 - '...making them an attractive model for the...' Line 77 - '...different exposure time points.' Materials and Methods: Line 81 - the authors might comment in the discussion about how widely these data might be extrapolated to other breeds/types of horses Line 82 - '...were enrolled in the study.' Line 84 - would add a comma after 'Italy)'; would remove the commas after 'collected' and 'consent' Line 88 - would remove 'symptoms and' from this sentence; what signs of laminitis, specifically, were evaluated (would list them)? Also, would specify whether the limb was a front or hind limb. Line 89 - would add a comma after 'joint'; '...within 45 minutes of slaughter and..' Line 101 - would use 'lamellae' instead of 'laminae' consistently throughout Line 114 - '48 hour incubation' Lines 116-117 - would include how survival was evaluated here (not just in the Results section) Line 117 - '...used for the rest of the study.' Line 125 - '...integrity of the samples.' Line 130 - '...lamellar site were considered valid measurements.' Lines 130-131 - I'm not sure what the authors are trying to say here, it's a bit unclear; consider rewording this sentence. Line 136 - '...then embedded in paraffin, and 3-um slices were mounted...' Line 142 - 'epidermal' Line 144 - '...or severe and extensive (Grade...' Lines 146 - 147 - This sentence is unclear - what particular characteristic of the SEL was evaluated? Line 149 - how long after the force testing did RNA extraction occur? Was RNA extracted from any samples that were not subjected to force testing? Can the authors comment on the likely influence of this testing on the mRNA concentrations of some of their target genes in the sample tissue (i.e., are they known to be influenced by stretch, if sufficient time had elapsed between force testing and RNA extraction)? Line 157 - can remove the comma after 'USA)' Line 162 - '...MMP-14, and TIMP-2 transcripts...' Line 164 - 'Equus caballus' Line 169 - '...gene accession numbers, and...' Line 174 - '...(ACTB), and...' Line 175 - how was this determined? Line 177 - '...triplicate, and template...' Line 179 - can remove comma after 'averaged' Line 187 - 'All the data were analyzed with a commercially available software program...' Lines 190-193 - 'The separation force test and q-PCR data were analyzed using one-way ANOVA and Tukey's...' Line 192 - How were the distributions of the histology score data analyzed? Line 196 - '...was accepted at values...' Results: Lines 200-201 - Were any more sensitive/quantifiable measures of autolysis used to evaluate this tissue? This seems like a very subjective assessment. Line 202 - Can the authors include a figure displaying representative histologic sections? Line 203 - '...in a few samples...' Line 204 - 'lamellae' Line 205 - '...changes, and histological...' Line 206 - 'structures; all samples...' Lines 208-209 - '...in the preliminary assay, samples were kept in DMEM+ for all subsequent experiments.' Line 212 - '...on all samples...' Line 213 - '...at predetermined time points.' Line 215 - '...separation of the HE lamellar structures could be... Line 221 - '...can be observed.' Line 222 - 'However, these differences did not reach the level of statistical significance.' Line 225 - 'methylglyoxal'; '(N=45 per time point)' Line 235 - '...due to alteration of...' Lines 237, 238, 241 - do the authors mean 'epidermal' instead of 'epithelial' in these instances? Lines 251-252 - 'Nevertheless, these differences did not reach the level of statistical significance.' Line 256 - would avoid commenting on trends, if possible Line 266, 270 - 'MG' is missing in the last line of the legend for Figure 6 and Figure 7 Line 267 - 'Tissue inhibitor of metalloproteinases-2 (TIMP-2) gene...' Discussion: Line 274 - would add a comma after 'solution'; can the authors discuss any other papers that have attempted to characterize/use a lamellar explant model of laminitis, for the purposes of comparing their model to others? (I believe there are at least a few out there currently.) Line 284 - '...but according to our results, FBS and L-glutamine are necessary for tissue preservation.' Line 290 - '...hoof lamellae, rendering them less robust and unable...' Line 291 - '...untreated samples are instead able to bear and remain intact.' Line 301 - would add a comma after 'features' Line 311 - would add a comma after 'concentration' Line 315 - would remove comma after 'detected' Line 320 - '...has been shown to be regulated in breast cancer, where...' Line 321 - '...with increased risk of malignancy...' Line 326 - '...in the past, yielding important...' Line 333 - would remove comma after 'intermediates' Line 341 - '...subjects show higher levels...' Line 342 - '...compared with those of healthy subjects.' Line 347 - 'Escherichia coli' - should be italicized Line 348 - 'induce the laminitis process...' Line 350 - '...suggests that endotoxins are unlikely to induce laminitis...' Line 351 - 'more likely merely contribute to the...' Lines 358-359 - can the authors clarify here how their approach and evaluations mitigated the effects of variable diffusion of nutrients and oxygen in explants? Line 362 - '...following their harvest, and that sections could be used for ex vivo studies.' Line 367 - would list specific MMP's here Lines 367-368 - I don't think that MMP mRNA expression necessarily correlates well with enzymatic activity, so would probably refine this statement a bit. Line 367-368 - 'Further studies are needed to advance our understanding...' Figure 1 - 'Sample preparation'; 'Histology/histology' Figure 2 - 'Outer'; 'Lamellae' Figure 4 - 'Histological score' Figures 5, 6, and 7 - 'Concentration of Methylglyoxal (mM)' Reviewer \#2: Congratulations for this important research work. It is not easy to culture hoof explants of horses mainly by secondary bacteria contamination. I only suggest you to replace the figure 1 by another one more explicative and less tedious. I also suggest to add a set of photographs showing the method for explant procurement in a sequential fashion (step by step). This technique is very important and useful for equine laminitis researchers around the world. \*\*\*\*\*\*\*\*\*\* 6\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No Reviewer \#2: No \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <figures@plos.org>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0253840.r002 Author response to Decision Letter 0 3 May 2021 Response letter manuscript PONE-D-20-06143 Journal requirements: • The manuscript has been formatted according to Journal’s style • The specification about funding has been added and the previous funding statement has been corrected • The overlapping texts have been corrected according to the suggestions. o <https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3605361/> � this reference (Miretti et al., 2017) has been added along the text and in reference list. Thank You for the suggestion because we forgot to cite a paper of one of the co- authors. o <https://onlinelibrary.wiley.com/doi/full/10.1002/dmrr.2811> � this paper was already present in the reference list and we added a new reference along the text. o <https://beva.onlinelibrary.wiley.com/doi/abs/10.2746/042516406778400565> � we are really sorry but we can not add this reference because we did not consult it to write the paper. We don’t know why there’s an overlapping. o <https://www.sciencedirect.com/science/article/abs/pii/S0165242711000201?via%3 Dihub> � this paper was already present in the reference list and we added a new reference along the text. Reviewer 1 Dear Reviewer, on behalf of all Authors, I would like to thank You for the time that you have spent to revise our manuscript. We appreciate Your efforts to improve our manuscript. We accepted all Your suggestions, and we provided the best answers that we could to the questions and comments that You addressed. Here I provide a point-to-point response. When rephrasing was suggested, I have just stated “fixed”. When a longer answer was needed, I provided an explanation. Line 1: fixed Abstract • Line 15: fixed • Line 16: fixed • Line 18: fixed • Line 19: fixed • Line 20: fixed • Line 21: fixed • Line 22: fixed • Line 23: fixed • Line 26: fixed • Line 27: fixed Introduction • Line 33: fixed • Line 37: fixed • Line 38-39: fixed • Line 41: fixed • Line 46: fixed • Line 48: fixed • Line 49: fixed • Line 50: fixed • Line 52: fixed • Line 54: fixed • Line 56: fixed • Line 58: fixed • Line 59: fixed • Line 60: fixed • Line 61: fixed. To answer to the Reviewer’s question “ has this substance been shown to appear in equine serum/plasma following induction of a carbohydrate overload model of laminitis? That would seem to be central to the premise here, given the proposed route of exposure of the lamellae to this substance in vivo” some sentences were added in order to clarify (L61-69 in the revised version). • Line 64-65: To answer to the Reviewer’s question “Would include some context- specific criteria where lactate may serve as a biomarker of laminitis risk (not laminitis itself) - certain situations are associated with hyperlactatemia but not laminitis risk (such as intense aerobic exercise), so would restrict this in some way to situations in which this is more likely to be the case (e.g., sepsis)” some sentences were added (L75-80 in the revised version). • Line 67: fixed • Line 70: fixed • Line 72: fixed • Line 77: fixed Material and methods • Line 81: According to Reviewer’s comment: “the authors might comment in the discussion about how widely these data might be extrapolated to other breeds/types of horses”, Authors provided a short explanation in discussion session (L 389-391 in the revised version). • Line 82: fixed • Line 84: fixed • Line 88: To answer to the Reviewer’s questions: “hind limb or front limb? List signs of laminitis that were excluded?” Authors added that front limbs were collected (L 99 in the revised version) and wrote the evaluated signs and symptoms (L 103-104 in the revised version). • Line 89: fixed • Line 101: fixed in this line and along the text • Line 114: fixed • Lines 116-117: According to Reviewer’s comment: “would include how survival was evaluated here (not just in the Results section)” a short sentence was added (L 131-132 in the revised version). • Line 117: fixed • Line 130: fixed • Lines 130-131: According to Reviewer’s comment: “I'm not sure what the authors are trying to say here, it's a bit unclear; consider rewording this sentence”, Authors rephrased (L 147-149 in the revised version). • Line 136: fixed • Line142: fixed • Line 144: fixed • Lines 146-147: According to Reviewer’s comment: ”how long after the force testing did RNA extraction occur? Was RNA extracted from any samples that were not subjected to force testing? Can the authors comment on the likely influence of this testing on the mRNA concentrations of some of their target genes in the sample tissue (i.e., are they known to be influenced by stretch, if sufficient time had elapsed between force testing and RNA extraction)?” Authors added some short sentences in lines 147-150 and 169-171 in order to explain that samples were immediately treated after the separation force test. According with reviewer's comment could be interesting evaluate changing in gene expression after the force test, but when the samples are maintained in culture condition for additional time after the test as reported by other studies (REF: Wu JH et al, Bone Joint Res. 2017 Mar;6(3):179-185; Qin TW, Sun YL, Thoreson AR, et al.. Effect of mechanical stimulation on bone marrow stromal cell-seeded tendon slice constructs: a potential engineered tendon patch for rotator cuff repair. Biomaterials 2015;51:43-50.The revitalisation of flexor tendon allografts with bone marrow stromal cells and mechanical stimulation: An ex vivo model revitalising flexor tendon allografts). In our case, we consider negligible the influence of the force test on the gene expression due to the closed timing between the test and RNA extraction. • Line 157: fixed • Line162: fixed • Line 164: fixed • Line 169: fixed • Line 174: fixed • Line 175: According to Reviewer’s question “how was this determined?” Authors answer that GAPDH was selected as housekeeping gene for the stable Cq values among the different experimental conditions. The correction is provided in lines 194-197 of the revised version. • Line 177: fixed • Line 179: fixed • Line 187: Authors tried to provide the suggested correction. We hope to have interpreted well the comments. • Line 190-193: fixed • Line 192: According to Reviewer’s question: “How were the distributions of the histology score data analyzed?” the test’s name (L210-211 in the revised version) and a short explanation (L 269-270 in the revised version) were added. • Line 196: fixed Results • Lines 200-201: According to Reviewer’s comment: “Were any more sensitive/quantifiable measures of autolysis used to evaluate this tissue? This seems like a very subjective assessment.”, Authors tried to rephrase and explain better (L 221 in the revised version). • Line 202: According to Reviewer’s question “Can the authors include a figure displaying representative histologic sections?”, Authors provided pictures to show the differences between well preserved hoof explant samples (using DMEM+) and other samples that were considered autholitic and not eligible for the further experiments. Pictures are presented as Figure 3 (L 232-238 in the revised version). The other pictures have been renumbered as consequence. • Line 203: fixed • Line 204: fixed • Line 205: fixed • Line 206: fixed • Lines 208-209: fixed • Line 212: fixed • Line 213: fixed • Line 215: fixed • Line 221: fixed • Line 222: Authors tried to provide the suggested correction. We hope to have correctly interpreted the comments. • Line 225: fixed • Lines 237, 238, 241: According to Reviewer’s question: “do the authors mean 'epidermal' instead of 'epithelial' in these instances?” the answer is: Yes. Authors provided a correction along the text (Lines 255, 256 and 259 in the revised version). • Lines 251-252: fixed • Line 256: According to Reviewer’s comment: “would avoid commenting on trends, if possible”, only statistically significances were considered. • Line 266-270: fixed • Line 267: fixed Discussion • Line 274: fixed. According to Reviewer’s comment: “can the authors discuss any other papers that have attempted to characterize/use a lamellar explant model of laminitis, for the purposes of comparing their model to others? (I believe there are at least a few out there currently.)”, short sentences were added (L325-329 in the revised version). • Line 284: fixed • Line 290: fixed • Line 291: fixed • Line 301: fixed • Line 311: fixed • Line 315: fixed • Line 320: fixed • Line 321: fixed • Line 326: fixed • Line 333: fixed • Line 341: fixed • Line 342: fixed • Line 347: fixed • Line 348: fixed • Line 350: fixed • Line 351: fixed • Line 358-359: According to Reviewer’s question “can the authors clarify here how their approach and evaluations mitigated the effects of variable diffusion of nutrients and oxygen in explants?” Authors added a short sentence to clarify (L380-381 in the revised version). • Line 362: fixed • Line 367: fixed • Lines 367-368: According to Reviewer’s comment: “I don't think that MMP mRNA expression necessarily correlates well with enzymatic activity, so would probably refine this statement a bit”, the sentence has been refined. • Lines 367-368: fixed Reviewer 1 also requested the following modifications on Figures: • Figure 1 - 'Sample preparation'; 'Histology/histology’: Figure 1 has been changed and designed according to the Reviewer 2 comment, considering also the corrections proposed by Reviewer 1. • Figure 2 - 'Outer'; 'Lamellae’: Figure has been corrected according to the Reviewer’s comment • Figure 4 - 'Histological score: Figure has been corrected according to the Reviewer’s comment • Figures 5, 6, and 7 - 'Concentration of Methylglyoxal (mM)': all the Figures have been corrected according to Reviewer suggestion. Reviewer 2: Reviewer 2 wrote: “Congratulations for this important research work. It is not easy to culture hoof explants of horses mainly by secondary bacteria contamination. I only suggest you to replace the figure 1 by another one more explicative and less tedious. I also suggest to add a set of photographs showing the method for explant procurement in a sequential fashion (step by step). This technique is very important and useful for equine laminitis researchers around the world”. Dear Reviewer, we are glad that our paper seems interesting for You. On behalf of all Authors, I would like to thank You for the time that you have spent to revise our manuscript. We accepted all Your suggestions, and here I provide the answers to Your comments. • According to Reviewer suggestion, Figure 1 was changes. Authors hope that this new design is easier to read and more attractive for readers. • Pictures: a panel of 4 pictures has been added as supplementary material ("Other") in order to show the different passages of the procedure. 10.1371/journal.pone.0253840.r003 Decision Letter 1 Singh Kanhaiya Academic Editor 2021 Kanhaiya Singh This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 26 May 2021 PONE-D-20-06143R1 Effect of sugar metabolite methylglyoxal on horse keratinocytes: an ex vivo model for laminitis PLOS ONE Dear Dr. Vercelli, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. Please submit your revised manuscript by Jul 10 2021 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <plosone@plos.org>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript: A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols>. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at <https://plos.org/protocols?utm_medium=editorial- email&utm_source=authorletters&utm_campaign=protocols>. We look forward to receiving your revised manuscript. Kind regards, Kanhaiya Singh, Ph.D Academic Editor PLOS ONE Journal Requirements: Please review your reference list to ensure that it is complete and correct. If you have cited papers that have been retracted, please include the rationale for doing so in the manuscript text, or remove these references and replace them with relevant current references. Any changes to the reference list should be mentioned in the rebuttal letter that accompanies your revised manuscript. If you need to cite a retracted article, indicate the article’s retracted status in the References list and also include a citation and full reference for the retraction notice. Additional Editor Comments (if provided): Please address the suggestions made by Reviewers 2 and 3. \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation. Reviewer \#1: (No Response) Reviewer \#2: All comments have been addressed Reviewer \#3: (No Response) \*\*\*\*\*\*\*\*\*\* 2\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Yes Reviewer \#2: Yes Reviewer \#3: Yes \*\*\*\*\*\*\*\*\*\* 3\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes Reviewer \#2: Yes Reviewer \#3: Yes \*\*\*\*\*\*\*\*\*\* 4\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes Reviewer \#2: Yes Reviewer \#3: Yes \*\*\*\*\*\*\*\*\*\* 5\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: Yes Reviewer \#3: Yes \*\*\*\*\*\*\*\*\*\* 6\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: The author have addressed the majority of the questions and comments that I had included in a review of the previous version of this manuscript, and I believe that it is improved. I have a few additional questions, by line, below: Title - 'Effect of the sugar metabolite...'; would use 'equine lamellar explants' instead of 'horse keratinocytes' in the title, to more accurately describe the tissue that was used; '... model of laminitis' Abstract: Line 18 - '...and excessive levels absorbed into peripheral blood could be delivered to the foot and lead to...' Line 23 - can remove the comma after '14' Line 25 - can remove 'a' after 'mimicking' Line 27-28 - this sentence (the one that begins with 'In the same...') is incomplete and should be reworded Introduction: Line 41 - '...resembling the human diabetic...' Line 49 - please add a space between 'from' and 'ingestion' Line 55 - '...caecal pH, often followed by colic and laminitis...' Line 59 - '...AGEs accumulate in significant amounts in the hoof lamellar tissue in the acute...' Line 61 - 'using the hyperinsulinemic model...' Line 63 - '...MG) causes the formation of AGEs, leading...' Line 66 - '...intermediate, it is converted...' Line 72 - 'In equines, feeding large amounts of fermentable carbohydrates with subsequent acidosis increases the level of plasma D-lactate...' Line 75 - '...D-lactate can be considered a...' Line 84 - '...making them an attractive model...' Line 87 - '...of MG on equine lamellar explants in an ex vivo...' Materials and Methods: Line 99 - '...typical stance, increased digital pulse, and increased hoof capsular warmth).' Line 100 - '...the right fore distal limb...' Line 112 - '...dermal lamellae, and the bone.' Line 120 - can remove 'of' in front of 'Fetal' Line 126 - 'Sample' Line 142 - would use 'lamellae' instead of 'laminae' throughout the manuscript Line 143 - 'broken at' Line 149 - 'One sample from each time point and from each MG concentration...' Line 150 - would add a comma after 'paraffin'; '...slices were mounted...' Line 153 - '...microscope, and images were captured...' Line 160 - '...evaluated for lesions of the secondary...' Line 164 - 'RNA-Later solution (Ambion), and the stratum lamellatum was disrupted using...' Line 177 - '...MMP-14, and TIMP-2 transcripts, qPCR...' Line 179 - 'designed from Equus caballus...' Line 200 - '...were designed from Equus caballus...' Line 205 - 'one-way ANOVA and Tukey's multiple... Line 207 - '...their distributions were analyzed using two-way ANOVA...' Results and discussion: Line 221 - '...all structures; all samples were...' Line 225 - 'characteristics' Line 236 - 'predetermined' Line 245 - '...for 48 h can be observed.' Line 258 - '...of all layers and were statistically...' Line 267 - 'concentrations' Line 280 - can remove the comma after 'concentration' Line 296 - '...as a suitable viable model system for investigating...' Line 307 - 'with our results, FBS and...' Line 309 - '...period, supporting the viability of the samples...' Line 310 - 'Other papers have evaluated the possibility of using hoof explant samples but focus their attention on different metabolic goals (e.g....' Line 312 - 'To the authors' knowledge, this is the first paper focusing specifically on the role of MG (...' Line 315 - 'to modification of the hoof lamellae...' Line 316 - '...that untreated samples are able to bear...' Line 317 - '...confirmed that higher concentrations of MG and a longer...' Line 319 - '...were confirmed by histological...' Line 322 - 'The results of gene expression of MMPs...' Line 325 - '...matrix remodeling, more severe histological features, and...' Line 337 - '...starch overload or other causes, increases...' Line 346 - can remove the comma after 'MMP2' here Line 349 - 'yielding' Line 363 - '...various pathologies; for example...' Line 370 - '...not sufficient to induce laminitis in vitro.' Line 373-374 - '...more likely contributing to the disease's...' Line 383-385 - '...(Breton horses) that were relatively young; it would be interesting to apply the same ex vivo model of laminitis to even older horses of different breeds, such as pony breeds, to assess any differences...' Line 388 - '...could be used for ex vivo studies.' Line 393 - would use 'supported' instead of 'confirmed' here Supplemental data: Line 519 - would use 'lamellar' instead of 'laminar' here Reviewer \#2: Congratulations. This is an important research work. In general, you addressed the concerns raised during the review process. Reviewer \#3: In this paper, the authors are examining the effects of excessive methylglyoxal on the hoof lamellar structure with an aim to identify the mechanism behind the Laminitis disease. They use an ex vivo experimental design and perform various types of analysis - macroscopic analysis, histological analysis, separation force test and gene expression. They identify that high levels of methylglyoxal could induce irreversible damage in the hooves which mimicks laminitis in an ex vivo model. Overall, the findings of the manuscript are well-supported by the data and the methods used are appropriate. The authors have also addressed the reviewers comments adequately. However, there are some points to consider that I have outlined below: 1\. In the revised manuscript, the authors have renamed the ‘Results’ section as ‘Results and Discussion’, however, the discussion has not been incorporated with the results. It is still presented as a separate section (minus the heading). The layout of this section will have to be changed significantly if the authors want to keep the ‘Results and Discussion’ heading. If not, I would recommend adding the ‘Discussion’ heading and keeping ‘Results’ separate. Minor concerns: 1\. In the abstract, the authors mention ‘Microscopical and histological analysis’, however, in the rest of the manuscript, there is ‘Macroscopical analysis’. Please correct accordingly. 2\. In the abstract, the second last line starting with ‘In the same samples…’ (Line 29), does not make sense, it seems incomplete. Please edit. 3\. Line 45, ‘…influences and weight/ pressures…’ needs to be edited for coherence. 4\. Lines 48-51, need to be changed to make sense. Right now, it seems like a collection of terms. 5\. Line 57, ‘this means’ should be ‘which means’. 6\. Line 64, ‘hoof lamellar tissue level’, the word ‘level’ can be removed. 7\. Line 70, ‘highly reactive intermediate is’, should be ‘highly reactive intermediate, it is’. 8\. Line 78, ‘sequent acidosis’, do the authors mean ‘subsequent acidosis’? 9\. Line 79, ‘increase the level of plasma levels’, should be ‘increase the plasma levels of’. 10\. Line 81, ‘also because’ is redundant. Use either also, or because. 11\. Line 89, ‘attracting model’ should be ‘attractive model’. 12\. Line 100, ‘for this study purpose’ can be rephrased to, ‘for the purposes of this study’. 13\. Line 105, ‘typical pain position and digital pulse and hot hoof’, one ‘and’ can be removed. 14\. Line 107, ‘within 45 minutes of slaughter’ should be ‘within 45 minutes’, since the sentence is starting with ‘following slaughter’. 15\. Line 114, Mungall and Pollitt is missing an in-text citation. 16\. Line 131, should be ‘Fig S1’ to make it consistent with the rest of the manuscript. 17\. Line 133, do the authors mean ‘Sample survival’? 18\. Line 141, for ‘controls (K)’, reference the figure number. 19\. Line 144, ‘the’ before structural integrity should be removed. 20\. Line 150, do the authors mean ‘broken’, instead of ‘brocken’? 21\. Line 172, change RNA later to RNAlater (as per manufacturers label). 22\. Line 198 and Line 202, move Cq definition from line 202 to line 198. 23\. Line 207, table 1 is not in line with the text. 24\. Line 211-214, do not make sense. Which softwares? Which companies? Please edit. Line 211, remove ‘a’ before ‘commercially’ if you are referring to more than 1 software. 25\. Line 216, ‘Turkey’s’ should be ‘Tukey’s’. 26\. Line 222, define SD. 27\. Line 228, should be ‘performed for all samples’. 28\. Line 232-233, separate the sentences. ‘All samples’ should be a different sentence. 29\. Line 248, should be ‘predetermined’. 30\. Line 254-255 is very confusing. Please clarify. 31\. Line 271-272, ‘and resulted statistically different’, please clarify. 32\. Line 277, what do you mean by definition? 33\. Line 280, should be ‘concentrations’. 34\. Fig 6-8 legend, the gene descriptions should be in the text and not in the figure legends. 35\. Line 328-329, please rephrase for coherence. 36\. Line 330, should be ‘In our knowledge’. 37\. Line 340, please edit for coherence. 38\. Line 344, edit spelling to ‘correlated’. 39\. Line 345, should be ‘resulting in structural weakness’. 40\. Line 367, please define the ‘entire process’ so as to make the sentence complete. 41\. Line 369, edit spelling to ‘yielding’. 42\. Line 378, the acronym AGEs does not need to be defined again. 43\. Line 391, the word ‘bacteria’ should be before E.coli and not after. 44\. Line 395, rephrase for coherence. 45\. Line 403-406, rephrase for coherence. 46\. Line 410, remove ‘an’ before ex vivo. 47\. Line 412, time-dependent would be more appropriate than ‘time-related’. 48\. Figure 1, The authors can consider changing ‘force test’ to ‘separation force test’. It would then be consistent throughout the manuscript. 49\. Figure 4, table headings are missing. 50\. Figure 4,6,7,8, the legend for time needs to be consistent throughout. Either 24h, 48h or T24h, T48h. 51\. Figure 6-8, the Y-axis should be Cq instead of Ct, for consistency between figures and text. \*\*\*\*\*\*\*\*\*\* 7\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No Reviewer \#2: No Reviewer \#3: No \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <figures@plos.org>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0253840.r004 Author response to Decision Letter 1 7 Jun 2021 Response to Reviewers Journal requirements Dear Editor, As required in your email, the following items are under submission: • Response to Reviewers • Revised Manuscript with Track Changes • Manuscript • Cover letter An updated and more precise financial disclosure statement is presented in our cover letter. All Figures have been checked using PAGE online system and are submitted according to PAGE system revision. Tables have been formatted according to <https://journals.plos.org/plosone/s/tables> References have been numbered along the text in order at first citation and reference list has been updated consequently. Two references have been delated from the reference list (Bierhaus et al., 2012 and Kyaw-Tanner et al., 2012) because they were not present in-text. The in-text citation of “Visser 2008” (L199) has been removed because it does not exist. References have been organized and formatted in Vancouver style using Zotero software. Reviewer comments Reviewer 1 Authors would like to thank R1 for all the efforts made to improve our paper. Authors accepted all suggestions that have been addressed and corrections have been provided along the text. Reading once again the revised manuscript, we correct some minor issue that are listed at the end of this letter. Here we copied R1 comments (italics) and we provide the answers. Reviewer \#1: The author have addressed the majority of the questions and comments that I had included in a review of the previous version of this manuscript, and I believe that it is improved. I have a few additional questions, by line, below: Title - 'Effect of the sugar metabolite...'; would use 'equine lamellar explants' instead of 'horse keratinocytes' in the title, to more accurately describe the tissue that was used; '... model of laminitis': corrections have been provided according to the comments. Abstract: • Line 18 - '...and excessive levels absorbed into peripheral blood could be delivered to the foot and lead to...' � correction has been provided according to Reviewer’s comment. • Line 23 - can remove the comma after '14'�correction has been provided according to Reviewer’s comment. • Line 25 - can remove 'a' after 'mimicking'� correction has been provided according to reviewer’s comment. • Line 27-28 - this sentence (the one that begins with 'In the same...') is incomplete and should be reworded �according to reviewer’s comment the sentence has been erased. R3 suggested the same. Introduction: • Line 41 - '...resembling the human diabetic...'. � correction has been provided according to reviewer’s comment. • Line 49 - please add a space between 'from' and 'ingestion' �Correction has been provided according to reviewer’s comment. • Line 55 - '...caecal pH, often followed by colic and laminitis...'. � Correction has been provided according to reviewer’s comment. • Line 59 - '...AGEs accumulate in significant amounts in the hoof lamellar tissue in the acute...'. � Correction has been provided according to reviewer’s comment. • Line 61 - 'using the hyperinsulinemic model...'� Correction has been provided according to reviewer’s comment. • Line 63 - '...MG) causes the formation of AGEs, leading...' � Correction has been provided according to reviewer’s comment. • Line 66 - '...intermediate, it is converted...': � Correction has been provided according to reviewer’s comment. • Line 72 - 'In equines, feeding large amounts of fermentable carbohydrates with subsequent acidosis increases the level of plasma D-lactate...'. � Correction has been provided according to reviewer’s comment. • Line 75 - '...D-lactate can be considered a...'. � Correction has been provided according to reviewer’s comment. • Line 84 - '...making them an attractive model...’ �Correction has been provided according to reviewer’s comment. • Line 87 - '...of MG on equine lamellar explants in an ex vivo...'. � Correction has been provided according to reviewer’s comment. Materials and Methods: • Line 99 - '...typical stance, increased digital pulse, and increased hoof capsular warmth).' � Correction has been provided according to reviewer’s comment. • Line 100 - '...the right fore distal limb...’ � Correction has been provided according to reviewer’s comment. • Line 112 - '...dermal lamellae, and the bone. � Correction has been provided according to reviewer’s comment. • Line 120 - can remove 'of' in front of 'Fetal'. � Correction has been provided according to reviewer’s comment. • Line 126 - 'Sample'. � Correction has been provided according to reviewer’s comment • Line 142 - would use 'lamellae' instead of 'laminae' throughout the manuscript �Correction has been provided according to reviewer’s comment in this line and along the text and captions. • Line 143 - 'broken at'. � Correction has been provided according to reviewer’s comment. Also, R3 suggested the same correction. • Line 149 - 'One sample from each time point and from each MG concentration...'. � Correction has been provided according to reviewer’s comment. • Line 150 - would add a comma after 'paraffin'; '...slices were mounted...'. � Correction has been provided according to reviewer’s comment. • Line 153 - '...microscope, and images were captured...'. � Correction has been provided according to reviewer’s comment. • Line 160 - '...evaluated for lesions of the secondary...'. � Correction has been provided according to reviewer’s comment. • Line 164 - 'RNA-Later solution (Ambion), and the stratum lamellatum was disrupted using...' � Correction has been provided according to reviewer’s comment. R3 suggested ‘RNAlater’ instead of 'RNA-Later’. We checked the manufactured label, and we wrote ‘RNAlater’. • Line 177 - '...MMP-14, and TIMP-2 transcripts, qPCR...'. � Correction has been provided according to reviewer’s comment. • Line 179 - 'designed from Equus caballus...'. �Correction has been provided according to reviewer’s comment. • Line 200 - '...were designed from Equus caballus...'. � Correction has been provided according to reviewer’s comment. • Line 205 - 'one-way ANOVA and Tukey's multiple.... �. Correction has been provided according to reviewer’s comment. • Line 207 - '...their distributions were analyzed using two-way ANOVA...'. � Correction has been provided according to reviewer’s comment. Results and discussion: • Line 221 - '...all structures; all samples were...'. � Correction has been provided according to reviewer’s comment. • Line 225 - 'characteristics'. � Correction has been provided according to reviewer’s comment. • Line 236 - 'predetermined'. � Correction has been provided according to reviewer’s comment. • Line 245 - '...for 48 h can be observed. � Correction has been provided according to reviewer’s comment. • Line 258 - '...of all layers and were statistically...' �Correction has been provided according to reviewer’s comment. • Line 267 - 'concentrations'. � Correction has been provided according to reviewer’s comment. • Line 280 - can remove the comma after 'concentration'. � Correction has been provided according to reviewer’s comment. • Line 296 - '...as a suitable viable model system for investigating...’ � Correction has been provided according to reviewer’s comment. • Line 307 - 'with our results, FBS and...' �Correction has been provided according to reviewer’s comment. • Line 309 - '...period, supporting the viability of the samples...' � Correction has been provided according to reviewer’s comment. • Line 310 - 'Other papers have evaluated the possibility of using hoof explant samples but focus their attention on different metabolic goals (e.g....'. � Correction has been provided according to reviewer’s comment. • Line 312 - 'To the authors' knowledge, this is the first paper focusing specifically on the role of MG (...' � Correction has been provided according to reviewer’s comment. • Line 315 - 'to modification of the hoof lamellae...’ � Correction has been provided according to reviewer’s comment. • Line 316 - '...that untreated samples are able to bear...'. � Correction has been provided according to reviewer’s comment. • Line 317 - '...confirmed that higher concentrations of MG and a longer...' � Correction has been provided according to reviewer’s comment. • Line 319 - '...were confirmed by histological...'. �. Correction has been provided according to reviewer’s comment. • Line 322 - 'The results of gene expression of MMPs...' � Correction has been provided according to reviewer’s comment. • Line 325 - '...matrix remodeling, more severe histological features, and...' � Correction has been provided according to reviewer’s comment. • Line 337 - '...starch overload or other causes, increases...' � Correction has been provided according to reviewer’s comment. • Line 346 - can remove the comma after 'MMP2' here. Correction has been provided according to reviewer’s comment. • Line 349 - 'yielding' �Correction has been provided according to reviewer’s comment. • Line 363 - '...various pathologies; for example...' � Correction has been provided according to reviewer’s comment • Line 370 - '...not sufficient to induce laminitis in vitro.' �. Correction has been provided according to reviewer’s comment. • Line 373-374 - '...more likely contributing to the disease's...' � Correction has been provided according to reviewer’s comment. • Line 383-385 - '...(Breton horses) that were relatively young; it would be interesting to apply the same ex vivo model of laminitis to even older horses of different breeds, such as pony breeds, to assess any differences...'. � Corrections have been provided according to reviewer’s comment. • Line 388 - '...could be used for ex vivo studies.' � Correction has been provided according to reviewer’s comment. • Line 393 - would use 'supported' instead of 'confirmed' here. � Correction has been provided according to reviewer’s comment. Supplemental data: • Line 519 - would use 'lamellar' instead of 'laminar' here. �The correction has been provided according to the comment in L519 and in Fig S1. Reviewer 2 Reviewer \#2: Congratulations. This is an important research work. In general, you addressed the concerns raised during the review process. Author would like to sincerely thank R2 for all the efforts made to improve our paper. We are pleased to hear that R2 is satisfied about the corrections that have been provided in the first revision round. Reading once again the revised manuscript, we correct some minor issue that are listed at the end of this letter. Reviewer 3 Authors would like to thank R3 for all the efforts made to improve our paper. Authors accepted all suggestions that have been addressed and corrections have been provided along the text. Reading once again the revised manuscript, we correct some minor issue that are listed at the end of this letter. Here we copied R3 comments (italics) and we provide the answers. Reviewer \#3: In this paper, the authors are examining the effects of excessive methylglyoxal on the hoof lamellar structure with an aim to identify the mechanism behind the Laminitis disease. They use an ex vivo experimental design and perform various types of analysis - macroscopic analysis, histological analysis, separation force test and gene expression. They identify that high levels of methylglyoxal could induce irreversible damage in the hooves which mimicks laminitis in an ex vivo model. Overall, the findings of the manuscript are well-supported by the data and the methods used are appropriate. The authors have also addressed the reviewers comments adequately. However, there are some points to consider that I have outlined below: 1\. In the revised manuscript, the authors have renamed the ‘Results’ section as ‘Results and Discussion’, however, the discussion has not been incorporated with the results. It is still presented as a separate section (minus the heading). The layout of this section will have to be changed significantly if the authors want to keep the ‘Results and Discussion’ heading. If not, I would recommend adding the ‘Discussion’ heading and keeping ‘Results’ separate. � Thank you for your suggestions. Authors would like to maintain the two sections separated. For this reason, headings have been changed. Minor concerns: 1\. In the abstract, the authors mention ‘Microscopical and histological analysis’, however, in the rest of the manuscript, there is ‘Macroscopical analysis’. Please correct accordingly � According to reviewer’s comment, correction has been provided. 2\. In the abstract, the second last line starting with ‘In the same samples…’ (Line 29), does not make sense, it seems incomplete. Please edit. � the sentence has been edited. 3\. Line 45, ‘…influences and weight/ pressures…’ needs to be edited for coherence. � The sentence has been edited according to reviewer’s comment. 4\. Lines 48-51, need to be changed to make sense. Right now, it seems like a collection of terms. � sentence has been rephrased in order to be clearer 5\. Line 57, ‘this means’ should be ‘which means’. � Correction has been provided according to reviewer’s comment. 6\. Line 64, ‘hoof lamellar tissue level’, the word ‘level’ can be removed. � ‘Level ‘has been removed, also according to R1 suggestion. 7\. Line 70, ‘highly reactive intermediate is’, should be ‘highly reactive intermediate, it is’. � The same correction was proposed also by R1. Correction has been provided. 8\. Line 78, ‘sequent acidosis’, do the authors mean ‘subsequent acidosis’? � yes, thank you for your comment, Correction has been provided. 9\. Line 79, ‘increase the level of plasma levels’, should be ‘increase the plasma levels of’.�Correction has been provided according to reviewer’s comment. 10\. Line 81, ‘also because’ is redundant. Use either also, or because. � Correction has been provided according to reviewer’s comment. 11\. Line 89, ‘attracting model’ should be ‘attractive model’. �The same correction was proposed also by R1. Correction has been provided. 12\. Line 100, ‘for this study purpose’ can be rephrased to, ‘for the purposes of this study’. � The sentence has been rephrased according to the comment of R3. 13\. Line 105, ‘typical pain position and digital pulse and hot hoof’, one ‘and’ can be removed. � R1 asked to rephrase the sentence. We hope that R3 agrees with this version. 14\. Line 107, ‘within 45 minutes of slaughter’ should be ‘within 45 minutes’, since the sentence is starting with ‘following slaughter’. � thank you for your comment, Correction has been provided. 15\. Line 114, Mungall and Pollitt is missing an in-text citation. � All the in- text citations have been formatted according to the journal requirements and the reference list has been updated. According to this, also the comment about L114 has been fixed. 16\. Line 131, should be ‘Fig S1’ to make it consistent with the rest of the manuscript. � correction has been provided. Thank you for the suggestion. 17\. Line 133, do the authors mean ‘Sample survival’? � The same correction was proposed also by R1. Correction has been provided. 18\. Line 141, for ‘controls (K)’, reference the figure number. � reference to figures has been added 19\. Line 144, ‘the’ before structural integrity should be removed. � ‘the’ has been removed 20\. Line 150, do the authors mean ‘broken’, instead of ‘brocken’?--\> yes, thank you for your suggestion. 21\. Line 172, change RNA later to RNAlater (as per manufacturers label). � R1 proposed a different correction (‘RNA-Later’) but we checked the manufacturer label and we accepted your suggestion. 22\. Line 198 and Line 202, move Cq definition from line 202 to line 198. � The Cq definition has been moved and the sentences have been rephrased in order to be clearer. 23\. Line 207, table 1 is not in line with the text. � all tables have been formatted according to the journal style. We hope that table 1 now is in line with the text. 24\. Line 211-214, do not make sense. Which softwares? Which companies? Please edit. Line 211, remove ‘a’ before ‘commercially’ if you are referring to more than 1 software. � thank you for your suggestion. The sentence has been edited and corrections have been provided. 25\. Line 216, ‘Turkey’s’ should be ‘Tukey’s’. � Correction has been provided. 26\. Line 222, define SD. � SD has been defined. 27\. Line 228, should be ‘performed for all samples’. � thank you for your suggestion. Correction has been provided. 28\. Line 232-233, separate the sentences. ‘All samples’ should be a different sentence. � R1 asked to change ‘,’ with ‘;’ to separate the sentences. We hope that R3 agrees with R1. 29\. Line 248, should be ‘predetermined’. � The same correction was proposed also by R1. Correction has been provided. 30\. Line 254-255 is very confusing. Please clarify. � these two lines have been modified according to the comments provided by reviewers in the fist round. Nevertheless, Authors checked the first manuscript that was submitted and tried to rephrase the sentences in order to be clearer and more related to the original meaning. 31\. Line 271-272, ‘and resulted statistically different’, please clarify. � sentence has been rephrased according to R1 comment. We hope that R3 agrees with R1. 32\. Line 277, what do you mean by definition? � no, in this case, we mean the definition of the histological image. 33\. Line 280, should be ‘concentrations’. � ‘s’ has been added according to R1 and R3comments. 34\. Fig 6-8 legend, the gene descriptions should be in the text and not in the figure legends. � the definitions have been removed from figure captions. 35\. Line 328-329, please rephrase for coherence. � some corrections have been provided according to the R1 comments. We hope that R3 agrees with R1. 36\. Line 330, should be ‘In our knowledge’. ��some corrections have been provided according to the R1 comments. We hope that R3 agrees with R1. 37\. Line 340, please edit for coherence. � some corrections have been provided according to the R1 comments. We hope that R3 agrees with R1. 38\. Line 344, edit spelling to ‘correlated’. � thank you. Correction has been provided. 39\. Line 345, should be ‘resulting in structural weakness’. �thank you for your comment. Correction has been provided. 40\. Line 367, please define the ‘entire process’ so as to make the sentence complete. � we add ‘laminitic’ 41\. Line 369, edit spelling to ‘yielding’. � also R1 suggested this correction. Correction has been provided. 42\. Line 378, the acronym AGEs does not need to be defined again. � definition has been removed. 43\. Line 391, the word ‘bacteria’ should be before E.coli and not after.--\> Thank you for your comment. This correction has been fixed. 44\. Line 395, rephrase for coherence. � also R1 suggested a rephrasing here. We hope that R3 agrees with R1. 45\. Line 403-406, rephrase for coherence. � also R1 suggested a rephrasing here. We hope that R3 agrees with R1. 46\. Line 410, remove ‘an’ before ex vivo. � ‘an’ as been delated according to R1 and R3 suggestions. 47\. Line 412, time-dependent would be more appropriate than ‘time-related’. � the word has been changed according to R3 suggestion. 48\. Figure 1, The authors can consider changing ‘force test’ to ‘separation force test’. It would then be consistent throughout the manuscript. � Figure 1 has been changed according to R3 suggestion. 49\. Figure 4, table headings are missing. � the complete table has been inserted. 50\. Figure 4,6,7,8, the legend for time needs to be consistent throughout. Either 24h, 48h or T24h, T48h. � Fig. 4,6-8 are now consistent and the same wording (T24 h and T 48 h) has been used along the text. 51\. Figure 6-8, the Y-axis should be Cq instead of Ct, for consistency between figures and text. � Y axis label has been corrected according to R3 comment. Authors comments • Reading once again the manuscript, we have found some typos and double-spaced words: we have fixed everything. • Some references have been delated. An explanation is provided at the beginning of this letter and in the cover letter. • L292: Fig 9 does not exist. Thus a correction has been provided. • L305: ‘/maintaing’ has been removed. • L356: ‘ alpha oxoaldehydes’ instead of ‘alphaoxoaldehydes’ • In the whole document, MMP-2, MMP-14 and TIMP-2 (using ‘-‘) instead of MMP2, MMP14 and TIMP2 have been written. This was done in order to be coherent along the text, tables and figures using the same wording and according to a suggestion of the first round of revision. • Table 2: the title was missing. We added it. 10.1371/journal.pone.0253840.r005 Decision Letter 2 Singh Kanhaiya Academic Editor 2021 Kanhaiya Singh This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 15 Jun 2021 Effect of sugar metabolite methylglyoxal on equine lamellar explants: an ex vivo model of laminitis PONE-D-20-06143R2 Dear Dr. Vercelli, We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements. Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication. An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at <http://www.editorialmanager.com/pone/>, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to- date. If you have any billing related questions, please contact our Author Billing department directly at <authorbilling@plos.org>. If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact <onepress@plos.org>. Kind regards, Kanhaiya Singh, Ph.D Academic Editor PLOS ONE Additional Editor Comments (optional): Reviewers' comments: 10.1371/journal.pone.0253840.r006 Acceptance letter Singh Kanhaiya Academic Editor 2021 Kanhaiya Singh This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 19 Jul 2021 PONE-D-20-06143R2 Effect of sugar metabolite methylglyoxal on equine lamellar explants: an ex vivo model of laminitis Dear Dr. Vercelli: I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department. If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact <onepress@plos.org>. If we can help with anything else, please email us at <plosone@plos.org>. Thank you for submitting your work to PLOS ONE and supporting open access. Kind regards, PLOS ONE Editorial Office Staff on behalf of Dr. Kanhaiya Singh Academic Editor PLOS ONE [^1]: The authors have declared that no competing interests exist.

# Introduction *Chlamydia trachomatis* (serovars D-K and lymphogranuloma venereum serovars L1-L3) are agents of human sexually transmitted disease, whereas ocular infections with *C*. *trachomatis* serovars A-C can lead to blindness. *C*. *trachomatis* is a member of a larger *Chlamydiaceae* family that contains numerous species that have likely co-evolved with a eukaryotic host for \>700 million years. All *Chlamydia* spp. are Gram-negative obligate intracellular bacteria that possess a conserved, biphasic developmental cycle. Development is initiated when infectious particles termed elementary bodies (EBs) invade host cells and differentiate into noninfectious, vegetative forms termed reticulate bodies (RBs). RB growth is eventually accompanied by asynchronous conversion of RBs to EBs. Subsequent exit from the host cell is mediated by lysis or extrusion. Intracellular development occurs entirely within a parasitophorous vacuole termed an inclusion. Chlamydiae develop effectively segregated from the host cytosol since the inclusion membrane is passively impermeable to molecules \>520 Da. Despite this physical separation, *Chlamydia* spp. are capable of directly modulating host cell biology. Members of the *Chlamydiaceae* all express a type III secretion system (T3SS) to promote survival from within a protected niche. Similar to systems in other T3S-expressing pathogens, the chlamydial T3SS is a multi-protein nanomachine capable of secreting and subsequently translocating (hereafter collectively referred to as secretion) anti-host proteins termed effectors, (T3SE) directly into an associated eukaryotic cell. The chlamydial T3SS is present, and apparently active throughout development. EBs contain abundant levels of effectors required for invasion, and secretion can be detected within minutes of attachment to a host cell. Described effectors first secreted during invasion include the <u>t</u>ranslocated <u>a</u>ctin-<u>r</u>ecruiting <u>p</u>hosphoprotein TarP, the <u>t</u>ranslocated <u>e</u>arly <u>p</u>hospho<u>p</u>rotein TepP, and the human Ahnak interacting protein designated CT694. Subsequent to entry, chlamydiae begin to secrete a number of identified effectors, the most abundant of which are the <u>inc</u>lusion membrane (Inc) class of effectors. Many Inc effectors interact with elements of host vesicular transport pathways while others likely play key roles in maintaining inclusion membrane architecture through interactions with other Incs. The continued accumulation of Incs in the expanding inclusion membrane indicates that T3S is likely active in RBs until completion of the developmental cycle. Interaction with T3S chaperones and *in silico* analyses have been used successfully to identify chlamydial T3S substrates. However, the use of surrogate T3SSs has been perhaps most efficacious in discovering putative chlamydial effectors. Several recent studies have leveraged *Yersinia* or *Shigella* T3SS in large-scale screens for T3S substrates. These studies and others have yielded a long list of potential effectors that require further validation. Confirmation of secretion by *Chlamydia* typically employs the use of indirect immunofluorescence assays to detect localization of the effector within or beyond the inclusion membrane. This approach requires the generation of effector-specific antibodies. In addition, the assay can be confounded by low abundance of a given effector. Therefore, there is a need for an efficacious system to easily detect secretion of chlamydial proteins during infection. Reproducible transformation of *Chlamydia* with a stably-maintained shuttle vector has overcome a significant barrier that impeded efficient progress in investigating chlamydial infection biology. This approach has enabled development of second generation vectors expressing fluorescent proteins or enabling conditional gene expression. This approach has already been used to express epitope-tagged effectors for efficacious detection of Inc secretion. Inc proteins have the advantage of being easily detectible due to concentration in the inclusion membrane. We regard it likely that detection of low-abundance, effectors that do not concentrate in a particular compartment will require signal amplification for detection. Numerous enzymatic tags have been employed in other T3SSs to demonstrate cytosolic localization of secreted effectors. The use of β-lactamase (BlaM) translational fusions, originally developed to detect pathogenic *Escherichia coli* effector secretion, has proven to be a convenient and sensitive tool for detection of bacterial protein secretion. Infected cells are treated with the cell-permeant reagent CCF2-AM which is subsequently converted to a membrane-impermeant molecule by esterases in the host cytosol. CCF2-AM is composed of the flourophores coumarin and fluorescein. Excitation of intact CCF2-AM at 409 nm results in green fluorescent via fluorescence energy transfer (FRET). Secretion of a given bacterial protein-BlaM fusion into the host cytosol is readily indicated when CCF2-AM is cleaved by the BlaM moiety to disrupt FRET and yield blue fluorescence. This approach is particularly desirable for an obligate intracellular pathogen such as *Chlamydia* since evidence of secretion can be detected directly in the absence of host cell lysis. We therefore designed a two-step vector system that would enable ectopic expression of T3SE-β-lactamase chimeras. We provide proof-of-principle evidence herein that this system allows the robust detection of T3SE secretion in a tissue-culture infection model. We focused efforts on characterization of CT695. This putative T3SE is secreted by the *Yersinia* T3SS and binds the chlamydial T3S chaperone Slc1, yet secretion by chlamydiae has not been confirmed. We reveal for the first time that *C*. *trachomatis* CT695 is secreted by chlamydiae at multiple stages of the developmental cycle. # Methods ## Cell cultures and organisms *C*. *trachomatis* serovar L2 (LGV 434) was cultivated in HeLa 229 epithelial cell monolayers (ATCC CCL-1.2; American Type Culture Collection, Manassas, VA), routinely maintained at 37°C in an atmosphere of 5% CO₂/95% humidified air in RPMI-1640 containing 2 mM L-glutamine (GibcoLife Technologies Corporation, Grand Island, NY) supplemented with 10% (vol/vol) heat-inactivated fetal bovine serum (HIFBS; Sigma-Aldrich Company, St. Louis, MO). Where appropriate, intrinsically fluorescent chlamydiae were generated by labeling bacteria with CellTracker Red CMPTX (Life technologies) as described. All EBs were purified from HeLa cells by centrifugation through MD-76R (diatrizoate meglumine and diatrizoate sodium injection U.S.P.; Mallingckrodt Pharmaceuticals, Mulhuddart, Ireland) density gradients (DG-purified) as previously described and were used as the infection source for all experiments. Assessment of chlamydial growth was accomplished by enumeration of progeny EBs from 24 hr cultures as described. Chemically competent *E*. *coli* 10-beta (NEB, Ipswich, MA) was used for routine cloning, and *dam*^-/*dcm*^- *E*. *coli* (NEB) was used to propagate plasmids prior to transformation of chlamydiae. Where appropriate, 50 μg/ml carbenicillin was used for *E*. *coli* selection while 1.0 μg/ml cycloheximide and 0.6 μg/ml Penicillin G sodium (PenG) was used during chlamydial transformations. ## Vector construction pGFP::SW2 was generously provided by Ian Clarke (University of Southampton). This template was modified using custom PCR primers (; IDT, Coralville, IA). Sequence encoding mCherry was amplified from pmCherry-C1 vector using forward and reverse primers mC@GFP F and mC@GFP R, respectively. This was used to replace the GFP gene in pGFP::SW2 by insertion/deletion PCR as described to produce pMC::SW2. Sequence encoding chloramphenicol drug resistance was amplified from pACD4K-C-loxP using forward and reverse primers SalI+AscI+Chlor F and SalI+AscI+Chlor R, respectively. In order to construct pL2dest, SalI restriction enzyme and Quick Ligation Kit (NEB) were used to digest and ligate the chloramphenicol drug resistance amplicon into pMC::SW2, simultaneously introducing AscI restriction sites around the chloramphenicol open reading frame. pUC19 was used as the backbone for construction of the β-lactamase translational fusions. The *Neisseria meningitidis* promoter was amplified from pGFP::SW2 and inserted into pUC19 by insertion/deletion PCR using forward and reverse primers NmP@puC F and NmP@pUC R, respectively, producing pUCNmP. *ct694-*, *ct695-*, *ct696-*, *euo-*, *groEL-*, and *tarp-bla* fusions were constructed by amplifying each open reading frame from *C*. *trachomatis* serovar L2 genomic DNA preparation, and by inserting each amplicon between the *Neisseria meningitidis* promoter and the full-length β-lactamase gene of pUCNmP by insertion/deletion PCR (primers listed). Each fusion was amplified from pUCNmP with primers NmP+BlaFus+AscI F and R, and inserted into pL2dest by AscI digestion followed by Quick Ligation (NEB). Loss of chloramphenicol resistance was used as a convenient marker for successful cloning of BlaM fusions in *E*. *coli*. Final constructs were transformed into *dam*-/*dcm*- *E*. *coli*, and plasmids were purified using a QIAfilter Plasmid Maxi Kit (Qiagen, Valencia, CA) prior to transformation into *C*. *trachomatis* L2. Q5 High-Fidelity DNA Polymerase (NEB) was used for all PCR amplifications and direct DNA sequencing (ACGT, Inc) was used to confirm all constructs. ## Chlamydial transformation *C*. *trachomatis* L2 were transformed as described with modifications. 2.6 x 10⁶ EBs and 2 μg of plasmid DNA were mixed in 50 μl CaCl₂ buffer (10 mM Tris pH 7.4 and 50 mM CaCl₂) and incubated at room temperature for 30 minutes. Each mixture was suspended in 2 ml Hanks Balanced Salt Solution (HBSS; Life Technologies) and applied to a 10-cm² well containing a monolayer of confluent McCoy cells. Monolayers were infected by centrifugation at 900 x g for 1 hr at room temperature, after which HBSS was replaced with 2 ml RPMI + 10% FBS medium without drugs. At 7 hours post infection (hpi), cultures were supplemented with cycloheximide and PenG. Cells were harvested 48 hpi with a cell scraper and centrifuged at 20,000 x g for 30 min at 4°C. The pellet was suspended in 1 ml HBSS, centrifuged at 200 x g for 5 min at 4°C, and the supernatant was used to infect a new confluent monolayer of McCoy cells in a 10 cm² well by centrifugation (900 x g, 1 hr, room temperature). Immediately after infection, medium containing both cycloheximide and PenG was added. Cells were harvested 48 hpi with a cell scraper, and the process of centrifugation and infection was repeated until transformed *Chlamydia* were recovered (typically 1 to 3 rounds of reinfection). In order to ensure clonal isolates, transformed *C*. *trachomatis* strains were diluted in HBSS and applied to confluent McCoy monolayers grown in 384-well plates (Greiner Cell Culture Microplate, catalog number 781091) at a concentration of one IFU for every 100 wells (approximately four inclusions per 384-well plate). Monolayers were infected by centrifugation with *C*. *trachomatis* at 900xg for 60 minutes at room temperature. Infection was allowed to continue in the absence of drug selection for seven days. Wells containing *C*. *trachomatis* (roughly one well for every 100) were then scraped with a p200 tip, and each isogenic population was then applied to new monolayers in T75 flasks for further expansion with antibiotic selection. ## Expression assays For assessment of gene expression, RNA was harvested at indicated times using the Aurum Total RNA Mini Kit (Bio-Rad, Hercules, Ca.) according to the manufacturer’s instructions. RNA was converted to cDNA using the QuantiTect Reverse Transcription Kit (Qiagen). Transcript levels were determined by quantitative real-time PCR using the Bio-Rad CFX96 Real-Time Systen (Bio-Rad), iTaq Universal SYBR Green Supermix (Bio-Rad), and appropriate primers. All quantitative real-time PCR primers were confirmed to amplify with efficiencies \>95%. For assessment of ectopically expressed protein levels, samples were obtained from HeLa cells infected with *C*. *trachomatis* at an MOI of 1. Cells were gently harvested with ice cold PBS and immediately concentrated by the addition of trichloroacetic acid to 10% (v/v) and centrifugation at 20,000 x g for 30 min at 4°C. Protein pellets were analyzed by SDS-PAGE electrophoresis followed by immunoblot analysis using β-lactamase-specific antibodies (Pierce, Rockford, IL) or α-Hsp60 (Santa Cruz, Dallas, TX) as a loading control. Proteins were visualized by peroxidase-conjugated secondary antibodies followed by development with ECL Prime (GE Healthcare, Pittsburgh, PA). ## BlaM Secretion Assay The presence of fusion proteins in the cytosol of infected HeLa cells was observed directly with the use of the GeneBLAzer In Vivo Detection Kit (Invitrogen). HeLa monolayers were cultivated on glass cover slips to a confluence of ca. 75% and infected with *C*. *trachomatis* L2 expressing various β-lactamase-fusion proteins. CCF2-AM substrate was applied 24 hpi for 30 minutes, samples were fixed in 4% paraformaldehyde, and fluorescence was observed using a Leica TCS SP5 laser scanning confocal microscope. Images were processed equivalently using Adobe Photoshop CS2 version 9.0 (Adobe Systems, San Jose, CA). ## BSA-EGTA release assay Cell-free release of secreted proteins from EBs was accomplished essentially as described. Briefly, volumes of 5 x 10⁷ EBs were suspended in 50 mM acetate buffer and one replicate was supplemented with bovine serum albumin (BSA; Sigma) and EGTA pH 7.4 to 5 μM final concentration for each. EBs were incubated for 2 hrs at 37°C and bacteria were pelleted by centrifugation at 20,000 x g for 15 min. Proteins from bacterial pellets and cell-free supernatants were precipitated using trichloroacetic acid and subsequent pellets were suspended in equal volumes of SDS-PAGE solublization solution. Supernatant material was loaded at 5X bacterial pellets, proteins were resolved via SDS- PAGE, and probed in immunoblots with α-TarP, α-Hsp60 (Santa Cruz), and α-CT695 (described below). Proteins were visualized by peroxidase-conjugated secondary antibodies and chemiluminescence development. ## Fluorescence microscopy localization assays Localization of CT695 and TarP was determined via indirect immunofluorescence using CT695-specific antibodies or α-TarP. Full-length, His-tagged CT695 was used as antigen for production of antibodies. The coding sequence for *C*. *trachomatis* L2 CT695 was amplified using Q5 DNA polymerase and primers sets (5’-GGGGACAAGTTTGTACAAAAAA GCAGGCTTCAG TAGCATAAGCCCTATAGGGGGG-3’ and 5’-GGGGACCACTT TGTACAAGAAAGCTGG GTCCTATTAGATATTCCCAACCGAAGAAGG-3’) for transfer into the GATEWAY (Life Technologies) entry vector pDONR-221. Donor sequence was mobilized into pDEST-17 and constructs were verified via DNA sequencing (GENEWIZ). His-Tagged CT695 was expressed in *E*. *coli* BL21-Al (Invitrogen), and protein was purified to homogeneity via passage of lysates over TALON affinity resin (Clontech, Mountain View, CA). Polyclonal antibodies were raised in female New Zealand White rabbits as previously described. To assess invasion- related secretion of endogenous CT695, HeLa cultures were infected for 1 hr with CMPTX-labeled *C*. *trachomatis* L2 at an MOI of ca. 10. Cultures were thoroughly washed and fixed for 20 min by treatment with 4% paraformaldehyde. Samples were permeablized by treatment with 0.1% Triton X100 in Tris-buffered saline supplemented with 5% BSA. *Chlamydia* were visualized via either intrinsic CMPTX label or with MOMP-specific antibodies. All images were acquired by epifluorescence microscopy using a 60x apochromat objective plus 1.5x intermediate magnification on a TE2000U inverted photomicroscope (Nikon, Melville, NY) equipped with a Retiga EXi 1394, 12-bit monochrome CCD camera (QImaging, Surrey, BC, Canada) and MetaMorph imaging software version 6.3r2 (Molecular Devices, Downington, PA). Images were processed equivalently using Adobe Photoshop CS2 version 9.0 (Adobe Systems). ## Statistical Analysis All presented data are representative of a minimum of three independent experiments. Quantitative data were generated from experiments containing triplicate biological replicates. All data are shown as mean of these replicates with 1 standard deviation. Calculation of standard deviation of the mean and assessment via Student’s t-test statistical analyses were performed using GraphPad Prism 6 version 6.04 (Graphpad Software Inc., La Jolla, CA). # Results We began by assessing whether CT695 is encoded within a dedicated T3S-related locus. *C*. *trachomatis* CT695 is encoded immediately downstream from the validated T3SE CT694 and upstream from the hypothetical CT696. Flanking gene orientation and a predicted rho-independent transcription terminator upstream from *ct694* indicate that *ct694*, *ct695*, and *ct696* could comprise an operon. Transcriptional profiling of *C*. *trachomatis* serovar D via microarray indicated that *ct694* is expressed much later than *ct695* or *ct696*. However, deep-sequencing-based transcriptome analysis of *C*. *trachomatis* L2 revealed a common transcriptional start site for *ct694* and *ct695*, raising the possibility of polycistronic message. We assayed temporal gene transcription by RT-PCR from total culture RNA harvested from *C*. *trachomatis* L2 infected HeLa cells 6, 15, or 24 hpi. Samples were normalized to *rpoD* and transcript levels were assayed for *ct694*, *ct695*, and *ct696*. Basal levels of message were detected for all three genes as early as 6 hpi. As expected, these levels were increased (ca. 10-fold) at each of the later time-points. We next addressed transcriptional linkage using the previously reported approach using RT-PCR and gene-spanning primer sets. A correctly sized amplicon (2.2 kb) was detectible only in the presence of RT and *ct694*-*ct695* spanning primers at 24 hpi. No product was detected using *ct695*-*ct696* spanning primers even though these primers were capable of yielding the 2.4 kb product using DNA template (data not shown). A faint band was detected in the 6 hpi sample, but the apparent size was below 1.5 kb. Since qRT-PCR indicate the presence of message at earlier time- points, we used gene-spanning primers and qRT-PCR as a more sensitive means to detect polycistronic message. In agreement with the RT-PCR data, an amplicon containing *ct694* and *ct695* was apparent at 24 hpi. Although a small (2.75-fold) increase over background was detected for *ct694*/*ct695* at 15 hpi, this was not statistically significant. No product was detected for *ct695-ct696*. Taken together, these data are consistent with mid-cycle expression of *ct694*, *ct695*, and *ct696*. However, *ct694* and *ct695* are likely transcribed separately from *ct696*, and *ct694*/*ct695* expression at 15 hpi is likely due to individual promoters and not co-transcription. Transcriptional linkage of *ct694* and *ct695*, coupled with previously reported secretion by the heterologous T3SS and association with the chaperone Slc1, predicts that CT695 is secreted by chlamydiae. We wanted to take advantage of the newly acquired ability to transform *Chlamydia* in order to construct a reporter system that would facilitate assessment of protein secretion during infection. The entry vector pUCNmP contains the *N*. *meningitidis* promoter (NmP) described by Wang, et al. positioned upstream from the complete TEM-1 β-lactamase coding sequence. This enables insertion of any chlamydial gene using insertion PCR to create an in-frame fusion with BlaM. PCR primers flanked with the AscI recognition-site sequence are then used to amplify the construct, followed by digestion and ligation with pL2dest, a derivative of pGFP::SW2 with GFP in place of mCherry and an engineered AscI restriction site. The coding sequences for CT694, CT695, and CT696 were PCR-amplified from *C*. *trachomatis* L2 and mobilized into pUCNmP to create translational fusions with the downstream βlaM gene. Similar constructs containing Tarp and Euo or GroEL were generated as positive and negative secretion controls, respectively. Entry clones were subsequently transferred into pL2-dest and used to transform *C*. *trachomatis* L2. PenG selection was applied, and transformed chlamydiae were isolated by limiting dilution. None of the constructs appeared to impact chlamydial growth since progeny IFUs were unchanged at 24 hpi compared to untransformed chlamydiae. HeLa cells were infected with respective strains and whole culture RNA or protein was isolated at 24 hpi to assess for trans-gene expression. qRT-PCR of total RNA was employed to assess respective transcript levels in comparison to mCherry message. Similar levels of each message were detected, consistent with the use of a constitutive Nm-promoter. While mCherry transcript appeared more abundant that *groEL-bla* and *tarp-bla*, these differences were not statistically significant. Whole-culture proteins were probed in immunoblots with BlaM-specific antibodies to visualize individual chimeric proteins and Hsp60-specific antibodies as a loading control. Both CT694-BLA and GroEL-BLA were detected in abundant amounts. Euo-BLA, TarP-BLA, and CT695-BLA were readily detectible but at comparably lower amounts. We were unable to detect CT696-BLA despite multiple attempts. With the exception of CT696, we therefore reasoned that chimeric proteins were expressed at sufficient levels to progress to secretion assays in *Chlamydia*-infected cells. We utilized the established FRET disruption via BlaM-mediated cleavage of CCF2-AM to test the ability of chimeric proteins to gain access to the host cytosol. HeLa cells were infected for 24 hrs and then loaded with CCF2-AM for 30 min. Cultures were paraformaldehyde fixed and processed for confocal immunofluorescence. In each case, chlamydial inclusions were visualized by mCherry-derived signal. As expected, mature chlamydial inclusions were detectible in each expression strain. Disruption of FRET results in blue signal, and robust signal was detected in HeLa cytosols when CT694-BLA, CT695-BLA, or the positive control TarP-BLA were expressed by chlamydiae. Only green fluorescence—indicative of intact CCF2-AM—was prominently detected when either GroEL-BLA or Euo-BLA were expressed. The apparent lack of blue signal in the presence of CT696-BLA was inconclusive given our inability to detect this chimeric protein. We did not detect significant green or blue signal within the inclusion lumen or within bacteria. Hence it is likely that eukaryotic cytosolic esterases cleave CCF2-AM to preclude passage across the inclusion membrane. These data indicate that the BlaM reporter system is applicable during *Chlamydia* infection and indicate that, similar to CT694 and TarP, CT695 is a secreted chlamydial protein. While the BlaM reporter indicates secretion, this approach does not indicate specific protein localization. We generated antibodies specific for full-length CT695 to examine CT695 immunolocalization and confirm secretion of endogenous protein. Antiserum specificity was confirmed by probing lysates of uninfected HeLa cells, *C*. *trachomatis*-infected HeLa cells or density-gradient-purified *C*. *trachomatis* EBs. Interestingly, CT695 migrated as a doublet with the lower band being most abundant in EB lysates. Despite a faint cross-reactive band apparent in HeLa lysates, these antibodies appeared specific and were used in subsequent analyses. TarP is packaged in EBs and low levels of this effector can be released by treatment of purified EBs with BSA and EDTA. To test whether CT695 behaved similarly, EBs were mock-treated or treated with BSA/EDTA followed by centrifugation to generate an EB-containing pellet and a cell-free supernatant. These fractions were probed in immunoblots with our CT695-specific antibodies. Material was probed with α-Tarp as a positive control or α-Hsp60 as a negative control. All proteins were detected in EB pellets, and CT695 reproducibly migrated as a doublet. Similarly to TarP, CT695 was detected in the cell-free supernatant only after treatment with BSA and EDTA. Hsp60 was not released from EBs by this treatment. We next asked whether immunolocalization of CT695 resembled that reported for TarP and CT694 during infection. HeLa cells were infected with intrinsically-labeled *C*. *trachomatis* L2 at an MOI of ca. 10 for 1 hr, fixed with paraformaldehyde, and processed for analysis via epifluorscence. Red-labeled EBs marked the position of chlamydiae while MOMP, TarP, and CT695 were detected using antibodies (green). While typical background staining was visible, both TarP- and CT695-specific signal was typically detected concentrated immediately adjacent to EBs in a pattern consistent with secretion of these proteins. In contrast, MOMP-specific signal overlapped with red EBs. Taken together, these data indicate that EBs are loaded with CT695, and this pool of protein can be secreted during the invasion process. Finally, we extended our analysis to test localization of CT695 during a later stage of development. HeLa cells were infected with *C*. *trachomatis* L2 and processed for immunofluorescence microscopy at 24 hpi. Staining with α-CT695 revealed signal that co-localized with Hsp60-stained chlamydiae as well as in a rim-like staining pattern typical of inclusion membrane staining. In contrast, TarP-specific staining was confined to intra-inclusion chlamydiae. The pattern was consistent with staining of EBs since inclusion membrane-localized RBs seemed to lack significant staining with α-TarP. TarP staining is comparable to CT694 which can only be detected co-localizing with bacteria at this time-point. Host syntaxin 6 has been previously shown to associate with the *C*. *trachomatis* inclusion membrane. We therefore stained inclusions with syntaxin-6 and CT695-specific antibodies to confirm inclusion membrane association of CT695. The rim-like pattern of CT695 signal overlapped with syntaxin 6-specific signal, indicating that CT695 likely accumulates in the host cytosol adjacent to the inclusion membrane. These data indicate that CT695 is secreted at later stages of development where it can associate with the inclusion membrane. # Discussion Prior to development of techniques to genetically manipulate chlamydiae, detection of protein secretion during infection traditionally involved the use of specific antibodies to examine protein localization via indirect immunofluorescence. Detection of the putative effector in the inclusion membrane or extra-inclusion spaces represented the sole indicator that a protein was secreted by chlamydiae. The recently acquired ability to reproducibly transform *Chlamydia* with a stably-maintained shuttle vector has opened the door to more efficacious approaches to directly test for protein secretion. For example, ectopic expression of epitope-tagged Inc proteins has surmounted the need to generate antigen-specific antibodies for a putative secreted protein. This approach was also used to confirm secretion of the type II secretion substrate CPAF. These data suggest that effector-reporter fusion proteins represent a useful approach for examination of protein secretion in a tissue-culture infection model. Fusion of TEM-1 β-lactamase to secretion substrates was originally developed to examine T3SE secretion in pathogenic *E*. *coli* and has become widely used to study secretion by extracellular and facultative-intracellular bacteria. We wondered whether this approach could be adapted to an obligate intracellular pathogen such as *Chlamydia*. In support of this concept, BlaM fusions have been used successfully to examine protein secretion by *Coxiella burnetti*. We chose to focus initial proof-of-principle experiments on the locus containing *ct694-ct696*. CT694 is a validated T3SE that is secreted during the invasion process and interacts with host cell AHNAK. We thought it likely that CT695 could also be secreted by chlamydiae since this protein can be secreted by the heterologous *Yersinia* T3SS. In addition, CT695 shares a common transcriptional start site and chaperone with CT694. We did not detect secretion of CT696 in *Yersinia*, yet de Cunha, et al reported indeterminate results leaving open the possibility of low-level secretion of CT696 by yersiniae. Hence this locus was ideal for the development of our chlamydial secretion reporter system since it contained a *bona fide* secretion substrate, a putative secretion substrate, and a questionable substrate. To accomplish our goal, we generated a two-vector system to first enable general construction of BlaM fusions with any chlamydial gene downstream from a constitutive *N*. *meningitidis* promoter. We chose the constitutive Nmp promoter since it has been functionally validated in *Chlamydia* and to avoid any regulatory complications that could arise from ectopic-overexpression from endogenous promoters. Standard primer sets enable amplification and subsequent AscI-mediated cloning into a *Chlamydia* shuttle vector. Transformation of *Chlamydia* is selected for by resistance to PenG. Constitutive expression of mCherry serves as an easily identifiable indication of transformation and as an internal control for normalization of gene expression. Genes encoding CT694, CT695, and CT696 were mobilized into *Chlamydia* using this system. We also included the T3SE TarP as a positive control and two non-secreted proteins, Euo and GroEL, as negative controls. The presence of neither vector-only nor chlamydial genes interfered with chlamydial development. This was true for Euo, which has been shown to repress gene transcription. However, we anticipate that this result is likely specific to the chlamydial gene being expressed. We recommend that growth be directly assessed for each new construct since we cannot rule out the possibility that ectopic expression of other chlamydial genes could alter chlamydial development. Ectopic expression of genes in the presence of endogenous, genomic copies of respective genes was another potential complication since competition for the secretion apparatus or chaperones could have interfered with secretion. In agreement with previous studies overexpressing IncD, this did not appear to be a factor. Indeed, trans-encoded overexpression of T3SE is routine in other T3SS-expressing bacteria. Although Tarp, CT694, and CT695 share Slc1 as a secretion chaperone, most T3SE contain both chaperone-dependent and independent secretion signals. Therefore, it is perhaps not surprising that overexpression of secretion substrates in *Chlamydia* is a productive approach. We cannot exclude the possibility, however, that competition among recombinant and endogenous proteins could result in false negative results in this assay. With the exception of CT696, all constructs were well expressed in *Chlamydia*. Although *ct696-bla* transcript was produced at levels comparable to the other constructs, protein levels were below detectable limits. The precise cause of this is unclear. However, we recognize the possibility that certain proteins which are natively expressed at extremely low levels may lack the translational machinery to allow for expression of additional constructs regardless of transcript levels. This result highlights the possibility that ectopic expression may not be possible for all chlamydial gene products. Regardless, results for the remaining constructs were conclusive. BlaM fusion to CT694, CT695, and TarP all resulted in blue signal indicative of cytosolic CCF2-AM cleavage. Hence, these proteins were clearly secreted into the host cytosol. Our vector contains an additional vector-encoded *blaM* conferring penicillin resistance, that could have confounded results. For example, chlamydial lysis in conjunction with an unexpectedly permeable inclusion membrane could have led to spurious BlaM in the HeLa cytosol. However, Euo and especially the abundant GroEL BlaM fusions did not yield significant blue signal. In addition, an extra copy of BlaM did not confound a similar approach in *C*. *burnetti*. Finally, strains have been passaged at least 8 times without loss of the intact plasmid (data not shown). Therefore recombination among multiple gene copies does not seem to be an issue. We currently have no means to confirm that secretion by *Chlamydia* is dependent on the T3SS. Any genetic lesion rendering T3S inactive is likely to be lethal to the bacteria. Although chemical inhibitors of type III secretion such as salicylidene acylhydrazides have been employed, they appear to not specifically target T3SS. Based on secretion in heterologous T3SS we can only infer this pathway for deployment. Evidence for secretion of TarP and CT694 has been restricted to invasion. Our results are clearly consistent with continued secretion of TarP- and CT694-containing fusion proteins later in development. Whether this finding reflects temporal secretion patterns for endogenous proteins remains unclear. The T3SS is clearly active throughout chlamydial development, and it is possible that forced expression of TarP and CT694 could result in atypical timing for secretion. However, we were able to detect endogenous CT695 at later times since the protein was concentrated at the inclusion membrane. Since CT694 and CT695 can be transcriptionally linked, it is plausible that CT694 is also secreted during later development. Although immunoblot revealed detectible levels of CT694 throughout development, detection of endogenous protein via immunolocalization was likely confounded by low abundance and/or the lack of effector concentration in a specific cellular compartment. How CT695 may be contributing to chlamydial infection remains to be determined. We detected evidence of endogenous CT695 secretion during invasion and subsequent development. We conclude that, similar to TarP, TepP, and CT694, CT695 is involved in early events necessary for chlamydiae to gain entry and-or establish an intracellular replication niche. Unlike, TarP, TepP, and CT694, ectopic expression of CT695 in yeast did not result in an overt phenotype that would give hints with regard to function. The apparent localization of CT695 adjacent to the inclusion membrane is interesting. CT695 does not contain predicted trans-membrane domains and may associate with membranes through interactions with other proteins or via direct association with lipids. CT694 contains a membrane localization domain found in effectors such as *Yersinia* YopE and *Pseudomonas* ExoS. It is therefore possible that CT695 could associate with membranes via a similar mechanism. Regardless, our immunolocalization studies imply that CT695 is likely a multifunctional effector necessary at multiple stages of chlamydial development. While the BlaM reporter system does not provide information regarding effector localization, there are several advantages to using this approach. *Chlamydia* employ T2S, T3S, and T5S to deploy host-interactive proteins and estimates based on current findings suggest as many as 80 proteins in the chlamydial secretome. Therefore, there is certainly a need for an approach to screen for secreted proteins in the context of a chlamydial infection. Although we used fixed samples for microscopy, secretion can easily be visualized in live cells using this reporter. This opens numerous possibilities that include quantitative and kinetic studies of effector secretion and translocation. In addition, the BlaM reporter system has been employed during animal infection studies to discriminate cell types susceptible to effector injection or separate infected from bystander cells. This system would also provide an efficacious platform to study the nature of T3 secretion signals. All of these approaches are adaptable for the study of *Chlamydia* pathogenesis. We conclude that use of BlaM fusion constructs will prove to be an efficacious approach for the study of protein secretion by chlamydiae. # Supporting Information We thank Dr. Ian Clarke (University of Southampton) for kindly sharing pGFP::SW2. We also thank Josh Ferrell and Dr. K. Wolf critical reading of the manuscript. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: KEM KAF. Performed the experiments: KEM KAF. Analyzed the data: KEM KAF. Contributed reagents/materials/analysis tools: KEM KAF. Wrote the paper: KEM KAF.

# Introduction Pancreatic cancer is the 4^th leading cause of cancer death in the United States and 5^th in the United Kingdom, with most cases being categorized as either metastatic or locally advanced at first presentation. As potentially curative surgical resection can be performed in only 15–20% of pancreatic cancer patients, the treatment goal for the majority of these patients is palliative in nature. For more than 15 years, the current standard of care for advanced disease has been chemotherapy with gemcitabine alone (G), after it was shown in a phase III randomized control trial (RCT) to offer greater symptom relief with a modest 1-year survival advantage (18% versus 2%) when compared to 5- fluorouracil. Since then, a number of phase II and III RCTs have attempted to improve the gemcitabine anti-tumour activity through gemcitabine-based combinations with cytotoxic and/or targeted agents such as capecitabine, oxaliplatin, erlotinib, and cisplatin. Recent trials have also compared gemcitabine alone to gemcitabine plus nab-paclitaxel (GnP), and a combination regimen without gemcitabine consisting of folinic acid, fluorouracil, irinotecan hydrochloride and oxaliplatin (FOLFIRINOX). The trial of G versus GnP found statistically significant hazard ratios (HRs) for overall survival (OS) in favour of the GnP combination. The safety analysis found that serious life-threatening toxicity was not increased with GnP and that adverse events were acceptable and manageable. Thus, the authors concluded that GnP may be considered as a new standard of treatment for advanced pancreatic cancer. In the FOLFIRINOX trial, survival was significantly better in the FOLFIRINOX group, but with an increased occurrence of adverse events. The study concluded that FOLFIRINOX should also be considered as a first-line option for advanced pancreatic cancer patients; however, due to safety concerns, it should be reserved for patients younger than 75 years of age and with a good performance status. No currently ongoing trials directly compare GnP and FOLFIRINOX. While the addition of these two chemotherapy regimens and their improvement in survival represent significant recent progress over gemcitabine monotherapy, the most effective chemotherapy strategy in clinical practice remains to be determined. As direct comparison of combination therapies has been tested mostly against single agent gemcitabine as the control arm in most clinical trials, the relative effectiveness of the various regimens remains unclear. In these instances, multiple treatment comparisons (MTC) can be used to synthesize evidence from RCTs using both direct (head-to-head) and indirect (using a common comparator) comparisons. MTC are valuable tools that are frequently employed by healthcare decision makers such as the National Institute for Health and Clinical Excellence and the Canadian Agency for Drugs and Technologies in Health, where their usage is gaining widespread acceptance. The aim of this study was to perform Bayesian MTC in order to determine the most effective treatment for advanced pancreatic cancer, taking into account the efficacy and safety profiles of each regimen. Through our analysis, we were able to achieve this goal. # Methods ## Literature Search We conducted a systematic literature review through the MEDLINE, EMBASE, and Cochrane Centre Register of Controlled Trials databases, as well as ASCO meeting abstracts up to and including May 23, 2013. Trials were limited to first-line treatment in pancreatic cancer or adenocarcinoma patients. Studies were limited to randomized controlled trials (RCTs) that used one of the following chemotherapy regimens: G, G + fluorouracil (GF), G + capecitabine (GCap), G + S1 (GS), G + cisplatin (GCis), G + oxaliplatin (GOx), G + erlotinib (GE), GnP, and FOLFIRINOX. These regimens were determined a priori by the authors, as they are clinically the most commonly considered treatments for advanced pancreatic cancer with prior studies suggesting possible benefits to patients. The outcomes of interest included OS, progression-free survival (PFS), and grade 3/4 toxicities. RCTs that did not include patients with advanced pancreatic cancer were excluded. Non-randomized trials and those concerning other malignancies, such as neuroendocrine tumours or lymphoma, were excluded. Trials comparing radiotherapy, hormonal, or gene therapy, and those comparing chemotherapy to no treatment (best supportive care) were excluded. No language restrictions were imposed. The articles that were not freely available to us were requested from the authors. ## Screening Two independent authors reviewed the literature search results and included studies that met the prespecified eligibility criteria. When reports overlapped or were duplicated, we retained the study with the most recent data that could be used in the meta-analysis. Discrepancies were resolved by consensus or by a third author. Our review has been reported using the PRISMA reporting guidelines. ## Data Abstraction and Analysis Data recorded included the following: first author, publication year, study location, regimens being compared, number of patients randomized to each treatment arm, median age of patients, percentage of patients with performance status of ECOG 0, 1, or 2 and the percentage of patients with locally advanced or advanced disease respectively was recorded. The treatments were sorted into categories based on the regimen: G, GF, GCap, GS, GCis, GOx, GE, GnP, and FOLFIRINOX. Risk of bias assessment was performed using the Cochrane risk of bias tool. The data extracted from each study included the following: OS, PFS, objective response rate (ObRR), and the occurrence of adverse events (febrile neutropenia, neuropathy, fatigue, and diarrhea) for all the chemotherapy regimens. If median values for PFS and OS were available, they were also recorded. If the HRs for OS and PFS were detailed in the publication, they were extracted directly, along with 95% confidence intervals (CIs) from Cox regression. Otherwise, HRs were calculated using the methods outlined by Parmar et al. A two-tailed p\<0.05 value was recorded whenever available to determine whether a statistically significant difference was detected between the two regimens being compared. Two independent authors extracted data and discrepancies were reviewed by a third author to reach consensus. ## Statistical Analysis We first made pairwise comparisons of regimens from the trials based on direct evidence only. We then performed MTC in a Bayesian model. The MTC combined direct and indirect evidence for specific pairwise comparisons and allowed data across a range of regimens to be compared in a simple network. Bayesian methods combine likelihoods, as a function of the parameters with a prior probability distribution based on previous knowledge, to obtain a posterior probability distribution of the parameters. The posterior probabilities provide a straightforward way to calculate the most effective treatment in the absence of head-to-head trials. By plotting the posterior densities of the direct, indirect, and network estimates, direct and indirect evidence can be combined to provide a network estimate and a single effect size. This effect size has increased precision than that of any one type of evidence alone. The Bayesian approach has undergone significant development in recent years and is able to monitor convergence in posterior distribution and reflect the uncertainty in estimating heterogeneity, offering significant improvements over the frequentist random-effects model, which cannot estimate that uncertainty. In more complex networks, especially those involving multi-armed trials, Bayesian approaches are more developed and more accessible than their frequentist counterparts. Analyses were done using Bayesian Markov Chain Monte-Carlo (MCMC) sampling in WinBUGS, version 1.4.3 and reported according to the Quality of Reporting of Meta-analyses (QUOROM) and International Society for Pharmacoeconomics and Outcomes Research (ISPOR) guidelines. In WinBUGS, 3 chains were fit with 40,000 burn-ins and 40,000 iterations each. Assessment of convergence was done using model diagnostics, such as trace plots and the Brooks-Gelman-Rubin statistic. Model fit was determined based on the residual deviance and deviance information criterion (DIC) for each outcome measure. The random effects model was used for OS, PFS, and ObRR because the residual deviance was less than the number of unconstrained data points and the deviance information criterion for each of these outcome measures favoured this model over the fixed effects model. Fixed effects were used in reporting toxicities because the residual deviance and DIC favoured this model. We used the following non-informative prior distributions: uniform (0,2) for standard deviation of the random effects model and normal (0, tau  = 0.0001) for log\[HR\]s. Non-informative priors were used because this allowed the trial data to inform the results, rather than letting strong priors dictate the results. The primary endpoint was OS and the secondary endpoints were PFS and ObRR. OS and PFS were summarized as log\[HR\], ObRR and toxicities were summarized as log\[Odds Ratio\]. Effect sizes are described with 95% credible regions (CRs), since “credible” is a more appropriate term than “confidence” when conducting Bayesian MTC. Consistency between direct and indirect evidence was assessed by comparing direct pairwise comparison estimates to the results generated in the MTC. Probability of each regimen being the best among all regimens were computed by ranking the relative efficacies of all regimens in each iteration and then calculating the proportion of each regimen being ranked first across all iterations. In order to assess the comparability of included studies, between- study heterogeneity was estimated and reported using the I² statistic; the value of I² lies between 0% and 100%, where 0% indicates no observed heterogeneity and larger values show increasing heterogeneity. Based on the HR results of the MTC, we attempted to project the survival of patients receiving each of the regimens and compared the results to the median OS of G. Projected median OS was calculated using a median OS of 5.65 months for G as reported by Buris et al. Survival was estimated based on the MTC results and the methods presented by Altman and Andersen. # Results ## Literature Search Results shows a flow diagram of the selection process for the studies included in our meta-analysis. 1269 studies were identified from the literature search, 386 studies were excluded because they were duplicates, and 801 were excluded after the abstracts were reviewed based on the prespecified criteria. Of the 82 studies that underwent full text review, 25 were excluded because they were an abstract of a full-included study, 22 had a different comparison arm, 4 were secondary analyses, 2 were quality of life studies, 2 were pooled analyses, 1 study was not randomized, 1 was a review, 1 was a tumour marker study, 1 was a safety analysis, and 1 study was excluded because it was retrospective. Twenty- two studies were identified to be included in this review,. 16 studies, involving 5488 randomized patients contained sufficient data to be included in the quantitative synthesis (meta-analysis). The studies included in the meta- analysis consisted of 15 manuscripts and 1 ASCO meeting abstract, which was subsequently published as a full manuscript. The subsequent publication was reviewed and the results were verified and found to be identical to the results reported in the original abstract. ## Study Quality A summary of the risk of bias for each included study can be found in and. All included studies were randomized and 12 out of the 16 studies followed intention-to-treat analysis for the primary endpoint, thus minimizing selection bias and attrition bias, respectively. Only one study had blinding of patients or personnel. Although blinding of outcome assessors was not explicitly indicated, 13 studies had OS as the primary endpoint, which would not be influenced by the outcome assessor. Therefore there is a low risk of detection bias in these studies. Allocation concealment was not mentioned in any of the studies, so some potential selection bias may be present. ## Trial Characteristics The chemotherapy regimens used in the included studies were G vs. GF (three studies), G vs. GCap (three studies), G vs. GS (three studies), G vs. GCis (seven studies), G vs. GOx (two studies), G vs. GE (one study), G vs. FOLFIRINOX (one study), GCap + GOx (one study), and G + GnP (one study). The treatment strategy network is shown in. All trials included in the meta-analysis reported median PFS and OS. There was no significant clinical heterogeneity between the studies based on the patient characteristics and outcomes in the G reference arm (median PFS  = 3 to 4 months, median OS  = 6 to 7 months). ## Comparison of Regimens The outcomes assessed in all the trials were OS, PFS, ObRR, and number of toxicity-related adverse events. Of the 16 trials that compared different regimens, seven found statistically significant differences in OS based on direct evidence only. These seven studies compared G alone to a different treatment arm. Direct comparisons detected statistically significant improvements in OS with GnP versus G (HR  = 0.72, \[95% CR 0.62–0.84\]), GCap versus G (HR  = 0.86, \[0.75–0.98\]), GE versus G (HR  = 0.82, \[0.69–0.97\]), FOLFIRINOX versus G (HR  = 0.57, \[0.45–0.72\]), GOx versus G (HR  = 0.87, \[0.76–0.98\]), and GS versus G (HR  = 0.80, \[0.66 to 0.96\]). These results can be seen in. Statistical heterogeneity (I²\>35%) was found only for the comparisons of GCis versus G (seven studies, I² = 64%) and GF versus G (three studies, I² = 62%) for OS. The direct comparisons for PFS with I² values are shown in. Through our Bayesian MTC, HR comparisons were made of OS and PFS to compare all the regimens simultaneously. The results of the MTC were similar to the results seen in direct pairwise comparisons. For OS, the results of the Bayesian MTC found that the probability that FOLFIRINOX was the best regimen was 83%, while it was 11% for GnP and 3% for GS and GE, respectively. For PFS, the Bayesian MTC found an 80% probability that FOLFIRINOX was the best regimen. shows the probabilities of each treatment regimen being the best in terms of OS. The probabilities for PFS can be seen in. The next best regimens according to the calculated probabilities are GnP, GE, and GS. The OS HR for FOLFIRINOX versus GS was 0.72 \[0.48–1.11\], FOLFIRINOX versus GnP was 0.79 \[0.50–1.24\], and FOLFIRINOX versus GE was 0.70 \[0.44–1.10\], where HRs are given with 95% CRs. The PFS HR for FOLFIRINOX versus GS was 0.78 \[0.47–1.40\], FOLFIRINOX versus GnP was 0.68 \[0.37–1.27\], and FOLFIRINOX versus GE was 0.61 \[0.33–1.15\]. Projected survivals were estimated comparing each regimen to G. The projected median OS ranged from 5.8 months for GCis and 9.9 months for FOLFIRINOX. The number needed to treat (NNT) at 6 months and 1 year relative to G have been shown in. The NNT at 1 year ranges from 5 for FOLFIRINOX to 146 for GCis. These estimates will be helpful in clinical decision-making and providing information to patients. Odds ratio (OR) comparisons were made of ObRR to compare all the regimens simultaneously. The Bayesian MTC found a 58% probability that FOLFIRINOX is the best regimen in terms of ObRR, while it was 33% and 8% for GnP and GS respectively. The ObRR HR \[95% CR\] for FOLFIRINOX versus GnP is 1.59 \[0.74–2.94\]. The probabilities that each treatment regimen is the best in terms of ObRR are shown in. The toxicity-related adverse events assessed in this study were febrile neutropenia and grade 3/4 fatigue, neuropathy, and diarrhea, as these are the most clinically relevant treatment related toxicities. ORs with 95% CRs were reported for each comparison with sufficient direct evidence available to make network estimates. Based on cross-trial comparisons, there was no obvious difference in toxicities for FOLFIRINOX and GnP. The raw numbers of toxicities from each included study can be found in. When comparing the direct pairwise comparisons to the results generated from the MTC, we found that the results are consistent. # Discussion ## Key Findings and Implications Based on the analysis of both the direct evidence and MTC, FOLFIRINOX had the highest probability of being the best regimen in terms of both OS (83%) and PFS (80%). In our study, selected comparisons of FOLFIRINOX with the regimens that had the next highest probabilities were also conducted. These results provide further evidence, albeit indirect, that FOLFIRINOX may be the most effective regimen in the treatment of advanced pancreatic cancer. Although this meta- analysis allows for network comparisons of FOLFIRINOX with other chemotherapy regimens, further large prospective trials with FOLFIRINOX and the other regimens, especially GnP, would ideally be performed to confirm these results. For over the past 15 years, gemcitabine monotherapy has been the standard of care in many countries for the treatment of metastatic pancreatic cancer based on its modest clinical efficacy. Although the tumor response rate and survival benefit of gemcitabine is modest, its favorable toxicity profile and ease of administration has led to its wide spread and continued use. Many studies have attempted to improve on the efficacy of gemcitabine by adding either another chemotherapeutic agent or a targeted agent. However, the vast majority of the phase III studies conducted in this setting have been remarkably negative with the exception of the addition of erlotinib and more recently, nab-paclitaxel. Although the gemcitabine and erlotinib study demonstrated a statistically significant overall survival benefit in favour of the combination, the modest improvement in survival and higher toxicity likely influenced a more broad adoption of this regimen. In addition, a population-based study conducted in 2012 examined the tolerance and effectiveness of FOLFIRINOX at three institutions. The median PFS and OS reported in this study were 7.5 and 13.5 months respectively. The PFS and OS from this study were actually higher than those from the pivotal randomized trial by Conroy et al. However, this may be attributed to the fact that the population-based study included patients with all stages of pancreatic cancer, while the Conroy study enrolled only those with metastatic disease. With respect to adverse events, the observed rate of febrile neutropenia in the population- based study was 4.9%, which is similar to the rate observed in the Conroy study (5.4%), which suggests that the results of the clinical trial may be generalizable to an uncontrolled setting. This population-based study concluded that FOLFIRINOX was clinically effective in the treatment of advanced pancreatic adenocarcinoma and that the toxicity profile of the regimen does not outweigh the benefits in terms of ObRR and survival. Although FOLFIRINOX demonstrates the best overall survival, progression-free survival, and objective response rate as per the large Phase III Trial, it is important to note that this regimen has a higher toxicity profile. When comparing the safety profiles of FOLFIRINOX and GnP from two separate clinical trials, the rate of febrile neutropenia in patients treated with FOLFIRINOX was 5.4%, while it was 3% in the GnP group. G-CSF was administered in 42.5% of patients receiving FOLFIRINOX and in 26% of patients receiving GnP. In addition, it is important to note that the FOLFIRINOX study excluded patients older than 75 years of age and those with an ECOG performance status of 2. Therefore, FOLFIRINOX may be more challenging to prescribe in elderly or frail patients and caution should be taken in these cases. Ongoing prospective population-based studies are being performed to assess the efficacy and safety of FOLFIRINOX outside of clinical trials, which will provide further real life experience of the regimen. In addition, no population-based studies conducted to evaluate the survival benefit and toxicity of GnP so further research should be done in order to compare FOLFIRINOX with GnP in clinical practice. ## Strengths and Limitations There are a number of strengths of the current MTC. For example, a comprehensive and robust search strategy was used, with data being extracted by two authors independently to ensure accuracy. Although MTC allow indirect comparisons to be made, these indirect estimates may be influenced by potential biases and uncertainties. Multiple-treatment comparison meta-analysis should be interpreted with caution and specifically, the underlying assumptions of homogeneity and consistency of studies across the network should be carefully scrutinized. In our study, heterogeneity between studies was indeed assessed and reported using I² values. Although some heterogeneity was noted in the comparisons of GCis versus G and GF versus G, all studies in included in the meta-analysis were comparable in terms of patient characteristics and outcomes in the G reference arm (median PFS  = 3–4 months, median OS  = 6–7 months). The HRs from direct pairwise comparisons and the MTC were also compared and found to be consistent. A limitation of our analysis was the small number of studies included which is a reflection of the landscape of the medical evidence. For many of the comparisons, data was extracted from only one trial so any biases or limitations from that study were more likely to affect the conclusions drawn from the MTC. Another limitation of this method is that it is based on published group data, rather than individual patient information. Individual patient data may allow for more patterns to be seen in terms of risk factors, however, it would still remain difficult to make strong inferences in such a complex network of treatments. Both the FOLFIRINOX and nab-paclitaxel trials included only those patients with metastatic pancreatic cancer in contrast to the other gemcitabine combination studies, which enrolled both metastatic and locally advanced pancreatic patients. One of the reasons behind this shift in patient profile of advanced pancreatic studies were the recommendations of a group of experts convened in 2009 by the National Cancer Institute in the United States based on the well described differences in survival between those with locally advanced and metastatic disease. Unfortunately, this difference in patient population across the trials included in our study could not be accounted for. However, given that the inclusion of locally advanced patients tends to magnify the overall and progression free survival, we do not expect this difference in the patients included in the studies to significantly influence our observed results. As RCTs directly comparing FOLFIRINOX and GnP, or other existing regimens are unlikely to be conducted in advanced pancreatic cancer in the future due to both commercial and scientific reasons, indirect comparisons such as ours may represent the best possible level of evidence as to which regimen is best. Such indirect evidence may still in fact be informative in terms of both clinical and policy decision-making. # Conclusions Our meta-analysis reviewed and analyzed the existing high-quality evidence for treating advanced pancreatic cancer in an MTC, which help synthesize evidence and may inform decision-making in the absence of direct pairwise comparisons. Based on our MTC, FOLFIRINOX appears to be the most effective regimen, however, direct pairwise comparisons are warranted to definitively address. Existing uncertainties of the relative effectiveness of FOLFIRINOX, as well as the potential toxicities and long-term effects suggest that further clinical trials and longitudinal studies are needed. # Supporting Information Chris Cameron and Dr. Sharon Straus for their guidance on Bayesian statistical analysis and comments on the manuscript. **Previous Presentations of the Manuscript**: Presented in part in the poster discussion at the Annual Meeting of the American Society of Clinical Oncology 2013, Chicago, USA. [^1]: Keya Shah has read the journal's policy and the authors of this manuscript have the following competing interests: Yoo-Joung Ko declared that he received research support and honoraria from Sanofi-Aventis, and Celgene; however, he has no stock ownership. The remaining authors have no competing interests to declare. This does not alter the authors' adherence to PLOS ONE policies on sharing data and materials. [^2]: Conceived and designed the experiments: KC YK DC. Performed the experiments: KC KL KS HL YK. Analyzed the data: KC KL KS. Wrote the paper: KC KS KL HL YK. Revised the manuscript: DC.

# Introduction Protein *N*-myristoylation is one of the predominant forms of lipid modification that occurs on eukaryotic and viral proteins. This modification is the attachment of myristic acid, a 14-carbon saturated fatty acid, to the N-terminal Gly of proteins. In general, myristic acid is cotranslationally attached to the N-terminal Gly residue after removal of the initiating Met. A stable amide bond links myristic acid irreversibly to the protein. In addition to cotranslational protein *N*-myristoylation, it is now established that posttranslational *N*-myristoylation can also occur on many caspase-cleavage products such as Bid, actin, gelsolin, and p21 activated kinases 2 (PAK2), in which proteolytic cleavage by caspase causes the exposure of an internal *N*-myristoylation motif. Both cotranslational and posttranslational *N*-myristoylation is catalyzed by *N*-myristoyltransferase, a member of the GCN5-related *N*-acetyltransferase superfamily of proteins. Many *N*-myristoylated proteins play critical roles in regulating cellular structure and function. These include proteins involved in a wide variety of cellular signal transduction pathways, such as guanine nucleotide binding proteins, protein kinases, phosphatases, E3-ubiquitin ligases, Ca²⁺ binding proteins, and cytoskeletal regulatory proteins. In addition to proteins related to cellular signal transduction pathways, recent studies have revealed that protein *N*-myristoylation occurs on many disease- associated proteins. In many cases, the functions of these *N*-myristoylated proteins are regulated by reversible membrane binding mediated by protein *N*-myristoylation. Thus, protein *N*-myristoylation has been recognized as a protein modification that occurs mainly on cytoplasmic proteins, and only very few integral membrane proteins have been demonstrated to be *N*-myristoylated so far. Previously, we showed that SAMM50, a mitochondrial outer membrane protein, is *N*-myristoylated, and this lipid modification is required for proper targeting of SAMM50 to mitochondria. SAMM50 is an integral membrane protein with a β-barrel structure, and it is a central component of the sorting and assembly machinery (SAM) necessary for the assembly of β-barrel proteins in the mitochondrial outer membrane. SAMM50 has been shown to interact with MIC19 (CHCHD3, MINOS3) and MIC25 (CHCHD6, CHCM1), *N*-myristoylated or putatively *N*-myristoylated mitochondrial intermembrane space protein, to form the mitochondrial intermembrane space bridging (MIB) complex that plays a critical role in the structure and function of mitochondria. However, the relative amount of these *N*-myristoylated proteins present in the cell, or the role of protein *N*-myristoylation in the function of these proteins has not been fully elucidated. These two proteins are components of mitochondrial contact site and cristae organizing system (MICOS) complex that plays critical role in cristae formation of the inner mitochondrial membrane and communication with the outer mitochondrial membrane. MICOS complex interacts with components of the SAM complex of the outer membrane to form MIB complex. In this study, we examined protein *N*-myristoylation occurring on four human mitochondrial proteins, SAMM50, TOMM40, MIC19, and MIC25. Among them, SAMM50, MIC19, and MIC25 are MIB components and TOMM40 is a major component of the translocase of the mitochondrial outer membrane (TOM complex) acts as a general import pore for most mitochondrial precursor proteins. As a result, it was revealed that all of these four proteins are *N*-myristoylated proteins. Analysis of intracellular localization indicated that protein *N*-myristoylation plays a critical role in mitochondrial targeting and membrane binding of two MIB components, SAMM50 and MIC19, but not those of TOMM40 and MIC25. Immunoprecipitation experiments using specific antibodies revealed MIC19, but not MIC25, to be a major *N*-myristoylated binding partner of SAMM50. Several molecules of MIC19 seemed to bind to each SAMM50 molecule. It was also revealed that 25 kDa MIC19 and its \~32 kDa isoform were ubiquitously expressed in mammalian cells. However, the expression of MIC25 differs depending on the cell type, and only very low level of expression was detected in human HeLa cells. Immunoprecipitation experiments using a stable transformant of MIC19 revealed that protein *N*-myristoylation of MIC19 plays a critical role in the interaction between MIC19 and SAMM50, as reported previously. Thus, protein *N*-myristoylation occurring on two mitochondrial MIB components, SAMM50 and MIC19, plays a critical role in mitochondrial targeting and protein-protein interaction between these two MIB components. # Materials and methods ## Materials Human cDNAs (Flexi ORF clones) were purchased from Promega (Madison, WI, USA). Transdirect insect cell, an insect cell-free protein synthesis system, was obtained from Shimadzu (Kyoto, Japan). Trident Membrane Protein Extraction Kit was from GeneTex (Irvine, CA, USA). A T7-Scribe standard RNA IVT kit was from CELLSCRIPT (Madison, WI, USA). \[³H\]leucine, \[³H\]myristic acid, and ECL prime Western blotting detection reagent were from GE Healthcare (Buckinghamshire, UK). ImmunoStar LD Western blotting detection reagent was from Wako Pure Chemical (Osaka, Japan). ENLIGHTENING was from PerkinElmer (Waltham, MA, USA). The dye terminator cycle sequencing kit, Lipofectamine LTX and Plus reagents, MitoTracker Red CMXRos, and Hoechst 33342 were from Life Technologies Corporation (Carlsbad, CA, USA). Anti-FLAG monoclonal antibody, anti-SAMM50 monoclonal antibody (WH0025813), anti-SAMM50 polyclonal antibody (HPA034537), anti-TOMM40 polyclonal antibody (HPA036231), anti-MIC19 polyclonal antibody (HPA042835), anti-MIC25 polyclonal antibody (HPA047673), and anti-rabbit IgG-FITC antibody were from Sigma (St. Louis, MO, USA). 13-tetradecynoic acid (Alk-Myr) was from Cayman (Ann Arbor, MI, USA). Azide TAMRA (Az-TAMRA) was from Click Chemistry Tools (Scottsdale, AZ, USA). Tris(2-carboxyethyl)phosphine hydrochloride (TCEP) and tris\[(1-benzyl-1*H*-1,2,3-triazol-4-yl)methyl\]amine (TBTA) were from Sigma (St. Louis, MO, USA). Protein G-HRP conjugate was from Bio-Rad (Hercules, CA, USA). X-ray film was from Eastman Kodak (Rochester, NY, USA). The other reagents used were from Wako Pure Chemical (Osaka, Japan), Daiichi Pure Chemicals (Tokyo, Japan) or Seikagaku Kogyo (Tokyo, Japan) and were of analytical or DNA grade. ## Prediction of protein *N*-myristoylation using prediction programs Two public WWW-server-based prediction programs for protein *N*-myristoylation, MYR Predictor (<http://mendel.imp.ac.at/myristate/SUPLpredictor.htm>) and Myristoylator (<http://www.expasy.org/tools/myristoylator/>), were used for the prediction of protein *N*-myristoylation. The entire amino acid sequences deduced from the nucleotide sequences of the ORFs were used as the query. ## Plasmid construction Nucleotide sequences of oligonucleotide primers used for plasmid construction are summarized in. Plasmid pTD1 (Shimadzu) was used for the insect cell-free protein synthesis system. Plasmids pTD1-SAMM50-, SAMM50-G2A-FLAG were constructed as described previously. Construction of the pTD1 plasmids including cDNAs encoding FLAG-tagged proteins of TOMM40 and TOMM40-G2A are summarized in. Construction of the pcDNA3 plasmids including cDNAs encoding FLAG-tagged or tag- free proteins of SAMM50, TOMM40, MIC19, MIC25 and their G2A mutants are summarized in. ## *In vitro* transcription reaction mRNAs encoding the cDNAs were prepared using a T7-scribe standard RNA IVT kit (CELLSCRIPT) in accordance with the manufacturer’s instructions. The synthesized mRNAs were purified by phenol-chloroform extraction and ethanol precipitation before use in the translation reaction. ## Cell-free protein synthesis The translation reaction was performed using an insect cell-free protein synthesis system (Shimadzu) in the presence of \[³H\]leucine or \[³H\]myristic acid as described previously. The mixture (composed of 6.2 μL insect cell lysate, 3.7 μL reaction buffer, 0.5 μL 4 mM methionine, 1.0 μL \[³H\]leucine \[1 μCi\] or 3.0 μL \[³H\]myristic acid \[20 μCi\], and 2 μL mRNA \[8 μg\]) was incubated at 26°C for 6 h. The translation products were then analyzed by SDS-PAGE and fluorography. ## Transfection of cells COS-1 (simian virus 40-transformed African green monkey kidney cell line, American Type Culture Collection) cells, HEK293T (a human embryonic kidney cell line) cells, HeLa cells, or HepG2 cells were maintained in Dulbecco’s modified Eagle’s medium (DMEM; Gibco BRL \[Palo Alto, CA, USA\]) supplemented with 10% fetal calf serum (FCS; Gibco BRL). Cells (2 × 10⁵) were plated onto 35-mm diameter dishes 1 day before transfection. pcDNA3 constructs (2 μg) containing cDNAs encoding FLAG-tagged or tag-free proteins were used to transfect the cells in each plate along with 2.5 μL Lipofectamine LTX and 2 μL Plus reagent in 1 mL serum-free medium. After incubation for 5 h at 37°C, the cells were re-fed with serum-containing medium and incubated again at 37°C for appropriate periods. Selection of cells stably expressing transfected cDNA was performed as described previously. ## Metabolic labeling of cells The metabolic labeling of cells with \[³H\]myristic acid or myristic acid analog (Alk-Myr) was performed as described previously. Cells (2 × 10⁵) were transfected with pcDNA3 constructs (2 μg) containing cDNAs as described above, and incubated at 37°C for 12 h. Then, they were washed once with 1 mL serum-free DMEM and incubated for 10 h at 37°C in 1 mL DMEM (+2% FCS) containing \[³H\]myristic acid (100 μCi/mL) or 25 μM Alk-Myr. Subsequently, the cells were washed three times with Dulbecco’s phosphate- buffered saline (DPBS), harvested and lysed with 200 μL of RIPA buffer (50 mM Tris-HCl (pH 7.5), 150 mM NaCl, 1% Nonidet P-40, 0.5% sodium deoxycholate, 0.1% SDS, protease inhibitors) on ice for 20 min. The radiolabeled samples were then analyzed by SDS-PAGE and fluorography. The samples labeled with Alk-Myr were reacted with Az-TAMRA by click chemistry, and then protein *N*-myristoylation was analyzed by in-gel fluorescence analysis. ## SDS-PAGE and fluorography The radiolabeled proteins were resolved by 12.5% SDS-PAGE, then the gel was fixed and soaked in ENLIGHTENING (PerkinElmer) for 20 min. Thereafter, the gel was dried under vacuum and exposed to X-ray film for an appropriate period. ## Cu(I)-catalyzed azide-alkyne cycloaddition (CuAAC) The cell lysates labeled with Alk-Myr (46 μL) were reacted with 4 μL of freshly premixed click chemistry reaction cocktail (1 μL Az-TAMRA \[5 mM\], 1 μL TCEP \[50 mM\], 1 μL TBTA \[5 mM\], and 1 μL CuSO₄•5H₂O \[50 mM\]) in a total reaction volume of 50 μL for 1 h at room temperature. After CuAAC, 500 μL of MeOH was added to the sample, then the samples were placed at -80°C overnight. After centrifugation at 15,000 rpm at 4°C for 30 min, the supernatant was removed. Thereafter, the pellet was washed with 500 μL of MeOH, and then dried in air. The samples were denatured by sonication in SDS-sample buffer, and then subjected to SDS-PAGE. In-gel fluorescence analysis of the gel obtained by SDS-PAGE was performed using a Typhoon FLA9500 (GE-Healthcare Bio- Sciences AB, Uppsala, Sweden). ## Western blotting Proteins were resolved by 12.5% SDS-PAGE and then transferred to an Immobilon-P transfer membrane. After blocking with non-fat milk, the membrane was probed with a primary antibody, as described previously. Immunoreactive proteins were detected specifically by incubation with protein G-HRP conjugate. The membrane was developed using ECL Prime or ImmunoStar LD western blotting detection reagent and detected using a MicroChemi Chemiluminescence Imaging System (Berthold Technologies, Bad Wildbad, Germany). The blots were quantified by densitometry using the software TotalLab Quant (TotalLab Limited, Newcastle upon Tyne, UK). ## Isolation of total membrane proteins and cytosolic proteins in cells Isolation of total membrane proteins and cytosolic proteins in intact or transfected COS-1 cells was performed using Trident Membrane Protein Extraction Kit (Gene Tex) according to the manufacturer’s protocol. The obtained total membrane protein fraction and cytosolic fraction were subjected to western blotting analysis. ## Immunofluorescence analysis and fluorescence microscopy Immunofluorescence analysis of transfected cells was performed 24 h after transfection. After staining with Hoechst 33342 and MitoTracker Red, the cells were washed with DPBS, fixed in 4% paraformaldehyde in DPBS for 15 min, and permeabilized with 0.1% Triton X-100 in DPBS for 10 min at room temperature, followed by washing with 0.1% gelatin in DPBS. The permeabilized cells were incubated with a specific antibody in DPBS for 1 h at room temperature. After washing with 0.1% gelatin in DPBS, the cells were incubated with anti-rabbit IgG-FITC antibody for 1 h at room temperature. After washing with 0.1% gelatin in DPBS, the cells were observed using a Leica AF7000 fluorescence microscope (Leica, Solmser, Germany). ## Immunoprecipitation Samples were immunoprecipitated with specific antibodies as described previously. ## Statistical analysis Statistical analysis was carried out using two-tailed *t* test (Microsoft Excel; Microsoft). The means of two distributions were considered significantly different if p \< 0.05. # Results ## Identification of protein *N*-myristoylation occurring on human SAMM50 and TOMM40 by cell-free and cellular metabolic labeling We showed previously that human SAMM50, a mitochondrial β-barrel membrane protein, is *N*-myristoylated, and protein *N*-myristoylation is required for the proper targeting of SAMM50 to mitochondria. When the N-terminal sequence of other human mitochondrial β-barrel membrane proteins was investigated, it was found that TOMM40, a major component of the translocase of the mitochondrial outer membrane (TOM complex), contained *N*-myristoylation motif at its N-terminus. In fact, both of two well-known *N*-myristoylation prediction programs (The MYR Predictor and Myristoylator) predicted TOMM40 to be an *N*-myristoylated protein. In addition, interspecies alignments of the N-terminal sequence of proteins revealed that the N-terminal *N*-myristoylation motif of TOMM40 is highly conserved among vertebrates, as is the case with SAMM50. To determine whether TOMM40 is *N*-myristoylated, cell-free and cellular metabolic labeling experiments were performed using cDNA coding for C-terminally FLAG-tagged TOMM40. For this analysis, non-*N*-myristoylated G2A mutant was constructed, in which Gly2 was replaced with Ala, and its susceptibility to protein *N*-myristoylation was compared with that of wild-type protein. As shown in, left panel, efficient protein expression (\[³H\]leucine incorporation) in the insect cell-free protein synthesis system was observed for the wild-type and G2A mutant forms of TOMM40-FLAG, as was the case with those of SAMM50-FLAG. Metabolic labeling with \[³H\]myristic acid revealed that efficient incorporation of \[³H\]myristic acid was observed in wild-type TOMM40-FLAG and SAMM50-FLAG, but incorporation was completely inhibited by replacing Gly2 with Ala (, right panel). In the case of cellular metabolic labeling experiments in COS-1 cells using a bioorthogonal myristic acid analog (Alk-Myr) followed by detection with click chemistry, results were obtained similar to those from the cell-free protein synthesis system, as shown in. These results clearly showed that TOMM40 and SAMM50 are *N*-myristoylated in *in vitro* and *in vivo* expression systems. ## Protein *N*-myristoylation of SAMM50 but not of TOMM40 is required for proper targeting to mitochondria In order to determine the physiological role of protein *N*-myristoylation occurring on human SAMM50 and TOMM40, the role of protein *N*-myristoylation in the intracellular localization of these two proteins was investigated by immunofluorescence staining of COS-1 (simian virus 40-transformed African green monkey kidney cell line) cells transfected with these cDNAs. Both of SAMM50 and TOMM40 are β-barrel proteins and contain a β-signal near the C-terminus that is required for β-barrel precursor recognition by the SAM complex. Thus, it is probable that the epitope-tagging at the C-terminus might affect the mitochondrial targeting of these proteins. Therefore, we used cDNAs encoding tag-free SAMM50 and TOMM40 to transfect COS-1 cells to investigate the intracellular localization of these two proteins. As shown in, a striking difference in the role of protein *N*-myristoylation in mitochondrial targeting was observed between SAMM50 and TOMM40. Immunofluorescence staining with the anti-SAMM50 antibody coupled with MitoTracker staining revealed that *N*-myristoylated SAMM50 exclusively localized to the mitochondria, whereas the non-myristoylated G2A mutant localized mainly to the cytoplasm. In contrast, in the case of TOMM40, both the wild-type and G2A mutant were found to localize exclusively to mitochondria. Thus, *N*-myristoylation of TOMM40 did not play a critical role in the mitochondrial targeting of TOMM40. In these experiments, endogenous SAMM50 and TOMM40 in control non-transfected COS-1 cells were not detected under the same experimental condition used for the analysis of the transfected COS-1 cells as shown in the lower panels of. In order to further clarify the role of protein *N*-myristoylation in the intracellular localization of SAMM50 and TOMM40, the role of protein *N*-myristoylation in the membrane binding of these two proteins was investigated by subcellular fractionation analysis. For this analysis, membrane protein extraction kit (Trident) was used to separate the total cell extracts into cytosolic and total membrane fractions. As shown in, endogenous SAMM50 and TOMM40 expressed in control non-transfected COS-1 cells were detected exclusively in total membrane fraction, whereas GAPDH, a cytosolic marker protein, was detected exclusively in cytosolic fraction. These data clearly indicated that the membrane proteins were successfully separated from the cytosolic proteins by using membrane protein extraction kit. As shown in, a striking difference in the role of protein *N*-myristoylation in membrane binding was observed between SAMM50 and TOMM40. Western blotting analysis using anti-SAMM50 antibody revealed that most of the overexpressed *N*-myristoylated SAMM50 localized to the membrane, whereas more than half of the non- myristoylated G2A mutant localized to the cytoplasm. In contrast, as for TOMM40, both the wild-type and G2A mutant were found to localize exclusively to the membrane. Thus, *N*-myristoylation of SAMM50 but not of TOMM40 plays positive role in the binding of protein to the membrane. ## Endogenous SAMM50 and TOMM40 expressed in mammalian cells is *N*-myristoylated To determine whether protein *N*-myristoylation was observed on endogenous SAMM50 and TOMM40 expressed in mammalian cells, metabolic labeling of endogenous proteins expressed in COS-1 cells with \[³H\]myristic acid followed by immunoprecipitation with a specific antibody against these two proteins was performed. As shown in , lane 2 and 3, \[³H\]myristic acid-labeled protein bands with the expected molecular masses of SAMM50 (50 kDa) and TOMM40 (40 kDa) were detected by SDS-PAGE and fluorography. When western blotting analysis of total cell lysates of COS-1 cells were performed, a specific protein bands with molecular masses of 50 kDa and 40 kDa were detected using anti-SAMM50 and anti-TOMM40 antibodies, respectively (lanes 1 and 3). In addition, the molecular masses of endogenous SAMM50 and TOMM40 were exactly the same as those observed with tag-free SAMM50 and TOMM40 overexpressed in COS-1 cells (lanes 2 and 4), suggesting that the \[³H\]myristic acid-labeled protein bands with molecular masses of 50 kDa and 40 kDa were endogenous *N*-myristoylated SAMM50 and TOMM40, respectively. ## Identification of *N*-myristoylated proteins specifically interacting with *N*-myristoylated SAMM50 expressed in mammalian cells As shown in, lane 2, several \[³H\]myristic acid-labeled protein bands with molecular masses smaller than that of SAMM50 (50 kDa) were detected in the immunoprecipitated sample obtained using the anti-SAMM50 antibody. No such extra \[³H\]myristic acid-labeled protein bands were observed in the same experiment performed with the anti-TOMM40 antibody (lane 3). Surprisingly, the level of \[³H\]myristic acid incorporation into a 25 kDa protein was much higher than that into SAMM50. Because it was reported that MIC19, an *N*-myristoylated mitochondrial protein with a molecular mass of 25 kDa, specifically interacted with SAMM50, we next studied whether the 25 kDa \[³H\]myristic acid-labeled protein band was MIC19 or not. When western blotting analysis of total cell lysates of COS-1 cells was performed, a specific protein band with a molecular mass of 25 kDa and a faint \~32 kDa protein band were detected using an anti-MIC19 antibody (lane 5). In addition, the molecular mass of major endogenous MIC19 was exactly the same as that observed with tag-free MIC19 overexpressed in COS-1 cells (lane 6), suggesting that the \[³H\]myristic acid-labeled protein band with a molecular mass of 25 kDa was endogenous *N*-myristoylated MIC19. When COS-1 cells were labeled with \[³H\]myristic acid and then immunoprecipitated with anti-MIC19 antibody, a specific protein band with a molecular mass of 25 kDa and a faint \~32 kDa protein band were detected by SDS-PAGE and fluorography, as shown in, lane 6. However, the \~32 kDa \[³H\]myristic acid-labeled protein band was not generated when a cDNA coding for tag-free MIC19 was transfected into COS-1 cells (lane 3). These results suggested that the \~32 kDa \[³H\]myristic acid-labeled protein band was not a translation product of the cDNA coding for 25 kDa MIC19 but might be an isoform of MIC19. The fact that both the 25 kDa and \~32 kDa \[³H\]myristic acid- labeled protein bands were detected in the immunoprecipitated sample obtained using the anti-SAMM50 antibody indicated that both of these MIC19 isoforms were associated with SAMM50 (lane 4). Thus, MIC19 and its \~32 kDa isoform were found to be major *N*-myristoylated binding partners of SAMM50. ## Identification and characterization of protein *N*-myristoylation occurring on human MIC25 In addition to MIC19, it was recently demonstrated that MIC25, a member of CHCH domain-containing protein family that has been predicted to be *N*-myristoylated, interacted with SAMM50. Therefore, we next studied whether MIC25 was contained in the immunoprecipitated sample obtained using the anti- SAMM50 antibody. Because protein *N*-myristoylation of MIC25 has not been verified experimentally, we first studied the protein *N*-myristoylation occurring on MIC25. For this analysis, MIC19 was used as a control *N*-myristoylated protein. As shown in, interspecies alignments of the N-terminal sequence of the proteins revealed that the N-terminal *N*-myristoylation motif of MIC25 was highly conserved among vertebrates, as was the case with MIC19. In addition, both *N*-myristoylation prediction programs (The MYR Predictor and Myristoylator) predicted MIC25 to be an *N*-myristoylated protein. To determine whether MIC25 is *N*-myristoylated, cellular metabolic labeling experiments were performed using cDNAs coding for wild-type and G2A mutant of C-terminally FLAG-tagged MIC25. As shown in, left panel, efficient expression of proteins with the predicted molecular mass (26 kDa plus 1 kDa FLAG) in COS-1 cells were observed with wild-type and G2A mutant of MIC25-FLAG, as was the case with those of MIC19-FLAG determined by western blotting analysis. Metabolic labeling experiments in COS-1 cells using bioorthogonal myristic acid analog (Alk-Myr) followed by detection with click chemistry revealed that efficient protein *N*-myristoylation was observed in wild-type MIC19-FLAG and MIC25-FLAG, but protein *N*-myristoylation was completely inhibited by replacing Gly2 with Ala, as shown in, right panel. These results clearly showed that MIC19 and MIC25 are efficiently *N*-myristoylated in transfected mammalian cells. ## Protein *N*-myristoylation of MIC19 but not of MIC25 is required for proper targeting to mitochondria To determine the physiological role of protein *N*-myristoylation occurring on human MIC25, the role of protein *N*-myristoylation in intracellular localization was investigated by immunofluorescence staining of COS-1 cells transfected with cDNA encoding tag-free MIC25. For this analysis, MIC19 in which protein *N*-myristoylation has been shown to play a critical role in mitochondrial targeting was used as a control. As shown in, a striking difference in the role of protein *N*-myristoylation in mitochondrial targeting was observed between MIC19 and MIC25. Immunofluorescence staining with the anti- MIC19 antibody coupled with MitoTracker staining revealed that *N*-myristoylated MIC19 exclusively localized to mitochondria, whereas the non-myristoylated G2A mutant localized mainly to the cytoplasm. In contrast, in the case of MIC25, both the wild-type and G2A mutant were found to localize mainly to cytoplasm and only a small amount of protein was found in mitochondria. Thus, *N*-myristoylation of MIC25 did not play a critical role in the mitochondrial targeting of this protein. In these experiments, endogenous MIC19 and MIC25 in control non-transfected COS-1 cells were not detected under the same experimental condition used for the analysis of the transfected COS-1 cells as shown in the lower panels of. In order to further clarify the role of protein *N*-myristoylation in the intracellular localization of MIC19 and MIC25, the role of protein *N*-myristoylation in the membrane binding of these two proteins was investigated by subcellular fractionation analysis using membrane protein extraction kit as performed in. As shown in, endogenous MIC19 and MIC25 expressed in control non-transfected COS-1 cells were detected exclusively in total membrane fraction, whereas GAPDH, a cytosolic marker protein, was detected exclusively in cytosolic fraction. In the case of detection of endogenous MIC25, the expression could not be detected by using conventional western blotting detection reagent (data not shown), and the MIC25 protein band was detected only when highly sensitive western blotting detection reagent (ImmunoStar LD) was used. These data indicated that the membrane proteins were successfully separated from the cytosolic proteins by using membrane protein extraction kit. As shown in, a striking difference in the role of protein *N*-myristoylation in membrane binding was observed between MIC19 and MIC25. Western blotting analysis using anti-MIC19 antibody revealed that most of the overexpressed *N*-myristoylated MIC19 localized to the membrane, whereas most of the non- myristoylated G2A mutant localized to the cytoplasm. In contrast, as for MIC25, many of the wild-type and G2A mutant were found to localize to the cytoplasm, and only a small portion of the protein was detected in the membrane. Thus, *N*-myristoylation of MIC19 but not of MIC25 plays critical role in the binding of protein to the membrane. ## Endogenous MIC25 expressed in mammalian cells is *N*-myristoylated To determine whether endogenous MIC25 expressed in mammalian cells is *N*-myristoylated, COS-1 cells were labeled with \[³H\]myristic acid and then immunoprecipitation was performed using the anti-MIC25 antibody. As a result, a faint \[³H\]myristic acid-labeled protein band with a molecular mass of 26 kDa was detected by SDS-PAGE and fluorography, as shown in, lane 2. The molecular mass of the \[³H\]myristic acid-labeled endogenous MIC25 was exactly the same as that observed with \[³H\]myristic acid-labeled tag-free MIC25 overexpressed in COS-1 cells, suggesting that the \[³H\]myristic acid-labeled 26 kDa protein was endogenous *N*-myristoylated MIC25 (lanes 2, 5 and 6). Because the strength of the \[³H\]myristic acid-labeled protein band of MIC25 was so weak compared with that of MIC19, it is probable that the level of protein expression of endogenous MIC25 was very low. To confirm this, the level of expression of endogenous MIC25 was compared with that of MIC19 by western blotting analysis using MIC19-FLAG, tag-free MIC19, MIC25-FLAG, and tag-free MIC25 expressed in COS-1 cells as standard proteins. As shown in, upper panels, lanes 1 and 4, similar amounts of MIC19-FLAG and MIC25-FLAG were expressed in transfected COS-1 cells. Whereas tag-free MIC19 and MIC25 were expressed at similar protein expression levels compared with FLAG-tagged MIC19 and MIC25, as determined using anti-MIC19 or anti-MIC25 antibodies, respectively (lower panels, lanes 1, 2, 4, and 5). In the case of endogenous MIC19 and MIC25, only low levels of protein expression of MIC25 were observed in total cell lysates of COS-1 cells compared with that of MIC19 (lower panels, lanes 3 and 6). We next compared the expression levels of endogenous MIC19 and MIC25 in 4 different mammalian cell lines (COS-1, HEK293T, HeLa, and HepG2). As a result, a significant difference in protein expression was observed between MIC19 and MIC25, as shown in. The 25 kDa MIC19 protein was expressed efficiently in all four cell lines. As for the \~32 kDa MIC19 isoform, relatively low levels of expression were observed in three human cell lines (HEK293T, HeLa, and HepG2) compared with monkey COS-1 cells (left panels). Concerning MIC25, a difference in protein expression levels was observed among the 4 cell lines; low expression levels were detected equally in COS-1, HEK293T, and HepG2 cells, whereas only very low expression levels were detected in HeLa cells, as shown in, right panels. ## Protein *N*-myristoylation of MIC19 is required for association of MIC19 with SAMM50 We next studied the role of protein *N*-myristoylation on the protein-protein interaction between SAMM50 and MIC19. As shown in, lane 1, association of SAMM50 with MIC19 was not observed in COS-1 cells transiently transfected with SAMM50 cDNA. Therefore, we tried to establish transformants stably expressing wild-type and G2A mutant of C-terminally FLAG-tagged SAMM50 or MIC19. As a result, stable transformants of MIC19-FLAG were successfully established, as shown in. However, as for wild-type and G2A mutant of SAMM50-FLAG, we tried more than five times but could not obtain COS-1 cells stably expressing these proteins. When the association of stably expressed MIC19-FLAG or MIC19-G2A-FLAG with endogenously expressed SAMM50 in the two stable transformants was studied by immunoprecipitation using anti-FLAG antibody, a striking difference was observed between two transformants, as shown in. In both transformants, FLAG-tagged MIC19 protein was efficiently immunoprecipitated with an anti-FLAG antibody (lanes 6 and 9). However, endogenous SAMM50 was detected only in the immunoprecipitated sample of the stable transformant expressing wild-type MIC19-FLAG (lane 6). Quantitative analysis of the amount of SAMM50 bound to MIC19-FLAG or MIC19-G2A-FLAG confirmed the efficient binding of SAMM50 to MIC19-FLAG but not to MIC19-G2A-FLAG, as shown in. These results clearly indicated that protein *N*-myristoylation of MIC19 is required for the association of MIC19 with SAMM50. # Discussion Protein *N*-myristoylation has been recognized as a protein modification that occurs mainly on cytoplasmic proteins. In eukaryotic cellular proteins, only very few integral membrane proteins have been demonstrated to be *N*-myristoylated. In this study, we first characterized protein *N*-myristoylation occurring on two human mitochondrial membrane proteins, SAMM50 and TOMM40. SAMM50 and TOMM40 are β-barrel proteins that reside within the outer membrane of mitochondria. SAMM50 is a central component of the SAM complex necessary for the assembly of β-barrel proteins in the mitochondrial outer membrane. TOMM40 is the protein conducting channel of the translocase of the outer mitochondrial membrane (TOM) acts as a general import pore for most mitochondrial precursor proteins. As for mitochondrial localization of TOMM40 and SAMM50, specific mitochondrial targeting signal has not been determined. Both of these proteins are β-barrel proteins and contain a β-signal that is required for β-barrel precursor recognition by the SAM complex. However, it was reported that the β-signal is not required for mitochondrial targeting of yeast β-barrel proteins. Thus, so far, the mechanisms for specific mitochondrial targeting of TOMM40 and SAMM50 remains to be elucidated. Metabolic labeling experiments using transfected mammalian cells performed in this study revealed that both SAMM50 and TOMM40 are *N*-myristoylated. It should be noted that N-terminal *N*-myristoylation motifs were observed exclusively in vertebrates. Therefore, SAM50 and TOM40 in yeast are not subjected to protein *N*-myristoylation. Analysis of intracellular localization in wild-type and a non-myristoylated G2A mutant of human proteins expressed in mammalian cells by immunofluorescence microscopic analysis revealed that protein *N*-myristoylation plays critical role in the mitochondrial targeting of SAMM50, but not that of TOMM40. In accordance with these observations, subcellular fractionation analysis revealed that protein *N*-myristoylation of SAMM50 but not of TOMM40 plays positive role in the binding of protein to the membrane. β-barrel proteins do not enter the outer mitochondrial membrane from the cytosolic side but rather are transferred via the TOM complex into the intermembrane space and are subsequently directed to the outer membrane. Similar to other mitochondrial precursors, β-barrel proteins are first recognized by TOM receptors before they are imported across the outer membrane with the help of the TOM complex. Therefore, it is probable that protein *N*-myristoylation of SAMM50 positively affected the interaction of SAMM50 with TOM receptors. In fact, non- myristoylated G2A mutant of SAMM50 did not localize to a membranous compartment but to the cytoplasm (Figs). In contrast to SAMM50, protein *N*-myristoylation of TOMM40 did not affect mitochondrial targeting and membrane binding of TOMM40. The mechanism for the difference in the role of protein *N*-myristoylation on the mitochondrial targeting and membrane binding of these two proteins is not clear. When protein *N*-myristoylation occurring on endogenous proteins expressed in COS-1 cells was analyzed by immunoprecipitation using specific antibodies, a striking difference was observed in the immunoprecipitated proteins obtained using anti-SAMM50 and anti-TOMM40 antibodies. In the case of TOMM40, \[³H\]myristic acid-labeled 40 kDa single protein band was detected in the fluorogram. Western blotting analysis further confirmed the presence of TOMM40 in the immunoprecipitated sample obtained using the anti-TOMM40 antibody (, lane 7). As for SAMM50, in addition to a \[³H\]myristic acid- labeled 50 kDa SAMM50 band, multiple protein bands were detected in the fluorogram. Comparison of the molecular mass of these protein bands with those of the tag-free MIC19, and MIC25 protein expressed in COS-1 cells, or endogenous \[³H\]myristic acid-labeled proteins in COS-1 cells immunoprecipitated using anti-MIC19 or anti-MIC25 antibodies clearly indicated that major *N*-myristoylated binding partner of SAMM50 is MIC19 and its \~32 kDa isoform, but not MIC25 (Figs,). So far, the isoform of MIC19 has not been reported. Thus, this is the first report indicating the presence of a MIC19 isoform with higher molecular mass. Of note, much higher incorporation of \[³H\]myristic acid was detected in the 25 kDa MIC19 protein band as compared with the 50 kDa SAMM50 band. Because protein *N*-myristoylation occurs at a single N-terminal Gly residue in each *N*-myristoylated protein, it was suggested that several molecules of *N*-myristoylated MIC19 bound to each SAMM50 molecule. Coiled-coil-helix-coiled-coil-helix domain (CHCHD)-containing proteins are nucleus-encoded small mitochondrial proteins with important functions. Nine members have been identified in this protein family. All CHCHD proteins have at least one functional CHCHD, which is stabilized by two pairs of disulfide bonds between two helices. CHCHD proteins have various important pathophysiological roles in mitochondria and other key cellular processes. MIC19 and MIC25 are part of the CHCHD protein family. Phylogenic analysis indicated that human MIC19 and MIC25 resulted from a gene duplication at the root of the vertebrates. Thus, they are both orthologous to MIC19 in yeast and likely the result of the whole genome duplication at the root of the vertebrates, a pattern that has been observed for mitochondrial proteins. Concerning mitochondrial localization of MIC19 and MIC25, neither canonical cleavable N-terminal mitochondrial targeting signal (presequence) nor other specific mitochondrial targeting signal has been determined. However, as for MIC19, it was reported that both protein *N*-myrstoylation and CHCHD located near the C-terminus were essential for proper mitochondrial targeting of this protein. In this case, however, the mechanisms by which protein *N*-myristoylation and CHCHD direct the mitochondrial targeting have not been elucidated. MIC19 is an *N*-myristoylated inner membrane-bound protein. MIC19 interacts with the peripheral surface of the mitochondrial contact site and cristae organizing system (MICOS) complex, facing the intermembrane space, where its presence is necessary for cristae formation and communication with the outer mitochondrial membrane. As for MIC25, it was predicted to be *N*-myristoylated, however, direct biochemical evidence for the protein *N*-myristoylation has not been demonstrated. As shown in, the *N*-myristoylation motif of MIC25 is highly conserved among vertebrates. In the present study, we showed that exogenously and endogenously expressed MIC25 is efficiently *N*-myristoylated. Interestingly, however, most of the *N*-myristoylated MIC25 overexpressed in COS-1 cells was detected in the cytoplasm but not in mitochondria (Figs). Thus, protein *N*-myristoylation does not play critical role in the mitochondrial targeting or membrane binding of MIC25. The mechanism for the difference in the role of protein *N*-myristoylation on the mitochondrial targeting and membrane binding of MIC19 and MIC25 is not clear. The very large MICOS complex is located in the inner membrane and connects with proteins in the outer membrane. To date, eight MICOS subunits in mammals (MIC10, MIC13, MIC14, MIC19, MIC23, MIC25, MIC27, and MIC60) have been identified. Most of the MICOS subunits are integral membrane proteins in the mitochondrial inner membrane, however, MIC19 and MIC25 are peripheral membrane proteins. MIC19 and MIC25 are both CHCHD proteins and are characterized by the presence of CX₉C motifs with two cysteines that stabilize coiled-coil-helix domains by forming disulphide bonds. These physicochemical properties are associated with scaffold functions, reflecting the structural and protein- protein interaction roles of these proteins. MIC19 is a 26 kDa protein associated with the inner membrane, interacting with MIC60 through the coiled coil helix motif in the C-terminal domain, and with SAMM50 through the N-terminal domain. Therefore, MIC19 bridges MIC60 and SAMM50, contributing to the formation of contact sites. This structure is called the MIB complex, and it is crucial for the maintenance of cristae and the assembly of respiratory chain complexes. The fact that MIC19 interacts with SAMM50 through the *N*-myristoylated N-terminal domain and that SAMM50 exposes an N-terminal polypeptide transport-associated domain into the intermembrane space suggested that N-terminal *N*-myristoylation of both MIC19 and SAMM50 might be involved in the specific interaction of MIC19 and SAMM50. Concerning the role of protein *N*-myristoylation of MIC19 in the interaction of MIC19 and SAMM50, it was previously reported that MIC19 transiently expressed in HEK293 cells bound to endogenous SAMM50 in an *N*-myristoylation dependent manner. On the contrary, in the present study it was shown that MIC19 transiently expressed in COS-1 cells could not bind to endogenous SAMM50, as shown in, lane 3. The reason for the discrepancy between these observations was not clear. However, in this study, MIC19-FLAG stably expressed in COS-1 cells with an expression level similar to that of the endogenous MIC19 was found to bind to endogenous SAMM50 in an *N*-myristoylation-dependent manner, suggesting that the stoichiometry of the expressed proteins in the cells might affect the interaction of MIC19 with SAMM50. In order to determine the intracellular localization of stably expressed MIC19-FLAG and MIC19-G2A-FLAG, we performed immunofluorescence analysis using anti-FLAG antibody. As a result, different from the experimental results obtained with transiently expressed MIC19 and MIC19-G2A, both of stably expressed MIC19-FLAG and MIC19-G2A-FLAG were found to localize specifically to mitochondria as shown in. Thus, it is probable that, in this experimental system, the lack of *N*-myristoylation did not affect the mitochondrial localization but affected the protein-protein interaction occurring in the mitochondria. Thus, it was confirmed that protein *N*-myristoylation of MIC19 plays a critical role in the MIC19-SAMM50 interaction. It is quite probable that *N*-myristoylation of SAMM50 also plays an important role in these interactions. Unfortunately, a stable transformant of SAMM50 has not been established even after multiple attempts. Thus, the role of protein *N*-myristoylation of SAMM50 on the SAMM50-MIC19 interaction remains to be elucidated. MIC25 is a 26 kDa protein also associated with the mitochondrial inner membrane. It has been poorly characterized thus far, but a recent immunoprecipitation study revealed the physical interaction of MIC25 with SAMM50 and MIC60 to sustain cristae structure. However, another study suggested that the most important subunits of the MIB complex in human mitochondria are Mic60, Mic19, and SAMM50, and that MIC25 is a peripheral subunit of the MICOS complex because its depletion did not affect cristae morphology. In the present study, immunoprecipitation experiments using specific antibodies revealed that MIC19, but not MIC25, was a major *N*-myristoylated binding partner of SAMM50. It was also revealed that 25 kDa MIC19 and its \~32 kDa isoform were ubiquitously expressed in mammalian cells. However, the expression of MIC25 differed depending on the cell-type, and only a very low level of expression was detected in human HeLa cells. In fact, it was reported that MIC25 does not play a critical role in the formation and stability of cristae in HeLa cells. Thus, although MIC19 was established as a crucial component of MICOS and MIB, MIC25 seems to fulfill a more peripheral function, as reported previously. Schematic representation of the intracellular localization and protein-protein interactions of SAMM50, TOMM40, and MIC19 is shown in. Protein *N*-myristoylation has been recognized as a protein modification that occurs mainly on cytoplasmic proteins, and the functions of these *N*-myristoylated proteins are often regulated by reversible membrane binding to the plasma membrane mediated by protein *N*-myristoylation as observed in the case of src family tyrosine kinases and G protein α subunits. In the present study, we showed that four mitochondrial proteins, two integral outer membrane proteins (SAMM50, TOMM40) and two mitochondrial intermembrane space proteins (MIC19, MIC25), were *N*-myristoylated. In addition, protein *N*-myristoylation was found to play a critical role in the mitochondrial targeting of two MIB components (SAMM50 and MIC19), and the protein-protein interaction between MIC19 and SAMM50. Thus, the role of protein *N*-myristoylation in the intracellular targeting and physiological function of *N*-myristoylated proteins seems to be much more diverse than reported previously. # Supporting information [^1]: The authors have declared that no competing interests exist.

# Introduction Appropriate prescribing of antimicrobials is crucial for individual patients to increase the chance of therapeutic success and to prevent spread of antimicrobial resistance (AMR) on a broader scale. For this reason, governments and healthcare institutions have developed and implemented antimicrobial stewardship (AMS) programs to improve appropriate prescribing. Local antimicrobial guidelines help physicians to prescribe appropriate antimicrobial therapy. However, guidelines change and increasing complexity of care requires easily accessible and frequently updated guidelines. Printed booklets and digital documents may not be sufficient for this purpose. In the age of information technology (IT), many processes within the healthcare system have been digitized or automated and IT has become an intrinsic part of modern medicine. IT interventions such as electronic health records (EHR), clinical decision support systems (CDSS), and antimicrobial drug approval systems increase guideline adherent prescribing. Such tools assist in a more timely intravenous to oral switch, and decrease overall antimicrobial consumption. Nevertheless, appropriate prescribing of antimicrobials can still be improved. AMS is important for general practice and hospitals, but prevalence of antimicrobial resistant microorganisms is the highest in hospitals, even in countries with overall low resistance rates. Furthermore, reserve antimicrobials are mainly used in hospitals. With the introduction of smartphones, applications (apps) can be accessed without the necessity of a non-mobile desktop and can simultaneously provide a framework to integrate CDSSs. Besides accessibility, apps offer several other advantages such as the most up to date content, short start-up time and administrator privileges to inform users of specific updates. In Europe and North America the number of unique mobile phone subscribers was respectively 85% and 84% at the end of 2017. In the same year over 318.000 mobile health (mHealth) apps were available in app stores. The majority of healthcare workers utilizes mHealth apps (77.2%) on a regular basis in the United States. Although smartphone apps have high potential for becoming a key component of AMS programs, user experience, uptake and effect on prescription of antimicrobials have not been systematically reviewed to the best of our knowledge. The aim of this study was to systematically review antimicrobial stewardship apps and their impact on prescribing by physicians treating in- hospital patients # Methods This systematic review was performed in accordance with the guidelines of Preferred Reporting Items for Systematic Reviews and Meta-Analyses: The PRISMA Statement.. ## Eligibility criteria Studies which focused on AMS app use by physicians treating in-hospital patients were assessed for eligibility. Studies focusing on smartphone or tablet apps and antimicrobial therapy published from January 2008 until February 28^th 2019 were included. The year 2008 was chosen since the two most popular app stores, App Store (iOS) and the precursor of Google Play (Android) were launched that year. We included randomized controlled trials (RCT), non-RCTs, time series, before-after studies and qualitative studies. Excluded were studies solely considering antimicrobial prophylaxis, only including patients younger than 18 years of age, case reports, conference papers, editorials, letters to editor and reviews or meta-analysis. Language was no exclusion criterion. We excluded studies which only described app use in the general practice and outpatient setting. In these settings, prevalence and severity of infectious diseases, available antimicrobials and routes of administration differ significantly compared to an in-hospital setting. ## Search strategy and review design EMBASE, MEDLINE (Ovid), Cochrane Central, Web of Science and Google Scholar databases were searched for relevant quantitative and qualitative studies published before February 28^th 2019. The search strategy was developed together with an information specialist and specified for each database. Search terms included “antimicrobial”, “anti-infective agent”, “prescription”, “application” and “mobile phone”. Search results were imported to Endnote (Clarivate Analytics, Philadelphia, PA, USA). After removal of duplicate studies, two investigators (RH and DF) independently screened all articles on title and abstract. Articles were included for full text review if selected by either investigator. In case of doubt, articles were included for full text review. Both investigators independently assessed full texts for eligibility and extracted data. Disagreements were resolved in discussion with a third investigator (AV). ## Study outcome Primary study outcomes are process indicators such as number of downloads, average monthly use and guidelines assessed, adherence to guidelines and user experience. Secondary study outcomes are drug consumption, susceptibility and costs. ## Data analysis Data was extracted using standardized forms. The quality of included studies was independently assessed by two investigators (RH and DF). Disagreements were resolved by discussion with a third investigator (AV). To assess the five different study designs we used five risk of bias assessment tools. Due to large variations in study design and outcome parameters, study outcome could not be pooled and used for meta-analysis. # Results ## Study characteristics Thirteen studies met the eligibility criteria and were evaluated in this systematic review. Primary outcomes were process indicators such as downloads, average app use and time spent per guideline, evaluated in seven studies. Changes in adherence to guidelines and user experience were analysed in four and five studies, respectively. Antimicrobial consumption was evaluated in two studies. In ten of the thirteen studies the app was custom built for the study. ## Study quality Emphasis on app dissemination and use, impact of app use and user experience resulted in quantitative, qualitative and mixed methods study designs such as: uncontrolled before-after (five), controlled before-after (two), interrupted time series (one), cross-sectional (six) and qualitative studies (three). Quality was evaluated with the corresponding tools. Study designs varied greatly due to different metrics studied, e.g. app dissemination and use, impact of app use and user experience. Overall, methodological study quality was considered low to moderate for the before-after, interrupted time series and cross- sectional studies and moderate to high for qualitative studies. In most studies participants and outcome assessors were not blinded. Furthermore, outcomes were usually measured at one time point. Finally, the amount of eligible physicians enrolled was generally unclear as well as the loss to follow-up because information on user retention was lacking. ## Process indicators Seven observational studies reported analytics of app use. Five studies evaluated total number of downloads. All registered an increase of downloads during their study periods (3–14 months). A study that assessed an app containing all hospital antimicrobial guidelines recorded an increase in average monthly app accessions during a 29-month period. In contrast, the monthly app accessions decreased over 3 months in the study of Yoon *et al*. which assessed an app containing only two guidelines, the treatment of community-acquired pneumonia and urinary tract infection. Additionally, clinicians accessed the guidelines more frequently by app than by desktop in all three studies evaluating number of accessions. One study reported a decrease in time spent per individual guideline over the course of the study possibly demonstrating familiarization with the app. The most frequently accessed guidelines were those outlining treatment for respiratory, skin & soft tissue and genitourinary infections. ## Adherence to guidelines Four studies analysed whether empirical prescribing of antimicrobials was according to guidelines, such as choice of drug, dose, interval and route of administration. The study of Charani *et al*. in which all antimicrobial guidelines were implemented in the app reported a significant increase in adherence to guidelines in surgical wards and a non-significant increase in general medicine wards. This increase persisted after six and twelve months. Two studies reported an significant increase in adherence to guidelines for pyelonephritis and uncomplicated diarrheal diseases respectively during the intervention period. One study showed increased adherence to the community- acquired pneumonia guideline in one hospital, but not in the other. Also, no change in adherence to the UTI guideline was shown in any of the three participating hospitals. Documentation of stop/review dates and indication for starting antimicrobials in the medical charts was evaluated in one study. No change in documentation of stop/review dates was reported during or after the intervention period. Remarkably, documentation on the reason to start antimicrobials decreased significantly during intervention and this sustained during follow up measurements over the next two years. ## User experience In five studies user experience was analysed by means of interviews, focus groups or surveys. The app was considered easy to use by 77.4%, 88.9% and \>90.0% and useful by 71.0%, 85.2% and \>90.0% of the users in before-after surveys with 31, 27 and 112 respondents, respectively. In one survey, 59 respondents reported app use increased their knowledge base regarding antimicrobial prescribing, while 81% reported app use helped them adhere to the guidelines. Another survey reported 68% of the 31 respondents found app use time saving Interviewees as well as \>90% of 112 survey respondents favoured the app guidelines over the web viewer or paper guidelines. Discomfort using the app in front of patients or colleagues due to a sense of unprofessionalism was mentioned by 20.0% of 59 survey respondents and 35.7% of 14 interviewees but this was not experienced by others. Frequent app use was inversely associated (survey respondents (SR) 106; risk ratio (RR) 0.03; confidence interval (CI) 0.0018–0.5; p = 0.0002) with preferring senior physician advice over antimicrobial guidelines, while frequent app use encouraged users to discuss incorrect prescribing by colleagues (SR 92; RR 3.8; CI 1.5–9.7; p = 0.005). Furthermore, app use was associated with a 1.1 point (p = 0.04) higher change in knowledge score of antimicrobial prescribing in 62 medical students and junior physicians compared to the control group and improved awareness of antimicrobial stewardship (SR 91; RR 6.8; CI 2.1–21.7; p = 0.001). ## Drug consumption, susceptibility and costs Monthly average antimicrobial drug consumption was the focus of one study conducted in Brazil. The app used in this study contained guidelines advising against use of some antimicrobials (e.g. carbapenems) while encouraging use of others (i.e. aminoglycosides) based on cost and susceptibility profile. After app introduction, the use of aminoglycosides and cefepime increased significantly while the use of piperacillin/tazobactam and meropenem decreased significantly. However, it should be noted that during the study period piperacillin/tazobactam was replaced by cefepime within the guideline for hospital-acquired infections. Furthermore, a significant increase in susceptibility to meropenem (73%–83%, p \< 0.05) and polymyxin (69%–83%, p \< 0.05) and a significant decrease in susceptibility to cefepime was described post-intervention (62%–57%, p \< 0.05). In the year of implementation, a significant reduction of \$296,485 USD (p\<0.05) in antimicrobial drug costs was attained compared to the pre-implementation year. In a different study the optimal approach to promote safe use of beta-lactam antibiotics in inpatients with a history of penicillin allergy was evaluated. Penicillin and cephalosporin use increased in the intervention period after introduction of a decision support app containing a decision tree for beta-lactam antibiotics (50% of 199 patients) compared to the standard of care period (38% of 148 patients). In the app intervention period odds of treatment with penicillin and cephalosporin were significantly increased (aOR 1.8% \[95% CI 1.1, 2.9\]). # Discussion In this systematic review, 13 studies which assessed antimicrobial stewardship smartphone apps in the hospital setting were analysed. Several studies measured different outcomes, applied different designs and varied in quality. In the reviewed studies, AMS apps were increasingly used or downloaded after implementation in five studies, guideline adherent prescribing of antimicrobials increased overall significantly in four studies and in one study this resulted in significantly less resistance to some antimicrobials and to a significant decrease in total drug costs. In general, users favoured the app based guidelines over web or paper versions in two studies, but some reported that app use in front of patients or colleagues felt unprofessional in three studies. Overall, although of varying quality, the studies indicate that AMS apps might increase guideline accessibility and offer physicians a friendly and efficient way of using antimicrobial guidelines. Content of all but one app in the reviewed studies focused solely on therapy to improve the prescribing of antimicrobials according to the guidelines. To evaluate appropriate prescribing of antimicrobials, different outcome parameters were selected: choice of antimicrobials, dose, dosing interval and route of administration. One study also included indication and stop/review date documentation. This variation in outcome parameters reflects the difficulty in defining appropriate antimicrobial prescribing. Since Gyssens *et al*. first described quality indicators (QI) for appropriate antimicrobial prescribing in 1992, QI’s have been added and debated in the infectious diseases community without reaching consensus, although many different quality indicators have been proposed. In the studies reviewed, a limited set of outcome parameters was evaluated, leaving out many insightful quality indicators of appropriate antimicrobial prescribing, such as switch from intravenous to oral therapy and timely initiation of antimicrobial therapy. Clearly defining quality indicators is essential in order to prevent interpretive bias and to compare studies promoting prescribing interventions. A factor that was not always taken into account but should be considered for each AMS app is how users experience them. Several studies show that physicians with different backgrounds are enthusiastic about apps, find them user-friendly and helpful in their work and would recommend them to colleagues. In the studies reviewed, some physicians regarded smartphone use unprofessional in front of patients. However, a study focussing on outpatients found that more than half of patients were “fine” when physicians used their smartphone to access information during consultations, and thirteen percent reported it was “not fine”. Initially used for calling and messaging, the increasing number of features and applications helped evolve the mobile phone into a possible valuable multifunctional tool for personal and professional use. AMS apps could help to educate students, but also junior and senior physicians, in antimicrobial use. Although for students the education on antimicrobials and AMS varies within and between countries, globally, the knowledge of appropriate antimicrobial prescribing of final-year medical students is limited. Additionally, prescribing errors are prevalent among junior physicians who are usually responsible for the majority of drug prescriptions in hospitals. The reviewed studies showed that AMS apps have additional educational value by improving knowledge of medical students and junior physicians on the prescribing of antimicrobials. Furthermore, as some of the younger smartphone using doctors have become attending physicians, smartphones will be an increasingly used tool for the prescription of antimicrobials. The effect of clinical decision support systems (CDSS) on antimicrobial prescribing in the healthcare setting such as hospitals have been evaluated in many studies including several randomized controlled trials and systematic reviews. Many are designed to improve guideline adherent prescribing of antimicrobials and are integrated into the EHR. Alternatively, almost all studied apps in our systematic review are standalone facilitating easy implementation in hospitals, are low cost and pose no risk in regard to losing patient data. However, a standalone system lacks patient specific data such as allergies, lab results, microbiological test results and previous treatment with antimicrobials which are mandatory to assess before antimicrobials can appropriately be prescribed. Overall, CDSS have a positive effect on adherence to guidelines for antimicrobial treatment and decreased antimicrobial consumption. Yet, similar to our findings, these outcomes were not unanimous. Furthermore, the studies we reviewed lacked important clinical outcomes reported in the studies evaluating CDSS such as mortality, length of stay and time to therapy. As for AMS apps user experience is an important stepping stone for successful CDSS implementation. In spite of many studies on CDSS and its impact on antimicrobial prescribing the need for good quality studies remains for this IT-intervention too. ## Strengths & limitations This systematic review has some strengths and limitations. One of the strengths is the clear start date of studies on this subject which coincides with the date app stores launched. A limitation is the exclusion of studies focusing on general practice or outpatient care. Therefore, we cannot draw conclusions on the advantage of AMS app use in these settings. Since the overall methodological study quality varied considerably and comparison between studies was limited due to large variations in study design and outcome parameters, only cautious conclusions could be drawn. Currently, the impact of a smartphone app on antimicrobial prescribing by physicians in hospitals is being evaluated in an international randomized trial. # Conclusions In this systematic review, the crossroad of healthcare and smartphone technology was explored. Smartphones may be used to improve knowledge of antimicrobial stewardship, to access antimicrobial guidelines and thereby improve important aspects of healthcare. During implementation of AMS apps, physician opinions and app uptake should be considered to optimize its impact. The small number of studies on AMS apps illustrate the novelty of this research area. Additionally, the quality of the data was limited. High quality, randomized, multi-centre studies including robust clearly defined clinical, microbiological and process outcomes are needed to evaluate the impact of AMS apps on antimicrobial prescribing and its role within healthcare. # Supporting information The authors thank Gerdien B. de Jonge for helping develop the search strategy. [^1]: The authors have declared that no competing interests exist. [^2]: ‡ These authors also contributed equally to this work.

# Introduction In coastal waters, strong diel pH variations are often observed due to high biological production, upwellings or riverine input. Superimposed on such natural pH variations, global oceans are being acidified due to continuously increasing atmospheric CO₂ concentration and its subsequent increased dissolution, which will lead to an overall pH drop in the oceans of 0.3–0.4 units by 2100. In combination with eutrophication and hypoxia events, acidification may occur at a faster rate in coastal waters compared to open oceans. Therefore, short-term pH fluctuations may become larger with progressive ocean acidification. Such changes in the chemical environment may affect cytosolic pH (and references therein) and cell homeostasis. Therefore, it is of significance to examine the physiological responses of marine organisms to rapid pH changes in addition to their adaptive and evolutionary responses. Marine diatoms contribute about 40% of oceanic primary productivity, playing a key role in carbon export to deep ocean waters and consequent regulation of global climate change. In addition, diatoms also play an important role in the marine food web due to their high abundance and wide size distribution. Therefore, how diatoms respond to ocean acidification (OA) or multiple environmental forcings is of general significance. However, documented findings on the effects of OA have been controversial, showing OA either caused enhanced growth, had neutral effects or inhibited growth of diatoms. As one of the most sensitive mechanisms to changes in pCO₂, the CO₂ concentrating mechanism (CCM) has been suggested to be associated with a range of physiological processes; consequently down regulation of CCMs under OA could have beneficial or detrimental effects under different light levels or treatment regimes. In coastal waters and upwelling areas with highly fluctuating pH, diatoms are often the dominant representative phytoplankton group. In these areas, pH fluctuations can exceed the 0.4 units predicted for the end of the century (and the same is true for corresponding CO₂ levels. While they experience pH variations, the cells usually suffer from frequent CO₂ limitation, over short time scales, that can be a selective pressure for phytoplankton with active CCMs. It is still unclear how diatoms may regulate their physiology to respond to rapid changes in pH and related changes in carbonate chemistry, and maintain a balance between photosynthetic efficiency and energy cost. CCM related genes have been shown to respond to CO₂ changes within 1 hr when cyanobacteria were exposed to CO₂-free conditions. However, since the downstream syntheses of CCM components should take longer if changes in gene expression are involved, how fast physiological responses to changes in carbonate chemistry occur remains unknown. Since diatoms might activate anti-stress mechanisms quickly to cope with the fast chemical changes in coastal waters, to maintain CO₂ supply when cells encounter carbon shortage. We therefore chose *Thalassiosira pseudonana*, a model diatom species frequently used for research in the past decades, to study whether diatoms can regulate physiological processes over a short timescale to respond to fast changes in pH/pCO₂. This species is distributed in coastal waters as well as open oceans. When grown under OA conditions, the growth rate shows little response under low light but suffers from inhibition under high light levels. Here, we show that this diatom can modulate its photosynthetic physiology to quickly respond to pH/pCO₂ changes, which may provide an advantage that allows it to outcompete other species in coastal waters. # Materials and Methods ## Species and culture condition *Thalassiosira pseudonana* (CCMP 1335), originally isolated from Moriches Bay, Long Island, USA, was inoculated in Aquil artificial seawater medium with a salinity of 35‰, and grown in triplicate independent cultures (500 mL Erlenmeyer flasks) in a plant growth CO₂ chamber (HP1000G-D, Ruihua) at 20 ± 0.1°C with a 14 h:10 h light: dark cycle. The cultures were illuminated at 120 μmol photons m^-2 s^-1 provided by cool white fluorescent lamps, and partially renewed with CO₂ pre-equilibrated medium every day to maintain cell concentrations within a range of 8×10⁴ to 3×10⁵ cell mL^-1. Target pH (*p*CO₂) values in the cultures and the fresh medium were achieved by bubbling air (390 μatm) or pre-mixed air-CO₂ mixtures (1000 μatm) within the plant growth chamber, which controlled the high CO₂ level with a variation of less than 30 μatm. Cultures were acclimated under the respective CO₂ level for nine days (\~20 generations). ## Determination of seawater carbonate chemistry pH was measured prior to and after the daily dilution as well as at the middle of the light period by a pH meter (pH510, Oaklon) which was calibrated with NBS buffer, to assure the stability of the carbonate system in cultures. Dissolved inorganic carbon (DIC) was sampled regularly and measured with an automatic system (AS-C3, Apollo SciTech Inc.) using an infrared gas analyzer (IRGA; Li- Cor7000, Li-Cor) after the samples were filtered into a syringe without any water/air CO₂ exchange. Samples of culture medium (0.5 mL) were acidified using phosphoric acid, and any subsequently released CO₂ was quantified by the IRGA. Parameters of the carbonate system were calculated as described in. The carbonate chemical parameters were maintained at relatively stable levels, with pH changes between dilutions being less than 0.04 units ## Chlorophyll fluorescence measurements To assess the photochemical responses of the cells to changes in the seawater carbonate chemistry, 50 mL sub-culture of LC- or HC-acclimated cultures was aseptically taken from each culturing bottle during the middle of the light period, and placed in a water bath to control temperature at \~20°C. 5 mL culture (LC-acclimated or HC-acclimated) was then dispensed into 20 mM Tris buffered medium with different pH levels (pH_NBS, i.e. 7.82, 8.10, 8.37 and 9.50), and illuminated in the same growth chamber. Rapid light curves (RLCs) were then measured using a Xenon-Pulse Amplitude Modulated Fluorometer (XE-PAM, Walz). To assess the contribution of diffusive CO₂ entry to photosynthetic efficiency, sub-samples were pipetted directly from illuminated cultures during the middle of light period, dispensed into 20 mM Tris buffered seawater medium that was pre-adjusted to target pH with or without ethoxyzolamide (EZ), a membrane permeable inhibitor of carbonic anhydrase, at a final concentration of 200 μmol L^-1 (pre-tests showed that 150 μmol L^-1 is a sufficient concentration). The RLC was determined as relative electron transport rate (rETR) in response to eight different and increasing actinic PAR intensities, with a 10 s duration for each increment separated by a 0.8 s saturating pulse (4000 μmol m^-2 s^-1). ## Determination of activity of carbonic anhydrase Cells grown under the two pCO₂ levels were harvested by gentle filtration onto a polycarbonate membrane, washed and re-suspended (cell concentration around 4×10⁶ mL^-1) in DIC free seawater that was buffered with 20 mM barbitone at pH 8.4. The CA_e (external CA) and CA_total (including periplasmic and intracellular CA) activity of the intact cells was determined by an electrometric method according to. A 5 mL cell suspension (for CA_ext) or cell suspension disrupted with a sonicator (102C Branson, for CA_total) with amplitude set at 30%, then added to a reaction cuvette and stirred. The disruption of cells was confirmed under a microscope. The time required for the pH to drop from 8.2 to 7.2 after the addition of 2 mL ice-cold CO₂-saturated pure water was recorded. The temperature during the reaction was controlled at 4°C. ## Calculations and statistical analysis The rETR was calculated as: rETR = Ф_PSII×0.5×PFD (μmol e^- m^-2 s^-1), where Ф_PSII is the photochemical quantum yield of PSII in light, PFD is the actinic PAR intensity (μmol photons m^-2 s^-1), and the factor 0.5 accounts for approximately 50% of all the absorbed energy being allocated to PSII. Rapid light curves were fitted according to as follows: y = x / (ax² + bx + c), Where y is the rETR, x is the photon flux density of actinic light (μmol photons m^-2 s^-1), a, b and c are the adjustment parameters. P_m, α and E_k was calculated as: $\text{P}_{\text{m}}\mspace{2mu} = \mspace{2mu} 1/\left( \text{b} + 2\sqrt{\text{ac}} \right)$, α = 1/c, E_k = P_m/a Activity of the carbonic anhydrase (in enzyme units, EU) was calculated as follows: 10×\[(T_b/T_c)-1\] / 10⁶ cells, where T_b and T_c were the times in seconds for the pH drop with or without cells, respectively. Statistical differences among treatments were tested with one-way ANOVA using a Tukey test by Origin 8.0, and the significance level was set at p = 0.05. # Results The rapid light curves (RLCs) showed typical patterns of photosynthesis versus irradiance; relative electron transport rate (rETR) increased with increasing light (1), while significant inhibition of rETR by EZ was observed for RLCs measured at both pH 7.82 and pH 8.37. rETR values of LC-acclimated and HC- acclimated cells treated with EZ (LC-acclimated +EZ or HC-acclimated +EZ cells) were significantly lower than those of the controls (without EZ) (p\<0.001), and recovered partially after 22–100 min incubation. The rETR of LC-acclimated cells was similar to that of HC-acclimated cells when measured at pH 7.82, while when samples were assayed at pH 8.37, the rETR of HC-acclimated cells were lower than that of LC-acclimated cells in the first 22 min , but achieved similar values after 44 min incubation. After 100 min incubation, rETR values of EZ treated samples was still lower than the control treatments when assayed at pH 7.82, while no significant differences were found among samples that were assayed at pH 8.37. Similar responses were observed at pH 8.10; rETR rates were significantly inhibited by EZ but recovered partially during the incubation (Fig A–D) (p\<0.001). However the responses were completely different at pH 9.50 where essentially all the DIC was available only as HCO₃^-. Here no differences were found among treatments, and no further recovery occurred during the incubation (Fig E–I). The maximal relative electron transport rate (rETR_m) measured at pH 7.82 remained relatively stable for LC-acclimated and HC-acclimated cells, values for both being around 128±4 μmol e^- m^-2 s^-1 during the 100 min incubation. The rETR_m of LC- acclimated +EZ cells showed a similar pattern, with relatively lower values around 113±2 μmol e^- m^-2 s^-1. While the rETR_m of HC-acclimated +EZ cells was around 73% of that of LC- acclimated +EZ cells for the first 22 min, then recovered quickly to similar values as LC-acclimated + EZ cells for the subsequent 78 min of incubation. There were obvious differences between LC-acclimated and HC-acclimated cells at pH 8.37 when CO₂ availability was only 25% of that at pH 7.82. rETR_m values of LC-acclimated cells measured at pH 8.37 decreased slightly but significantly during the 100 min incubation (p\<0.01), while HC- acclimated cells had rETR_m values significantly lower (by 20%) than those of LC-acclimated cells at T₀ (p\<1E-4), but increased linearly to reach similar values to those of LC-acclimated cells by the end of the incubation. The rETR_m values of HC-acclimated +EZ cells were 19% lower than those of LC-acclimated +EZ cells, but during the incubation, all increased gradually to final values that were close to those of non EZ treated samples. The E_k values of LC-acclimated and HC-acclimated cells at pH 7.82 remained relatively stable during the incubation, at 780±43 μmol m^-2 s^-1. The E_k of LC-acclimated +EZ cells was 619±26 μmol m^-2 s^-1 initially but increased slightly to 672±24 μmol m^-2 s^-1 then remained stable for the subsequent incubation (p = 0.058). The E_k of HC-acclimated +EZ cells was the lowest at T₀, being around 77% of that of LC-acclimated +EZ cells, but increased gradually and reached a similar value to that found in LC-acclimated +EZ cells at the end of incubation. The E_k measured at pH 8.37 remained relatively stable for the whole incubation at around 500–650 μmol m^-2 s^-1 with some variations but no clear trend. The values of light utilization efficiency (α, initial slope of RLC) of all treatments were around 0.18 e^- photon^-1 at T₀ under pH 7.82, and remained stable in most treatments for the subsequent incubation. While values for HC-acclimated +EZ cells decreased significantly from 0.17±0.009 to 0.15±0.005 e^- photon^-1 at T22 (p\<0.05), and then increased gradually to a similar value as the other treatments. α of LC-acclimated and HC-acclimated cells at pH 8.37 was similar as to that at pH7.82, and remained stable for the whole incubation period, at around 0.17, though α of LC-acclimated +EZ and HC-acclimated +EZ cells was lower at T₀, at 0.15±0.007 and 0.11±0.008 e^- photon^-1, respectively. While EZ treated cells all gradually increased to a similar value to that found in non-EZ samples at T₁₀₀. RLC parameters measured at pH 8.10 and pH 9.50 are shown in. The rETR_m values across the assay pH showed a decreasing pattern with increasing pH levels, at T₀, rETR_m values of LC-acclimated cells was similar to those of HC-acclimated cells at pH 7.82. While HC- acclimated values were lower than those of LC-acclimated at higher pH, around 80% of LC-acclimated at pH8.37, but all decreased to the same value, around 60 μmol e^- m^-2 s^-1, when measured at pH 9.50. rETR_m of HC-acclimated EZ cells was lower than LC-acclimated EZ, except for at pH 9.50 where both decreased to similar values to those of LC- acclimated or HC-acclimated cells. At T₁₀₀, rETR_m of LC- acclimated and HC-acclimated cells were similar to those at T₀, and dropped from 125±2 to 38±2 μmol e^- m^-2 s^-1 when the assay pH gradually increased from 7.82 to 9.50. rETR_m values of EZ treated cells were slightly lower than for non EZ samples, and decreased with increasing pH to a value of 38±2 μmol e^- m^-2 s^-1 at pH 9.50. The light utilization efficiency values of LC- acclimated and HC-acclimated cells were similar and stable, around 0.18 e^- photon^-1, while the EZ treated samples had lower values at pH8.10 and pH8.37, but were similar to non-EZ samples at pH7.82 or pH9.50. The light utilization efficiency values at T₁₀₀ were similar and stable for all treatments at pH 7.82 to pH 8.37, while they decreased sharply by 50% at pH 9.50. No differences were found among CO₂ or EZ treatments at T₁₀₀. The whole cell carbonic anhydrase activity (CA_total) was 0.18±0.03 EU (10⁶ cells)^-1 for LC-acclimated cells, while that of HC- acclimated cells was much less (p\<0.01), around 0.09±0.05 EU (10⁶ cells)^-1. Extracellular carbonic anhydrase (CA_e) activity was detectable but very low for both LC-acclimated and HC-acclimated cells, at 0.02(±0.01) and 0.01 (±0.004) EU (10⁶ cells)^-1 respectively. # Discussion Fluctuations of environmental factors are regarded as providing selection pressure for dominant species of phytoplankton. As the oceans continue to absorb anthropogenic CO₂, the buffering capacity of seawater will decrease, therefore phytoplankton cells are likely to experience increasing acidic stress with progressive ocean acidification. This will be superimposed on the rapid and significant fluctuations in pH and CO₂ concentration experienced by phytoplankton in upwelling regions and near-shore or estuarine environments. In the present study, by comparing the physiological responses of a diatom, grown at two *p*CO₂ levels, to rapid changes of pH, we concluded that *T*. *pseudonana* could respond quickly to changes of pH, to maintain a steady state supply of CO₂. When LC-acclimated cells are exposed to high pH (limited CO₂) environments, they could immediately achieve a relatively high electron transport rate, while there was an obvious lag before HC-acclimated cells responded to high pH. This may provide this diatom with a competitive advantage in coastal waters. Short-term exposure to low pH/high CO₂ should lead to negligible enhancement of photosynthesis of the LC-acclimated cells due to a sufficient supply of CO₂ by the CCM, recognizing that in *T*. *pseudonana* a biochemical CCM based on C4 photosynthesis, rather than the biophysical CCM found in most other microalgae may operate. The diatom tested in this study, when acclimated to LC conditions, can achieve higher electron flow than HC- acclimated cells, which could be attributed to the higher uptake of CO₂ for the photosynthetic machinery. Once the whole cellular enzymatic conversion between CO₂ and HCO₃^- was blocked by the inhibitor EZ, the rETR sharply decreased due to an inhibited CO₂ supply; in such a case, the photosynthetic rate should solely depend on diffusive entry or active uptake of CO₂. The higher rETR of LC-acclimated +EZ cells compared to that of HC-acclimated +EZ cells, suggests that when CO₂ was the only source of inorganic carbon, LC-acclimated cells had a higher efficiency in active CO₂ uptake. In *T*. *pseudonana*, the C4 pathway could play an important role during the acclimation under low CO₂ or the diel cycle, since previous studies have demonstrated that *T*. *pseudonana* has the ability to concentrate CO₂ via the C4 (or a C3-C4 intermediate) pathway for photosynthesis. Indeed, our results are in agreement with a recent study, which revealed that C4 pathway-related genes were up-regulated during an acclimation under reduced CO₂ availability, indicating a fundamental role for the C4 pathway in *T*. *pseudonana* photosynthesis, especially in a low CO₂ environment. The diminishing differences of rETR_m with time among the treatments with or without EZ in HC-acclimated or LC-acclimated cells , indicated that there was an inducible mechanism operational for the carbon uptake once the cells encountered CO₂ shortage, especially for EZ treated cells, and that this allowed recovery of photosystem II. Since light energy is essential for the operation of the CCM, when exposed to high pH with limited supply of CO₂, the cells have to allocate extra energy for carbon transport or synthesis of C4 compounds, resulting in a lower light utilization efficiency (α). Even when major components of CCMs, extra- and intracellular carbonic anhydrase, were blocked with sufficient inhibitor, rETR_m values of EZ cells could recover by \~40% to \~90% of non-EZ treated samples within 100 min, indicating that the cells could modulate their physiological processes to adapt to extreme conditions, including e.g. the up-regulation of the C4 pathway, reallocation of light energy to CCM, active CO₂ uptake, to maintain steady state CO₂ supply. Phytoplankton experience fluctuating environmental changes due to their temporal-spatial distributions. Riverine input, mixing or cyclones could stimulate the growth of some species which have strong nutrient accumulating mechanisms, while fluctuating sunlight would influence photosynthetic carbon fixation in surface seawater depending on levels of solar radiation and phytoplankton cells circulating in the water column will be exposed to rapidly changing light levels. Therefore, phytoplankton species may have acquired, during evolution, sophisticated mechanisms to modulate their physiology to respond to fast environmental changes. As revealed recently, the chain forming diatom could maintain photochemical performance, even exposed to acute pH changes. CO₂ is the major factor regulating CCM activity; while some diatoms rely on CO₂ diffusion more than active transport of bicarbonate across the plasma membrane, the varied CO₂ availability in coastal waters may affect their photosynthesis. Based on the findings of this work, we see that diatoms like *T*. *pseudonana* can cope with frequent fluctuations of pH/CO₂, and maintain CO₂ supplies at a steady state for photosynthesis. This can be considered as an advantage for these species, allowing them to dominate coastal waters. # Supporting Information This study was supported by National Natural Science Foundation (41120164007, 41430967, 41206091), Joint project of NSFC and Shan-dong province (Grant No. U1406403), Strategic Priority Research Program of CAS (Grant No. XDA11020302), Program for Chang-jiang Scholars and Innovative Research Team (IRT_13R51), SOA (GASI-03-01-02-04), MELRI1403, and the Fundamental Research Funds for the Central Universities (20720150076). JB’s work on climate change effects on algae has been funded by the Australian Research Council and his visit to Xiamen was supported by ‘111’ project from Ministry of Education. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: YW KG. Performed the experiments: YW. Analyzed the data: YW. Contributed reagents/materials/analysis tools: YW. Wrote the paper: YW JB KG.

# Introduction Hyaline cartilage has long been recognized as a tissue with limited capacity for repair. Articular cartilage loss can precipitate whole-joint pathology with associated pain and dysfunction that may necessitate surgical intervention. Resurfacing of the joint with functional hyaline cartilage is therefore an attractive treatment strategy. Implantation of osteochondral autograft plugs has been described in dogs with encouraging results but is associated with donor site morbidity and is usually limited to treating lesions ≤ 2 cm².. Osteochondral allograft transplantation procedures follow the same principles of osteochondral restoration as osteochondral autografting, but the graft is recovered from a qualified tissue donor of the same species. The use of an osteochondral allograft allows for resurfacing of larger defects using site- specific donor tissue and avoids donor-site morbidity. Cartilage is immune- privileged such that donor hyaline cartilage with intact matrix is not associated with immune system rejection responses. The subchondral bone component of an osteochondral allograft, however, is not immune-privileged and the recipient response must be dampened by limiting osteochondral allograft bone volume and removing donor marrow elements using pressurized lavage prior to implantation. Large, multi-surface or bipolar (apposing surfaces) osteochondral allograft transplantation has been performed for human patients for more than 40 years. Although success rates appear variable, reports suggest that use of osteochondral allografts with high chondrocyte viability at time of transplantation, autogenous bone marrow aspirate concentrate pretreatment of donor bone, accurate graft cutting and implantation techniques, and compliance with procedure-specific rehabilitation protocols can consistently result in highly successful 3- to 4-year outcomes. One of the major technical challenges in effectively transplanting bipolar osteochondral allografts involves consistent, precise preparation of the donor graft and recipient bed. Maintenance of appropriate joint alignment is important for functional joint movement, and successful osseous integration by creeping substitution is facilitated by sufficient subchondral bone contact and graft stability. Osteochondral allograft subchondral bone thickness between 3–8 mm has been advocated to confer structural integrity while minimizing recipient immune stimulation. Currently, grafts and recipient beds are manually prepared using common orthopedic instrumentation such as saws, reamers, burrs osteotomes, rasps, and rongeurs in a “freehand” technique. While this technique can be effective, the process is user-dependent, not standardized, and subject to human error. Three-dimensional (3D) planning and custom surgical guide printing are being used with increasing frequency in human and veterinary orthopedic surgery. The use of patient-specific osteotomy guides and instrumentation is associated with a high degree of accuracy and improved surgical efficiency. In dogs and cats, custom 3D printed guides have been used to facilitate limb deformity correction, fracture repair, and spinal surgery. Given the documented advantages of 3D virtual planning and 3D printed custom guides in other orthopedic applications, this technology may present similar benefits for performing bipolar osteochondral allograft transplantation. If this technology allowed for a standardized method for femoral head and acetabular osteochondral allograft transplantation that resulted in stable fixation, recapitulation of normal articular alignment, and successful osseous integration, biological hip joint restoration may become a suitable treatment option for the large number of dogs affected by disabling hip disorders. The objective of this study was to develop a system of custom 3D-printed guides that would allow precise shaping of donor and recipient femoral heads and donor acetabula for bipolar osteochondral allograft transplantation for the coxofemoral joint in dogs. We hypothesized that use of a custom guide system would result in faster osteochondral allograft and recipient site preparation, more precise osteochondral allograft-recipient fit, and improved accuracy in restoration of normal femoral head and neck version, inclination, and length compared to graft and recipient preparation performed without the use of guides. # Materials and methods ## Subjects This study was approved by the Institutional Animal Care and Use Committee (University of Florida IACUC Protocol \#201710005). Ten mixed breed dog cadavers weighing 20–38 kg were used for this study. Pre-study power analyses were performed to determine the number of dogs needed to test our hypotheses with alpha = 0.05 and minimum power of 0.8. Using one-sample paired t-test size analysis, we determined that a minimum of 5 pairs of dogs was needed for testing our hypotheses. The dogs were skeletally mature, but age was otherwise unknown. There were seven intact male dogs and three female dogs who’s reproductive status was unknown. The dogs were sourced from local animal shelters and had been euthanatized for purposes unrelated to this study. ## Preoperative imaging Standard hips-extended ventrodorsal radiographs of the pelvis and both femurs were obtained to confirm the joints were appropriately sized and did not have advanced hip dysplasia or other coxofemoral pathology. The pelvis and pelvic limbs were then imaged using computed tomography (CT) (160 Slice Toshiba Aquillion CT Scanner, Cannon Medical Systems, Tustin, CA, USA). Helical volume data (slice thickness of 0.5 mm and 0.3 mm slice overlap) was acquired. The bone algorithm was used for all 3D reconstructions and analysis. ## Group allocation and 3D modeling 3D image processing and guide design were performed using 3D medical image processing software (Materialise Medical Imaging Software Suite, Materialise NV, Leuven, Belgium). Bone window volume Digital Imaging and Communications in Medicine (DICOM) files were imported into imaging software (Mimics, Materialise Medical Imaging Software Suite, Materialise NV, Leuven, Belgium) for segmentation and 3D image reconstruction of the individual femora and pelves. Stereolithography (.stl) files of the bones were then imported into 3D modeling software (3Matic, Materialise Medical Imaging Software Suite, Materialise NV, Leuven, Belgium) for development of the virtual operative plan and surgical guide design. Once 3D images of each dog were created, acetabular diameter was used to match donor and recipient pairs. Acetabular diameter was measured on axial projections as the distance between the cranial and caudal acetabular border bilaterally for each dog. To ensure an appropriate fit, each dog’s mean acetabular diameter (average length from cranial to caudal pillars, bilaterally) was used to pair donors with recipients, with an acceptable donor acetabulum diameter 1–4 mm smaller than the recipient acetabulum. Each hip (left or right) was randomly assigned in matched pairs to either the *guide group* or *freehand group*, where custom surgical guides were used during surgery for the guide group and the freehand group surgery was performed without guides. ## Femoral guide design A biplanar osteotomy of the femoral head aligned along the physeal scar was planned to mimic the topography of the capital physis, simplify the osteotomy, and preserve the structural integrity of the femoral neck. Virtual 3D models of both the donor and recipient femoral heads were superimposed by aligning four anatomic landmarks, or points, located along the physeal scar on each femoral head. The craniomedial point was located where the physeal scar curved distally on the craniomedial aspect of the scar. The craniolateral point was located at the most proximal extent of the scar where the scar crossed over the proximal femoral neck. The caudomedial point was located on the caudomedial aspect of the physeal scar, where the scar was most proximal. The caudolateral point was located on the caudolateral side of the physeal scar, at the most distal point of the scar. Two planes were established based on the anatomic landmarks described above. A craniodistal-to-caudoproximal plane was created using the craniomedial, caudomedial, and craniolateral points. A cranioproximal-to-caudodistal plane was created using the caudomedial, caudolateral, and craniolateral points.. These planes served as planned common osteotomies in a chevron configuration that would be performed on the donor and recipient specimens. A donor femoral osteotomy guide was designed to conform to the topography of the donor femoral head and neck, and featured osteotomy shelves to guide sagittal saw placement when performing the osteotomies. The shelves were positioned such that the subchondral thickness was \< 8 mm in resultant ostectomized femoral head segment. A recipient femoral osteotomy guide was designed to conform to the topography of the femoral head and proximal femoral neck. The guide included osteotomy shelves that would accommodate a sagittal saw blade. The osteotomy shelves were designed using the same planes as the donor guide, resulting in a complementary recipient bed for the prepared donor graft. ## Acetabular guide design Commercially available acetabular reamers (BioMedtrix, Whippany, New Jersey, United States) were used to prepare the recipient acetabular site in the manner of artificial total hip replacement. In order to match the hemispherical osteotomy created by acetabular reamers, a donor acetabular cutting jig was designed to be used in conjunction with a pneumatic surgical drill (Hall 5058–01 Surgairtome Two, ConMed, Utica, New York, United States) and burr to sequentially contour the subchondral bone. The articular surface of the donor acetabulum was secured to a central post in the center of the base of the jig. The topography of the convex surface of the central post conformed to the articular lunate surface and acetabular fossa. A circular guide rail was designed to guide a burr during shaping of the donor acetabulum. The guide rail was secured to the base of the jig and could be rotated 360° around the central post that held the donor acetabulum. A burr support was designed to allow the drill and burr to move along the circular guide rail. The burr support contained a set-screw that allowed adjustment of burr height in relation to the central post, thereby adjusting the graft thickness. A 3D-printed version of the virtually planned donor acetabular graft, or *check acetabulum template*, was used to determine the appropriate height of the burr. By moving the burr back and forth along the guide rail and rotating the guide around the base of the cutting jig, a hemispherical donor acetabular graft of appropriate subchondral bone thickness could be fashioned. All guides and bone models were 3D printed using a fused deposition modeler additive manufacturing printer (Fortus 450mc, Stratasys, Eden Prairie, Minnesota, United States). Biocompatible plastics used in printing included polyetherimide (Ultem^™ 1010 resin, Stratasys, Eden Prairie, Minnesota, United States), polycarbonate ISO (PC-Iso, Stratasys, Eden Prairie, Minnesota, United States), and acetyl-butyl-styrene (ABS M30i, Stratasys, Eden Prairie, Minnesota, United States). The plastic used for each guide was dictated by availability of material at the time of printing. ## Surgery All surgical procedures, including osteochondral allograft preparation and implantation, were performed by the same board-certified small animal surgeon (*SEK*) to limit variation in technique. Recipient femoral site preparation time recorded, which was the duration from recipient femoral head and neck exposure until completion of the osteotomies. Time to prepare each femoral and acetabular osteochondral allograft and the recipient femoral site was also recorded. Recipient acetabular site preparation time was not included for comparison as the same commercially available total hip replacement acetabular reaming instrumentation and technique were used for both groups. Implantation and fixation times were not included for comparison as the same technique was used for both groups. ### Guide group procedure Cadavers were placed in lateral recumbency and a craniodorsal approach to the coxofemoral joint was made. The coxofemoral joint was luxated and the recipient femoral guide was applied to the femoral neck and secured using a 0.9 mm Kirschner wires A sagittal saw was used to create biplanar osteotomies with the guide shelves. The acetabulum was reamed in accordance with manufacturer recommendations for total hip replacement after femoral head resection. The osteochondral allograft femoral osteotomy guide was applied to the femoral neck and calcar region of the donor femur and secured using 0.9 mm Kirschner wires. Biplanar converging osteotomies were made along the osteotomy shelves to create the femoral osteochondral allograft. The printed donor acetabulum template was applied to the acetabular guide central post and maximum burr advancement was determined. The printed donor acetabulum template was removed, and the harvested donor acetabulum was secured to the guide central post and surrounding bone was secured to the base using 1.1 mm Kirschner wires. Several broad cuts were made using a sagittal saw, and then the arc and burr were used to finish contouring the spherical subchondral surface. The osteochondral allografts were secured in the recipient sites using Kirschner wires. Acetabular osteochondral allografts were secured with two wires: one at the level of the caudal pillar and the other at a more craniodorsal location The femoral osteochondral allograft was secured to the recipient using parallel Kirschner wires aiming from the fovea capitis distally down the center of the femoral neck. The coxofemoral joint was reduced. No additional procedures were done to maintain the reduction other than routine closure of the joint capsule, tendon, fascia, subcutaneous tissues and skin. ### Free-hand procedure After approaching the contralateral coxofemoral joint and luxating the femoral head in the same manner as the guide group, biplanar osteotomies were made in the femoral head at the level of the physeal scar using a sagittal saw to excise the articular surface. The osteotomies were made referencing the same landmarks outlined in the guide group but were performed without the assistance of guides. The acetabulum was reamed in standard fashion. Donor osteochondral allografts were manually shaped using burrs, rongeurs, and a sagittal saw. The biplanar donor femoral head and neck osteotomy configuration was based on the four landmarks around the physeal scar as described for the guide-group; however, the osteotomies were performed without the assistance of guides. Donor acetabular preparation was performed freehand with the same spherical subchondral bone shape objective as the guide group. This was accomplished by first making broad cuts with a sagittal saw to remove the medial acetabular cortex, ilium, and ischium, exposing the acetabular subchondral bone. Progressively smaller sections of bone were removed using the saw. Rongeurs were used to remove prominent edges and smooth out a spherical convex subchondral surface. Final maximal subchondral bone thickness was measured to ensure it did not exceed 8 mm. Osteochondral allografts were secured in the same manner as done in the guide group coxofemoral joints and closure was routine. ## Postoperative imaging and assessment Postoperative pelvic CT scans were acquired using the same methods as described for pre-operative planning. The size of the gap at the donor-recipient interface and femoral head alignment were calculated from the CT images. Acetabular alignment was not measured as the recipient acetabular bed was reamed without guide use for both groups, and the guide system was not designed to assist in acetabular orientation. *Gap volume* was defined by using a thresholding algorithm in segmentation software (Mimics, Materialise Medical Imaging Software Suite, Materialise NV, Leuven, Belgium) with a maximum of 250 Hounsfield units to segment the gap between donor and recipient bone. The resultant gap.stl file was imported into 3D modeling software (3Matic, Materialise Medical Imaging Software Suite, Materialise NV, Leuven, Belgium) to quantify the volume in cubic millimeters. *Femoral neck inclination* was measured by calculating the angle between the axes of femoral neck and femoral diaphysis. Firstly, a best fit sphere encompassing the femoral head was generated by selecting the 3D mesh triangles that made up the femoral head a best fit sphere was generated based on the selected triangle mesh. Similarly, femoral diaphyseal axis was defined as the central axis of a best fit cylinder by selecting the 3D mesh triangles of the diaphysis. A femoral neck axis was defined by a line extending from the center of the femoral head sphere through the center of the femoral neck and intersecting the diaphyseal axis. *Femoral neck length* was obtained by measuring the distance from the center of the femoral head to the intersecting axes of the femoral neck axis and diaphyseal axis. The angle between the femoral neck axis and the diaphyseal axis in the frontal plane represented femoral neck inclination. *Femoral neck version* was determined by measuring the angle created between the femoral neck axis and the transcondylar axis in the transverse plane. Total surgical time, defined as donor and recipient preparation, and gap analysis data were compared between groups. A mirror image of the virtual surgical plan from the custom guide hip was created to provide target parameters for the contralateral side. Femoral neck inclination, version, and length were compared to the virtual surgical plan for both groups and the absolute value difference for each parameter was calculated and compared between groups. All data sets were normally distributed on Shapiro-Wilks test, therefore parameters were compared using a 1-tailed paired t-test. Data is expressed as mean and standard deviation for each parameter. Significance was determined at p\<0.05. # Results Total surgical time preparing grafts and recipient beds for the guide group was longer than for the freehand group (p = 0.014). When compared to free-hand preparation, mean donor femoral preparation time was 2 minutes longer and mean recipient preparation time was 10 minutes longer when using guides (p = 0.011 and p = 0.001, respectively). No difference in acetabular preparation time was noted between groups. Gap volume at the acetabular and femoral donor-recipient interface was not different between groups. Mean difference between the planned and postoperative version angle was 6.2° lower for the guide group when compared to the freehand group (p = 0.025). Mean femoral neck length was 2 mm closer to the plan when using guides than when performing surgery freehand (p = 0.037). Accuracy for femoral angle of inclination was not different between groups. # Discussion This pilot study evaluated the accuracy of bipolar coxofemoral osteochondral allograft transplantation using 3D printed guides and freehand technique in cadaveric dogs. While guides did not confer an advantage in surgical time, donor-recipient interface gap, or femoral inclination, improved accuracy for femoral version and femoral neck length was achieved when using guides when compared to the freehand technique. Accurate restoration of version and femoral neck length could be clinically relevant, as these parameters are important for hip function and stability, and for success of artificial total hip arthroplasty in dogs. We hypothesized that use of our custom guide system would result in faster osteochondral allograft and recipient site preparation compared to freehand techniques; however, freehand preparation was more efficient than using guides. Both increases and reductions in surgical times have been reported with 3D printed custom surgical guides for several orthopedic applications in humans. In our study, we noted that positioning and securing the guides required careful attention to detail to ensure they were applied correctly, which was more time consuming than anticipated. Conversely, the brief visual assessment prior to osteotomy execution translated to a faster procedure. The difference in mean total graft and recipient preparation time in our study was only 10 minutes, which may be clinically acceptable if guides consistently improve accuracy of key aspects of osteochondral allograft transplantation. The volume of the gap present between donor and recipient subchondral bone was quantified since intimate contact between donor and recipient tissues is recommended for graft stability and eventual osseous integration by creeping substitution. We found no statistically significant differences in femoral and acetabular gap volumes between the two techniques evaluated in this study. In both groups, the mean gap volume between femoral graft and recipient was ≤ 131 mm³ and there was ≤ 1190 mm³ for the acetabular graft and recipient. While multiple studies state the importance of accurate graft- recipient fit for biomechanics and osseous union, no published study has quantified minimum donor-recipient gap sizes for appropriate integration to the authors’ knowledge. Without an evidence-based benchmark, it is unclear if the gaps noted in the current study would have clinical biomechanical relevance or impact graft incorporation. Accuracy of restoring femoral version and femoral neck length were superior in the guide group compared to the freehand surgery group. Excessive femoral anteversion following hip replacement in dogs can predispose to patellar and coxofemoral luxation. Malalignment has been cited as a cause of mechanical failure of osteochondral allografts in humans. Proximal femoral malalignment can lead to complications following total hip arthroplasty in dogs including impingement, luxation, and medial patellar luxation. Malalignment of hip prostheses in humans has been shown to precipitate alterations in contact mechanics, increase polyethylene wear, and predispose to implant failure. In our study, femoral version deviated from the virtual plan by an average of 0.8° in the custom guide group and 7° in the freehand group. While we cannot assume that the biomechanical properties of a hip osteochondral allograft are the same as a prosthesis, malalignment of osteochondral allografts could potentially have similar long-term detrimental effects. As such, use of custom surgical guides may reduce risk for major complications following bipolar coxofemoral joint osteochondral allograft transplantation compared to use of freehand technique. The limitations of this study are primarily related to use of cadavers, lack of functional and biomechanical testing, and small sample size. Additionally, dogs used in this study had little or no coxofemoral pathology. Clinically, dogs that would be considered candidates for bipolar coxofemoral osteochondral allograft transplantation are more likely to have anatomical alterations and/or osteoarthritis. Remodeling and osteophytosis can obscure anatomic landmarks, and custom surgical guides may provide greater advantages in clinical cases than our study findings suggest. Greater statistical strength from an increased sample size could uncover trends or significance not appreciated in this study. We did not design and test a guide for the recipient acetabular bed preparation and therefore acetabular orientation was not compared between groups. Lastly, these five cadavers represent our initial experience in bipolar coxofemoral osteochondral allograft in dogs. Increased familiarity with the procedure would likely yield improved outcomes. In this pilot study using a custom guide system for bipolar coxofemoral osteochondral allograft transplantation, we found that use of guides resulted in longer procedure times; however, guide use also increased surgical accuracy. Although bipolar coxofemoral osteochondral allograft transplantation procedures have not yet been reported clinically in dogs, this study serves as a foundation for developing the technique prior to implementation in clinical cases. Further studies will be required to evaluate the potential benefits of custom guides in dogs with a wider spectrum of anatomic variability and pathology. We would like to thank Ms. Cat Monger and Mr. Tom DeHaan for their technical support with cadaver surgery as well as the students of the University of Florida’s EML 5598 class of Fall 2017 for their engineering contributions. 10.1371/journal.pone.0244208.r001 Decision Letter 0 Tang Simon Yue-Cheong Academic Editor 2021 Simon Yue-Cheong Tang This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 16 Oct 2020 PONE-D-20-20362 Three-dimensional-printed custom guides for bipolar coxofemoral osteochondral allograft in dogs PLOS ONE Dear Dr. Kim, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. Please submit your revised manuscript by Nov 30 2020 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <plosone@plos.org>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript: A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols> We look forward to receiving your revised manuscript. Kind regards, Simon Y Tang, PhD Academic Editor PLOS ONE Journal Requirements: When submitting your revision, we need you to address these additional requirements. 1\. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at <https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main _body.pdf> and <https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_titl e_authors_affiliations.pdf> 2\. Please provide the full name of the IACUC, along with the approval number, in the manuscript. 3\. Please clarify in your Methods section the origin of the cadavers and how these were provided. 4.We note that you have stated that you will provide repository information for your data at acceptance. Should your manuscript be accepted for publication, we will hold it until you provide the relevant accession numbers or DOIs necessary to access your data. If you wish to make changes to your Data Availability statement, please describe these changes in your cover letter and we will update your Data Availability statement to reflect the information you provide. \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Yes \*\*\*\*\*\*\*\*\*\* 2\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes \*\*\*\*\*\*\*\*\*\* 3\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes \*\*\*\*\*\*\*\*\*\* 4\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes \*\*\*\*\*\*\*\*\*\* 5\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: PONE-D-20-20362 In this pilot study, authors created and tested a 3D printed surgical guide for osteochondral allograft transplantation in hip joints of dog cadavers. Authors tested a number of relevant parameter but only show moderate success with some negative findings. Despite a number of limitations, the study provides a platform for further investigation in development of custom surgical guide in veterinary medicine for osteochondral allograft transplantation. Overall, this is a very interesting paper and its technical concepts are well-conceived. It is written very well other than a few minor concerns noted below: Abstract Abstract is well-written and study design is presented clearly. However, result paragraph needs some details of actual data. For instance, p-values are shown but mean or 95% CI values are not shown. It would be helpful if the data can be included in the abstract. Introduction line 88, add “that” after hypothesized… Methods Sex of the animals not directly indicated. Age of the animals also need to be added if known. The sample size is very low, and it would be helpful if a power analysis is shown. This is important since low sample size is certainly one of the reasons as to why results for some important parameters did not reach formal level of significance. Results In this section, the actual data is missing. While this reviewer appreciates the citation of results in a Table, it would be helpful to add important numerical data in text, whereas p-values are presented by differences are not. For all Figures author stated color coding to direct the attention of the reader to a specific location, however this would be problematic on a black/white print. Could author use arrows, circles or similar to point our location of interest. In Fig. 5 pink color needs to be pointed by arrows, although it is clear on color print. Discussion Line 26: please replace “initial” with “pilot’ In Discussion, please add a summary paragraph at the end. Currently Discussion ends with limitations, this distracts from the main message. \*\*\*\*\*\*\*\*\*\* 6\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <figures@plos.org>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0244208.r002 Author response to Decision Letter 0 24 Nov 2020 Dear Editor and Reviewer, We thank you for your time and constructive comments regarding our submission entitled Three-dimensional-printed custom guides for bipolar coxofemoral osteochondral allograft in dogs. Please find our changes below: Editor Comments: 1\. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. RESPONSE: Formatting has been corrected to the best of our knowledge. Please let us know if there are further formatting issues. 2\. Please provide the full name of the IACUC, along with the approval number, in the manuscript. RESPONSE: Requested IACUC details have been added to paragraph 1 of the “Materials and methods” section. 3\. Please clarify in your Methods section the origin of the cadavers and how these were provided. RESPONSE: Requested cadaver details have been added to details have been added to paragraph 1 of the “Materials and methods” section. 4.We note that you have stated that you will provide repository information for your data at acceptance. Should your manuscript be accepted for publication, we will hold it until you provide the relevant accession numbers or DOIs necessary to access your data. If you wish to make changes to your Data Availability statement, please describe these changes in your cover letter and we will update your Data Availability statement to reflect the information you provide. RESPONSE: DOI has been provided Reviewer Comments: Abstract Abstract is well-written and study design is presented clearly. However, result paragraph needs some details of actual data. For instance, p-values are shown but mean or 95% CI values are not shown. It would be helpful if the data can be included in the abstract. RESPONSE: The Abstract now contains the means for each of our significant findings. Introduction line 88, add “that” after hypothesized… RESPONSE: Line 111: “that” has been added following “hypothesized” Methods Sex of the animals not directly indicated. Age of the animals also need to be added if known. The sample size is very low, and it would be helpful if a power analysis is shown. This is important since low sample size is certainly one of the reasons as to why results for some important parameters did not reach formal level of significance. RESPONSE: The methods section now includes the sex of the animals, approximate age (skeletally mature), and our pre-study power analysis. Results In this section, the actual data is missing. While this reviewer appreciates the citation of results in a Table, it would be helpful to add important numerical data in text, whereas p-values are presented by differences are not. RESPONSE: Means of our significant findings are now included in the text of the results section as well as in Table 1. For all Figures author stated color coding to direct the attention of the reader to a specific location, however this would be problematic on a black/white print. Could author use arrows, circles or similar to point our location of interest. In Fig. 5 pink color needs to be pointed by arrows, although it is clear on color print. RESPONSE: Figures 1 and 5 and their corresponding figure captions have been amended to account for black/white print Discussion Line 26: please replace “initial” with “pilot’ RESPONSE: - Line 399: “Initial” has been changed to “pilot” In Discussion, please add a summary paragraph at the end. Currently Discussion ends with limitations, this distracts from the main message. RESPONSE: - The Conclusion paragraph is now the last paragraph of the Discussion and has a summary sentence. 10.1371/journal.pone.0244208.r003 Decision Letter 1 Tang Simon Yue-Cheong Academic Editor 2021 Simon Yue-Cheong Tang This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 7 Dec 2020 Three-dimensional-printed custom guides for bipolar coxofemoral osteochondral allograft in dogs PONE-D-20-20362R1 Dear Dr. Kim, We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements. Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication. An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at <http://www.editorialmanager.com/pone/>, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to- date. If you have any billing related questions, please contact our Author Billing department directly at <authorbilling@plos.org>. If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact <onepress@plos.org>. Kind regards, Simon Yue-Cheong Tang, PhD Academic Editor PLOS ONE Additional Editor Comments (optional): Reviewers' comments: 10.1371/journal.pone.0244208.r004 Acceptance letter Tang Simon Yue-Cheong Academic Editor 2021 Simon Yue-Cheong Tang This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 2 Jan 2021 PONE-D-20-20362R1 Three-dimensional-printed custom guides for bipolar coxofemoral osteochondral allograft in dogs Dear Dr. Kim: I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department. If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact <onepress@plos.org>. If we can help with anything else, please email us at <plosone@plos.org>. Thank you for submitting your work to PLOS ONE and supporting open access. Kind regards, PLOS ONE Editorial Office Staff on behalf of Dr. Simon Yue-Cheong Tang Academic Editor PLOS ONE [^1]: The authors have declared that no competing interests exist.

# Introduction Influenza is a serious global healthcare issue and there are approximately one billion cases annually. Influenza is also associated with between 290,000 to 650,000 deaths every year. Seasonal influenza is a substantial cause of a large number of lower respiratory tract infections like pneumonia or bronchitis. The World Health Organisation (WHO) have prioritised influenza as a key area for future development, in their recently published global strategy. In this document, the WHO have outlined a number of recommended actions which relate to supporting influenza prevention, control and preparedness internationally. A key influenza prevention strategy is promoting the annual uptake of the influenza vaccination amongst at-risk populations and healthcare professionals. Encouraging influenza vaccination uptake amongst healthcare professional groups has been a key prevention strategy for many years. The annual vaccination is recommended by global healthcare institutions due to its role in reducing influenza and transmission between patients. The vaccination also reduces the occurrence of healthcare professional absenteeism, subsequent staff shortages and reduced quality of care. Despite this, influenza immunisation uptake can be challenging with vaccinations being administered to approximately 74% of front- line healthcare professionals in England and 80% of front-line workers in the USA. There have been a number of research studies which have examined healthcare professional views on influenza vaccination and two important themes have emerged as to why healthcare professionals do not receive the influenza vaccination. The first reason relates to healthcare professional misconception about the influenza vaccination, its effectiveness and safety. The second reason is around healthcare professional ability to access the free vaccination. In response to this, there have been a range of interventions utilised to increase vaccination uptake amongst healthcare professionals including the use of educational strategies, organisational flu campaigns, incentivisation and adoption of vaccination champions. Nursing students are one group that are at a high risk of exposure to seasonal influenza. Despite this, they are a marginalised group and are often absent in practice guidelines about influenza and empirical research investigation on the topic. In the UK, current National Health Service (NHS) advice about who should receive the influenza vaccination, does not explicitly mention nursing students. To our knowledge, there are no national or international recommendations for nursing students about the influenza vaccination. As a consequence, knowledge and subsequent uptake of influenza vaccination is potentially low amongst nursing students. Research indicates that between15% to 50% of nursing students will receive annual influenza vaccination, however due to the paucity of research in this area, combined with small numbers of participants in these published studies, the data is not generalisable. While research is limited, it is postulated that the rationale for low influenza vaccine uptake amongst healthcare professional students is due to limitations in knowledge, professional misconceptions and ease of access. An increasingly effective and innovative method of educating professionals is through the use of ‘serious games’. Unlike traditional entertaining games, the ‘serious game’ has been designed with a specific educational purpose in mind. The ‘serious game’ is considered as an entertaining tool with a purpose of education, where players cultivate their knowledge and practice their skills through gaming. In the context of infection control and prevention, there has been evidence to suggest that a ‘serious game’ can enhance healthcare education and practice. The aim of this study was to evaluate the effect of a ‘serious game’ about influenza, on nursing student attitude, knowledge and uptake of the influenza vaccination. The objectives of this study were as follows: 1. To examine nursing student perceptions about their attitudes and understanding of influenza before and after playing the game. 2. To learn if playing the ‘serious game’ increased nursing student knowledge about influenza. 3. To establish if playing the ‘serious game’ correlated with increases in flu vaccination uptake amongst nursing students. # Methods ## Ethics This study was approved by the School of Nursing and Midwifery’s Research Ethics Committee at Queen’s University Belfast in July 2018 (2.GMitchell 09.18M1.V1). Participants did not provide verbal or written consent but were informed that they were under no obligation to complete any of the questionnaires. Participants gave their consent to complete the questionnaire when they actively accessed the survey web links. ## Design, setting and population This study took place at one university in Northern Ireland. All university students (n = 1306) who were undertaking a BSc Honours Degree in Nursing were provided with access to the ‘serious game’ between 1^st September 2018 and 31^st March 2019. The nursing students were enrolled in one of the four programmes; adult nursing, mental health nursing, children’s nursing or learning disability nursing. All students received access to the same version of the ‘serious game’. In total 430 nursing students, from year 1, 2 and 3 (32.92%), played the game completed a short 8-item questionnaire about their perceptions and attitudes to influenza throughout September 2018 to March 2019. From this sample of 430 nursing students, 356 (82.79%) went on to complete a further 2-item questionnaire about their uptake of the influenza vaccination, after access to the game had ended, in April 2019. In addition to these two short questionnaires, we conducted a separate 40-item pre and post knowledge questionnaire with a homogenous sample of first year adult nursing students (n = 145) to determine if knowledge about influenza improved in this cohort. Year one students were selected on the basis that they were likely to have less knowledge about influenza compared to their peers in year two and year three. The year one students in this sample had received twelve weeks of standard nurse education at the university and undertaken two clinical placements, each lasting 6 weeks, as part of their nursing programme prior to completing a pre-knowledge questionnaire about influenza in August 2018. A follow-up post-knowledge questionnaire about influenza was administered to this cohort in April 2019. In total 124 students (85.52%) completed both knowledge questionnaires. ## Intervention The game, known as ‘Flu Bee Game’ was developed by Focus Games Ltd in 2017. The game has been used to promote knowledge about influenza and encourage vaccination uptake amongst healthcare professionals at multiple international settings. The game has been associated with improving vaccination uptake in multiple settings but has yet to be tested amongst nursing students. The Flu Bee Game is an HTML5 web application with a supporting website. The game works on any device through a web browser and only takes a few minutes to play. Players answer random questions, from an existing question bank, about influenza and vaccination. If they get a question correct, they build a ‘honeycomb path’ that leads them to ‘Queen Bee’ status. Players of the Flu Bee Game can share their success on the game’s leader board and invite colleagues to play via social media. The serious game presents players with influenza facts and challenges common myths associated with vaccine hesitancy. These common myths include statements like “I’m healthy and so do not need a flu vaccine”, “I’m 12 weeks pregnant so cannot have the flu vaccine” or “I don’t want to risk getting the flu from the vaccine”. The Flu Bee Game takes approximately 90 seconds to play and players can have multiple attempts, as questions are randomly generated. Players receive feedback and further information on each question they answer in the game. The overall objective of this serious game is to create awareness about influenza, dispel myths associated with the influenza vaccine and increase uptake the vaccination. ## Consent and recruitment All students (n = 1306) received information about the ‘serious game’ and this study by a person unrelated to the project in August 2018 via email. Students were provided with a web-link to the game and informed that they could access the game any time throughout September 2018 to March 2019. All students from years 1 to 3 could also opt to complete a voluntary 8-item questionnaire immediately after playing the game. All students received three follow-up emails, at the end of September 2018, the end of November 2018 and the end of January 2019, to remind them of the availability of the ‘serious game’ and accompanying questionnaire. These reminders were also sent by a person unrelated to the project. Students did not provide any personal or demographic information in their responses but could report their university email address to be contacted about a further 2-item questionnaire, about their uptake of influenza vaccination, in April 2019. A second 2-item questionnaire was emailed to all nursing students who had played the game, completed the first questionnaire and provided their email address. In addition to these questionnaires, the authors worked with Focus-Games Ltd to design a 40-item knowledge questionnaire about the myths associated with influenza and influenza vaccination. Face validity was tested with 12 nursing students, who were not part of the cohort, prior to administration. A homogenous group of year one nursing students were invited to complete this 40-item questionnaire before they received access to the ‘serious game’ in August 2018 and then again in April 2019 once access to the game had ended. This cohort of year one nursing students received both questionnaires via email by a person unrelated to the study. Students did not have to sign written consent forms but were informed that they were under no obligation to complete any of the questionnaires. It was assumed that students gave their consent to complete a questionnaire when they actively accessed the survey web links. Student participants were required to use their own laptop, computer tablet or mobile phone to complete the questionnaires. Questionnaires were completed by students in their own time and not during any timetabled classes. ## Data collection All nursing students (n = 1306) were eligible to participate in the first 8-item questionnaire. This questionnaire, designed by the authors, sought to examine nursing student perception of their knowledge about influenza, likelihood of getting vaccinated and importance of promoting the vaccine amongst their patients after playing the game. This was achieved using Likert scale items with participants asked to select an option for each question; ranging from very poor, poor, average, good to very good. In total 430 nursing students (32.92%) completed this questionnaire, then in April 2019, 401 nursing students from this sample agreed to be contacted via email to receive a follow-up 2-item questionnaire about their uptake of influenza vaccine. Subsequently 356 nursing students went on to complete this second questionnaire. Finally, a cohort of year one nursing students (n = 145) were purposively selected to participate in a pre- and post-knowledge questionnaire about influenza after receiving access to the ‘serious game’. Overall, 124 nursing students (85.52%) completed a 40-item pre and post questionnaire, designed by the authors, about myths associated with influenza and influenza vaccination. All nursing students that completed both the pre and post knowledge questionnaires were automatically entered into a prize raffle and three winners received a complimentary stay at a hotel in Belfast. While other validated measures were available, the authors developed their own questionnaires to answer the research question. The rationale for this was due to a combination of factors including the nature of the intervention (a digital game), its duration (each play taking approximately two minutes), the sample (nursing students) and the information the authors wanted to glean. ## Data analysis Descriptive statistics were used to illustrate the findings from the 8-item questionnaire and the 2-item questionnaire to measure nursing student perceptions about their attitudes and understanding of influenza and subsequent vaccination uptake. The pre and post knowledge questionnaires, administered to the cohort of nursing students, were analysed using paired t-tests to establish if the ‘serious game’ increased student knowledge about influenza. # Results ## 8-Item questionnaire about attitudes and perception The 8-item questionnaire measuring influenza attitudes and perceptions, was completed by 430 nursing students. Of the respondents who completed the 8-item questionnaire 53.3% were from year one (n = 229), 24.4% were from year two (n = 105) and 22.3% were from year three (n = 96). Of these participants, 36.7% (n = 158) had received the influenza vaccine the year prior and 63.3% (n = 272) had not received the vaccination to date. Nursing students perceived that their knowledge about influenza and the vaccination was very good (n = 36/8.37%), good (n = 164/38.14%), average (n = 188/43.72%), poor (n = 32/7.44%) or very poor (n = 10/4.30%) prior to playing the game. Immediately after completing the game, nursing student perception of their knowledge increased with 91.4% of students perceiving their knowledge to be either good (n = 187/43.49%) or very good (n = 206/47.91%). In relation to their willingness to receive the influenza vaccination, 39.7% (n = 171) stated that prior to playing the game they did not intend to receive the vaccination. After playing the game, this number decreased to 7.4% (n = 32). The number of students who stated they definitely would receive the influenza vaccination doubled from 29.5% (n = 127) pre-game, to 58.6% (n = 252) post-game. Finally, as it pertains to student perception about the importance of promoting the influenza vaccination to their patients and public, 44.0% (n = 189) felt this was very important pre-game and this increased to 83.3% (n = 358) post- game. Less than 1% of the sample (n = 2) believed the influenza vaccination was either slightly important or not important post-game. Descriptive statistics from this 8-item questionnaire can be viewed in. ## 2-Item questionnaire about influenza vaccine uptake In total 356/401 nursing students (88.8%) went on to complete the second questionnaire. The first questionnaire item determined if the respondent had received the influenza vaccination during the period of 1^st September 2018 to 31^st March 2019. Overall, 47.8% (n = 170) of students who played the game received the vaccination and 52.2% did not (n = 186). Of the students who did not receive a vaccination they were asked to complete the second item of the questionnaire to provide a reason for their decision. The most common reason selected by nursing students was related to a lack of time (33.9%). The remaining reasons were confusion about where to receive vaccination (23.7%), concerns about receiving the vaccination (21.5%), forgetting to get vaccination (15.1%) and being told they had to pay for their vaccination by their general practitioner (5.8%). ## 40-Item knowledge questionnaire about influenza and vaccination A total of 145 nursing students were purposively selected to participate in a pre- and post-knowledge questionnaire about influenza after receiving access to the ‘serious game’. This sample was selected to determine the extent to which the game improved knowledge about influenza in a homogenous group. From these 124 nursing students (85.52%) completed a 40-item pre and post questionnaire. The questionnaire comprised of 27 true or false questions and 13 multiple choice questions, each with four available answers. Participants scored 1 point for every question correctly answered. Overall, nursing students scored an average of 68.6% before receiving access to the game and 85.2% after, demonstrating a statistically significant increase (p\<0.001) using paired t-tests. The Pearson’s correlation coefficient was 0.594, indicating a moderate positive relationship between nursing student’s knowledge about influenza and playing the game. The most significant increases in knowledge that were noted post-game related to the amount of time it took to become fully protected from the influenza after the vaccination (10–14 days); the coverage of the influenza vaccination in relation to Australian Flu and Swine Flu; that most people who have influenza do not have symptoms, and that vaccination is recommended for pregnant women. The increase in knowledge across these questions ranged from 47.2%-70.3% post-test. Increases in knowledge was recorded across 34 of the 40 items. The remaining six items demonstrated a marginal decrease of 3.6%-1.4% Descriptive statistics from this 40-item questionnaire can be viewed in and the dataset can be viewed in the supporting information. # Discussion To our knowledge, this is the first study that has examined the impact of a ‘serious game’ on nursing student knowledge and vaccination uptake. Pre- intervention, 36.7% of the sample had received the influenza vaccination in the preceding season. This low vaccination uptake is reflective of other research studies which have examined influenza vaccination uptake amongst this population group. Post-intervention, vaccination uptake increased to 47.8% amongst the sample. While this modest increase is encouraging, it is evident that despite improvement in the attitudes and knowledge about influenza, more than half of the sample did not go on to receive their vaccination. Due to the nature of this study, it is difficult to judge the extent to which the serious game led to an increase in uptake. The study invitation, email reminders and clinical experiences of nursing students during this research are acknowledged as possible confounders. The main reasons why nursing students did not receive their influenza vaccine were lack of time, uncertainty of where to receive the vaccination and concerns about receiving the vaccination. These reasons are reflective of the international literature on the reasons why qualified healthcare professionals do not receive influenza vaccination. While there are no recommended vaccination uptake targets for nursing students, the Public Health Agency in England recently set the target of having 90% of its healthcare professional workforce vaccinated. The target is 75% in Europe and the USA. These figures would suggest that influenza vaccination of nursing students is a priority area due to their absence in influenza vaccination guidelines, empirical research and literature, combined with their apparent low uptake of the vaccination. There has been a plethora of research studies which have demonstrated that provision of education, to address professional misconceptions about the influenza vaccination, have been associated with improvement in knowledge and subsequent uptake. This study was reflective of this literature and provided all nursing students with six-months access to a ‘serious game’ about influenza. The students who participated in this study perceived their knowledge, likelihood to get the vaccination and likelihood to recommend the vaccination to patients and the public had improved after playing the game. In terms of the 40-item knowledge questionnaire, provided to a cohort of year one nursing students, we found highly statistically significant changes in level of knowledge after receiving access to the ‘serious game’. Incidentally, this cohort of nursing students did not receive any additional education about influenza or influenza vaccination during their nursing programme in the six-month period they had access to the ‘serious game’. The use of ‘serious games’ has already been associated with significant improvement in participant knowledge in previous studies. Despite these positive findings, vaccine uptake remained below half after this study. While education appears to be a supportive factor for increasing influenza vaccination uptake, providing an environment where nursing students can easily receive the influenza vaccine appears equally important. In the United Kingdom, front line healthcare professionals often receive their influenza vaccination at their place of work, for example at a staff influenza session within their hospital. This was not the case for the nursing students included in this sample because many were attending university during the flu vaccination season, were not attending a clinical placement and therefore were responsible for organising their own vaccination with their general practitioner or local pharmacy. While this study focused on influenza, the findings are also of interest in the context of the COVID-19 pandemic. Presently several COVID-19 vaccines are currently in human trials or have been rolled out. Despite the availability of COVID-19 vaccination, recent international research has suggested a high potential for vaccine hesitancy amongst the global population. Vaccine hesitancy is now a global concern and presents a substantial obstacle to achieving community immunity from COVID-19. Governments, healthcare services and patient advocacy groups are now tasked with building vaccine literacy amongst all members of society and this research suggests that the use of gamification may be one such supportive evidence-based strategy. There were some limitations to this study. While the study sample was large comparable to similar research, approximately two in three students did not participate. In addition, our sample was not equally spread, with more than half of our sample made up from year one nursing students. This makes generalisability of findings to similar settings difficult. With consideration to knowledge, this study only examined this in year one students, and it is therefore difficult to determine the impact of the serious game on knowledge of year 2 and year 3 nursing students. The study design would have also been strengthened had the serious game been compared to an alternative intervention or control. Finally, this study may have also been strengthened had it used validated instruments that have previously been used to measure attitudes to influenza and vaccination-related knowledge. Despite these limitations, this study makes a key contribution to a limited evidence-base and demonstrates how provision of a ‘serious game’ is very likely to improve knowledge of influenza and subsequent uptake of the vaccination. # Conclusions The research highlights the importance of equipping nursing students with education about influenza. Improvements in influenza knowledge is likely to encourage more nursing students to receive the influenza vaccination, an action that will help prevent influenza transmission between healthcare professionals and patients. This study also demonstrates how provision of a ‘serious game’ can provide nursing students with an innovative learning tool which has been associated with highly statistically significant improvements in knowledge. # Supporting information We wish to thank the nursing students who participated in this study and Focus Games Ltd who agreed for us to carry out this independent evaluation and publish the findings prior to commencing data collection. # Declarations All authors meet at least one of the following criteria (recommended by the ICMJE: <http://www.icmje.org/ethical_1author.html>) and have agreed on the final version. 10.1371/journal.pone.0245389.r001 Decision Letter 0 Duplaga Mariusz Academic Editor 2021 Mariusz Duplaga This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 11 Nov 2020 PONE-D-20-24096 Evaluation of a ‘serious game’ on nursing student knowledge and uptake of influenza vaccination. PLOS ONE Dear Dr. Mitchell, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. Please submit your revised manuscript by Dec 26 2020 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <plosone@plos.org>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript: A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols> We look forward to receiving your revised manuscript. Kind regards, Mariusz Duplaga, Ph.D., M.D., Ass. Prof. Academic Editor PLOS ONE Journal Requirements: When submitting your revision, we need you to address these additional requirements. 1\. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at <https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main _body.pdf> and <https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_titl e_authors_affiliations.pdf> 2.Thank you for including your ethics statement: This study was approved by the School of Nursing and Midwifery’s Research Ethics Committee at Queen’s University Belfast in July 2018. Please provide additional details regarding participant consent. In the ethics statement in the Methods and online submission information, please ensure that you have specified (1) whether consent was informed and (2) what type you obtained (for instance, written or verbal, and if verbal, how it was documented and witnessed). If your study included minors, state whether you obtained consent from parents or guardians. If the need for consent was waived by the ethics committee, please include this information. If you are reporting a retrospective study of medical records or archived samples, please ensure that you have discussed whether all data were fully anonymized before you accessed them and/or whether the IRB or ethics committee waived the requirement for informed consent. If patients provided informed written consent to have data from their medical records used in research, please include this information. Once you have amended this/these statement(s) in the Methods section of the manuscript, please add the same text to the “Ethics Statement” field of the submission form (via “Edit Submission”). For additional information about PLOS ONE ethical requirements for human subjects research, please refer to <http://journals.plos.org/plosone/s/submission-guidelines#loc-human-subjects- research>. \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Yes Reviewer \#2: Partly \*\*\*\*\*\*\*\*\*\* 2\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 3\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 4\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 5\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: The study describes the evaluation with a pre- and post-design of a serious games aimed to increase knowledge about influenza vaccination (and, consequently, vaccination uptake) among nursing students. The paper is very well written and the study is timely. Major comment: \- The design of the study is a bit complex. You had two homogenous cohorts, but they were not really comparable since they were not administered the same questionnaires. Is this the case? This might be a limitation and should be stated as such in the Discussion. If I misunderstood the design of the study and the completion of the questionnaires, this means that the description is not clear and that the Methods section should be better written. The reader gets lost with the different questionnaires, cohorts, timeline… Overall, it is not clear why you recruited two cohorts. Minor comments: \- Since you have the chance to revise your paper before publication, I suggest to mention in either the Introduction or the Discussion Covid-19. This would add some practical implications from your study to the current pandemics’ situation. \- Please justify why you have created ad-hoc questionnaires to measure attitudes to influenza and vaccination-related knowledge even if some validated instruments already exist. \- In the Introduction you mention that in the UK vaccinations are administered to 70% front-line healthcare professionals. Do you have access to any data about real uptake? The percentage of 70% is quite high, especially compared to other European countries. Please confirm this high uptake. \- Include p values in Table 1. \- Lack of time is the main reason for not getting vaccinated. This could be better addressed in your Discussion. It seems that knowledge is not that important to get vaccinated… \- In Table 2, provide the answers for “Which of these treatments does not treat influenza”. This stand-alone item is difficult to interpret. Reviewer \#2: This is a potentially interesting study on the impact of a serious game intervention on influenza vaccination. The paper's main strength is the fact that it seems to be the first to assess this type of intervention in the context of influenza vaccination. It's main weaknesses are the potential confounders that make it hard to ascertain a causal link between exposure to the game and the observed increase in vaccination uptake, and the lack of comparison to alternative (possibly cheaper and less time-consuming than a serious game) interventions. In general, the paper is well written and the study's findings are clear. The methods section is also clear, but the subsection on the intervention should have described the game in more detail. A better description of the game would help the reader assess possible confounding factors, and form a better idea of what other kinds of intervention might be comparable to the intervention assessed in this study. The soundness of attitudes and perception questionnaire is debatable. I found it doubtful that the respondent could have formed an unbiased opinion of the state of their knowledge and attitude regarding vaccination after playing the game. The authors should discuss this point. It also strikes me that it is hard to judge the extent to which any change of attitude, reflected in the higher vaccination uptake, is due to the game, or to other factors surrounding the study. It could be due, for instance, to the pre- study questionnaire acting as a reminder or raising awareness to vaccination, irrespectively of the game intervention. This issue should be discussed. I also found it somewhat surprising that the authors did not include an alternative intervention for comparison. Might a leaflet or a short video have been as effective as the serious game? \*\*\*\*\*\*\*\*\*\* 6\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: **Yes: **Ilaria Montagni Reviewer \#2: **Yes: **Saturnino Luz \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <figures@plos.org>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0245389.r002 Author response to Decision Letter 0 21 Dec 2020 This has been uploaded as a separate document with this submission. For convenience it is also noted below: Response to Reviewers PONE-D-20-24096: Evaluation of a ‘serious game’ on nursing student knowledge and uptake of influenza vaccination. PLOS ONE Please include the following items when submitting your revised manuscript: • A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'. • A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'. • An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. Journal Requirements: When submitting your revision, we need you to address these additional requirements. 1\. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at <https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main _body.pdf> and <https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_titl e_authors_affiliations.pdf> Thank you for this. We have now reviewed these style requirements and amended our submission accordingly. 2.Thank you for including your ethics statement: This study was approved by the School of Nursing and Midwifery’s Research Ethics Committee at Queen’s University Belfast in July 2018. Please provide additional details regarding participant consent. In the ethics statement in the Methods and online submission information, please ensure that you have specified (1) whether consent was informed and (2) what type you obtained (for instance, written or verbal, and if verbal, how it was documented and witnessed). If your study included minors, state whether you obtained consent from parents or guardians. If the need for consent was waived by the ethics committee, please include this information. Thank you. We have included the following additional information about consent: Participants did not provide verbal or written consent but were informed that they were under no obligation to complete any of the questionnaires. Participants gave their consent to complete the questionnaire when they actively accessed the survey web links. Review Comments to the Author Reviewer \#1: The study describes the evaluation with a pre- and post-design of a serious games aimed to increase knowledge about influenza vaccination (and, consequently, vaccination uptake) among nursing students. The paper is very well written and the study is timely. Major comment: \- The design of the study is a bit complex. You had two homogenous cohorts, but they were not really comparable since they were not administered the same questionnaires. Is this the case? This might be a limitation and should be stated as such in the Discussion. If I misunderstood the design of the study and the completion of the questionnaires, this means that the description is not clear and that the Methods section should be better written. The reader gets lost with the different questionnaires, cohorts, timeline… Overall, it is not clear why you recruited two cohorts. Thank you for this feedback. We have clarified this in the manuscript and provided clearer rationale about why a specific cohort of participants (year one students) was selected and noted this as a limitation in the manuscript as they are not comparable with the first cohort of participants (that included year 1, 2 and 3 students). Minor comments: \- Since you have the chance to revise your paper before publication, I suggest to mention in either thm e Introduction or the Discussion Covid-19. This would add some practical implications from your study to the current pandemics’ situation. Thank you. We have included information about COVID-19 and the potential of translating this approach within the discussion section as suggested: While this study focused on influenza, the findings are also of interest in the context of the COVID-19 pandemic. Presently several COVID-19 vaccines are currently in human trials or have been rolled out. Despite the availability of COVID-19 vaccination, recent international research has suggested a high potential for vaccine hesitancy amongst the global population\[52\]. Vaccine hesitancy is now a global concern and presents a substantial obstacle to achieving community immunity from COVID-19\[53\]. Governments, healthcare services and patient advocacy groups are now tasked with building vaccine literacy amongst all members of society and this research suggests that the use of gamification may be one such supportive evidence-based strategy. \- Please justify why you have created ad-hoc questionnaires to measure attitudes to influenza and vaccination-related knowledge even if some validated instruments already exist. Thank you for this comment. We have included the following paragraph to provide clarity: While other validated measures were available, the authors developed their own questionnaires to answer the research question. The rationale for this was due to a combination of factors including the nature of the intervention (a digital game), its duration (each play taking approximately two minutes), the sample (nursing students) and the information the authors wanted to glean. We have noted the following within the limitations section of the manuscript: Finally, this study may have also been strengthened had it used validated instruments that have previously been used to measure attitudes to influenza and vaccination-related knowledge. \- In the Introduction you mention that in the UK vaccinations are administered to 70% front-line healthcare professionals. Do you have access to any data about real uptake? The percentage of 70% is quite high, especially compared to other European countries. Please confirm this high uptake. We have rechecked this reference and this was correct (for England – not the whole UK: (739,187/1,051,851 of front line healthcare workers: 70.3%). We have taken the opportunity to update our statistics for England and this is now 74.3% (791,112/1,065,017). \- Lack of time is the main reason for not getting vaccinated. This could be better addressed in your Discussion. It seems that knowledge is not that important to get vaccinated… Thank you. We have included the following: Despite these positive findings, vaccine uptake remained below half after this study. While education appears to be a supportive factor for increasing influenza vaccination uptake, providing an environment where nursing students can easily receive the influenza vaccine appears equally important. In the United Kingdom, front line healthcare professionals often receive their influenza vaccination at their place of work, for example at a staff influenza session within their hospital\[15\]. This was not the case for the nursing students included in this sample because many were attending university during the flu vaccination season, were not on a clinical placement and therefore were responsible for organising their own vaccination with their general practitioner or local pharmacy. \- In Table 2, provide the answers for “Which of these treatments does not treat influenza”. This stand-alone item is difficult to interpret. Thank you for this comment. We have included the options to this question: Antibiotics, antiviral medications, influenza vaccination or keeping hydrated/staying warm. Reviewer \#2: This is a potentially interesting study on the impact of a serious game intervention on influenza vaccination. The paper's main strength is the fact that it seems to be the first to assess this type of intervention in the context of influenza vaccination. It's main weaknesses are the potential confounders that make it hard to ascertain a causal link between exposure to the game and the observed increase in vaccination uptake, and the lack of comparison to alternative (possibly cheaper and less time-consuming than a serious game) interventions. Thank you for your review. We have included the following sentence in our limitations section in response to your feedback: The study design would have also been strengthened had the serious game been compared to an alternative intervention or control. In general, the paper is well written and the study's findings are clear. The methods section is also clear, but the subsection on the intervention should have described the game in more detail. A better description of the game would help the reader assess possible confounding factors, and form a better idea of what other kinds of intervention might be comparable to the intervention assessed in this study. Thank you for this comment. We have now provided the following information in this section:. Players of the Flu Bee Game can share their success on the game’s leader board and invite colleagues to play via social media. The serious game presents players with influenza facts and challenges common myths associated with vaccine hesitancy. These common myths include statements like “I’m healthy and so do not need a flu vaccine”, “I’m 12 weeks pregnant so cannot have the flu vaccine” or “I don’t want to risk getting the flu from the vaccine” \[49-50\]. The overall objective of this serious game is to create awareness about influenza, dispel myths associated with the influenza vaccine and increase uptake the vaccination. The soundness of attitudes and perception questionnaire is debatable. I found it doubtful that the respondent could have formed an unbiased opinion of the state of their knowledge and attitude regarding vaccination after playing the game. The authors should discuss this point. Thank you for this comment. We have included the following text in our limitations section with regards to the attitudes/perception questionnaire: Finally, this study may have also been strengthened had it used validated instruments that have previously been used to measure attitudes to influenza and vaccination-related knowledge It also strikes me that it is hard to judge the extent to which any change of attitude, reflected in the higher vaccination uptake, is due to the game, or to other factors surrounding the study. It could be due, for instance, to the pre- study questionnaire acting as a reminder or raising awareness to vaccination, irrespectively of the game intervention. This issue should be discussed. I also found it somewhat surprising that the authors did not include an alternative intervention for comparison. Might a leaflet or a short video have been as effective as the serious game? Thank you for this helpful comment. We have now acknowledged the potential confounders in the discussion: Due to the nature of this study, it is also difficult to judge the extent to which the serious game led to an increase in uptake. The study invitation, email reminders and clinical experiences of nursing students during this research are acknowledged as possible confounders. In terms of the control/comparator we have included the following in our limitations section: The study design would have also been strengthened had the serious game been compared to an alternative intervention or control \_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_ Thank you to the reviewers for taking the time to review this manuscript and providing their helpful feedback. We believe this process has supported us in strengthening the manuscript. 10.1371/journal.pone.0245389.r003 Decision Letter 1 Duplaga Mariusz Academic Editor 2021 Mariusz Duplaga This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 30 Dec 2020 Evaluation of a ‘serious game’ on nursing student knowledge and uptake of influenza vaccination. PONE-D-20-24096R1 Dear Dr. Mitchell, We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements. Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication. An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at <http://www.editorialmanager.com/pone/>, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to- date. If you have any billing related questions, please contact our Author Billing department directly at <authorbilling@plos.org>. If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact <onepress@plos.org>. Kind regards, Mariusz Duplaga, Ph.D., M.D., Ass. Prof. Academic Editor PLOS ONE 10.1371/journal.pone.0245389.r004 Acceptance letter Duplaga Mariusz Academic Editor 2021 Mariusz Duplaga This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 6 Jan 2021 PONE-D-20-24096R1 Evaluation of a ‘serious game’ on nursing student knowledge and uptake of influenza vaccination. Dear Dr. Mitchell: I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department. If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact <onepress@plos.org>. If we can help with anything else, please email us at <plosone@plos.org>. Thank you for submitting your work to PLOS ONE and supporting open access. Kind regards, PLOS ONE Editorial Office Staff on behalf of Dr. Mariusz Duplaga Academic Editor PLOS ONE [^1]: The authors have declared that no competing interests exist.

# Introduction Bats are substantially more vulnerable to collisions with wind turbines than birds, although the underlying reasons for collision mortalities remain unclear. The scale of the problem became apparent recently when, during a six-week period, an estimated 1,764 and 2,900 bat fatalities were recorded at two wind farms in West Virginia and Pennsylvania. Bat fatality was highest during late summer and fall when bats begin autumn migration and migratory species comprise the majority of fatalities at all wind farms studied to date. This is undoubtedly exacerbated by the placement of wind turbines on topographical features such as ridgelines and in forest corridors, which are used as migratory routes for several bat species. However, it remains unclear whether foraging bats, as well as migrating individuals, are also at risk from collisions. Thermal images of wind turbines appear to indicate that bats are attracted to and investigate both moving and static blades, and studies in Europe have also reported bats foraging close to turbine blades. The numbers of collision mortalities reported in America are much greater than in Europe, where mortality surveys have begun more recently. One of the problems with providing accurate data on bat deaths in Europe is the lack of a consistent methodology. Numbers in published reports range from 2–50 bats with each study including a different number of turbines, different survey methods and different time periods. However, 15 of the 35 species of European bat have been recorded as regular victims of turbine collisions, and an Intersessional Working Group of Eurobats cited 20 species thought to be at risk of collision. Current research in Europe is concentrated on the development of scientifically credible mortality estimates to assess the extent of the problem. Although this is clearly important, the rapid proliferation of wind turbines requires a more urgent response. Research is required on the underlying reasons behind these collisions and on potential methods to mitigate this increasing threat to endangered bat populations. Attempts at mitigating bird collisions with wind turbines have typically involved the application of visual stimuli to increase the conspicuousness of the turbine blades, but for bats, where audition is the primary sensory modality, this is clearly inappropriate. The design of an acoustic deterrent for bats, as used to mitigate cetacean entanglement in drift nets, is complicated by the intrinsic properties of ultrasound, which attenuates rapidly in air. Therefore in the absence of an efficient acoustic deterrent it is essential to investigate alternative sources capable of inducing aversive behaviour in bats. Researchers at Aberdeen University have observed for some time that bat activity is reduced in the vicinity of the Aberdeen civil Air Traffic Control (ATC) radar station despite the proximity of habitat types where bat activity would be expected. This raised the possibility that the radio frequency (RF) radiation associated with radar installations may elicit an aversive behavioural response in foraging bats. Radio frequencies occupy the portion of the electromagnetic spectrum between 3 kHz and 300 GHz. Absorption of energy in the range of 1 MHz–300 GHz results primarily in tissue heating by movement of ions and oscillations of dipole molecules resulting in a transfer of energy from the RF field to the biological medium. Short term RF exposure can produce a thermal burden in an organism that can result in significant behavioural and physiological changes, some of which may be harmful. Although behavioural effects of such exposure on humans include perception, aversion, work perturbation, work stoppage and convulsions, few field experiments have been carried out to ascertain the possible effects of high frequency electromagnetic radiation on wild animals. However, electromagnetic radiation can influence the development, reproduction, and physiology of insects, mammals, and birds. There is no current direct evidence to suggest that bats can detect or respond to electromagnetic radiation. However, we predict that if high frequency electromagnetic radiation exerts an aversive response in foraging bats, bat activity will be reduced in the vicinity of radar installations. The aim of the present study was to test this hypothesis. # Methods ## Study sites and sampling protocol In Britain, foraging bats are predominantly associated with areas where insect density is high: broadleaved woodland, particularly woodland edge, linear vegetation (tree lines and hedgerows) and riparian habitat. More open and intensively managed areas are avoided. Therefore in order to assess the impact of radar on foraging bats it was important to locate radar stations surrounded by habitat suitable for foraging. This negates the possibility that the absence of bats around radar stations is simply an artefact of the exposed location of the radar. Following initial reconnoitres, 10 radar stations were selected. These included four civil airport air traffic control (ATC) radar stations, three military ATC radars and three weather radars. Each selected radar station was surrounded by habitat with a high degree of heterogeneity, thereby facilitating the identification of sampling points along an electromagnetic gradient. At each radar station three sampling points were chosen within one of three habitat categories: riparian woodland, woodland edge and tree-lines. Each sampling point was matched for habitat type, habitat structure (e.g. height and length of isolated tree-line), dominant vegetation species, altitude and surrounding land class. Each successive sampling point was located at increasing distance from the radar station and subject to differing levels of electromagnetic field strength. A portable electromagnetic field meter (PMM 8053-Accelonix Ltd.) and isotropic field probe (EP-330 Isotropic E-Field probe-Accelonix Ltd.) were used to measure the electromagnetic field strength (EMF) of the radar in volts per metre (v/m) at three distances from the radar source. Radars emit a train of very brief pulses of high intensity followed by a silent interval for the echoes to return, therefore the peak intensity, at each sampling point, was recorded in one minute intervals over a thirty minute period and the average of these readings used to classify each site as follows: in close proximity to the radar (\<200 m) and subject to a high electromagnetic field (EMF) strength \>2 v/m, an intermediate sampling point within line of sight of the radar (200–400 m) and with an EMF strength \<2 v/m and a control site that was not in line of sight of the radar and registered an EMF of zero v/m (\>400 m). At each radar station bat activity was recorded contemporaneously within these three field strength categories. Paired sampling was used to control for variation in bat activity due to environmental parameters. Throughout the summer each radar station was surveyed on three occasions with three different sampling points monitored on each occasion resulting in a total of 90 samples, 30 of which were obtained within each field strength category. ## Bat activity recording At each radar station bat activity was recorded using three automatic bat- recording stations. Each automatic station consisted of a Batbox 3 heterodyne bat detector (Stag Electronics, Sussex, UK) linked to a count data logger (Gemini Data Loggers, UK Ltd, Chichester, UK) via an analogue to digital signal converter (Skye instruments, Ltd). The signal converter converts analogue signals from the bat detector into digital signals that can be recorded by the data logger. Every 0.5 seconds a positive or negative signal is sent to the data logger indicating the presence or absence of ultrasound respectively. The Batbox 3 was tuned to 50 kHz in order to detect each of the five breeding species of bat in Scotland (*Pipistrellus pipistrellus, Pipistrellus pygmaeus, Myotis daubentonii, Myotis nattereri* and *Plecotus auritus*). However *P. auritus* seldom emits calls loud enough to be detected and is therefore unlikely to be recorded. The recording stations were operational from sunset to sunrise and the data loggers were set to record bat active minutes (the number of minutes throughout the recording period that ultrasound was detected by the bat detector). The component parts of the system were housed in large plastic boxes with a hole cut for the bat detector microphones. Automatic recording stations were positioned on platforms 1.5 m above the ground and orientated perpendicular to the linear element (e.g. woodland edge, tree-lines, riparian woodland). In addition to the automatic recording of bat activity, a 30 minute transect was carried out at each sampling site using a bat detector (S-25 Ultrasound Advice, London) set to frequency division mode. This method of ultrasound transformation allows calls to be recorded in real time on audiocassettes and the number of recorded passes provides a quantitative assessment of bat activity. Bat detectors were linked to a tape recorder (Sony Walkman, Tokyo, WMD6C) containing metal cassettes. At each site, sound recording equipment was held at waist height and a 50 m transect was walked in a zigzag fashion back and forth across the site to avoid any bias in direction or placement of the detectors. The 30 minutes recording at each site was analysed using BatSound Pro software (Pettersson Elektronic AB, Uppsala Sweden). ## Statistical analysis Bat activity in sites subject to electromagnetic radiation (\>2 v/m) was compared to the control sites (0 v/m) using paired t tests. To avoid pseudoreplication, tests were carried out on the average of the three replicates at each radar site (n = 10). Data were log (log₁₀ (x+1)) transformed to achieve normality and homogeneity of variance. Analyses were carried out using Minitab version 14. The difference in bat activity between treatment and control groups was analysed further using general linear models (GLMs) in SPSS 12.0, including all relevant and recorded confounding variables. Radar type was included as a random factor and reproductive status (pre-lactation, lactation, post-lactation), temperature and EMF strength (high or intermediate) were included as covariates. Interactions were only included in the model when of direct relevance to the hypothesis being tested and interactions between confounding variables were not tested. # Results The automatic stations recorded a total of 3727 bat active minutes over 230 h of recording. A further 2979 bat passes were recorded during transects with the frequency division detector (45 h). As expected, the majority of passes (83%) were attributed to the two cryptic pipistrelle species: *Pipistrellus pipistrellus* and *Pipistrellus pygmaeus* (45% and 38% respectively) which are the most common and abundant bats in Scotland. A further 14% of bat passes were attributed to *Myotis daubentonii* and 3% to *Myotis nattereri*. ## Bat activity Total bat activity (bat active minutes) was higher in the control sites (0 v/m) when compared to sites exposed to a high level (\>2 v/m) of electromagnetic radiation. Paired t tests carried out on the log-transformed bat active minutes, recorded by the automatic bat recording stations, showed that bats were significantly more active in control sites when compared to high EMF sites (t = 4.41; n = 10; p = 0.003;). The number of bat passes (all species) recorded during transects with the frequency division detector was also higher in the control sites when compared to sites exposed to a high level of electromagnetic radiation. Paired t tests carried out on the log transformed number of bat passes, showed that bats were significantly more active in control sites when compared to high EMF sites (t = 4.86; n = 10; p = 0.001;). The general linear models showed that, when confounding variables were taken into account, the level of EMF strength within 400 m of the radar had no significant effect on the difference in bat activity recorded between treatment and control groups. However, radar type did have a significant effect with the difference in bat activity between treatment and control groups greatest in the vicinity of civil ATC radars and least in the vicinity of military ATC radars. The interaction of field strength\*radar type was not significant when added to the model (Bat active minutes: F_2,50 = 2.3, p = 0.11; bat passes: F_2,50 = 0.3, p = 0.76) indicating that each radar type had a similar effect at both high and intermediate EMF sites. # Discussion ## Bat activity and foraging effort Currently there have been no successful attempts to directly mitigate bat collisions with wind turbines. Attempts at deterring bats by the use of ultrasound have, as yet, been unsuccessful. Therefore the identification of alternative methods capable of inducing an aversive response in bats approaching turbine blades is of paramount importance. Our result have demonstrated that bat activity is significantly reduced in habitats exposed to an electromagnetic field (EMF) strength of greater than 2 v/m when compared to matched sites registering EMF levels of zero. Even at sites with lower levels of EMF exposure (\<2 v/m), bat activity was reduced in comparison to control sites. However, the difference in bat activity between treatment groups and control groups varied between radar types with results more equivocal in the vicinity of military ATC radars. It is possible that this may be explained by the characteristics and operating times of the individual radar units concerned, sensitive information which was not available. Despite this the overall reduction in bat activity within habitats exposed to high and intermediate EMFs supports our hypothesis that the electromagnetic radiation associated with radar installations can exert an aversive behavioural response in bats. This raises questions regarding the mechanisms by which bats could perceive electromagnetic radiation and why they would avoid EMF exposure. We propose two mechanisms through which electromagnetic fields could induce an aversive behavioural response in foraging bats: thermal induction leading to an increased risk of overheating and hyperthermia, and echolocation interference - the auditory microwave hypothesis. ## Thermal effects of EMF exposure Studies investigating the behavioural response of laboratory animals to the presence of electromagnetic fields have provided substantial insight into the most probable mechanism of interaction of these fields with intact organisms. This mechanism relates to the generation of heat in the skin that results in the activation of thermal sensors in the tissues and central nervous system. Studies of human thermal sensation generated by RF exposures, reinforce the conclusion that behavioural changes observed in RF-exposed animals are thermally motivated. Indeed, it has been demonstrated in the laboratory that measured elevations of surface and deep body temperatures often accompany specific behavioural changes. The effect advances from the threshold of perception, through intermediate steps, to an extreme thermal insult, grand mal seizures, and finally death. In this respect, exposure to an RF field differs little from exposure to conventional sources of thermal energy or inhospitable thermal environments. For the majority of animals a short period of overheating constitutes a much greater hazard than does an equivalent degree of cooling. A rise of only a few degrees above the optimum temperature is quickly fatal. The wing membranes of bats present a large surface area over which radiation might be absorbed, increasing heat load to the animal. This, combined with the heat energy produced during flight makes bats particularly susceptible to overheating, which can be fatal in experimental conditions between 38–39°C. Furthermore, observations of captive bats have noted their aversion to even a moderate infra-red heat source. Therefore it is possible that thermal induction, resulting from EMF exposure in the vicinity of radar installations, may provide an inhospitable thermal regime for foraging bats, which could vary from discomfort to hyperthermia depending on EMF strength and the duration of exposure. ## Auditory microwave hypothesis Although the mammalian ear has no sensitivity to electromagnetic waves at microwave frequencies (300 MHz–300 GHz) human auditory perception of radio frequency energy has been reported since the 1940s. It is now widely accepted that the auditory perception of microwaves is a result of thermoelastic expansion. The absorption of the energy in the RF pulse leads to a rapid thermal expansion resulting in a thermoelastic wave. This wave is then propagated through the soft tissues of the head until it reaches the fluid-filled inner ear, where it is transduced into a sound pressure wave leading to excitation of auditory neurons in the kHz range. Laboratory experiments have shown that the frequency of the induced sound is a function of head size and of the acoustic properties of the brain tissue. The estimated fundamental frequency of vibration in guinea pigs, cats and adult humans were 45, 38, and 13 kHz respectively. It is therefore not only plausible but probable that bats exposed to an RF pulse of sufficient power would effectively hear this pulse and the frequency detected would lie within the range of frequencies used for orientation, prey detection and capture for the majority of bat species. There is no evidence that the auditory perception of microwaves would act to deter foraging bats any more than the production of ultrasound at the same frequency. However if bats can perceive areas of high EMF exposure and experience an associated rise in internal temperature it provides a mechanism through which an aversive response may be elicited. ## Conclusions We have demonstrated that bat activity is reduced in habitats exposed to electromagnetic radiation when compared with matched sites where no such radiation can be detected. However without access to detailed specifications of individual radar units (including operational times and operating frequency) it is difficult to quantify this relationship further. To more fully define the impact of radar on foraging bats, and ascertain its value as a potential source of mitigation, field trials involving a mobile radar that can be introduced into areas of known bat activity are now required. If the parameters of an RF signal capable of inducing an aversive response in foraging bats could be characterised then this may offer a method of mitigating bat collisions with wind turbines. We are grateful to J.D. Pye, C.D MacLeod and W.Cresswell for comments on an earlier draft of this paper. Our attention was first drawn to the absence of bats around ATC radar by Laurent Preynat, and Michael Muir-Wright carried out a pilot study. [^1]: Conceived and designed the experiments: PR. Performed the experiments: BN. Analyzed the data: BN. Wrote the paper: BN. [^2]: The authors have declared that no competing interests exist.

# Introduction Habitat loss is the primary threat to biodiversity worldwide, but species show wide variation in their responses to habitat loss. This variation is often attributed to differences in species traits. However, most studies evaluating species responses to habitat loss have measured habitat amount as patch size, rather than evaluating the effects of habitat loss over the landscape (i.e. landscape scale study). In addition, most are limited to a narrow range geographical locations. Moreover, understanding the effects of species traits on species responses to habitat loss is often (unavoidably) confounded by correlations or synergistic interactions among the traits themselves. Therefore, we still do not know, in general terms, why some species or species groups are more sensitive to habitat loss than others. Dispersal ability is generally considered an important species trait influencing species response to habitat loss. Species with greater movement ranges are predicted to have higher colonization rates because they are able to access more habitat in a landscape. Moreover, since colonization rates are assumed to be correlated to immigration rates, local extinction probability is predicted to be lower for species with higher movement ranges, due to rescue of populations from low numbers by immigration. Metapopulation studies therefore typically assume that more mobile species should be less susceptible to habitat loss than less mobile species. However, some empirical studies have found the opposite, that more mobile species are more sensitive to habitat loss, possibly because they incur higher dispersal mortality. Therefore, the general relationship between mobility and tolerance to habitat loss is not clear. Reproductive rate could also influence species responses to habitat loss, but it is not often considered, at least in empirical studies. Simulation studies suggest that reproductive rate has a much larger effect on the amount of habitat required for population persistence than the per capita rate of emigration or dispersal ability ; lower reproductive rates are predicted to increase the amount of habitat required for population persistence. To our knowledge, there are only two empirical tests of this prediction in which habitat amount is measured at a landscape scale. Vance et al. found that forest bird species with lower reproductive rates require more habitat for a 50% probability of occurrence than do forest birds with higher reproductive rates. Similarly, Holland et al. found a negative association between reproductive rate and minimum habitat amount required for presence across a group of dead wood boring beetle species. These studies suggest that species with lower reproductive rates require more habitat in a landscape for population persistence than do species with higher reproductive rates. This is likely because for a given amount of habitat in a landscape, species with higher reproductive rates will rebound more quickly from population declines than species with lower reproductive rates. Moreover, a large number of offspring increases the number of potential colonists which increases colonization rates of unoccupied patches. Therefore, we predicted that species with lower reproductive rates are more sensitive to habitat loss than species with higher reproductive rates. The objective of this study was to determine the importance of mobility and reproductive rate in determining the responses of wetland species to wetland habitat loss. We used a meta-analytical approach to quantitatively synthesize the results of 90 studies conducted across the world that quantified the relationship between wetland amount in a landscape and wetland animal abundance. From these we obtained 426 responses to wetland loss for 220 wetland species including mammals, birds, reptiles and amphibians. We selected wetlands and wetland species for several reasons. First, wetlands are generally thought to function as metapopulations – and thus they provide an appropriate system to test whether mobility drives species responses to habitat loss. Moreover, wetland species are undergoing the largest wildlife population declines worldwide, primarily due to habitat loss, but are a relatively understudied ecological group in landscape ecology. Therefore, there is a need to provide a general synthesis on how wetland species are responding to wetland habitat loss, to prioritize landscape-scale conservation action. For example, if mobility is driving the response to habitat loss, the conservation focus should be on facilitating movement during high dispersal events or increasing connectivity in landscapes. In contrast, if reproductive rate is driving the response, the focus should be on supporting critical reproductive stages. To our knowledge, this is the first meta-analysis of population responses to habitat amount at the landscape scale (previous reviews were conducted at the patch scale), across taxa, and including the relationship to species’ mobility and reproductive rate. # Methods ## Study Selection Criteria We measured wetland loss as the amount of wetland in a landscape. We included studies that measured wetland amount as the percent of wetland area in a landscape (area-based buffers) or wetland connectivity (or isolation). We searched for studies that quantified the relationship between the amount of wetland in a landscape and population abundance of at least one wetland species in the Web of Science and ProQuest dissertation and theses databases on 01 December 2011 using the following keyword string: (wetland\* OR marsh\* OR swamp\* OR pond\*) AND (amount OR area OR isolat\* OR fragmentation) AND (amphib\* OR turtle\* OR reptile\* OR mammal\* OR bird\*) AND (abundance\* OR occurrence\* OR occup\* OR distribution) AND (species OR population\*) AND (landscape\*). No restriction on date was used. We limited our analyses to empirical studies that were conducted in wetlands, including natural wetlands (e.g. pond, marsh) and artificially created wetlands (e.g. stormwater basins, rice fields). For all studies, we assumed the authors accurately selected wetland habitat for each species; however if wetland amount at the landscape scale included land cover types other than wetlands (e.g. lakes), we contacted authors for clarification or excluded the study from analyses. We used a broad definition of “population abundance” to include population size (or relative abundance), population density (or relative density) and species occupancy (as an index of low vs. high abundance). We defined “wetland species” as any vertebrate (mammal, bird, reptile or amphibian) that uses wetlands as primary habitat for at least one part of its life cycle. We included species complexes that were fertile hybrids (e.g. *Pelophylax esculentus*) or two species that could not be distinguished (e.g. larval stages of *Ambystoma* spp) as one species in the meta-analysis. During our literature search, we recorded the number of articles identified and the number of studies included and excluded according to the Preferred Reporting Items for Systematic Reviews and Meta- Analyses (PRISMA) Statement. ## Data Extraction In our meta-analysis, an effect size represents the quantitative relationship between the amount of wetland in a landscape and population abundance for a given wetland species. To extract an effect size from each study, we first searched the paper for a test statistic for the effect of wetland amount on animal abundance, and/or summary statistics (e.g. mean and variance), and corresponding sample size that could be converted into an effect size. When these values were not reported, we calculated them using raw data if they were provided in the paper, or we could extract them from figures using GetData Graph Digitizer 2.25 (Fedorov, S. 2012, internet free software), or we could obtain them from the authors. When a single study reported results for more than one species, we entered each species effect size as an independent estimate. When a study combined abundance or occurrence data across species, such that values for individual species could not be extracted, we contacted the authors for raw data or excluded the study. We did not calculate an effect size for 36 (of 256) species that occurred in ≤10% of sampled landscapes or locations in each study, because effect size estimation is biased when there is a high proportion of absences. While this may have resulted in exclusion of species sensitive to habitat loss, in the absence of sufficient sample sizes we cannot include them in the analyses. When a single study presented more than one effect size for a given species such that different effect sizes representing responses of the same species to different wetland types, we averaged these estimates across wetland types to extract a single effect size for that species, to avoid non- independence (4 studies). When a single study presented data in multiple years using the same study design, we averaged estimates across years for continuous data or tallied the numbers of years present for occupancy data. When studies presented effects of wetland amount at multiple landscape scales (buffer sizes), we selected the largest estimate, on the assumption that this scale was closest to the scale at which wetland loss best predicts the species’ response (i.e. scale of effect, *sensu*). ## Study Design Moderators We identified three study design moderators to test if differences in study design influenced the magnitude and direction of the effect size, and to statistically control for such effects in remaining analyses. First, the effect size might vary depending on how wetland amount in a landscape was measured. We combined several measures of wetland amount, including simple area-based buffers and nearest-neighbour distances to more complex connectivity indexes based on the incidence function model, because these measures have been shown to be highly correlated and have similar performance in predicting ecological responses. These comparative studies also suggest that measures with more information about the amount of (occupied) habitat in the landscape are better predictors, and therefore we expected a priori studies using such measures would have larger effect sizes. We distinguished two study types, 1) amount-based studies, where wetland amount was calculated as the percent wetland area in a landscape or buffer surrounding the sampled wetland patch, or 2) configuration- based studies, where the configuration of wetland habitat was included in the calculation of the measure, such as the number of wetland patches in a landscape, nearest-neighbour distances, wetland proximity or wetland connectivity. For all study types, we applied the convention that each effect size extracted from a study should represent the population response of a species to increasing wetland amount in a landscape. However, for nearest- neighbour studies, a negative effect of increasing distance indicates that a species responded positively to closer wetlands, or equivalently, greater wetland habitat amount within the surrounding landscape. Therefore, we reversed the sign of the effect sizes extracted for nearest-neighbour studies to make them comparable to those extracted for all other studies representing the response to increasing wetland amount. Second, the relationship between sampling effort used to measure population abundance of a given species and the size of wetlands may influence the effect size observed. Studies where sampling effort increased in proportion to wetland size will observe a positive relationship between wetland size and abundance, simply because more area is searched in larger wetlands. If wetland patch size is positively correlated with total wetland amount in a landscape, this will inflate the effect of wetland amount because a greater amount of wetland was sampled in landscapes containing more wetland. Therefore, we categorized studies by the sampling effort as, 1) area-independent, where sampling effort was consistent across sampled wetlands, 2) area-dependent, where sampling effort increased in proportion to the wetland area, or 3) unknown, where the sampling effort was unknown and could not be obtained by contacting authors. When a study used a combination of more than one of these methods, we selected the sampling method that accounted for the majority of the data. Third, the effect of wetland habitat amount in a landscape could vary depending on whether the sampled wetland was included in the calculation of wetland area in a landscape (wetland amount). Prugh found that measures of habitat amount in a landscape, including area-based buffers, nearest-neighbour distances and connectivity, were better predictors of occupancy when focal patch area was included in the model. Therefore, we expected a priori that studies that did not include the sampled wetland patch area in wetland amount would have a lower effect size compared to studies that did include the sampled wetland patch area. ## Effect Size Calculations We selected the Pearson correlation coefficient *r* as our common estimate of effect size. When a study reported Spearman’s rank correlation coefficient (*ρ*), we converted *ρ* to *r* following. If studies did not report a correlation coefficient, we transformed published test statistics as follows. For studies with continuous measures of population abundance, we extracted *r* values by taking the square root of reported R² values from univariate linear regressions, and adding the sign of the slope. Note that we did not use partial R² values. When raw data were available, we calculated *r* for species with occupancy rates ≥0.7. Data sets with occupancy rates ≤0.7 did not meet normality assumptions of *r*; in this case we converted continuous data to occupancy data to determine an effect size. For studies that measured species occupancy or reported means and variances between two groups (e.g. mean wetland amount in occupied vs. unoccupied landscapes), we first calculated the standardized mean difference (ES_sm) following. We then converted each ES_sm to *r* following. We transformed all correlation coefficients to Fisher’s *z*-scale (ES*_Zr*) following. The next step was to obtain accurate and comparable sample sizes across studies. Meta-analysis weights each study by its inverse variance, based on the assumption that studies with greater precision provide a more accurate estimate of the true effect. The variance of ES*_Zr* is approximated as V*_Zr*  = 1/(*n*-3), where *n* is the total sample size of the study. This gives more weight to studies with larger sample sizes; however, this may overweight studies with pseudoreplication, a common problem in landscape ecology. For example, within a given study area, studies that selected spatially independent wetlands in non-overlapping landscapes at a landscape size (i.e. scale) based on a species’ biology may have a lower apparent sample size than studies that sampled as many wetlands as possible without consideration of spatial independence. In the latter, the sample size would be inflated due to non-independence of sample points because almost identical (i.e. overlapping) landscapes were incorrectly used as multiple independent observations. In addition, if neighbouring species sample points are closer together than the movement range of an individual, the same individual may be sampled more than once, again leading to violation of the assumption of independence. Therefore, we assessed the sample size of each study for pseudoreplication using process similar to that in Rytwinski and Fahrig, as follows. Our assessment was based on the assumption that each data point should represent a spatially independent sample. We assumed that an independent sample was equivalent to an independent individual in a spatially independent sampling location, such that it was unlikely the same individual was sampled at more than one wetland. We considered studies to have independent samples, and therefore accurate sample sizes, in two situations. First, studies that selected non-overlapping landscapes a priori based on the movement range of the species were assumed to be independent samples because it is unlikely that the same individual would be sampled in neighbouring sites. Second, in studies where each landscape represented the area around a sampled individual (e.g. a nest), the number of landscapes was already equivalent to the number of independent individuals. For all other studies, we adjusted sample size. When the distance between two sampling locations (e.g. wetlands) was less than the linear home range or territory size of the species, the two locations were counted as a single sample. For studies that compared population abundance in sampling locations to randomly selected locations where the species was known to be absent, and spatial information on these random locations was not available, the sample size was the number of spatially independent sampling locations plus one (to account for all random locations). After determining the adjusted sample size (*n_adjusted*) of each study, we calculated the inverse variance weight for each ES_Z*r* as *w = n_adjusted* - 3. Studies with *w* \<1 were excluded from the meta-analysis. Refer to and for studies included in the meta-analysis and associated country, species, effect sizes, adjusted sample sizes and study- design categories. ## Species Traits Our main objective was to test if mobility and reproductive rate could explain variation in species responses to wetland habitat loss. We collected data on these species traits from primary literature, theses and published species guides. We estimated mobility using two species traits, home range size and body size, which are both strongly correlated to dispersal distance across species of mammals and birds, independent of their migratory status. Home range size for mammals was indexed as mean annual home range area (ha). For birds, the mean annual breeding territory size (ha) was used for species that forage primarily within their breeding territory. For species that travel away from the nest site to forage (e.g. great blue heron, *Ardea herodias*), home range or foraging distance was reported in the literature rather than breeding territory. In these cases, mean annual home range area (ha) during the breeding season was used as a measure of territory size. When annual home range estimates were not reported, we used the mean foraging distance as a diameter to calculate a circular home range area (ha). For studies conducted during the non-breeding season (e.g. migration, overwintering), we used territory size or foraging distances during the same season when available. For reptiles (turtles and snakes) and amphibians (anurans and salamanders), a cross-species relationship between either home range or body size and dispersal distance has not yet been demonstrated. Therefore, we used adult annual movement ranges, which represented seasonal migration distances between breeding and summer (foraging) or overwintering sites. This is often called a home range, but is conceptually different than a breeding home range or territory in mammals or birds. However, it is the only measure of space use that we could consistently obtain from the literature across all species of reptiles and amphibians. We assumed that species with larger adult annual movement ranges are more mobile in general and thus have larger dispersal distances than species with smaller adult annual movement ranges. Moreover, since body size is generally correlated with dispersal distance across several vertebrate taxa (including amphibians;), it is also reasonable to assume that larger bodied reptile and amphibian species disperse greater distances than smaller bodied species (but see). Therefore, for reptiles and amphibians, we used mean annual home range area (ha) as an index of mobility, preferably estimated using the minimum convex polygon method. When minimum convex polygon estimates were not available, we estimated home range as a circular area (ha) with diameter equal to the mean seasonal migration distance. All home range estimates were averaged between the sexes. Body size was measured as body length (mean body length across both sexes in centimeters); body mass was not used as it was unavailable for 37% of species (primarily amphibians), which represented half of the effect sizes included in the meta- analysis. We estimated reproductive rate as the mean litter or clutch size multiplied by the mean number of litters or clutches per year. Species traits were taken from sources as close to the study region of each study as possible. When trait data were available over multiple years, we used the mean value. Where studies reported a range in values, we used the mid-point of these ranges. For species complexes that were comprised of two species that could not be distinguished, we took the average of the traits of the species comprising the complex. We classified effect sizes by taxonomic group at the class level: mammal, bird, reptile and amphibian, and by order within each taxonomic group. Since there were only two orders within amphibians, we also classified effect sizes by family for amphibians. Refer to for life-history trait values and for information sources. ## Meta-analysis We first conducted a random-effects meta-analysis using the DerSimonian-Laird method to determine the summary weighted-mean effect size of the overall response of wetland species abundance to wetland loss at the landscape scale. Under the random-effects model, the weight assigned (*w\**) to each effect size is the inverse of the sum of two variance components *w\**  = 1/(*w*+*T*²), where *w* (see above) is the unique sampling variance for each study (within-study error) and *T²* is the pooled variance of the true effects across all randomly selected studies (between- studies variance. We also calculated the heterogeneity in true effects (*Q* statistic), which we compared against a chi-square distribution, to test whether the total variation in observed effect sizes (*Q*_T) was significantly greater than that expected from sampling error (*Q*_E). We then tested if moderator variables can explain true variation in the effect sizes (*Q*_M), i.e. *Q*_T = *Q*_M+*Q*_E. All analyses were conducted using the ‘metafor’ package (version 1.7-0) in R 3.0. To test whether mobility, reproductive rate and/or study-level moderators explained a significant amount of heterogeneity in effect sizes, we performed a mixed-effects meta-regression using restricted maximum-likelihood estimation of heterogeneity. Since mobility and reproductive rate information was not available for all species, we removed all effect sizes with missing species trait data to have equal datasets. All species traits were log-transformed to meet test assumptions. We first tested if study design or taxonomic moderators influenced the effect sizes by performing univariate mixed-effects meta-analysis where, if study design or taxonomy explained significant heterogeneity in the effect sizes, we would then subset our data by that moderator variable to control for the effect of study design or taxonomy in analyses of the effects of mobility and reproductive rate. We then performed univariate mixed-effects meta- regression for home range, body length and reproductive rate. We assessed publication bias by a rank correlation (Kendall’s tau) test of the relationship between ES_Z*r* and *n* in association with visual inspection of a scatterplot between these two variables. # Results Although we found more than 200 studies that examined the effect of wetland habitat amount in a landscape on population abundance of wetland vertebrates, only 90 studies met the inclusion criteria. These 90 studies generated 426 effect sizes across 220 species and 16 countries. Studies were predominately from North America (60) and Europe (19), with remaining studies from Australia (5), Central and South America (3), Asia (2) and Africa (1). After removing effect sizes due to lack of information on mobility or reproductive rate, the total number of effect sizes was reduced to 334 across 137 species. The summary weighted-mean effect size from a random-effects meta-analysis across all taxa was 0.11 (95% CI: 0.089, 0.137; n  = 334), indicating an overall weak, positive effect of wetland amount in a landscape on wetland animal population abundance. The overall heterogeneity was *Q*  = 712.03 (p\<0.0001), indicating highly significant variation in species responses to wetland amount. There was no strong evidence of publication bias as there was a weak relationship between effect size and sample size (Kendall's tau  = 0.03, p  = 0.36), and a scatterplot between these two variables showed effect sizes were symmetrically distributed around the summary effect and produced a funnel-shape with greater variation in studies at low sample sizes. Mixed-effects meta-analysis across all taxa (n  = 334) showed that no study design moderator explained any significant heterogeneity in the effects (study type: *Q*_M  = 0.65, p  = 0.42; sampling effort: *Q*_M  = 4.34, p  = 0.11; sampled wetland area: *Q*_M  = 1.26, p  = 0.26;). Therefore we did not control for study design in analyses of the influence of mobility and reproductive rate on species responses to habitat amount. Reproductive rate and body length explained significant heterogeneity in the effects of wetland amount on animal population abundance across all taxa (*Q*_M  = 18.83, p\<0.0001; *Q*_M  = 16.02, p\<0.0001, respectively;). Since the correlation between reproductive rate and body length was high (*r* = −0.72, p\<0.0001, n  = 334), we performed a multiple meta- regression to test for independent effects of each moderator while accounting for the presence of the other (*Q*_M  = 20.76, p\<0.0001;). After controlling for the effect of body length, reproductive rate was negatively related to the effect of wetland amount on population abundance (ES_Z*r* = −0.03; 95% CI: −0.061, −0.002; p  = 0.03;). In other words, species with lower reproductive rates were more sensitive to wetland amount in a landscape than species with higher reproductive rates. In contrast, after controlling for the effect of reproductive rate, the effect of body length was not significant (ES_Z*r*  = 0.06; 95% CI: −0.025, 0.137; p  = 0.15;). Home range size did not explain any significant heterogeneity in effect sizes (*Q*_M  = 0.54, p  = 0.46, n  = 334;). The correlation between home range and reproductive rate was *r* = −0.06, p  = 0.26, and between home range and body length was *r*  = 0.25, p\<0.0001. Since home range did not explain significant heterogeneity and home range size information was missing for 92 of the total 426 effect sizes extracted across all taxa, we removed home range and re-analyzed the effect of body length and reproductive rate with a larger dataset (n  = 421). The multiple meta-regression containing body length and reproductive rate explained a significant amount of heterogeneity in the effect sizes of the larger dataset (*Q*_M  = 25.38, p\<0.0001). Consistent with the analysis with 334 effect sizes (above), reproductive rate was negatively related to the effect of wetland amount on population abundance (ES_Z*r* = −0.03; 95% CI: −0.061, −0.007; p  = 0.02) and body length had no significant effect (ES_Z*r*  = 0.06; 95% CI: −0.021, 0.134; p  = 0.16). Refer to for descriptive statistics of species traits across taxa. The effect of the amount of wetland habitat in a landscape on animal abundance varied by taxonomic class (*Q*_M  = 25.17, p\<0.0001;). The weighted- mean effect size for mammals and birds was greater than that of reptiles and amphibians. The correlation between measures of mobility and reproductive rate differed across taxa, with negative correlations for mammals and birds, and positive correlations for reptiles and amphibians. Since there was large heterogeneity of effect sizes within each taxon separately (birds: *Q*  = 168.96, p\<0.0001, n  = 115; reptiles: *Q*  = 35.32, p  = 0.048, n  = 24; amphibians: *Q*  = 412.24, p\<0.0001, n  = 189), we tested for effects of mobility and reproductive rate within taxa. Refer to for descriptive statistics of species traits for each taxonomic group. For mammals, there were too few effect sizes (n  = 6) to meaningfully test for effects of species traits. For birds, differences in study design did not any explain any significant heterogeneity in effect sizes (study type: *Q*_M  = 0.85, p  = 0.36, sampling effort: *Q*_M  = 0.12, p  = 0.94, sampled wetland area: *Q*_M  = 0.49, p  = 0.45;). Bird effect sizes did not vary significantly by Order (*Q*_M  = 9.89, p  = 0.20;). Reproductive rate and body length explained significant heterogeneity in the effects of wetland amount on population abundance of birds (*Q*_M  = 6.09, p  = 0.014; *Q*_M  = 3.86, p  = 0.049, respectively;). A multiple meta-regression containing both body length and reproductive rate significantly explained heterogeneity (*Q*_M  = 7.75, p  = 0.021). Reproductive rate was negatively related to the effect of wetland amount on bird population abundance and the effect of reproductive rate was marginally significant (ES_Z*r* = −0.25; 95% CI: −0.506, −0.001; p  = 0.050) while the effect of body length was non-significant (ES_Z*r*  = 0.12; 95% CI: −0.056, 0.261; p  = 0.205). We were able to test for the effect of body mass, which was moderately correlated with body length in our bird dataset (r  = 0.47, p\<0.001). Body mass did not explain heterogeneity in effect sizes for birds (*Q*_M  = 0.315, p  = 0.58), nor did home range size (*Q*_M  = 0.052, p  = 0.82). For reptiles (turtles and snakes), response to wetland amount did not vary significantly by Order (*Q*_M  = 0.66, p  = 0.42;). Population-level effects of wetland amount on reptiles varied by study type (*Q*_M  = 6.10, p  = 0.01;); amount-based studies (area-based buffers) had a lower mean-weighted effect size (0.05) than configuration-based (isolation or connectivity) studies (0.31). However, we could not subset our data according to study type due to the small sample size (n  = 24). In any case, in the non- subsetted data, neither mobility nor reproductive rate explained any significant heterogeneity in the effect of wetland amount. For amphibians (anurans and salamanders), population-level effects of wetland amount did not vary significantly by Order (*Q*_M  = 0.183, p  = 0.668), but responses varying by Family was marginally significant (*Q*_M  = 18.081, p  = 0.054;). Therefore, we subsetted our data by Family and tested for effects of mobility and reproductive rate within the largest subset, which was the Family Ranidae (n  = 62). Within the Ranidae subset, no study design moderator or species trait explained significant variation in the effect of wetland amount. In the non-subsetted data, no study design moderator or species trait explained significant variation in the effect of wetland amount. # Discussion Contrary to the widely-held assumption in metapopulation and landscape ecology, our results suggest that dispersal ability is not a useful predictor of species sensitivity to habitat loss, at least for wetland vertebrates. When analyzed across all taxa, we found no evidence to support the prediction that animals with greater mobility were less sensitive to wetland habitat loss than species with lower mobility, when measured as home range size or body length. Analyses within each taxon also showed no effect of home range size or body length on species responses to wetland loss, as well as no effect of body mass for bird responses. Bowne and Bowers point out that despite the putative importance of mobility, there is very little evidence to validate the relationship between movement and population persistence. Consistent with our results, studies that compared the relative influence of mobility to other species traits found that home range or movement distances were weakly related to species responses to habitat loss. Moreover, studies that find strong effects of mobility are generally based on univariate models using indirect indices of mobility. We found that body length explained heterogeneity in species responses to wetland loss in univariate models across all taxa and within birds, but the effect of body length was no longer significant when reproductive rate was controlled for. Consistent with our results, there is generally little empirical support for a relationship between body size and sensitivity to habitat loss in vertebrates. A possible reason for the apparent lack of influence of mobility on species responses to habitat loss is that the effect of mobility varies. While several empirical studies have shown that more mobile species are less sensitive to habitat loss, other studies have found the opposite, that greater mobility decreases tolerance to habitat loss. If our meta-analysis included some species with mobility that positively influenced response to wetland amount and other species with mobility that negatively influenced response to wetland amount, it is possible that the two response types canceled each other out and resulted in no overall effect of mobility. Alternatively, no effect of mobility could have also resulted from a non-linear relationship with species responses to habitat loss. Thomas found that butterflies with intermediate mobility were more vulnerable to habitat loss than butterflies with either low or high mobility. However, a scatterplot of species mobility and species responses to wetland loss was highly scattered in our dataset, indicating a very weak relationship rather than a non-linear one. A second possible reason for the apparent lack of influence of mobility on species responses to habitat loss could be that mobility varies widely with landscape structure, such that the relative rankings of species’ mobility changes as the landscape changes. For example, translocation experiments of two forest bird species in three different landscape types (forested, timber harvested, agricultural) showed that the relative ability of each species to move in a landscape changed depending on landscape context. Ovenbirds (*Seiurus aurocapilla*; forest specialist) had greater return rates than white-throated sparrows (*Zonotrichia albicollis*; forest generalist) in forested landscapes, whereas the opposite was found in harvested or agricultural landscapes. Several studies have shown that species’ movement distances vary depending on landscape context – and that movement is a product of both species traits and landscape structure. For example, home range sizes of northern saw-whet owls (*Aegolius acadicus*;) and elk (*Cervus elaphus*;) increased by an order of magnitude as the amount of forest cover increased in a landscape. Similar to the results of our meta-analysis, Ferraz et al. found no effect of dispersal ability on patch occupancy responses of 55 tropical birds to forest patch isolation. They suggest as a possible explanation that species dispersal abilities change in disturbed landscapes, such that mobility estimated in continuous habitat is not useful to predict occupancy parameters in human-dominated landscapes. Fahrig suggested that species mobility cannot be estimated independently of landscape structure and that to test dispersal ability the landscape context should match the location where movement data were collected. We were unable to test mobility using home range estimates that matched the location of each study included in our meta-analysis since home range estimates are generally very limited. Lastly, it is possible that we did not find an effect of mobility because of high uncertainty in dispersal estimates for vertebrates. We attempted to reduce this problem by using two measures of mobility, home range area and body size, that are known to be highly correlated with dispersal distance. However, the error associated with such estimates is still likely higher than the error associated with estimates of reproductive rate. For example, home range estimates for some species included in the meta-analysis varied by two orders of magnitude both within and between populations. If true, this means we were *a priori* more likely to find effects of reproductive rate than mobility on species sensitivity to habitat loss. Furthermore, although home range size is strongly correlated with dispersal distance in mammals and birds, we assumed this relationship for reptiles and amphibians (see Methods). If the relationship is weaker for reptiles and amphibians than for mammals and birds, body size is a more uncertain measure of mobility for reptiles and amphibians. Body size is commonly used to index sensitivity to habitat loss in animal taxa. However, the exact inference one can make from a cross-species effect of body size ambiguous because body size is simultaneously correlated to several life- history attributes (reviewed in). For example, in addition to its positive correlation with dispersal, body size is positively correlated with area requirements and trophic level, and these relationships occur indirectly through a negative correlation with natural abundances and population fluctuations. We had expected that body size would be the best predictor of wetland vertebrate response to habitat loss because it is an indirect measure of several life- history mechanisms. However, we found that, in wetland vertebrates, the effect of body size was no longer significant once we had controlled for the effect of reproductive rate. Our results provide support for the hypothesis that animals with lower reproductive rates are more negatively affected by habitat loss than are animals with higher reproductive rates. When analyzed across all taxa, reproductive rate was the main explanatory variable for population-level responses to habitat loss, and the effect remained strong when controlling for body size. This suggests that reproductive rate affects species response to habitat loss, independent of its correlation with body size. Other studies have also found a greater effect of reproductive rate than movement-related traits. Simulation studies found that reproductive rate has a much larger effect on the amount of habitat required for population persistence than the per capita rate of emigration and dispersal ability. An empirical test of the relative effects of reproductive rate and mobility at the landscape scale found a strong negative association between reproductive rate and minimum habitat required for a group of dead wood boring beetles, whereas the effect of emigration rate was no longer significant once reproductive rate was controlled for. The only other empirical test of the effect of reproductive rate at the landscape scale that we have found, Vance et al., reported a strong negative cross-species relationship between reproductive rate and the amount of forest in a landscape for forest birds. In line with our results, a recent meta-analysis of road and/or traffic effects across the same vertebrate taxa found that reproductive rate explained a larger amount of variation in population responses of mammals and amphibians to roads than mobility (indexed as home range size) and body size. Both types of landscape change – increased road density or habitat loss – both result in the loss of individuals. Therefore, the mechanism linking the landscape change to reproductive rate is the same: higher reproductive rates compensate for increased mortality and reduce local extinction risk. In contrast, several patch scale studies have found no effect of reproductive rate on species response to patch size and isolation. In a large meta-analysis of patch occupancy of 785 animal species across several taxa, Prugh et al. found no effect of fecundity on species responses to patch area and isolation. Similarly, no effect of reproductive rate was found on patch occupancy rates of 25 mid- and large-sized Neotropical mammals in sites within fragmented compared to continuous forest landscapes or on the number of islands occupied by five lizard species. However, consistent with our results, a meta-analysis of patch area effects on butterfly species richness showed a negative effect of reproductive rate. Patch scale studies may not find an effect of reproductive rate if patch size is not correlated to habitat amount in the landscape. A greater amount of habitat in a landscape represents a greater number of potential colonists available to a patch via the mass effect that is the net flow of individuals from high abundance areas to low abundance areas. For given amount of habitat in a landscape, species with higher reproductive rates will on average produce more colonists than species with lower reproductive rates. This high influx of individuals (immigration) from surrounding habitat would increase local population size with increasing habitat in the landscape. This mechanism would only occur in patch-scale studies if the amount of habitat in the landscape is positively correlated with patch size. On the other hand, the relationship between reproductive rate and species response to habitat loss could occur through a correlation between reproductive rate and another unmeasured variable. For example, reproductive rate was found to be highly correlated with habitat and diet breadth in mammals, and niche breadth (composite of habitat and diet) was found to have a greater influence on patch occupancy rates of mammals and amphibians than body size. Similarly, feeding guild (index of diet) was most strongly related to fragmentation sensitivity in a review of Neotropical vertebrate (including mammals, birds, reptiles and amphibians) responses at the patch-scale compared to body size. However, in both Swihart et al. and Vetter et al., reproductive rate was not tested. Interestingly, a meta-analysis by Newbold et al. found that both generation length (surrogate for annual reproductive rate), habitat and diet breadth were in the top AIC models (ΔAIC \<2) explaining population-level responses of pan-tropical birds to surrounding landuse intensity. This suggests that the mechanism behind the effect of reproductive rate on population response to habitat loss, i.e. greater reproductive output and potential colonists, may act independently of any correlation with niche breadth, at least for birds. We are unable to test for an effect of niche breadth due to the lack of detailed habitat and diet information for most species included in our meta-analysis. We found that wetland mammals and birds were more sensitive to wetland amount in a landscape than were wetland reptiles and amphibians. This difference was also found by Prugh et al. in a meta-analysis on patch area effects. These results may reflect the fact that many wetland reptile and amphibian species require more than one habitat type in a landscape to complete their life-cycle (e.g. foraging, nesting, hibernation) to sustain populations (i.e. landscape complementation). In fact, the amount of forest in a landscape was found to be more important than wetland amount or connectivity for the occurrence of wetland reptiles, and many species of wetland-breeding amphibians have a strong, positive association with the amount of forest in a landscape –. If access to or quantity of complementary habitat is limited, local population sizes will be low despite high wetland amount in a landscape. In fact, reptiles and amphibians were found to be more susceptible to negative effects of roads than mammals and birds. Taken together, these results suggest that wetland reptiles and amphibians are less limited by the amount of wetland habitat in a landscape than are wetland mammals and birds. Other factors, such as landscape complementation or road mortality, may have stronger effects than wetland loss on abundance and distribution of wetland reptiles and amphibians. We present a comprehensive, worldwide review of wetland vertebrate responses to wetland habitat loss at a landscape scale; however some limitations need to be considered. First, although we included study level moderators to control for potential bias at the study level, effect sizes obtained are likely influenced by study area attributes and the scale selected for analysis. Prugh et al. found that patch area and isolation effects varied depending on the predominant land cover in a study area. While we did not find an effect of study area type (natural, agricultural, rural or urban) on responses to wetland amount (*Q*_M  = 4.63, p  = 0.20;), we could not assess whether other landscape variables (e.g. roads) confounded the effect of wetland loss. Moreover, if studies that conduced analysis at multiple scales are more likely to find the scale of effect for a given species, then studies that selected only one scale of analysis may systematically have lower effect sizes. However, we did not find an effect of the number of scales selected by a study on species response to wetland amount (*Q*_M  = 1.48, p  = 0.22;). At the review-level, although we attempted to include unpublished studies such as theses (4 studies) and government reports (2 studies) in our review, published literature was the primary data source in our meta-analysis, representing 93% of studies. While we did not find evidence of publication bias, our review may be biased towards species from geographical areas with high publication rates (North America, Europe). Lastly, since the majority of studies were conducted on birds and amphibians, our mammal and reptile results are less solid. # Conclusions Our synthesis shows that wetland habitat loss at the landscape scale has an overall negative effect on the population abundance of wetland vertebrates across many taxonomic groups and landscape contexts worldwide. Our results support the hypothesis that species with lower reproductive rates are more negatively affected by landscape-scale habitat loss than are species with higher reproductive rates. Surprisingly, we found no evidence that mobility influences species response to habitat loss. This implies that immigration and colonization rate is more strongly related to reproduction, which determines the total number of potential colonists, than it is to a species’ intrinsic mobility. From a conservation management perspective, our results suggest that priority should be placed on species with low reproductive rates. Also, our results suggest that conservation plans for declining wetland species should focus on actions aimed at increasing reproductive output, such as providing artificial nesting substrates or managing local wetland variables that increase reproductive success of target species (e.g. hydroperiod ; vegetation structure –). # Supporting Information We are grateful to all of the wetland researchers cited in Reference List S1 who responded to our inquiries and/or provided data. We also thank K. Henle and an anonymous reviewer for valuable comments that improved the manuscript. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: PEQ KEL LF. Performed the experiments: PEQ. Analyzed the data: PEQ. Wrote the paper: PEQ KEL LF.

# Introduction Intravenous thrombolysis with alteplase (recombinant tissue plasminogen activator) is the most effective therapy for acute ischemic stroke. However, the potential increased risk of thrombolytic-related hemorrhagic transformation (HT), including Asymptomatic Hemorrhagic Transformation (AHT), limits its application. The incidence of AHT after intravenous thrombolysis was found to be 9.9% from combined CT-based data of the NINDS rt-PA trial and the ATLANTIS trial. Thrombolytic-related symptomatic HT is associated with worse functional outcome. However, it is controversial whether AHT directly leads to negative prognosis. Some studies suggest that prognosis of patients with asymptomatic HT may be innocuous; some studies confirmed that patients with AHT compared with those without HT had a negative effect on functional outcome. Therefore, we hypothesized that thrombolytic AHT might be associated with negative neurologic recovery. The aim of this study is to identify risk factors of thrombolytic- related AHT as well as to examine the influence of AHT on short-term and long- term clinical outcome with TIMS-China registry. # Materials and Methods ## Subjects All of data were collected from the Thrombolysis Implementation and Monitor of acute ischemic Stroke in China (TIMS-China) registry, a national prospective stroke registry of thrombolytic therapy with intravenous alteplase in patients with acute ischemic stroke in China. In nationwide, 67 centers participated in this stroke registry between May 2007 and April 2012. Demographics on clinical characteristics, cranial CT scans, medical therapy and thrombolysis information were collected. All patients were followed up for 3 months. Eligible patients received intravenous thrombolysis within 4.5-hour. Patients with incomplete data and thrombolysis time window\>4.5h or unknown were excluded. Patients with NIHSS score ≥25were excluded. Patients with symptomatic HT were excluded from this study. Eligible patients were divided into two groups: the AHT group, which had asymptomatic hemorrhagic transformation; and the non-HT group, which had no thrombolytic-related hemorrhagic transformation. ## Ethics Statement The registry of TIMS-China was approved by ethics committee of Beijing Tiantan hospital in 2006. The data from the TIMS-China registry was anonymous before access and analysis by all researchers and patient treatment was assigned and determined independent of all researchers. All patients had written informed consent before being entered in the registry. ## Outcome analysis Follow-up CT scans for patients were collected at 24–36 hours and 7-day. The diagnosis of HT on brain images was determined independently by 2 neurologists blinded to clinical data. If there was a disagreement between 2 neurologists, a third neurologist was consulted, and a consensus decision was reached. Hemorrhagic transformation must be scored according to the ECASS II classification during all follow-up CT scans: hemorrhagic infarcts (type 1 or 2) or parenchymal hematomas (type 1 or 2). HT was also often divided into symptomatic HT and AHT groups based on the deterioration in neurologic status. AHT, according to ECASS II criteria, was defined as NIHSS score less than 4 points worsening from baseline in neurological status in the first 24 hours after thrombolysis with the presence of hemorrhagic transformation on the follow-up CT scan at 24-36h. The primary outcome measure collected included functional outcome at 90-day (functional independence was defined as a modified Rankin Scale (mRS) score of 0 to 6), NIHSS at 7-day, and mortality at 7-day and 90-day. ## Statistical Analysis Results were reported as mean ± SD or the frequency of categorical variables. Pearson chi-square (χ2) test was used for comparisons of categorical variables. Continuous variables were analyzed using t test or Mann-Whitney U test. Percentage proportions of outcome events were calculated by dividing the number of events by the total number of patients. For each outcome variable, a separate multivariable logistic regression was performed. Odds ratios (ORs) with its 95% confidence intervals (CI) were calculated by using non-HT group after thrombolysis as the reference group. Two-sided values of P less than.05 were considered statistically significant. All statistical analyses were performed using SAS software version 9.3 (SAS Institute Inc, Cary, NC). # Results 904 patients (62.8%) were enrolled from a patient population of 1440 from 67 centers across TIMS-China. Our exclusion criteria were as follows: (1) incomplete neuroimaging workups (n = 40), (2) loss to follow-up (n = 36), (3) thrombolysis time window\>4.5h or unknown (n = 312), (4) symptomatic HT (n = 24), (5) Patients with severe stroke with a NIHSS score \>25(n = 93). It had been reported that 89(9.6%) patients were transformed to AHT in China, 24 (2.5%) patients had symptomatic HT. The characteristics of the patients with AHT group and non-HT group after intravenous thrombolysis are shown in, which indicates that more patients with the history of atrial fibrillation, high NIHSS score and elderly age are with AHT after thrombolytic therapy and fewer patients with a history of transient ischemic attack are with AHT. Proportion of patients transformed with asymptomatic hemorrhage after thrombolytic therapy were as follow: HT2 subtype highest proportion (38 cases) \> followed HT1 subtype (30 cases) \> PH1 subtypes (15 cases) \> PH2 subtypes (6 cases). NIHSS score of asymptomatic patients with hemorrhagic transformation after thrombolysis (15.56±6.61) was higher than that of the patients without bleeding transformation after thrombolysis (11.96±6.90) (P\<0.001). The proportion of poor outcome in patients with AHT was 13.48% at 3 months after a stroke onset and the patients with non-HT was 11.04%. AHT deteriorated clinical outcomes of 90-day of follow-up (mRS score (0–1 grade) and mRS score (0–2) in patients after thrombolysis when we analyze the data without any adjustment for covariates. After adjustment for baseline relevant factors, no significant difference was found on the odds of 7-day (95% CI:0.692 (0.218–2.195), (P = 0.532) or 90-day mortalities (95% CI:0.548 (0.237–1.268), P = 0.160) and modified Rankin score(0–1) at 90-day (95% CI:0.798 (0.460–1.386), P = 0.423) or modified Rankin score(0–2) at 90-day (95% CI:0.732 (0.429–1.253), P = 0.116) or modified Rankin score(5–6) at 90-day (95% CI:0.375 (0.169–1.830), P = 0.116) between two groups. In AHT subgroup analysis, there was a more favorable clinical outcome in HI1and HI2 subgroups than PH1 and PH2 subgroups. # Discussion There were remarkably low rates of asymptomatic hemorrhagic transformation (4.5%) in the National Institute of Neurological Disorders and Stroke (NINDS) rtPA Stroke Study. A large observation study of intravenous thrombolysis, the Safe Implementation of Thrombolysis in Stroke-Monitoring Study (SITS-MOST), also showed low rates of asymptomatic hemorrhagic transformation (9.6%). In TIMS- China, its rate was 9.5%. However, there were higher rates of asymptomatic hemorrhage (39.6%) in the European Cooperative Acute Stroke Study (ECASS-II). Compared with ECASS-II, the earlier initiation of thrombolysis is the possible reason for the low asymptomatic hemorrhage rates in NINDS, SITS-MOST and TIMS- China. In addition, the direct comparison of the AHT rates among different studies may be hampered by variability in definitions, the earlier initiation of thrombolysis and different image examination time. In non-HT group, more patients had TIA history compared with AHT group, which was mainly explained by recurrent TIA increasing collateral circulation. In Asia, many hospitals choose a low-dose intravenous alteplas(IV-tPA) regimens because they believe that racial differences need to reduce dose of treatment in order to decrease rate of hemorrhage transformation. The result of J-MARS showed that 0.6 mg/kg IV-tPA achieved low rates of HT. TIMS-china registry demonstrated the standard-dose IV- tPA therapy was safe for Chinese patients with stroke, without any increased rate of HT or mortality compared with the low-dose Group. In our study, there is no difference of the dose IV-tPA between AHT group and non-HT group. Previous some studies have shown that thrombolytic-related AHT does not have a negative effect on the short-term or long-term functional outcome. AHT after thrombolysis has a suggestive of vascular recanalization and effective treatment. Moreover, it seems to be a beneficial treatment choice that antiplatelet agents are administered for AHT patients after intravenous thrombolysis. Our novel findings are that among patients who suffered acute ischemic stroke receiving intravenous thrombolysis, the risk of poor outcome for patients with AHT was 0.375-fold at 3 months than for patients without any HT; the risk of mortality for patients with AHT was 0.548-fold at 3 months than patients without any HT. These findings show that clinical outcomes after thrombolysis do not deteriorate in the presence of AHT. Moreover, there is a favorable trend in AHT group, which may be due to improved reperfusion after thrombolysis. Conversely, some studies suggest AHT may not be benign. A recent study showed AHT maybe have a negative effect. This differential effect may be due to the AHT populations coming from all ischemic stroke patients, while the AHT patients in our study are from ischemic stroke patients receiving intravenous thrombolysis. Additionally, the use of antiplatelet agents is not prescribed to some patients with AHT in above-mentioned study, which results in increased risk of ischemic stroke one-year recurrence. Our study suggests that elderly, cerebral embolism and higher NIHSS score increase risk of AHT. Similarly, previous studies have linked the factors to increased risk of HT. Different from mechanism of symptomatic HT, we speculate that AHT may link with small artery formation rather than large artery formation. It has been showed that thrombolysis after acute ischemic infarction may lead to artery reopening and collateral circulation reperfusion6. Moreover, the formation of collateral circulation may also improve the functional outcome. There is a favorable trend after thrombolysis-related AHT in our study supports this suggestion. Our study may have the following limitations. (1) Our sample size was relatively small, and our data exclude severe stroke(NIHSS\> or = 25), which maybe have a subtle effect (either positive or negative) of AICH outcome. (2) Another limitation is that the study did not follow the randomized, double-blind principles. Despite these limitations, our results have implications that post- thrombolytic AHT in China does have not a negative effect on clinical prognosis. # Supporting Information Special thanks to the physicians as well as all other personnel who dedicated themselves to the TIMS-CHINA project. We thank all participating hospitals, relevant clinicians, statistician, and imaging and laboratory technicians. The TIMS-CHINA investigators include: Yongjun Wang (Principle investigator), Beijing Tiantan Hospital; Xingquan Zhao, Beijing Tiantan Hospital; Yilong Wang, Beijing Tiantan Hospital; Xiaoling Liao, Beijing Tiantan Hospital; Chunxue Wang, Beijing Tiantan Hospital; Liping Liu, Beijing Tiantan Hospital; Chunjuan Wang, Beijing Tiantan Hospital; Yuesong Pan, Beijing Tiantan Hospital; Qi Bi, Beijing Anzhen Hospital; Liying Cui, Peking Union Medical College Hospital of Peking University; Yuheng Sun, Beijing Jishuitan Hospital; Maolin He, Beijing Shijitan Hospital; Dongsheng Fan, Peking University Third Hospital; Xiaojun Zhang, Beijing Tongren Hospital; Yansheng Li, Shanghai Renji Hospital; Shaoshi Wang, Shanghai First Municipal People’s Branch hospital; Wei Fan, Zhongshan Hospital of Fudan University; Chuancheng Ren, The Fifth People’s Hodpital of Shanghai, Fudan University; Zhenguo Liu, Xin Hua Hospital Affiliated to Shanghai Jiao Tong University; Xiaojiang Sun, The sixth People’s Hospital Affiliated to Shanghai JiaoTong University; Xu Chen, Shanghai 8th People’s Hospital; Qingke Bai, Pudong New Area People’s Hospital; Dexiang Gu, Shanghai Yangpu Area Shidong Hospital; Chunmei Hu, Shanghai Baoshan Area Center Hospital; Xin Li, Shanghai Yangpu Area Center Hospital; Qiang Dong, Huashan Hospital of Fudan University; Yan Cheng, Tianjin Medical University Gengeal Hospital; Bin Li, Dagang Oilfield Gengeal Hospital; Chen Li,Tianjin Fifth Center Hospital; Tongyu Wang, Bohai Oilfield Hospital; Liping Wang,Ninghe County People’s Hospital of Tianjin; Kun Zhao, Baodi District People’s Hospital of Tianjin; Dingbo Tao, The First Afflicated Hospital of Dlian Medical University; Fang Qu, Dalian Second People’s hospital; Jingbo Zhang, Dalian Third People’s hospital; Jianfeng Wang, Dalian Central hospital; Ying Lian, Dalian Economic and Technological Development District Hospital; Fang Qu, Shenying Military District General hospital of Chinese People’s Liberation Army; Ying Gao, National Traditional Chinese Medicine (TCM)Thrombus Treatment Center of Liaoning Province; Huashan Sun, Jilin Chemical Industrial Group General hopital; Jinying Li, Jilin Oilfield General Hospital; Guozhong Li, The First Clinical College of Harbin Medical University; Yulan Zhu, The Second Clinical College of Harbin Medical University; Zichao Yang, The Fourth Clinical College of Harbin Medical University; Jun Zhou, Mudan Jiang Second hospital; Minxia Guo, Shaanxi Provincial People’s Hospital; Qilin Ma, The First Hospital of Xiamen; Xiaoping Gao, Hunan Provincial People’s Hospital; Renbin Huang, Chenzhou First People’s Hospital; Bo Xiao, Xiangya Hospital of Centre-south University; Kangning Chen, Southwest Hospital; Li Gao, Chengdu Third Municipal People’s Hospital; Anding Xu, The First Affiliated Hospital of Jinan University; Ming Shao, The First Affiliated Hospital of Guangzhou Medical University; En Xu, The Second Affiliated Hospital of Guangzhou Medical University; Xiaoping Pan, Guangzhou First Municipal People’s Hospital; Yefeng Cai, Guangdong Provincial Hospital of Traditional Chinese Medicine; Yun Xu, Drum Tower Hospital Affiliated to Nanjing Medical University; KaiFu Ke, The Affiliated Hospital of Nantong University; Yuenan Kong,Wuxi Second People’s Hospital; Liangcai Ding, The Third Affiliated Hospital of Suzhou University; Yibin Cao, Tangshan Gongren Hospital; Yumin Liu, Zhongnan Hospital of Wuhan University; Kang Xu, Hubei Xinhua Hospital; Chengming Xing, Qingdao Municipal Hospital Group; Shaohua Su, Dezhou People’s Hospital; Guiru Zhang, Penglai People’s Hospital; Rongyuan Zheng, The First Affiliated Hospital of Wenzhou Medical University; Ming Zhai, First People’s Hospital of Yunnan Province; Yi Zhu, The Xinjiang Autonomous Region People’s Hospital; Yuanxin Liu, Autonomous Region TCM Hospital Affiliated to Xinjiang Medical University; Xiaoying Zhang, Hospital of Xinjiang Production & Construction Corps; Shizheng Wu, Qinghai Provincial People’s Hospital; Jinfeng Liu, Yangquan Coalmine Group General hospital. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: YW WJ LZ. Performed the experiments: WJ XL TC. Analyzed the data: WJ YP YW. Contributed reagents/materials/analysis tools: YP. Wrote the paper: WJ XL. [^3]: ¶ Membership of The TIMS-CHINA investigators is provided in the Acknowledgments.

# Introduction Maintenance of tissue homeostasis is a tightly regulated process and the loss of responsiveness to negative growth signals can alter this delicate balance, leading to cancer. This is especially evident in the breast, where epithelial cells undergo cycles of quiescence, proliferation, differentiation, and apoptosis during the menstrual cycle and as a result of pregnancy. Tight negative growth control of the mammary epithelial compartment is crucial, and disruption of the balance between mitogenic and anti-growth signals can leave this tissue susceptible to the formation of cancer. Therefore, delineating how negative growth control of breast epithelial cells is lost during tumor formation is essential to understand the pathogenesis of breast cancer. Insensitivity to negative growth signals is considered a hallmark of cancer cells. This means that the mechanism that conveys growth inhibiting signals from DNA damage or transforming growth factor β (TGF-β) to the cell cycle machinery is universally disrupted. In general, growth inhibiting signals arrest the cell cycle by blocking cyclin dependent kinase (CDK) activity through the actions of cyclin dependent kinase inhibitors (CKIs) such as p15Ink4b, p16Ink4a, p21Cip1, and p27Kip1. This in turn activates the retinoblastoma protein (pRB). In cell culture, disruption of an individual CKI, such as p21Cip1, can result in deregulated proliferation despite signals from negative growth regulatory stimuli such as DNA damage. However, negative growth regulators such as TGF-β are capable of inducing a cell cycle arrest in the absence of any one of these CKI proteins. Furthermore, ablation of individual CKI genes in mice has no effect on viability and is accompanied by surprisingly few developmental abnormalities. Interestingly, each knockout mouse strain develops tumors in a specific subset of tissues, but none of these mice develop mammary tumors,. In fact, pRB itself was previously considered dispensable for mammary development. This is surprising because pRB pathway components such as cyclin D1 are commonly amplified, and pRB itself is sometimes lost in human breast cancers and these are direct targets of CKI regulation. Thus, even though the current body of literature suggests that anti-growth effects greatly influence mammary epithelial cells, mouse models reveal layers of complexity that make it challenging to understand how negative growth responses protect mammary epithelial cells from aberrant proliferation and ultimately tumorigenesis. Activation of pRB maintains cells in G0/G1 and prevents inappropriate cell cycle entry. Breast epithelial cells frequently respond to anti-growth signals from DNA damage, exogenous growth factors like TGF-β, and other cellular stresses. All of these serve to activate pRB and arrest proliferation. Using a knock-in mouse model with a discrete defect in the pocket domain of *Rb1* (*Rb1^ΔL*), we have reported the remarkable discovery that pRB uses interactions with LXCXE-motif-containing proteins almost exclusively for growth arrest in response to stressful negative growth stimuli, and this mechanism is largely dispensable for cell cycle control in development allowing us to obtain viable animals with this defect. The lone circumstance in development where pRB interactions with LXCXE containing proteins, such as histone deacetylases, are critical is in mammary epithelium. In this tissue, ductal morphology and branching are normal, but there is mild hyperplasia of the luminal cell compartment. Our data indicates that, the *Rb1^ΔL* mutation results in an inability to silence transcription of E2F target genes and renders cells insensitive to anti-growth signals, including TGF-β, DNA damage, and other senescence cues. Furthermore, cells from these mice fail to induce growth arrest in response to ectopic expression of the CKI proteins p16Ink4a and p21Cip1. This suggests that there are likely many stress-induced negative growth signals to which cells from these mice are uniquely resistant. Since pRB-dependent negative growth control is dramatically reduced in the mammary glands of *Rb1^ΔL* mice, but viability and overall development are normal, these animals provide a unique opportunity to examine the impact of negative growth signals in suppressing mammary cancer. In order to understand the role of pRB LXCXE-dependent negative growth control during mammary tumorigenesis, we have crossed our mice with a *Wap-p53^R172H* transgenic strain. *Wap- p53^R172H*; *Rb1^ΔL/ΔL* females developed mammary tumors more frequently and at a significantly younger age than control mice, while showing no differences in histopathological characteristics or metastasis. This strongly suggests that loss of negative growth regulation in this model exacerbates cancer development at a very early stage. In contrast, co-expression of *Neu* and *Rb1^ΔL* did not accelerate mammary tumor initiation, progression, or metastatic dissemination. Surprisingly, the contrasting data between the two transgenic mammary tumor models indicates that failure to respond to negative growth signals by pRB affects cancer incidence in a context-dependent manner. # Results ## pRB-LXCXE interactions act as an initial barrier to tumor formation Our lab has previously generated a knock-in mouse model (*Rb1^ΔL*) to disrupt the LXCXE binding cleft on the retinoblastoma tumor suppressor protein. Loss of LXCXE-dependent interactions, between pRB and enzymes such as histone deacetylases, disrupts negative growth control in the mammary gland during development. This leads to hyperplasia of the ductal epithelium that is ubiquitously detectable in virgin animals between four and 16 weeks of age. The *Rb1^ΔL* mutation alone does not lead to spontaneous mammary cancer, perhaps because this mutation doesn't increase the cellular proliferation rate, but alters the response to negative growth cues. Since hyperplasia of ductal epithelia is considered a risk factor for human breast cancer, we postulated that the *Rb1^ΔL* mutation would exacerbate cancer pathogenesis when combined with other oncogenic changes that provide a proliferative signal. To explore this possibility, we first performed a soft agar colony assay using *Rb1^+/+* and *Rb1^ΔL/ΔL* mouse embryonic fibroblasts (MEFs) expressing a dominant negative form of p53 and an oncogenic allele of Ras (pLXSN dn p53/Ras^V12). Expression of these oncogenes resulted in a greater frequency of colonies in *Rb1^ΔL/ΔL* MEFs compared with controls. Furthermore, many *Rb1^ΔL/ΔL* colonies were also larger than control colonies, suggesting that cells lacking pRB-LXCXE interactions were able to transform earlier. Since the effect of the *Rb1^ΔL* mutation is exerted after only a few cell divisions, we interpret this phenotype to be caused by a defect in negative growth control rather than genome instability, which can also result from expression of the *Rb1^ΔL* allele. This provided proof of principle that pRB-LXCXE-dependent anti-growth effects can protect cells from oncogenic transformation. To validate that pRB-LXCXE interactions can exert a tumor suppressive effect in the mammary gland, we crossed our mice into the *Wap-p53^R172H* background. *Wap-p53^R172H* is a dominant negative form of p53 driven by the *whey acidic protein* promoter, which is expressed in the mammary gland during pregnancy and lactation. We bred cohorts of *Wap-p53^R172H*; *Rb1^+/+* and *Wap-p53^R172H*; *Rb1^ΔL/ΔL* females through five rounds of pregnancy in order to induce transgene expression. Since *Rb1^ΔL/ΔL* females are frequently unable to nurse their pups, all live pups were removed two days after birth to ensure consistent timing of transgene expression between genotypes. *Wap-p53^R172H* expression leads to genomic instability, so we reasoned that expression of *p53^R172H* during pregnancy and lactation would create random mutations. We expected that some of these mutations would drive tumorigenesis later, after the transgene was turned off, and this would allow us to assess how a diminished response to negative growth regulators affects mammary tumorigenesis. Both *Wap-p53^R172H*; *Rb1^+/+* and *Wap-p53^R172H*; *Rb1^ΔL/ΔL* females developed high grade mammary adenocarcinomas, characterized by high cytological variability. Many cells exhibited large cellular and nuclear size which is comparable to other studies using *Wap-p53^R172H* mice, indicating that the transgene was active in both genotypes. Furthermore, tumors from both genotypes displayed high but comparable rates of mitosis and similar rates of tumor growth ( and data not shown), demonstrating that the rate of proliferation was similar for *Wap-p53^R172H*; *Rb1^+/+* and *Wap-p53^R172H*; *Rb1^ΔL/ΔL* tumors. While mice from both genotypes developed similar types of tumors, we found an increased frequency of tumor formation in *Wap-p53^R172H*; *Rb1^ΔL/ΔL* females. 63.6% of *Wap-p53^R172H*; *Rb1^ΔL/ΔL* females compared to 44.4% of *Wap-p53^R172H*; *Rb1^+/+* females developed tumors over the course of the study. Importantly, loss of pRB-LXCXE interactions in the *Wap-p53^R172H* background also resulted in earlier tumor onset as detected by mammary palpation (Log rank test, *P* = 0.0238). Like the data from our soft agar colony assay, this suggests that pRB-dependent anti-growth control can act as a barrier to tumor initiation, and when it is lost in *Rb1^ΔL/ΔL* females *Wap-p53^R172H* can then drive tumour progression. To explore this concept further, we examined tumor-free mammary glands from our tumor-burdened mice. Some mammary glands had extensive lobuloalveolar development, preventing an assessment of abnormal proliferation by whole mount staining. However, we examined glands where tumors were not palpable among the remaining necropsied animals, and discovered that there was a small increase in the number of hyperplastic lesions in the *Wap-p53^R172H*; *Rb1^ΔL/ΔL* mice. Together with the increased frequency of tumors and shortened timing of tumor onset in *Wap-p53^R172H*; *Rb1^ΔL/ΔL* females, these data indicate that pRB-dependent anti-growth control acts as an initial barrier to tumor formation in the mammary gland. In addition, both *Wap-p53^R172H*; *Rb1^+/+* and *Wap-p53^R172H*; *Rb1^ΔL/ΔL* tumors were able to metastasize to the lungs, and there were no major differences in the appearance of the metastases, or the number, or size of these metastases. This is distinct from previous reports demonstrating that loss of pRB-mediated genomic stability leads to more aggressive tumors and increased metastasis. Because the frequency of cancer incidence and age of onset are exacerbated by the *Rb1^ΔL* mutation, but aspects of progression such as proliferation rate and metastasis are indistinguishable from controls, we conclude that reduced responsiveness to negative growth signals likely functions as a barrier to tumor initiation at a very early stage in cancer pathogenesis. ## Retinoblastoma-dependent negative growth effects on tumorigenesis are context-dependent The advantage of the *Wap-p53^R172H* model is that it introduces random mutations and this creates a selection for ones that can cooperate with defects found in *Rb1^ΔL/ΔL* mice. However, this also prevents us from knowing what the initiating oncogenic mutations were and how they normally engage negative growth responses that activate a pRB-LXCXE-dependent arrest. For this reason we also used a transgenic line that expresses a dominantly acting oncogene so that the origin of oncogenesis would be known. To determine how pRB- dependent responses to negative growth signals affect mammary tumorigenesis in this context, we crossed our mice into the *Tg(MMTVneu)202Mul* (herein referred to as *Neu*) background, where expression of the rat proto-oncogene *Neu* is driven by the *MMTV* promoter. These mice normally develop focal mammary tumors with a high rate of metastasis to the lung. We chose this transgenic line because *Neu* is known to activate the Ras pathway and our data in indicates that pRB-dependent growth arrest opposes it. Furthermore, expression of activated *Neu* in mice with disrupted TGF-β signaling results in reduced tumor latency. Conversely, when crossed to mice that overexpress TGF-β or a constitutively active TGF-β receptor, primary tumor formation is delayed or tumor growth is slowed. This anti-tumor effect is commonly attributed to TGF-β- induced cell cycle arrest, although this aspect of its signaling has not been testable in isolation before now. For these reasons *Neu* mice offer a relatively well characterized system in which to determine how important negative growth signalling is in opposing mammary cancer formation. Since TGF-β is a key negative regulator of proliferation in the mammary epithelial compartment and induces a G1 arrest in a pRB-LXCXE-dependent manner, we next wanted to characterize the *Neu*; *Rb1^ΔL/ΔL* genotype to ensure that this experimental system would allow us to address the role of negative growth responses during tumorigenesis as we expected. To this end we tested *Rb1^+/+*, *Rb1^ΔL/ΔL*, and *Rb1^ΔL/+* mammary epithelial cells (MECs) for their ability to respond to TGF-β–induced G1 arrest. This confirmed our previous results that TGF-β's cytostatic response is vastly diminished in *Rb1^ΔL/ΔL* cells. Surprisingly, *Rb1^ΔL/+* MECs have a similar defect in TGF-β growth control indicating that mutation of only one copy of *Rb1* is sufficient to abrogate its arrest mechanism. Consistent with this observation, examination of mammary epithelia in *Rb1^ΔL/+* virgin female mice revealed they have a similar degree of hyperplasia as we have previously reported for *Rb1^ΔL/ΔL* mice. The proto-oncogenic *Neu* in these young mice needs to be activated by a deletion in its extracellular domain, so the growth signals that will drive cancer formation are not evident at this early stage. Because of the defective response to negative growth signals caused by the *Rb1^ΔL* mutation, these mice create an ideal opportunity to assess how an oncogenic growth signal is opposed. Importantly, this can be determined by comparing *Neu* transgenic mice with wild type *Rb1* to either *Rb1^ΔL/+* or *Rb1^ΔL/ΔL* animals. To assess the importance of pRB-dependent negative growth control in suppression of primary tumor formation and growth, we followed cohorts of *Neu*; *Rb1^+/+*, *Neu*; *Rb1^ΔL/+* and *Neu*; *Rb1^ΔL/ΔL* females throughout their natural lives and palpated them weekly to determine the onset of mammary tumor formation. Unfortunately, the long latency before tumor formation resulted in excessive grooming in many of our mice and the need to euthanize them before palpable tumors formed. This was particularly true of the *Neu*; *Rb1^ΔL/ΔL* mice. However, the *Neu*; *Rb1^ΔL/+* animals have a similar response to TGF-β, indicating that they offer the same insight into how negative growth signals impact tumorigenesis in this transgenic model. In contrast to the *Wap-p53^R172H*; *Rb1^ΔL/ΔL* mice, there was no difference in tumor latency between the remaining *Neu*; *Rb1^ΔL/ΔL* females and females from the other two genotypes. The frequency of tumorigenesis in the *Rb1* mutant genotypes was also relatively unchanged from wild type (85.7% for *Rb1^ΔL/ΔL* and 77.4% for *Rb1^ΔL/+* vs. 90% for wild type). This result suggests that negative growth regulatory signals that depend on pRB do not significantly influence cancer pathogenesis in *Neu* transgenic mice. Because this was unexpected, we also investigated other tumor characteristics to determine if the *Rb1* mutant genotypes altered the tumor type of these mice in such a way that the direct comparison in is misleading. To this end, we classified the tumors histologically and discovered that they all fit the characteristics of solid or acinar carcinomas that have been reported previously for *Neu* mice, indicating that transgene function was not altered in the different genotypes. Our expectation from the *Wap-p53^R172H* cross is that negative growth responses are most important at the initiation step. However, Muraoka *et al.* found that overexpression of TGF-β did not affect tumor latency of *Neu* mice, but instead reduced tumor proliferation. For this reason we measured the number of mitotic figures in five randomly selected microscopic fields for each tumor as a means to compare proliferation and this revealed no significant differences. Furthermore, there were no significant differences in the final tumor volume. Lastly we investigated unaffected mammary glands from tumor-burdened animals for evidence of non-malignant nodules by whole mount preparations. Again, there were no statistically significant differences between the three *Rb1* genotypes and if anything there was a trend toward fewer nodules in mice bearing the *Rb1^ΔL* mutation. These findings are in stark contrast to those for the *Wap-p53^R172H* cross, where there was an earlier tumor onset in *Wap-p53^R172H*; *Rb1^ΔL/ΔL* females and slightly more hyperplastic nodules. Together, these data indicate that loss of pRB-dependent anti-growth control exacerbates *Wap-p53^R172H*–dependent tumor formation, but leaves *Neu*-driven tumorigenesis unaltered. This indicates that during cancer formation, the response to anti-growth signals by pRB is context-dependent. In an effort to better relate the combination of the *Wap-p53^R172H* and *Neu* transgenes with our *Rb1* mutant, we also investigated metastasis in *Neu*; *Rb1^+/+*, *Neu*; *Rb1^ΔL/+*, and *Neu*; *Rb1^ΔL/ΔL* female mice. This revealed that the number of lung surface metastases that formed during the 60 day period from initial palpation to euthanasia were similar. Furthermore, these metastatic lesions occupied a similar proportion of tissue volume when quantified microscopically in lung sections. Lastly, there were no apparent differences in histology between metastases from the respective genotypes. From these experiments it is clear that the *Rb1^ΔL* allele does not enhance the metastatic potential of mammary tumors whether they form in the *Neu* or *Wap-p53^R172H* backgrounds. This reveals that pRB can confer responsiveness to negative growth signals that limit p53^R172H tumorigenic effects. Surprisingly, this aspect of pRB function does not affect *Neu*-driven oncogenesis. Unexpectedly, these experiments reveal that loss of pRB-dependent negative growth responsiveness does not ubiquitously contribute to cancer formation in the mammary gland. # Discussion Using two transgenic mouse models of breast cancer, we have examined the importance of pRB-LXCXE interactions during cancer formation and progression. Our work has revealed that LXCXE-dependent anti-growth control can act as a barrier to oncogenic transformation in the mammary gland. Surprisingly, the *Rb1^ΔL* mutation exacerbated *Wap-p53^R172H*-induced tumor formation, while having no effect in the *Neu* transgenic background. The retinoblastoma protein has several important roles in the cell including induction of a cell cycle arrest in response to various stresses, as well as maintaining genome stability. While it is possible that the enhanced tumor phenotype observed here is the result of increased genomic instability, we do not consider this a likely possibility for several reasons. Recent work demonstrated that loss of pRB-mediated genome integrity manifests as more aggressive tumors characterized by altered histopathology compared to controls. Furthermore, they also exhibited increased metastasis. In contrast, we saw no change in tumor size, proliferation, or metastasis in *Wap-p53^R172H*; *Rb1^ΔL/ΔL* mice. The increase in tumor penetrance and earlier onset in the *Wap-p53^R172H*; *Rb1^ΔL/ΔL* background are more consistent with an inability to induce a growth arrest. We posit that pRB acts as a braking mechanism, protecting the cell from aberrant growth in response to stresses in this transgenic model. Mutations to the LXCXE binding cleft disrupt this braking system, leaving cells vulnerable to oncogenic stimuli, such as those generated in the *Wap-p53^R172H* model. The precise pRB LXCXE-dependent anti-growth mechanism that protects against *Wap-p53^R172H* driven tumors is unclear since mutant p53 acts as a generator of mutations that drive tumorigenesis almost a year after the transgene has been silenced. However, the exact source of anti-growth signals that fail to be received in *Wap-p53^R172H*; *Rb1^ΔL/ΔL* mice is not the key question; rather, the *Rb1^ΔL* mutation allows us to determine the circumstances when the response to negative growth signals are most important. We interpret our data to indicate that pRB responds to these signals to oppose oncogenic transformation in mammary epithelial cells at an early stage in tumorigenesis, before tumors become palpable. Several lines of evidence support this conclusion. First, prior analysis of *Rb1^ΔL/ΔL* fibroblasts demonstrates that cell size and length of G1 and S phases are similar to wild type cells, indicating that this *Rb1* mutant is phenotypically distinct from the increased rate of proliferation found in *Rb1^−/−* cells. Instead, *Rb1^ΔL/ΔL* cells have a defect in inducing cell cycle arrest in response to anti-growth signals. Second, in both the *Wap-p53^R172H* and *Neu* crosses, the rates of proliferation were comparable between control tumors and those lacking pRB-LXCXE interactions. In contrast, the median age of tumor onset was significantly lower for *Wap-p53^R172H*; *Rb1^ΔL/ΔL* females, compared to control mice. Similarly, only *Wap-p53^R172H*; *Rb1^ΔL/ΔL* tumor- free mammary glands displayed an increase in hyperplastic nodule formation. This suggests that the *Rb1^ΔL/ΔL* mutation impacts tumorigenesis at an early stage, rather than increasing proliferation in existing tumors. Taken together, we interpret these findings to indicate that pRB responds to anti- growth signals to act as a barrier to initial tumor formation by inducing cell cycle arrest; loss of this anti-growth pathway leaves mammary epithelial cells vulnerable to oncogenic transformation caused by p53^R172H. Unresponsiveness to negative growth signals is described as a hallmark of cancer, implying a ubiquitous need for it to be eliminated during tumorigenesis. There is little evidence of how these signals are generated and when they act to block aberrant proliferation. The evidence of their importance is largely implied by the frequency of their loss in cell cultures generated from tumors. Our genetic approach is a first step to gaining better insight into this mechanism. Our data from studying *Neu*-induced tumorigenesis implies that negative growth signals may be lost with no apparent effect on cancer incidence. However, comparing the two tumor models used here may be instructive in narrowing down the possibilities. Both *Wap-p53^R172H* and *Neu* (line 202) require accompanying or sporadic mutations in the transgene to initiate cancer. The *Neu* transgene requires the deletion of a small portion of the ectodomain to be become activate and drive cancer formation. This mechanism has been shown to be consistent between individual *Neu* tumors as similar deletions are generated each time. The resulting activated *Neu* receptors have relatively weak transforming activity in comparison with the original V664E mutant allele. For this reason the relative strength of the oncogenic signal may be an important factor. This is consistent with mammary tumors in *Wap-p53^R172H*; *Rb1^ΔL/ΔL* mice having incomplete penetrance; some animals may not acquire sufficiently strong mutations. In addition, mitogenic signals are known to elicit a negative growth response from TGF-β in mammary epithelium. For these reasons the strength of oncogenic growth signal is likely an important cue in determining if a negative growth arrest mechanism will need to be overcome during tumor formation. Future experiments will need to be designed to address this possibility. Understanding the context in which the response to negative growth signals is a key tumor suppressor mechanism is a challenging new question. However, determining what factors influence context *in vivo* will greatly influence our understanding of negative growth control in cancer, and will undoubtedly impact the classification and treatment of this disease in the future. # Materials and Methods ## Mouse strains The *Rb1^ΔL* mouse strain has been described previously. Analyses of *Rb1^ΔL/ΔL* mice were performed on a mixed 129/B6 background. *Wap-p53^R172H* mice were obtained from the Mouse Models of Human Cancer Consortium on an FVB background. These mice express the p53^R172H mutation driven by the *whey acidic protein* promoter. These mice were bred to the *Rb1^ΔL* mutation, creating a mixed 129/B6/FVB genetic background. *Wap-p53^R172H*; *Rb1^+/+* and *Wap- p53^R172H*; *Rb1^ΔL/ΔL* females were bred through five or six rounds of pregnancy to induce expression of p53^R172H. Live pups were removed at P2 to allow equivalent timing of transgene expression between genotypes. *Tg(MMTVneu)202Mul* (herein called *Neu*) mice express the wild type form of the rat *Neu* oncogene driven by the *murine mammary tumor virus* promoter (*MMTV*). These mice were obtained from Jackson Labs on an FVB background and were bred to the *Rb1^ΔL* mutation, creating a mixed 129/B6/FVB genetic background. Genotyping methods and PCR primers were provided by the suppliers, or are as outlined by Isaac, *et al.*. All animals were housed and handled as approved by the UWO animal use subcommittee (protocol 2007-058) and CCAC guidelines. ## Histology and mammary whole mounts Full necropsies were performed on tumor-bearing animals after 60 days or at the time of euthanasia. Mammary tumors, lung tissues, and any other tissues that appeared abnormal were fixed in formalin and sectioned as previously described. The mitotic indices were manually counted in 5 high-power fields of view (400×) for mammary tumors from each genotype. Lung metastases were identified by gross morphological analysis (surface metastases) and microscopic analysis (micrometastases). Percent metastatic surface area (SA) was calculated by measuring the total two dimensional area occupied by lung metastases in five hematoxylin and eosin (H&E) stained lung sections, divided by the total area of the lung in these sections using Volocity 4 software (Perkin Elmer, Waltham, MA). Analysis of hyperplasia in H&E stained sections of *Rb1^ΔL/+* mammary glands, as well as *Neu* expressing mammary glands, was as described before. For whole-mount analysis, unaffected mammary glands from tumor-burdened mice were removed, mounted on glass slides, and stained with Carmine Red using standard techniques. ## Primary cell culture assays Mammary epithelial cells were harvested and cultured as previously described. Cell culture experiments were carried out on passage 1 or 2 MECs. TGF-β1 growth inhibition assays were performed as previously described. ## Soft Agar Colony Formation Assay *Rb1^+/+* and *Rb1^ΔL/ΔL* MEFs were retrovirally transfected with the pLXSN dominant negative (dn) p53/Ras^V12 virus as previously described. Infected cells were then grown in soft agar according to standard protocols. Cells were allowed to grow for 2 weeks, at which time colonies were photographed and counted. The cut-off for scoring a colony as transformed was that its size needed to correspond with the volume of at least 5 cells (as judged by neighboring single cells). In this way we were confident that these colonies represented multiple cell divisions. The authors would like to thank numerous colleagues for advice during the course of this research. We are grateful to lab members who helped with tumor measurements and the CHRI histology facility for technical assistance. The pLXSN dn p53/Ras^V12 construct was provided by Piotr Sicinski (Dana-Farber, Boston). [^1]: Conceived and designed the experiments: SMF FAD. Performed the experiments: SMF FAD. Analyzed the data: SMF SC FAD. Wrote the manuscript: SMF FAD. [^2]: The authors have declared that no competing interests exist.

# Introduction Work-addicted individuals–so called workaholics–experience that they are only really “alive”, when working (e.g.,). This situation indicates the essence of a craving desire for work. Intensive working is the main strategy available for work-addicted individuals to feel alive. Thus, work-addicted individuals’ sense of self is contingent on their persistent working. This suggests that work-addicted individuals have rather low skills in volitional action control. At the same time, however, work-addicted people’s high persistence and self- discipline when performing work-related activities suggest strong skills in volitional action control. To solve this paradox, we take an action control perspective on work craving that differentiates two types of volitional action control: self-regulation and self-control. We integrate both theoretical approaches in order to derive our hypotheses. ## Work Craving Researchers have traditionally portrayed workaholism as a phenomenon that manifests almost entirely in obsessive-compulsive symptoms: obsessive thinking about work (intrusive thoughts that, for example, address the need to work, feeling that the job was not finishes, worry about not meeting a deadline) and a compulsive pattern of working behavior both behavioral and mental (e.g.,,). By contrast, other scholars argued that workaholism is an addiction disorder. Indeed, work-addicted individuals continue to engage in the same behavior (working) despite persistent adverse consequences. Moreover, in response to leisure time, they exhibit similar symptoms of withdrawal (including craving, restlessness, irritability, or exhaustion) as individuals who meet the criteria for addictive disorders, recognized by the APA (DSM-V,). Like substance-addicted individuals, work-addicted individuals also develop a tolerance for the rewards of addictive means (work) manifested by a need for markedly increased amounts of their intoxicant (work) to achieve a desired effect. There is evidence that behavioral addictions and substance addictions both share the same association with changes in the neural pathway of the reward system in the brain (,) as well as with neurocognitive deficits, like those in executive functioning. The theory of work craving, recently developed by Wojdylo,, tries to reconcile the seemingly contradictory positions regarding the obsessive- compulsive versus addictive nature of workaholism. Specifically, in contrast to previous theories of workaholism (e.g.,,), it emphasizes the importance of a synthesis of obsessive-compulsive and *addictive* elements for better understanding the nature of work addiction. According to work craving theory (,), work craving is defined as an *addictive disorder* (work addiction). Consequently, we use the terms “work craving” and “work addiction” interchangeably. Work craving comprises both mechanisms typical for addiction: *positive reinforcement* as well as *negative reinforcement*. Specifically, the theory assumes that work-addicted persons become pathologically focused on *pursuing reward* from engaging in the behavior of compulsive working and neurotic standards: achieving positive outcomes such as self-worth, and social acceptance (positive reinforcement) and avoiding aversive outcomes such as negative emotions (negative reinforcement). Work-addicted individuals struggle with refraining from the behavior, experience intense craving, and have difficulty resisting; they have limited awareness of the problems that have emerged as a result of their behavioral addiction. Thus, according to the work craving theory, the phenomenon of workaholism is only dysfunctional. There is no functional form of workaholism (for a broader discussion see:). It should be noted that for convenience, we use the terms “work-cravers” and “work-addicted individuals” interchangeably for those individuals scoring relatively high on the Work Craving Scale (WCS), taking into account that the WCS has been analyzed as a continuous and not as a categorical variable (see:, ,). In contrast, we use the term “obsessive-compulsive workers” for individuals scoring relatively high on scales that narrowly measure “workaholism” as an obsessive-compulsive phenomenon. From a work craving perspective, there are two ways in which compulsive working may support addictive work processes. First, low self-worth and high negative emotions may stimulate individuals to crave an obsessive-compulsive style of working and to strategically persevere in such an action in order to gain overcompensation of low self-worth and sustain emotional escape from negative symptoms. Second, low self-worth may guide individuals to crave very high, unrealistic standards of achievement (neurotic perfectionism) and strategically persevere in such an action in order to gain overcompensation of low self-worth and sustain emotional escape from negative symptoms. Wojdylo et al. provided empirical evidence for the craving/addictive nature of workaholism. The findings showed that work craving comprises four empirically distinct dimensions: (1) obsessive-compulsive desire for work (obsessive thinking and compulsive working); two hedonic symptoms including (2) overcompensation of low self-worth resulting from overworking, and (3) escape from negative affect (emotional relief) and withdrawal symptoms (e.g., restlessness, irritability, or exhaustion); as well as the cognitive symptom (4) neurotic perfectionism. There is also growing empirical evidence that work craving is embedded in a meaningful nomological network of related constructs and not conducive to health and well-being. For instance, work craving correlates positively with rumination, depression, and burnout as well as negatively with self-esteem and health,. Recent results also confirm the detrimental effect of work craving on health in a longitudinal design. It is therefore important to understand how work-addicted individuals regulate their unhealthy work style and what factors influence their “success” in doing so. A rather unexplored but important issue is how self-regulatory competencies are related with work craving (vs. work engagement) and, in turn, health problems. In studies where “workaholism” was measured only as an obsessive-compulsive phenomenon (and not as work addiction), subjective work experiences indicate that obsessive-compulsive workers and engaged workers both tend to work excessively hard (behavioral component). However, their respective work styles serve different overarching purposes. Whereas obsessive-compulsive workers are driven to work by strong obsessive self-imposed demands they cannot resist (i.e., adopt external standards of self- worth and social approval without fully identifying with them; *introjected regulation*), engaged employees are working under the influence of intrinsic motivation, a positive connection with work, and the ability to deal with the demands of their jobs (i.e., engage in an activity for its own sake and act with a full sense of volition; *autonomous regulation*). According to work craving theory, the emotion regulatory strategies of work cravers (i.e., the desire for self-worth compensatory incentives through overworking and the desire to escape from negative emotions) imply that work craving should derive from low rather than high self-regulatory competencies. Indeed, Wojdylo, et al. found that work cravers have deficits in self-regulating affect, whereas engaged workers have high competencies in self-regulating their emotions. Specifically, whereas work craving was related to a low ability to down-regulate negative emotions and to poor health, work engagement correlated with the high ability to up-regulate positive emotions and to good health. Importantly, a longitudinal study confirmed that deficits in self-regulation of emotions influence work craving and, in turn, the symptoms of psychological distress. These findings imply that work craving and work engagement may be associated with low versus high self-regulation competencies, respectively. This, in turn, inspires the question: How do work cravers manage to initiate and maintain their excessive work activities if they have poor volitional competencies? Their ability to focus on work-related goals suggests that work cravers also have strong volitional competencies. # Self-Regulation and Self-Control Kuhl’s theory of self-regulation provides a useful framework for understanding the volitional competencies of work cravers,,. According to self- regulation theory, the various volitional competencies individuals use to regulate their psychological functioning can be grouped into two basic modes of volition: self-regulation and self-control. *Self-regulation* is conceptualized as a “democratic” mode of volition in which individuals form goals in accordance with their needs and preferences (self-determination), regulate emotions in a context-adequate manner (self-motivation and self-relaxation), and flexibly resolve action-related conflicts. It is self-supportive and fosters an overview and integration of own needs, wishes, and goals. *Self-control*, in contrast, is conceptualized as an “authoritarian” mode of volition in which individuals maintain a specific goal by suppressing any tempting alternatives, even if these alternatives are self-congruent. It is self-suppressive, focused on goal maintenance and self-discipline, and supported by high planning abilities for specific action steps. Whereas self-regulation operates by activating the reward system, identification with goals, and self-motivation through positive emotions (e.g., “I do this because it is important”), self-control operates by activating the punishment system, conscious and depletive efforts, and self-motivation via negative emotions (e.g., “If I don’t do this I will feel ashamed”) (,). Consistent with this assumption, Kuhl and Fuhrmann show that participants who report high self-regulation in daily life are more efficient in maintaining a healthy diet when rewarding themselves for dietary success (e.g., eating more broccoli and less ice cream) whereas participants high in self-control are more efficient when punishing themselves for dietary failures. This empirical dissociation supports the assumption that self-regulation and self-control are two distinct modes of volition. Healthy psychological functioning requires both modes because self-regulation and self-control are adaptive under different conditions and have different (dis)advantages,. Especially the overly strong use of self-control entails a trade-off between short-term benefits such as resistance to temptation and long-term costs such as alienation from own preferences and perseverance of stress,. Self-regulation, in contrast, entails positive short- and long- term effects because it promotes recovery from stress and correlates with fewer reports of psychopathology,. In a study by Forstmeier and Rüddel, most of the self-regulation competencies correlated negatively with psychopathological measures such as depression as well as neuroticism, social inhibition, and excitability. However, to our knowledge, there is no systematic evaluation of the volitional skills of work-addicted individuals. # Integrating Theories of Work Craving and Self-Regulation The distinction between self-regulation and self-control helps to resolve the paradox that work cravers have both weak and strong volitional skills at the same time. On the one hand, it can be assumed that work cravers have poor emotion regulation skills and that working hard is their main way of coping with low self-worth and negative feelings (e.g.,). Our studies showed that work cravers report difficulties with self-relaxation in daily life. Other studies showed that obsessive-compulsive workers spend more time on work-related activities during the evening than non-obsessive-compulsive workers when experiencing negative emotions at the end of the workday,. This suggests that work cravers have weak self-regulation (deficits in self-maintenance). On the other hand, it can be assumed that work cravers have high volitional skills because they manage to work excessively hard and show extreme drive in implementing work-related goals–albeit in an obsessive-compulsive manner,. The studies showed that obsessive-compulsive workers tend to introject (rather than identify with) external requirements and perform actions in order to avoid guilt and anxiety. This suggests that work cravers have *strong self-control* and strong activation of the associated punishment system *(competencies in goal maintenance)*. Biebrich and Kuhl found the combination of low self-regulation and high self-control to yield coping responses that they consider to refer also to work addicted individuals: high action competence without inner security. Taken together, we tested the following assumptions: (*H1*) Work craving is associated with low self-regulation and high self-control. (*H2*) Work craving is associated with symptoms of psychological distress. (*H3*) Low self- regulation is associated with symptoms of psychological distress. Because of the mixed effects of self-control—positive short-term effects as well as negative long-term effects of (too much) self-control—, we did not make a prediction about the relationship between self-control and distress symptoms. (*H4*) Consistent with previous findings of Wojdylo and Baumann et al. on work craving, we expected work craving to mediate the relationship between volitional skills and distress symptoms. As depicted in, we expected a moderated mediation: Work craving mediates the relationship between low self-regulation and distress symptoms at high levels of self-control. We tested our hypotheses in two studies that are part of the Work Craving International Project (WCIP), a cohort research endeavor of Poland and Germany. The goal of the WCIP is to study the new conceptualization of workaholism as work craving and its underlying personality antecedents. # Study 1 ## Participants Researchers recruited 291 participants who were Polish employees (174 women = 59.79%) in different Polish companies (e.g., employees of business and information technology services, employees of the electric company, teachers, office and administrative workers, etc.). In each of the respective participating companies, about 90–95% of the employees participated. Their mean age was 41 years (range 22–63 years). Most participants held nonmanagerial (77.3%) rather than managerial (22.7%) positions; 75.6% of participants had completed higher education, 22.3% secondary education, and 2.1% vocational education. ## Ethics Statement We carried out the study procedures in accordance with the Declaration of Helsinki. We collected information anonymously and did not have access to identifying participant information. These project studies received approval from the German ethics committee before commencing with the investigations (Aufsichts- und Dienstleistungsdirektion Rheinland-Pfalz: ADD 51 111-32/129-12; Landesbeauftrage für Datenschutz und Informationsfreiheit Rheinland-Pfalz: LfD RLP [6.08.22.001:0363](http://6.08.22.001:0363)). The ethics committee in Poland confirmed that the study met applicable standards for the ethics of experimentation and research integrity. ## Materials ### Self-Regulation and Self-Control We used the Polish version of the Volitional Components Inventory (VCI; to assess self-regulation and self-control. The self-regulation scale consists of 12 items (Cronbach’s α =.89, in the present study) covering self-determination (*“I feel that most of the time I really want to do the things I do”*), self- motivation (*“When my enthusiasm drops off*, *I know exactly how to motivate myself again”*), and self-relaxation (*“I know exactly how to decrease my anxiety”*). The self-control scale consists of 8 items (Cronbach’s α =.83, in the present study) covering planning abilities (*“Before I approach a new task*, *I usually make a plan first”*) and anxious goal-orientation (*“In order to motivate myself I often imagine what would happen if I didn't finish the task on time”*). Participants rate the extent to which each statement applies to them on a 4-point scale ranging from 1 (*not at all*) to 4 (*completely*). Kuhl and Fuhrmann as well as Fröhlich and Kuhl reported extensive research on the reliability and validity of the scales. ### Work craving In order to measure work craving, the Polish version of the Work Craving Scale (WCS;) was administered. The questionnaire consists of 28 items (Cronbach’s α =.95, in the present study) covering an obsessive-compulsive desire for work (*“My desire for work overpowers me”*), overcompensation of low self-worth resulting from overworking, (*“Overworking makes me feel important”*), escape from negative affect (emotional relief) and withdrawal symptoms (*“I can relax only if I’m working hard”*), and neurotic perfectionism (*“Even though I perform a task very carefully*, *I feel that it is done not correctly enough*”). Participants indicate to what extent they (dis)agree with each statement on a 7-point scale ranging from 1 (*I completely do not agree)* to 7 (*I completely agree)*. For extensive research on the reliability and validity of the scales, see Wojdylo et al.. ### Psychological distress symptoms To assess mental health status we used the Polish version of the General Health Questionnaire (GHQ-28;). It consists of 28 items (Cronbach’s α =.91, in the present study) covering somatic symptoms (*“Have you recently felt that you are ill*?*”*), anxiety and insomnia (*“Have you recently lost much sleep over worry*?*”*), social dysfunction (*“Have you recently been taking longer over the things you do*?*”*), and severe depression (*“Have you recently felt that life is not worth living*?*”*). Participants rated items on four-point scales ranging from 1 to 4 with slightly different labels across items. Higher scores indicate lower general mental health and higher psychological distress. ## Design and Procedure Researchers recruited participants via an e-mail that was distributed throughout the companies that had given informed consent to take part in the study. The e-mail included an Internet link to a website that could be accessed by those workers interested in participating in an online survey on work behaviors and health. Data are based on anonymized self-report questionnaires administered during working hours. The participants were not reimbursed for their participation. ## Results ### Descriptive statistics and correlations presents the means, standard deviations, and correlations among the study variables in Study 1. All correlations were in the expected direction. Consistent with *H1*, work craving exhibited a significantly negative correlation with self-regulation and a significantly positive correlation with self-control. Consistent with *H2*, a significantly positive correlation was found between work craving and symptoms of psychological distress. Consistent with *H3*, we found a significantly negative correlation of self-regulation with psychological distress. Self-control did not significantly correlate with psychological distress symptoms. There were no significant relationships between work craving and age or gender. ### Moderated mediation analysis To test whether work craving mediated the negative relationship between self- regulation and symptoms of psychological distress at high levels of self-control (*H4*), we conducted a moderated mediation analysis with 5,000 bootstrap resamples using the SPSS macro (Model 8) described by Hayes. Using this procedure, we computed a point estimate and a 95% confidence interval (CI) for the moderated mediation effect. We entered self-regulation as an independent variable, self-control as a moderator, work craving as a mediator, and psychological distress as an outcome variable. ### Work craving In the model testing work craving as a dependent variable, there were significant main effects of self-regulation and self-control indicating that self-regulation was associated with lower and self-control with higher work craving. Consistent with expectations, the self-regulation x self-control interaction was significant. In, the interaction effect is illustrated for values of *M* ± 1 *SD* for both predictor variables. At low levels of self- control, work craving was low irrespective of the level of self-regulation (*ß* =.09, *t* = -1.14, *ns*). At high levels of self-control, in contrast, lower self-regulation was associated with higher work craving (*ß* =.29, *t* = -3.96, *p* \<.001). Similarly, at high levels of self-regulation, work craving was low irrespective of the level of self-control (*ß* =.11, *t* = 1.35, *ns*). At low levels of self-regulation, in contrast, higher self-control was associated with higher work craving (*ß* =.31, *t* = 3.99, *p* \<.001). Findings are consistent with our hypothesis that work craving is associated with a combination of low self-regulation and high self-control. ### Psychological distress symptoms In the model testing symptoms as a dependent variable, there were significant main effects of self-regulation and work craving on symptoms, indicating that self-regulation was associated with lower and work craving with higher symptoms. Self-control had no main or interaction effects on symptoms when controlling for work craving. The significance of the indirect effect of low self-regulation through work craving on symptoms was tested across low, moderate, and high levels of self- control with bootstrapped errors and 95% confidence intervals (CIs). Consistent with expectations (*H4*), the indirect effect of low self-regulation through work craving on symptoms was significant at moderate and high levels of self- control (because the limits of the 95% confidence interval did not include zero) but not at low levels of self-control. The indirect effect of the self-regulation x self-control interaction through work craving on symptoms was significant. This finding indicates that a combination of low self-regulation and high self-control may be problematic because it fosters work craving and, in turn, symptoms of psychological distress. # Study 2 In Study 2, we aimed at replicating our findings in a different sample of workers (e.g., managerial, administrative, and clerical employees in areas such as banking services, data analysis, office administration, etc.). In addition, we wanted to support the discriminant validity of work craving with respect to work engagement by showing their different volitional underpinnings. As mentioned earlier, work craving and work engagement have distinct regulatory mechanisms and outcomes: an obsessive-compulsive drive versus intrinsic motivation, low versus high self-regulation, and low versus high psychological health,. These findings imply that work engagement is related to high self- regulation competencies. Whereas the sole reliance of work craving is on self- control rather than self-regulation, we expected work engagement to be associated with high abilities in both self-regulation and self-control. Furthermore, we expected work engagement to be negatively associated with symptoms of psychological distress. We also expected work engagement to mediate the relationship between volitional skills and distress symptoms. ## Participants Researchers recruited 272 participants who were employees (158 women = 58.09%) in various Polish companies (e.g., banking services employees, data analysts, office workers, economists, etc.). In each of the respective participating companies, about 90–95% of the employees participated. Their mean age was 30 years (range 21–52 years). Most participants were employed in nonmanagerial (82%) rather than managerial (18%) positions; 71.3% of participants had completed higher education and 28.7% secondary education. ## Materials We used the same measures for self-regulation (Cronbach’s α =.86, in the present study), self-control (Cronbach’s α =.82, in the present study), work craving (Cronbach’s α =.95, in the present study), and psychological distress (Cronbach’s α =.88, in the present study) as in Study 1. ## Work engagement To assess work engagement, we used the Polish version of the Utrecht Work Engagement Scale (UWES;). The measure consists of 17 items (Cronbach’s α =.94, in the present study) covering vigor (*“At my work*, *I feel bursting with energy”)*, dedication, (*“I find the work that I do full of meaning and purpose”)*, and absorption (*“When I am working*, *I forget everything else around me”)*. Participants rated items on a 7-point frequency scale ranging from 1 (*never*) to 7 (*each day*). Higher scores indicator higher work engagement. ## Design and Procedure We used the same strategy to recruit participants as in Study 1. ## Results ### Descriptive statistics and correlations presents the means, standard deviations, and correlations among the study variables in Study 2. All correlations were in the expected direction. Consistent with *H1*, work craving exhibited a significant, negative correlation with self-regulation and a (marginally) significant and positive correlation with self-control. Work engagement, in contrast, was significantly and positively correlated with self-regulation and self-control. Consistent with *H2*, a positive correlation was found between work craving and psychological distress symptoms, and a negative correlation was found between work engagement and psychological distress symptoms. Consistent with *H3*, self-regulation correlated negatively with distress symptoms. There was no correlation between self-control and distress symptoms. There were no significant relationships between work craving and age or gender. ### Moderated mediation analysis We conducted the same analysis (SPSS macro Model 8;) as in Study 1 with the following change: In addition to work craving, we entered work engagement as a mediator into the analysis. ### Work craving In the model testing work craving as a dependent variable, there were significant main effects of self-regulation and self-control, indicating that self-regulation was associated with lower and self-control with higher work craving. Consistent with expectations and findings in Study 1, the self- regulation x self-control interaction was significant. At low levels of self- control, work craving was low and not related to self-regulation (*ß* =.13, *t* = -1.45, *ns*). At high levels of self-control, in contrast, lower self- regulation was associated with higher work craving (*ß* =.41, *t* = -5.05, *p* \<.001). Similarly, at high levels of self-regulation, work craving was low and not related to self-control (*ß* =.06, *t* =.71, *ns*). At low levels of self- regulation, in contrast, higher self-control was associated with higher work craving (*ß* =.34, *t* = 4.21, *p* \<.001). Findings are consistent with our hypothesis that work craving is associated with a combination of low self- regulation and high self-control. ### Work engagement In the model testing work engagement as a dependent variable, there was a significant main effect of self-regulation indicating that higher self- regulation was associated with higher work engagement. In addition, there was a significant self-regulation x self-control interaction. As illustrated in, at low levels of self-control, lower self-regulation was associated with significantly lower work engagement (*ß* =.41, *t* = 4.76, *p* \<.001). This relationship was less pronounced but still significant at high levels of self- control (*ß* =.20, *t* = 2.43, *p* \<.02). Similarly, at high levels of self- regulation, work engagement was high irrespective of the level of self-control (*ß* =.01, *t* = 0.07, *ns*). At low levels of self-regulation, in contrast, work engagement was low but reached moderate levels with higher self-control (*ß* =.21, *t* = 2.56, *p* \<.02). The finding is consistent with the assumption that work engagement has different volitional underpinnings than work craving. As summarized in, self-regulation is a necessary and sufficient volitional competency for work engagement. Whether or not employees are able to exert self- control does not affect their level of work engagement as long as they are able to self-regulate emotions. In contrast, sole reliance on self-control in the absence of self-regulation is associated with work craving. Finally, a combination of low self-regulation and low self-control is associated with low levels of both work styles and may be characterized as low work involvement. ### Psychological distress symptoms In the model testing symptoms as a dependent variable, there were significant main effects of work craving, work engagement, and self-regulation on symptoms, indicating that work craving was associated with higher symptoms whereas work engagement and self-regulation were associated with lower symptoms. Self-control had no main or interaction effects on symptoms when controlling for work styles. The significance of the indirect effect of low self-regulation through work styles on symptoms was tested across low, moderate, and high levels of self- control with bootstrapped errors and 95% confidence intervals (CIs). Consistent with expectations (*H4*), the indirect effect of low self-regulation on symptoms through increased work craving was significant at moderate and high levels of self-control (because the limits of the 95% confidence interval did not include zero) but not at low levels of self-control. Furthermore, the indirect effect of low self-regulation on symptoms through reduced work engagement was significant across all levels of self-control. The indirect effects of the self-regulation x self-control interaction through work styles on symptoms were significant for both variables, work craving and work engagement. Findings indicate that two different mechanisms partially mediated the relationship between low self-regulation and symptoms. First, in conjunction with high self-control, low self-regulation aggravated the unhealthy work style of work craving. Second, in conjunction with low self-control, low self-regulation undermined the healthy work style of work engagement. # Discussion The present studies aimed at contributing to a better understanding of workaholism: specifically, its addictive nature and self-regulatory mechanisms. First, using Wojdylo’s theory of work craving we outlined workaholism as a clearly pathological work style and extended it by hedonic and cognitive components to define work craving. It encompasses not only compulsive behaviors but also emotion-regulatory components characteristic of addiction disorders (i.e., a desire for self-worth compensatory incentives and relief from negative emotions through neurotic perfectionism). Second, we applied Kuhl’s, theory of self-regulation to differentiate two modes of volition: self-regulation (i.e., an “inner democracy” oriented at self-maintenance and integration of multiple needs, goals, and preferences) and self-control (i.e., an “inner dictatorship” oriented at goal maintenance and perseverance of a single goal). Finally, we integrated both theories and predicted work craving to be associated with a combination of low self-regulation and high self-control. In two samples of Polish workers (total of 563 participants), we found work craving to be associated with the expected combination: work cravers have a low ability to form self-congruent goals (low self-regulation) but a high ability to maintain goals and suppress tempting alternatives which is conducive to successful goal striving (high self-control). However, even successful goal striving does not promote well-being if goals are not integrated into the self and do not satisfy personal needs. Koole et al. refer to conditions in which people are chronically locked into self-control as *ego fixation* and review its role in alienation from emotional preferences, rigidity, overcontrol, rumination, and psychosomatic problems,. Consistent with this negative view on chronic self-control, our findings showed work craving to be associated with symptoms of psychological distress symptoms. Furthermore, work craving mediated the relationship between self-regulation deficits and symptoms at moderate and high levels of self-control. Thus, strong self-control did not buffer against the negative effects of weak self-regulation because it fueled an unhealthy work style. Metaphorically speaking, the dominance of self-control in the personality (“when self-control is burning”), together with deficits of self-regulation (“under the rubble of self-regulation”) forms the “inflammatory,” destructive mechanism of work craving resulting in reduced health. In light of these findings, one may wonder why work cravers keep on going. Previous findings show that obsessive-compulsive workers are avoidance motivated and try to prevent failure, guilt, shame, and anxiety,. At the same time, avoidance motivation is a strong source of self-control,. Based on our findings, it can be concluded that work cravers’ strong self-control skills make them highly efficient in avoiding negative emotions. This may explain why they remain so persistent in their work activities and why it is so difficult to change their unhealthy behavior (cf.,,). Furthermore, high self-control makes people suppress intuitive (dis)likes and bodily responses that may serve as “somatic markers” to guide people away from potentially dangerous and self- incongruent choices or toward successful and self-congruent options. On the one hand, this may lead people to become alienated from their intuitive dislike for aversive experiences. It may explain the obstinate persistence of work-addicted individuals’ destructive work habits despite negative consequences (e.g., distress symptoms, exhaustion). On the other hand, the lack of inner reference may explain why work cravers are not even satisfied with successful results, set neurotically high standards, and apply a social reference norm. Taken together, work cravers are characterized by their high action competencies without inner security (cf.). It is noteworthy that our findings were obtained in two heterogeneous general population samples: workers with different occupations and from different work environments. Work craving theory does not specify work environments or job characteristics. Instead, work craving is assumed to develop as a function of personality and may be utilized as a coping mechanism in nonwork contexts, for example, in college or after retirement. Items may have to be adjusted to allow a valid assessment of work craving in such contexts. Thus, work craving theory takes a personality perspective on work addiction. In support of the discriminant validity of work craving, we found work engagement to have a completely different profile of volitional competencies: high self-regulation and high self-control. Furthermore, higher work engagement was associated with lower psychological distress and mediated the relationship between self-regulation competencies and health. Thus, work engagement is clearly a healthy work style. Engaged workers’ self-regulatory ability to integrate the multitude of personal needs, wishes, and goals may also contribute to successful work-life balance. Taken together, engaged workers are characterized by high action competencies combined with inner security (cf.). Note that self-regulation and self-control show a significant positive correlation in our studies. This is because both are volitional competencies. Functionally distinct dimensions can correlate to the extent that they cooperate under certain conditions. In healthy individuals, self-control and self- regulation should cooperate whenever this seems reasonable, for example, when it makes sense to invest impulse control (as an example of self-control) after having made a self-congruent decision (as an example of self-regulation). Work engagement seems to indicate this type of healthy cooperation. Nevertheless, self-regulation and self-control are distinct volitional competencies because they may dissociate: one competency may be available in excess while the other is deficient. Work craving seems to indicate this type of dissociation and one- sided volitional action control (excessive self-control and deficient self- regulation). # Limitations and Future Perspectives Our study has several limitations that can be addressed in future research. First, our findings are cross-sectional and do not allow us to draw causal conclusions. However, self-regulation and self-control are relatively stable dispositions that develop from early childhood onwards,. Therefore, it is unlikely that the scores on the volitional measures were caused by work styles or health. In contrast, the causal direction of the association between health and work styles is less clear. However, in a longitudinal design, Wojdylo et al. showed that work craving predicted health decrements whereas health did not predict changes in work craving. Second, our outcomes are merely self-reports of psychological distress symptoms. In future studies, it would be informative to broaden the network of possible outcomes (e.g., job performance, work-life balance) and to extend measures to less obtrusive data (external ratings, behavioral indicators). Third, we have concentrated on a personality perspective in our studies. It would be informative to assess work environment factors in order to explore their role in strengthening or ameliorating volitional competencies and/or work styles. For instance, employees’ self-control and work craving may be fostered in a company where the policy and its managers focus on the task and time pressure. In contrast, a company that values benevolence may employ more autonomy-supportive managers who foster employees’ self-regulation and work engagement. Future studies can verify if a more supportive work environment may compensate employee’s self-regulation deficits and function as a moderator at different positions in the model depicted in. Finally, our findings have important practical implications. Forstmeier and Rüddel have indicated that improving volitional competencies is crucial for the efficacy of treatment of psychosomatic symptoms in psychotherapy. Thus, it can be concluded that improving volitional competencies is one of the most relevant issues in the prevention and treatment of work addiction and its effects on psychosomatic symptoms. Our findings highlight the importance of differentiating between self-regulation and self-control to understand the specific problems and resources of work-addicted individuals. Intervention programs should focus especially on increasing self-regulation competencies in employees such as: (a) generating many alternative ways to reach a goal, (b) expanding attention to the multitude of personal needs, wishes, and goals, (c) feeling and differentiating own emotions and body signals, and (d) training self-motivation and self- relaxation. Furthermore, organizations should encourage an organizational climate and leadership style that supports self-regulation and autonomy in their staff to prevent work craving and promote employees’ health,. # Supporting Information [^1]: The authors have declared that no competing interests exist. [^2]: **Conceptualization:** KW JK. **Formal analysis:** KW NB. **Funding acquisition:** KW. **Investigation:** KW. **Methodology:** KW NB. **Project administration:** KW. **Software:** KW NB. **Supervision:** KW. **Writing – original draft:** KW NB. **Writing – review & editing:** KW NB.

# Introduction Triple negative (TN) breast cancers are defined by lack of expression of estrogen, progesterone, and HER-2 (ERBB2) receptors. It is widely recognized that TN breast cancers have a poorer prognosis than any other subtypes of breast cancer. Given the lack of effective targeted therapies for TN breast cancer patients, a better understanding of the mechanisms of growth and invasion of these tumors and their interaction with the tissue microenvironment will provide valuable insight into developing novel approaches to lower the mortality associated with TN breast cancer. There is increasing evidence demonstrating both β-catenin-dependent (canonical) and independent (non-canonical) WNT-signaling pathways play a key role in regulating pathological processes by facilitating tumor growth, migration, and invasion. For example, Matsuda et al. reported that inhibition of the WNT- signaling pathway suppressed tumor growth in xenograft experimental systems in MDA-MB-231 cells. Genetic and histological studies demonstrated that changes in the WNT-signaling pathway genes are common in metaplastic breast cancer cells and enriched in epithelial-mesenchymal transition and stem cell phenotypes. Importantly, recent studies analyzing gene expression profiles suggest that specific classes of TN breast cancer cells express genes regulating tumor migration, invasion, and differentiation, including TGF-β signaling pathways, extracellular matrix (ECM) reorganization, and the WNT-signaling pathway. Consistent with this notion, immunohistochemical studies demonstrated that the canonical WNT-signaling pathway is activated in TN breast cancer cells compared to cells other of other molecular subtypes. Thus, it is important for generating therapeutic strategies by evaluating how signaling pathways regulate tumor migration and invasion in TN breast cancer cells. Recent studies have demonstrated that FH535 is a synthetic inhibitor of the canonical WNT-signaling pathway and inhibits growth of colon, lung, and hepatocellular carcinoma line but not normal fibroblasts, suggesting that the targeting of the canonical WNT-signaling pathway by small molecules could be a promising therapeutic approach for cancer cells in which this pathway is activated. Indeed, Vaid et al. reported that FH535 significantly inhibited malignant melanoma growth and migration. In this study, we hypothesize that the canonical WNT-signaling enhances tumor progression and metastasis in TN breast cancer cells. In order to test this hypothesis, we utilized FH535 as a tool to evaluate its ability to inhibit growth, migration, and invasion of TN breast cancer cell lines (MDA-MB-231 and HCC38). Here, we show results suggesting that targeting the canonical WNT-signaling pathway by compounds like FH535 may be a promising therapeutic approach for TN breast cancers. # Materials and Methods ## Cell Lines HCC38, T47D, MDA-MB-231, MCF-7, SK-Br3 cells were purchased from ATCC (Manassas, VA, USA) and maintained in RPMI1640 containing 10% FBS. ## Antibodies and Chemicals MOPC-21 (IgG₁), UPC-10 (IgG2a), and MOPC141 (IgG2b) were purchased from Sigma-Aldrich (St. Louis, MO, USA). Rat tail type I collagen, anti-CSPG4 (clone 9.2.27), anti-β1(clone P4C10), anti-α2 (clone P1E6), anti-α3 (clone P1B5) integrin, anti-α3 (clone P1D5), anti-α6 (clone NK1-GoH3), anti-αvβ3 (clone LM609) anti-actin, anti-Ki67, anti-MMP-2, anti-MMP9, and anti-NEDD9 antibodies were purchased from Millipore (Billerica, MA, USA). Anti-Axin, anti-phospho ³⁹⁷Tyr-FAK, anti-FAK, anti-phospho⁴¹⁶Tyr-Src, anti-Src, anti-p38MAPK, anti-phospho-p38 MAPK, anti-Erk1/2, anti-phospho-Erk1/2, and anti- CD44 (clone 156-3C11) antibodies were purchased from Cell Signaling Technology (Beverly, MA, USA). FITC-conjugated rabbit anti-mouse IgG and FITC-conjugated rabbit anti-rat IgG, HRP-conjugated rabbit anti-mouse IgG, and HRP-conjugated rat anti-rabbit IgG were purchased from Jackson Immunoresearch (West Glove, PA, USA). Antigen-retrieval solution was purchased from Biogenex (Fremont, CA, USA). Matrigel and rat tail type I collagen were purchased from BD Bioscience (Franklin Lakes, NJ, USA). FH535 was purchased from Calbiochem, and dissolved in DMSO at a concentration of 10 mM as a stock solution and used by diluting in DMSO to give a final concentration of 0.1% (v/v). Thus, control groups in this study contained DMSO at this concentration. For inhibiting adhesion, migration, and invasion of tumor cells by anti-integrin antibodies, we utilized each antibody at a final of concentration of 1 µg/ml. Other chemicals were purchased from Sigma-Aldrich unless otherwise mentioned. ## FACS Analysis Expression of integrin subunits was evaluated by FACS analysis according to the methods previously described. Briefly, cells were harvested with PBS containing 5 mM EDTA and immediately neutralized in FACS buffer (RPMI1640 containing 1% BSA and 0.025% NaN₃). Tumor cells (10⁵ cells) were incubated with 1 µg/ml of the primary antibodies by shaking for 1 hour at 4°C. After extensive washing with FACS buffer, cells were incubated with FITC-conjugated secondary antibody (1∶1000 dilutions) by shaking for 1 hour at 4°C. Cells were then washed with FACS buffer 5 to 6 times and fixed in PBS containing 1% paraformaldehyde. Expression of integrin subunit was measured on a FACS Calibur (Becton-Dickinson, NJ USA) by collecting 10,000 events and analyzed by CellQuest. ## Adhesion Assays Adhesion of tumor cells to type I collagen was evaluated as described previously with minor modifications. 96-well plates were coated with type I collagen at a concentration of 5 µg/ml and blocked with PBS containing 1 mg/m BSA. Cells were harvested with PBS containing 5 mM EDTA, washed, resuspended in RPMI1640 containing 1 mg/ml BSA (adhesion buffer) at a concentration of 10⁵ cell/ml. Cells (100 µl/well) were incubated on plates coated with type I collagen as described above for 1 hour in the presence or absence of FH535. Plates were washed with adhesion buffer 4–5 times and remaining cells were fixed followed by staining with crystal violet. After extensive washing to remove excess dye, cells were lysed with 100 µl of PBS containing 2% SDS and then the absorbance was measured at 550 nm. Anti-β1integrin antibody (P4C10), anti-α2 integrin antibody (P1E6), anti-α3 integrin antibody (P1B5) antibodies were used to ensure that tumor adhesion was mediated by α2β1 and α3β1 integrins. MOPC-21 antibody was used as an isotype-matched negative control antibody. Experiments were performed in quadruplicates and repeated three times. The representative data were shown as mean +/− S.D. of absorbance values. ## Migration Assays Tumor cell migration was characterized using the Transwell systems (Costar, 8.0 µm pore size) as described previously. Briefly, cells were harvested with PBS containing 5 mM EDTA, washed, and resuspended in serum-free RPMI1640 media at a concentration of 5×10⁵ cells/ml. An aliquot of 100 µl of the cell suspension was added into the upper chamber of Transwell system). Type I collagen was dissolved in serum-free RPMI1640 media at a concentration of 3 µg/ml and added into the lower chamber of the system. Antibodies (1 µg/ml) or FH535 (0.01 to1 µM) were added in both upper and lower chambers. Migration assays were performed at 37°C for 4–5 hours. The inserts were removed and then fixed in 10% formalin in PBS for five minutes and stained with 0.5% crystal violet for five minutes. The inserts were washed and the upper surface of the membranes was wiped with a cotton swab to remove non-migratory cells. Migrated cells were counted in three randomly selected fields. Each experiment was performed in triplicate and repeated three times. The results presented are shown as mean cell numbers +/− S.D. per mm². ## Invasion Assays Invasion of tumor cells into Matrigel was evaluated as described previously. Tumor cells were harvested and washed as described above, and resuspended in RPMI1640 containing 2% FBS at a concentration of 10⁶/ml. Transwells (Costar, cat# 3422, 8 µm pore-size) were coated with 100 µl of Matrigel at a concentration of 5 mg/ml overnight and washed with RPMI1640 containing 2% FBS to remove unpolymerized proteins. Type I collagen (3 µg/ml) were added to the lower chambers of the Transwell as an adhesive protein. Cells (100 µl/well) were incubated on Matrigel overnight at 37°C in the presence or absence of FH535 in both upper and lower chambers of the Transwell. Anti-β1integrin antibody (P4C10), anti-α2 integrin antibody (P1E6), anti-α3 integrin antibody (P1B5) antibodies were used as a control. MOPC-21 antibody was used as an isotype- matched negative control antibody. The inserts were removed and fixed, stained, as described above. Migrated cells underneath of the membranes were counted at three randomly selected fields. Each experiment was performed in triplicate and repeated three times. The represented results were shown as mean cell numbers +/− S.D. per mm². ## Western Blotting Cells were cultured in 6 well plates (7×10⁵ cells/well/1 ml) coated with 3 µg/ml of type I collagen for 4 hours in the presence or absence of FH535 (1 µM). Plates were washed with PBS and lysed in SDS-sample buffer. After fragmenting DNA by sonication, proteins were separated on SDS-PAGE and transferred onto Immobilon-P membrane as we described previously. The membranes were blotted with primary antibodies followed by HRP-conjugated secondary antibodies. The proteins were detected with enhanced chemiluminescence (ECL) reactions. For serial immunodetection by different antibodies, membranes were stripped and ensured that there was no carry over signals from the last detection by ECL. ## Co-immunoprecipitation Assays Co-immunoprecipitation assays were performed as described previously. Briefly, cells were cultured in dishes (10 cm diameter) overnight in RPMI1640 containing 10% FBS. Cells were briefly washed with PBS one time and then directly lysed in 1 ml of 100 mM Tris-HCl (pH7.4) containing 1% Brij35, 0.14 M NaCl, 1 mM CaCl₂, 1 mM MgCl₂, 1 mM MnCl₂, and a protease inhibitor cocktail. Cell lysates were centrifuged and the supernatants were precleared with mouse IgG-protein G-agarose for 4 hours by shaking at 4°C. The cleared lysates were immunoprecipitated with control IgG, anti-CD44, anti-β1, or anti-CSPG4 antibody-conjugated protein G-agarose for 4 hours by shaking at 4°C. The beads were extensively washed with Tris-HCl (pH7.4) containing 1% Brij35, 0.14 M NaCl, 1 mM CaCl₂, 1 mM MgCl₂, 1 mM MnCl₂, and a protease inhibitor cocktail to remove any proteins that nonspecifically bound on the beads. The bound proteins were released by boiling at 95°C in SDS-sample buffer under reducing conditions and separated on SDS-PAGE followed by western blotting analysis as described above. ## Collagen Gel Culture Cell growth was studied in three-dimensional collagen gels as described previously. Briefly, purified type I collagen gels were prepared by adding 10X RPMI, FBS (final concentration 10%) and neutralized by adding 1 N NaOH to give 2.5 mg/ml of type I collagen. Immediately, cells were added at a concentration of 5×10⁵ cells/ml and the collagen/cell mixture was added to wells, and warmed at 37°C for completing gelling for 4–5 hours. When gels had formed, culture media (RMPI1640 containing 10% FBS) was added and cultured for eight days by changing media every three days. FH535 was added in both gel and medium throughout the experiments at final concentrations as described in the text. ## Morphological Studies After eight days of incubation, cell/collagen matrix was fixed by immersion in PBS-10% formalin for 4–5 hours and embedded in paraffin. The paraffin section was serially cut at 5 µm and mounted on glass slides. Every seven sections, cell/collagen matrices were stained with hematoxylin and eosin to determine the presence of cells in the serial sections. The sections were deparaffinized and treated with antigen-retrieval solution according to the manufacturer’s instructions. Sections were stained with various antibodies and visualized by DAB staining with counter staining by DAPI to localize cells. The results are shown by the (mean +/− standard deviation) of % of positive cells (DAB-positive) out of the total cells (DAPI-positive) from three independent experiments. ## Gene expression Analysis The log2 transformed gene expression data of 514 breast cancers were downloaded from The Cancer Genome Atlas (TCGA) project data portal (<https://tcga- data.nci.nih.gov/>). PAM50 classification results of all samples were obtained from the TCGA breast cancer AWG group, distributed as follows: Luminal A (LumA) = 231, Luminal B (LumB) = 128, Her2-like = 58, and Basal = 97. All statistical analysis was performed using SAS and/or R software. ## Statistical Analysis Statistical significance was calculated by Student’s two-tailed t-test (paired). *P* values less than 0.05 were considered as statistically significant. For analyzing gene expression profiles, analysis of variance (ANOVA) was performed. # Results ## FH535 Inhibited Triple-negative (TN) Breast Cancer cells, MDA-MB-231 and HCC38, Migration Toward Type I Collagen Migration of breast cancer cells into surrounding connective tissue architecture is important for establishing metastasis. Type I collagen is a major component of connective tissues and migration toward this protein is implicated as a key step for invading into tissues. Previous studies demonstrated that tumor migration to type I collagen is mediated by either of α2β1or α3β1-integrin or both. Indeed, inhibitory antibodies against α2, α3, or β1integrin subunits significantly inhibited migration toward type I collagen using MDA-MB-231 and HCC38 cells (not shown). Under these experimental conditions, we tested FH535 for its ability to regulate the migration of HCC38 and MDA-MB-231cells to type I collagen. Our results demonstrated that FH535 inhibited migration in a concentration dependent manner and statistically significant inhibition was observed even at a concentration of 0.1 µM in both cell lines, consistent with the previous studies using human malignant melanoma cells. Previous studies demonstrated that FH535 is a potent inhibitor for the canonical WNT-signaling pathway without affecting the amount of β-catenin. When MDA-MB-231 cells were treated with FH535 at a concentration of 1 µM, the amount of β-catenin was not affected, nor was axin consistent with previous studies. The same treatment, however, reduced the expression of β-catenin while increasing the amount of axin in HCC38 cells. Given the key role of axin in regulating degradation of β-catenin, these results imply that FH535 may inhibit the canonical WNT- signaling pathway through the stabilization of axin, which leads to a degradation of β-catenin. Thus, regardless of the significant inhibition of migration in the presence of FH535 in both cell lines, these results suggest that FH535 may affect migratory abilities of these cell lines through different mechanisms. We, then, asked whether FH535 inhibited expression of integrin subunits that promote cell adhesion to type I collagen. Treatment of MDA-MB-231 cells with FH535 at a concentration of 1 µM did not affect expression of α2, α3, or β1 integrin subunit evidenced by FACS analysis. The same results were obtained when HCC38 cells were treated with FH535 (**not shown**). Consistent with these results, adhesion of MDA-MB-231 or HCC38 cells to type I collagen was not inhibited in the presence of FH535 at a concentration of 1 µM. These results demonstrate that FH535 inhibited cell migration without affecting adhesive abilities of cells to type I collagen, suggesting that signaling pathways important for promoting migration would be attenuated in the presence of FH535. ## FH535 Inhibited Invasion of MDA-MB-231 and HCC38 Cells In order to establish metastasis, tumor cells must transverse basement membrane to reach connective tissues. Invasion of tumor cells through matrigel has been used as a model system to evaluate migratory abilities of tumor cells through the basement membrane. Invasion of HCC38 and MDA-MB-231cells through Matrigel was significantly inhibited by anti- α2, α3 and - β1integrin antibodies, supporting that α2β1and α3β1 integrins play a key role in promoting tumor invasion into matrigel. These results are consistent with the fact that type IV collagen is one of the major components that specifically binds to α2β1 and α3β1 integrins expressed on tumor cell surfaces. Importantly, FH535 inhibited invasion of both MDA-MB-231 and HCC38 cells into matrigel in a concentration- dependent manner. As observed in the migration assays as described above, statistically significant inhibition of invasion was achieved even at a concentration of 0.1 µM. ## FH535 Regulated Growth of HCC38 in 3D type I Collagen Culture Although *in vivo* tissues are highly complex architecture consisting of ECM proteins, stromal fibroblasts, and soluble growth factors, recent studies suggest that the ECM could approximate tissues and provide a model for growth of various tumor cell including breast cancer cells,. In order to test if the canonical WNT-signaling pathway is involved in growth of breast cancer cells, various breast cancer cell lines (MDA-MB-231, HCC38, SkBr3, MCF-7, and T47D) were cultured in three dimensional (3D) type I collagen matrices as described previously. When HCC38 cells were cultured for eight days in the presence of FH535 at a concentration of 10 µM, cell proliferation was significantly inhibited compared to control cells. Similarly, growth of the other TN breast cancer cells, MDA-MB-231, was also significantly inhibited in the presence of FH535. Interestingly, proliferation of breast cancer cells with other from other subtypes, T47D (ER⁺, PR⁺, Her2-, SK-Br3(ER⁻, PR⁻, Her2⁺), or MCF-7(ER⁺, PR⁺, Her2⁻) was not inhibited by FH535 under our experimental conditions. These results suggest that FH535 selectively inhibited growth of TN breast cancer cells under conditions of an artificial three dimensional collagen matrix and thus this experimental system may provide a useful model for evaluating molecules that affect the growth of TN breast cancer. ## Expression of NEDD9 was Inhibited by FH535 in MDA-MB-231 Cells In order to characterize mechanisms of tumor migration, invasion, and growth which were inhibited by FH535, we evaluated expression of various proteins important for promoting these processes by western blotting analysis. Among proteins tested, our results suggest that expression of NEDD9 was markedly inhibited in the presence of FH535 in MDA-MB-231 cells. Previous studies suggest that activated FAK and Src phosphorylate NEDD9. However, FH did not affect expression or activation of these tyrosine kinases. Similarly, expressions of Pyk2 or p130Cas were not inhibited in the presence of FH535 (**not shown**). Importantly, FH535 inhibited neither phosphorylation nor protein expression of ERK1/2 or p38MAPK. These results suggest that the canonical WNT-signaling pathway would regulate specific gene expressions in MDA-MB-231 cells without affecting integrin-mediated cell adhesion followed by the activation of FAK and Src. Despite of the significant inhibition of migration and growth of HCC38 by FH535, we did not observe inhibition of NEDD9 in these cells. Given the fact that NEDD9 promotes tumor cell migration and invasion, these results suggest that NEDD9 plays a key role in facilitating MDA-MB-231 cell migration, invasion, and growth. It has been reported that NEDD9 associates with integrin-mediated signaling pathways by scaffolding various signaling molecules important for promoting migration and growth. In order to test whether NEDD9 associates with cell adhesion receptors in MDA-MB-231, we performed co-immunoprecipitation assays as described in materials and methods. Cell lysates were precleared and precipitated with control antibody, anti-CD44, anti-β1integrin, or anti-CSPG4 antibodies. The precipitates were separated on SDS-PAGE and blotted with anti- NEDD9 antibody. NEDD9 was specifically detected in anti-CSPG4 precipitate but not in precipitates of the other cell adhesion receptors. Thus, our results suggest a novel model in which NEDD9 and CSPG4 form a molecular complex in MDA- MB-231 cells that stimulates formation of a signaling complex that contains various kinases such as FAK and Src, which play a key role in promoting migration, invasion, and proliferation in a type I collagen matrix. ## Gene Expression of CSPG4 in Human Breast Cancer Tissues One-way between-subjects analysis of variance (ANOVA) was performed on the CSPG4 gene expression, as a function of breast cancer subtypes for basal, Her2-like, Luminal A, and Luminal B subtypes as determined by PAM50. The assumption of normality of error was met for each subtype. The assumption of homogeneity of variance was not met with Brown-Forsythe-test F (3, 510) = 37.63, (*p* = \<.0001); however, using the Welch test of means to adjust for heterogeneity of variance did not change the final results. Therefore, the results of ANOVA are shown. We observed that there was a significant difference in the expression of CSPG4 among the subtypes, F (3, 510) = 24.62, (*p*\<.0001). Post hoc pairwise comparisons were performed using the Bonferroni adjustment. Tumors from patients with the basal subtype had significantly higher CSPG4 expression than the patients with other subtypes, with all the *p* values \<.0001. In addition, Luminal A tumors also had a significantly higher CSPG4 expression than Luminal B tumors (p = 0.0006). # Discussion It is widely recognized that triple negative (TN) breast cancers have a poorer prognosis than any other subtypes of breast cancer. In order to develop strategies and therapeutic reagents for this phenotype of breast cancers, we tested FH535 a compound known to inhibit invasion, migration, and growth in vitro in other types of cancer. In this study, we utilized MDA-MB-231 and HCC38 cells to evaluate the canonical WNT-signaling pathway in triple negative (TN) breast cancers since this pathway is activated evaluated by LRP6 expression, suggesting that the WNT-signaling pathway is activated in these cell lines. In this study, we demonstrated that FH535 inhibited migration, invasion, and growth of breast cancer cell lines. In a three-dimensional collagen matrix, which is considered as a model system for evaluating cancer growth in vivo, FH535 selectively inhibited TN breast cancer cell lines (i.e. HCC38 and MDA-MB-231) but not breast cancer cell lines from other subtypes (i.e. SK-Br3, T47D, and MCF-7). These results suggest that the canonical WNT-signaling pathway could be a target for patients diagnosed with TN breast cancer. Previous studies suggest that FH535 inhibits the transcription of TCF4, which is a critical protein regulating gene expression in colon cancer cells. We did not observe similar inhibition by quantitative RT-PCR analysis in HCC38 cells (not shown). Instead, our results suggest that FH535 treatment reduced the protein levels of total β-catenin, which was different from what was found in previous studies. As one mechanism for reducing β-catenin protein, we demonstrated that FH535 treatment increased the levels of axin. Axin has been characterized as a tumor-suppressor gene and plays a key role in inhibiting the canonical WNT- pathway by forming molecular complexes with other proteins such as GSK3 and APC. In contrast, FH535 did not affect expression of β-catenin or axin in MDA-MB-231 cells, consistent with the previous studies. Thus, it is possible that FH535 would exert its inhibitory activity for migration, invasion, and growth through distinct mechanisms in a cell-type dependent manner. Enhanced migration and invasion through connective tissue is one of the key steps for establishing metastasis. Since type I collagen is a major component of connective tissues, we tested FH535 as a potential inhibitors in migration and growth in a type I collagen matrix. In addition to the inhibitory activity of FH535 on cell migration, our results demonstrated that FH535 significantly inhibited invasion into basement membrane (matrigel) in both MDA-MB-231 and HCC38 cells, consistent with the previous studies with human melanoma. Our results suggest that the canonical WNT-signaling pathway would play a key role in tumor dissemination from the primary tumor by enhancing migratory and invasive abilities of tumor cells *in vivo*. Invasion of breast cancer cells into matrigel is mediated by complex molecular mechanisms involving matrix degradation and/or cell shape change and includes activation of small GTPases and cytoskeletal reorganization. For example, it has been reported that collagenases such as MMP-1, MMP-13, and MMP-14 play a key role in promoting invasion and metastasis possibly by facilitating pericellular degradation of matrix proteins. Recently, Poincloux et al, proposed a unique model for invasion of MDA-MB-231 cells into matrigel, in which traction forces for cell movement in the matrix are generated by actomyosin-mediated formation of molecular complexes of β1integrin and f-actin on the cell surface. In this regard, our current research focus is to characterize the mechanisms of the canonical WNT-pathway signaling that induce expression of matrix degrading enzymes and/or organizes cytoskeletal architectures in HCC38 and MDA-MB-231 cells toward a better understanding of cancer cell migration and invasion. Recent studies suggest that three-dimensional (3D) cell culture models emulates a more physiologically relevant microenvironment. For example, down-regulation of β1integrin and growth factor signaling pathways in 3D culture systems resulted in reversion of the malignant phenotypes which were demonstrated as growth arrest and reformation of tissue polarity. Keely and colleagues reported that a tumor microenvironment consisting of a type I collagen matrix regulates breast cancer proliferation and differentiation. In this study, we demonstrated that FH535 significantly inhibited growth of TN breast cancer cell lines (i.e. HCC38 and MDA-MB-231) but not cell lines from other subtypes, suggesting a selectivity of growth inhibition of FH535 in breast cancer cells. Interestingly, FH535 did not induce apoptosis in MDA-MB-231 nor HCC38 cells evaluated staining with Apoptag (**not shown**). The mechanisms of tumor growth in a 3D type I collagen matrix is not clear at present, however, the requirement of signaling pathways to promote growth is likely dependent on cell types. For example T47D cells grow in the matrix through the activation of GTPase (i.e. Rho) and Rho kinase (ROCK). In human breast cancer tissues, immunohistochemical studies suggest that the canonical WNT signaling pathways are activated in TN compared to other subtypes. Given the fact that FH535 is an inhibitor for the canonical WNT signaling pathway, our results suggest a key role of this pathway in facilitating progression and metastasis of TN breast cancers. NEDD9 has been evaluated in tumor progression in various tumor types including breast, lung, and colon cancers. The fact that the NEDD9-null genetic background significantly limits mammary tumor initiation in mice strongly supports the notion of a key role of NEDD9 in promoting breast cancer progression and metastasis. We demonstrated that FH535 inhibited the expression of NEDD9 by western blotting analysis in MDA-MB-231 cells, consistent with the recent studies identifying NEDD9 as a downstream target of the canonical WNT-signaling pathways. Importantly, activation or expression of tyrosine kinases demonstrated to phosphorylate NEDD9 such as FAK and src were not affected by the treatment with FH535. These results suggest that the expression of NEDD9 would be responsible for promoting migration, invasion, and growth of MDA-MB-231 cells. Recent studies suggest that changes in the expression of NEDD9 were a potent prometastatic stimulus in various tumor cells. Thus, our results support the notion that NEDD9 promotes migration and invasion of TN breast cancer cells and would be a promising target for therapy of TN breast cancer. Previous studies demonstrated that integrin and proteoglycan expressed in tumor cells promotes migration and invasion. In this study, we demonstrated that NEDD9 was specifically precipitated with CSPG4 but not with CD44 or β1integrin, suggesting that NEDD9 would form a molecular complex with CSPG4 in MDA-MB-231 cells. CSPG4 was first identified in human melanoma as a melanoma specific antigen and has been evaluated as a target for melanoma therapy. We previously reported that CSPG4 plays a key role in promoting spreading, migration, invasion, as well as growth of malignant human melanoma cells. CSPG4 is also expressed in various cancer cells including breast cancer cells and anti-CSPG4 antibody has been demonstrated to inhibit breast cancer growth and metastasis, suggesting a key role in promoting breast cancer progression and metastasis. Thus, it is possible that CSPG4 expressed on breast cancer cells would transduce signals by forming a complex with NEDD9 for promoting migration, invasion, and growth in surrounding tissues. We are currently identifying key domains of NEDD9 that binds to CSPG4 to further understand the mechanisms of cancer progression and metastasis. Recent studies suggest that CSPG4 plays a key role in maintaining stem cell phenotypes in various tumor types such as melanoma, glioblastoma, and breast cancer cells. The characterization of CSPG4 interaction with NEDD9 may provide insights for understanding mechanisms of growth and survival of these tumor stem cells. Using the TCGA breast cancer gene expression data and the PAM50-based subtypes, we confirmed that the basal subtype breast cancers express significantly higher levels of CSPG4 transcripts compared to tumors of other subtypes as described previously. In addition, there is an apparent higher variance of the CSPG4 expression level in the basal subtype compared to the levels in other subtypes, which may be useful in the study of the heterogeneity of the basal subtype and/or the mechanism for the regulation of CSPG4 expression. Furthermore, Luminal A tumors express significantly higher CSPG4 compared to the Luminal B tumors. It will be interesting to find out whether there is a correlation between CSPG4 expression level and the outcome, which can be examined later when sufficient outcome data are available from the TCGA breast cancer project. Although the mechanism of expression of CSPG4 in basal and Luminal A tumors was not fully characterized, these results suggest that CSPG4 may serve as a target not only for basal but also Luminal A breast tumors. In this study, we demonstrated a pivotal role for the canonical WNT-signaling pathway in enhancing tumor migration, invasion, and growth of TN breast cancer cells, MDA-MB-231 and HCC38. We identified NEDD9 as a potential downstream target of the canonical WNT-signaling pathway and showed that it form a complex with CSPG4 in MDA-MB-231 cells. We are currently identifying target molecules of FH535 in HCC38 by gene expression analyses. Recent studies suggest that the non- canonical WNT-signaling pathway plays a key role in promoting brain metastasis in breast cancer. Thus, more comprehensive characterization of both the canonical and non-canonical WNT-signaling pathways by genomic and biological studies of breast cancer cell lines are clearly required for generating better treatments for patients who diagnosed with TN breast cancer. The authors thank Dr. Patricia Keely (University of Wisconsin, Madison) for guidance with the three dimensional collagen gel assays. We thank to Brenda Deyarmin and Jennifer Kane for advising histology and immunohistochemistry procedures. We thank to Allyson Valente for running qRT-PCR reactions. The opinion and assertions contained herein are the private views of the authors and are not to be construed as official or as representing the views of the Department of the Army or the Department of Defense. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: JI. Performed the experiments: JRL RC JD. Analyzed the data: JI RE SS CL YC HH. Wrote the paper: JI. Critical discussion and revising the draft: RJM CDS.

# Introduction One of the main diseases that affects horses worldwide is popularly known as “pigeon fever”, and it is characterized by the formation of painful abscesses in the pectoral region resembling a pigeon’s breast and sometimes is accompanied by fever. The disease may occasionally be developed in two other forms, infection with purulent abscess formation in the internal organs, which is the most severe form of the disease, or ulcerative lymphangitis, which is the least frequent clinical form and is characterized by the infection of peripheral lymph vessels, severe swelling of the limbs and lameness. The causative agent of the disease is *Corynebacterium pseudotuberculosis* biovar *equi*, a pleomorphic Gram-positive bacteria that, is facultative intracellular, and nitrate-reducing. The reservoir for infection is soil. The transmission of disease occurs through contact with contaminated soil or pus from an infected horse via fly vectors, breaks in the skin or by inhalation.. *C*. *pseudotuberculosis* is also the causative agent of caseous lymphadenitis in small ruminants. However, the strains that cause this disease are grouped into another biovar called *ovis*. At first, the distinction between the two biovars, *ovis* and *equi*, was based mainly on the nitrate reduction capability and results of restriction fragment length polymorphisms (RFLP). Although the incidence of disease in horses suffers strong temporal variations, recent data showed an increase in the number of infections in the USA characterizing this species as a re-emergent animal pathogen. Geoclimatic conditions strongly influence the appearance of disease outbreaks, e.g., high surface land temperature (\>35°C), dry soil, among other factors. These geoclimatic characteristics indicated that outbreaks are favored in hot climate areas with low rainfall. In the USA, high incidence rates of the disease have been described in Texas, Colorado California and several other states. This problem is not restricted to North America, however, and several other strains have been isolated from animals around the world, such as in Chile and Egypt. Until now, no phenotypic or genotypic characteristic of biovar *equi* could be related to the different forms of infection that can be caused by the pathogen. Britz and colleagues evaluated phenotypic characteristics of equine isolates, such as reducing arabinose, sucrose, dextrin or nitrate, but the results showed no statistical correlation between these phenotypes and the location and extent of lesions caused by the bacteria. On the other hand, the genotypic characteristics can best be accessed and analyzed by sequencing the genome of the *equi* strains. Recently, several complete genomes of this biovar were published providing a large amount of biological data that can be used for comparative genomics studies. These analyses have been widely employed in bacterial taxa with a large number of genomes sequenced to determine molecular features that contribute to the virulence and pathogenicity of these species. Comparative genomics enables the differentiation of genotypes based on the genome identity level, allows the characterization of pangenome and their subsets (shared and singletons genes), permits the identification of the evolutionary process (loss and gain of genes), and several other applications. Through comparative genomics, Soares and colleagues identified significant differences in the composition of pathogenicity islands among biovars *ovis* and *equi* of *C*. *pseudotuberculosis* and demonstrated the high clonal behavior of strains that infect small ruminants and a greater plasticity in strains belonging to the biovar *equi*. In this context, this study aimed to sequence and characterize the genomes of seven new strains of the biovar *equi*, isolated from horses diagnosed with *C*. *pseudotuberculosis* infection in the state of California, USA. A comparative genomic analysis was carried out with these seven genomes, together with twelve genomes of the biovar *equi* available in the National Center for Biotechnology Information (NCBI) database, totaling 19 genomes analyzed. Comparative analyses included: pangenome calculation of biovar *equi*; phylogenomic clustering of *Corynebacterium* genus; prediction of genomic islands (GEIs) and prophages; and SNP-based phylogeny. # Results and Discussion ## Major genomic features The main genomic features of the seven strains sequenced in this study are shown in. These characteristics are very similar to other strains of the same species previously sequenced that were isolated from different hosts such as horses cows and even humans. All these strains showed a high GC content (\~ 52%) and carried important virulence factors such as phospholipase D and pili-formation proteins, among other factors. After assembly, the contigs of the seven strains were ordered in a high-quality scaffold with the MAUVE program using the MB11 strain as a reference because its genome order was validated with an optical map. All contigs smaller than 500 bp that were not aligned with their respective scaffolds were evaluated to ensure representation of genes in these genomes. The contigs were analyzed for the presence of coding DNA sequences (CDSs). The contigs with predicted CDSs were subsequently included in the files used in the comparative analysis. Although the genomes had possible gaps in their sequences, their sizes were very similar to other complete genomes of the same species. Of the seven strains described, two were responsible for severe infectious processes that culminated in the death of the animal (MB122 and MB154). The strain MB122 was isolated from the liver of an Arabian horse that was 8 years old with septicemia and had microbial growth in at least three different internal organs. The MB154 strain was isolated from an English thoroughbred that was 14 years old and also had suspected sepsis and bacterial growth in at least three internal organs. The thoroughbred did not respond to antimicrobial therapy with trimethoprim/sulfamethoxazole (TMS). Three other strains were isolated from animals with severe internal infection (MB278, MB295, and MB302). These animals recovered fully after antimicrobial treatment. In contrast, the strains MB44 and MB336 showed less severe infections without the need for antibiotic treatment (Pectoral abscesses). Initially, a comparative ring was generated with BRIG v.0.95 software (BLAST Ring Image Generator) to analyze the identity between the seven sequenced genomes using blastn. Five other genomes of the biovar *equi* from the strains isolated in California (MB20, MB11, MB14, MB30, and MB66) were also used in addition to the genomes of the phylogenetically related species *Corynebacterium ulcerans* BR-AD22 and *Corynebacterium diphtheriae* NCTC 13129. In the rings of, it is possible to notice that the 12 genomes of biovar *equi* showed high identities (between 90–100%) without the formation of large visible gaps. From the analysis of 16S rDNA, *C*. *ulcerans* has been described as a species of the genus *Corynebacterium* that is more closely related to *C*. *pseudotuberculosis* and, consequently, showed several genomic regions of high identity with the isolates from California. On the other hand, *C*. *diphtheriae* notoriously showed a lower identity level. ## Insights on the genetic content of GEIs and prophages The detection of GEIs was performed with the program GIPSy v.1.1.2. Two types of GEls were detected and analyzed in this study, Pathogenicity Islands (PAIs) and Resistance Islands (RIs). The PAIs contain a large number of virulence genes grouped into blocks that play important roles in the pathogenicity of the microorganism. In principle, eighteen PAIs were identified in six of the seven sequenced genomes. Only the MB336 strain showed 21 PAIs. The PAIs presenting a prediction score of "weak" were not taken into consideration. These numbers were higher than the 16 PAIs previously identified by Soares and colleagues. The number of virulence factors of each sequenced genome is presented in. The *C*. *pseudotuberculosis* MB44 was the strain with a higher number of detected virulence factors. To confirm the detection of PAIs, the genome of the strain MB122, which showed 18 PAIs, and the genome of the strain MB336, which showed 21 PAIs, were compared and analyzed by blastn with the Artemis Comparison Tool (ACT). In the graph obtained with the ACT, it is possible to observe that the genes belonging to the unique islands from MB336 were also present in the genome of the strain MB122 and other isolates from California, which suggest that the conservation of these PAIs is higher than that observed by GIPSy. These PAIs contained important virulence factors for *C*. *pseudotuberculosis*, such as the cluster of Oligopeptide Permease (Opp) genes, which has been shown to have great importance in the pathogenicity of species, such as *Streptococcus* group A (GAS) and *Bacillus thuringiensis*. These findings suggest that the virulence of *equi* biovar is directly related to its high plasticity, and the genomic sequencing of new genomes may assist in determining a more robust set of virulence factors. The number of RIs was quite variable. A total of 17 RIs was predicted for the strains MB122, MB278, and MB295, while strains MB44, MB154, MB302, and MB336 showed 16 RIs, 22 RIs, 15 RIs and 19 RIs, respectively. The numbers of genes encoding proteins involved in antibiotic resistance are described in. Interestingly, the strains with the highest number of RIs had a lower number of detected resistance genes. Among them is the MB154 strain, which causes a severe infection with septicemia. Despite these genotypic differences described above, Rhodes and colleagues found no significant phenotypic differences in the antimicrobial resistance profiles of strains of *C*. *pseudotuberculosis* isolated over 16 years (1996–2012). The nucleotide sequences of RIs for the seven sequenced genomes and other five genomes of bacteria isolated in California were recovered with Artemis software and analyzed using fragmented all-against-all comparison with the Gegenees software v. 2.2.1. The comparative results of Gegenees are presented as a heatmap containing the percentage of identity between the samples. This heatmap was analyzed with SplitsTree4 v.4.14.2 for the calculation of a Neighbor-Joining dendrogram. The strains with the largest number of islands (MB154 and MB336) were more closely related and grouped in a distinct clade. These results indicated, despite the lower number of resistance genes for these two strains, the genes were clustered into a greater number of islands and, therefore, were acquired from a larger number of horizontal gene transfer events when compared to the other strains, generating greater variability. However, this gained genetic content was not enough to influence a change in the antimicrobial resistance profile. A high number of genes related to resistance against antibiotics in *C*. *pseudotuberculosis* biovar *equi* were detected including beta-lactamase enzymes, membrane permeases, efflux pumps and others. Two predicted PAIs carry the gene clusters *spaA* and *spaD* responsible for the expression of pili proteins. A Pilus is a filamentary cell surface structure that provides adhesive and invasive functions to the cell of the pathogen. Genes of both clusters were obtained from the genome of *C*. *pseudotuberculosis* 1002 biovar *ovis*. Genes were compared by blastn with the nucleotide sequence of the complete genomes of strains 1002 (biovar *ovis*), MB11 (biovar *equi*, caused ulcerative lymphangitis), MB14 (biovar *equi*, infection with internal abscess formation), MB30 (biovar *equi*, infection with external abscess formation), and CIP 52.97 (the outmost taxon of biovar equi). The comparison was performed using the web tool Simple Synteny. Those genes with a minimum coverage cutoff of 50% were identified and were represented in. The *spaD* cluster showed high conservation in all strains analyzed, while *spaA* cluster presented some genomic rearrangements. Strains of the biovar equi showed a deletion in the region that should contain the sortase gene *srtA*. A gene prediction followed by manual curation identified *srtA* as a pseudogene in all strains of biovar equi. Sortase enzymes are essential for the assembly and anchoring of pilus structure in the bacterial cell wall, and this type of arrangement has been demonstrated by Soares and colleagues for 16 genomes of *C*. *pseudotuberculosis*. All the three strains of biovar equi analyzed have the same arrangement of pili genes which potentially rejects the possibility that these genes may have a major influence on the type of infection caused by the pathogen. In addition to the detection of PAIs and RIs, the prophages were identified with the online tool PHAST. Six incomplete prophages were detected containing both bacterial and phage genes that were interleaved. These regions showed high conservation both structurally (position of genes) and functionally (high identity) in the twelve genomes of Californian isolates. Of the 66 genes detected in prophages, 24 were uncharacterized proteins (approximately 36% of the total genes). One of the conserved prophages was related to phage E3 of the *Geobacillus* sp., an extremophile genus commonly isolated from soil samples, in the main environmental habitat of *C*. *pseudotuberculosis*. This prophage contains the sequence of the alternative sigma factor M (*sigM*), which is involved in bacterial adaptation to changes in the extracellular environment, such as oxidative stress. Interestingly, downstream of *sigM* in prophage E3 we found the *trxB* and *trxA1* genes, which encode thioredoxin enzymes that have an antioxidant function and are regulated by *sigM*. Combating oxidative stress is necessary for the persistence of *C*. *pseudotuberculosis* in the host organism because the bacteria behaves as an intracellular parasite. After phagocytosis, they are exposed to the harsh environment of the phagosome, which contains a high concentration of harmful oxidizing agents to the cell of the invading pathogen. In a second incomplete prophage, OH2, the *xerC* gene coding for a phage integrase containing the tyrosine recombinase domain was detected. The recombinases are involved in the process of excision/integration of mobile genetic elements in bacterial genomes and thus are responsible for genome rearrangements and acquisition and/or gene silencing. Nevertheless, the *xerC* gene is in a genomic region highly conserved without visible genomic rearrangements. Downstream *xerC* gene is a non-characterized protein containing a domain of the M23 family of metallopeptidases, a group of endopeptidases that catalyze the hydrolysis of the bacterial peptidoglycan, leading to cell lysis. These proteolytic enzymes are important virulence factors of the pathogen because they must overcome other bacterial cells of the host microflora to succeed in the invasion process during infection. The presence of a metallopeptidase, besides a recombinase gene, is conserved even in temporally and spatially distant genomes such as *C*. *pseudotuberculosis* CIP 52.97, which indicates an ancient and widespread event of horizontal gene acquisition within the genus *Corynebacterium*. ## Phylogenomic inference and SNP-based clustering Two approaches were performed to evaluate the genomic and evolutionary similarity of the species *C*. *pseudotuberculosis*. The first used PGAP V.1.11 software, which took into account the core genome to calculate dendrograms based on the Neighbor-Joining model. The unrooted tree of indicates a high distinction between the species *C*. *pseudotuberculosis*, and the rest of the environmental and pathogenic species of the genus *Corynebacterium*. *C*. *vitaeruminis*, *C*. *diphtheriae*, and *C*. *ulcerans* converged, respectively, to form an evolutionary path to the species *C*. *pseudotuberculosis*. *C*. *ulcerans*, therefore, is the species of the genus most closely related to the species *C*. *pseudotuberculosis*, corroborating the phylogenetic analysis of 16S rDNA. The nucleotide identity between these two species can be observed in the rings of. No phenetic distinction between species that cause internal and external abscesses could be observed within the clade of *C*. *pseudotuberculosis* biovar *equi*. However, it is evident that there was greater proximity between the strains isolated in California in comparison with the other species from biovar *equi*. It is worth noting that the strains CIP 52.97 and 1/06-A are more closely related to each other and form an outer clade, distinct from other strains of the biovar. The CIP 52.97 strain is the one with greater temporal distance from the other analyzed strains being isolated in 1952 and kept in the Pasteur Institute Collection. In the second approach, only the strains belonging to the biovar *equi* were compared using Gegenees v.2.2.1 software. In this analysis, the genomes were fragmented in blocks of 200 bp, and these blocks were compared all-against-all with blastn, generating a heatmap based on standardized blast values. From these values, a Neighbor-Joining dendrogram was calculated with the SplitsTree4 software, as described above for the RI analyses. As shown in, the analysis with Gegenees indicated CIP 52.97 and 1/06-A as the outer strains of biovar *equi*. It was also observed that the strains MB154 and MB122, which caused more severe infectious conditions and challenging treatment, were closely related in both phylogenomic analyses. However, no genome region of these two strains significantly differed from the genomes of other strains that caused less severe infections with external abscess formation. As expected, the topography of the dendrogram calculated only for the RIs was quite different from that observed in the dendrogram calculated for the entire genomic sequence. While the dendrogram of RIs grouped species with similar clinical history, the second dendrogram distinguished the species isolated in California from the rest of the strains of biovar *equi*, analogous to that observed in the tree shown in. A clustering analysis based on the SNPs detected in the core genome of genus *Corynebacterium* was performed. Maximum likelihood method was used to calculate the tree. As the gene density in prokaryotes is very high, SNPs can drastically change protein structures and consequently modify the bacterial phenotype. However, the topography of the SNP-based tree was similar to that constructed using a Neighbor-joining method based on the gene content of the core genome. These results emphasize the phenetic and cladistic distance between the isolates of California and the other strains of biovar *equi*. ## Pangenome evaluation of biovar *equi* The α value of Heap's Law was calculated based on the results of the pangenome obtained in PGAP V.1.11 to characterize the genomic plasticity level of biovar *equi*. Eighteen genomes available in the NCBI database (seven of them sequenced in this study) were used. Previously, Soares and colleagues demonstrated that *C*. *pseudotuberculosis* had an open pangenome (α value = 0.89) and the *equi* biovar had higher plasticity compared to biovar *ovis*. The analysis of Soares and colleagues was performed with the EDGAR software. The pangenome graph shown in indicates that, even after a significant increase in the number of sequenced genomes from biovar *equi*, the value of α = 0.849 indicated that this pangenome was still open. This means that the sequencing of new genomes will possibly add new genes not described in this pangenome. The pangenome of biovar *equi* consists of 3,183 genes, with a core genome of 1,355 genes, which was less than half of the genes described in the pangenome (approximately 42.5%) and is a result that enhances the high genetic variability of this biovar. This percentage of the core genome is one of the lowest reported in the literature compared with the pangenomes of other bacterial species, such as *Escherichia coli* that has 44% of core genome, *Pseudomonas syringae* with 64%, *Streptococcus pneumoniae* with 74%, and *Listeria monocytogenes* with 80%, among others. It is possible to note in the flower graph of that the number of singletons was highest in strains CIP 52.97, 1/06-A and 316. However, the number of singleton genes is quite low in the remaining genomes; e.g. singleton genes of MB44 correspond to only 2% of the total predicted CDSs. Thus, doubletons (genes shared by two strains) and other accessory genes contribute more significantly than singletons to the high variability of the biovar. The singleton genes were classified into categories of the Clusters of Orthologous Groups (COGs) using the Batch Web CD-Search Tool (<https://www.ncbi.nlm.nih.gov/Structure/bwrpsb/bwrpsb.cgi>). Most genes were not classified into COGs and therefore their biological functions are unknown. The most represented COG categories were M—Cell wall/membrane biogenesis and L—replication, recombination and repair. Some substantial differences were observed between the genomes of strains isolated in California and the genomes of strains CIP 52.97 and 1/06-A, which were phylogenetically more distant within biovar *equi*. Therefore, the strains with complete genomes, MB14, CIP 52.97 and *C*. *diphtheriae* NCTC 13129, were compared with blastn, and the results were visualized with ACT. Five insertion- regions were detected in the genomes of isolates from California. It is also possible that these regions were deleted from the genome of CIP 52.97. One of these regions was found upstream of the gene *mas*. This gene is a virulence factor encoding the enzyme mycocerosic acid synthase, which is responsible for the production of phenolic glycolipids. These substances are the main components of the waxy material of the cell wall of mycobacteria and are largely responsible for its pathogenicity. The insertion/deletion observed contained six genes that encode uncharacterized proteins and a seventh gene coding for a protein with a resolvase domain, which is very common in endonucleases involved in DNA recombination. Interestingly, *C*. *diphtheriae* has a prophage sequence of approximately 15 Kbp inserted in the same genome site. This sequence showed no similarity to the sequence found at the same location for the isolates from California. Therefore, this is probably a common recombination site of *Corynebacterium* genus, where the events of horizontal gene transfer that occasionally happened in *C*. *pseudotuberculosis* and *C*. *diphtheriae* were evolutionarily independent. Five genes of uncharacterized proteins were analyzed with the online tool InterProScan to detect conserved protein domains. Two CDSs had no predicted domains. The other three had a catalytic domain of phage integrase (IPR013762), a P-loop containing nucleoside triphosphate hydrolase (IPR027417) and a HNH nuclease domain (IPR003615). The latter is commonly found in bacteriocins, which are proteins secreted by microorganisms that have antimicrobial activity against closely related bacterial strains. Thus, it is probably an important factor that contributes to the long persistence of *C*. *pseudotuberculosis* in environments outside the host, such as soil. The identification of an uncharacterized protein containing a phage integrase domain indicates that *C*. *pseudotuberculosis* genomes probably have a larger number of incomplete prophages than just the six identified by PHAST. Moreover, these results emphasize the need to analyze the uncharacterized proteins of *C*. *pseudotuberculosis*, as these may play important roles in the pathogenesis and persistence of the bacteria in the environment or the host. In addition to the features described above, a peptidase S8A classified in the family of subtilisin-like serine protease (SRSP) was found upstream of gene *mas*, which has been previously described as an important virulence factor in *C*. *pseudotuberculosis* FRC41, one of the few strains isolated from human hosts. In the strain CIP 52.97, this virulence factor is a pseudogene due to the presence of a deletion in the genome and, hence, is probably non-functional, while in the MB14 strain this CDS is complete, without the presence of frameshifts. Five other serine proteases were found in the genomes of isolates from California, which is a number larger than that described for the human strain FRC41. These findings demonstrated the vast repertoire of virulence factors present in the genomes of the Californian isolates. # Conclusions In this study, seven new genomes of the re-emerging pathogen *C*. *pseudotuberculosis* biovar *equi* were sequenced and assembled in high-quality scaffolds with a size of 2.3 Mbp. These genomes showed molecular characteristics that were very similar to the complete genomes of other strains from biovar *equi*. Comparative genomics results demonstrated that *C*. *pseudotuberculosis* carries an extensive repertoire of virulence factors and resistance genes in its genome, e.g., peptidase enzymes, phage integrases, recombination endonucleases, micoside synthesis enzymes, bacteriocins with antimicrobial activity, beta-lactamases, antibiotic efflux pumps and other molecular factors that shape the host-pathogen interaction process. The pangenome demonstrated that the genetic variability of the biovar was greater than previously thought. No genotypic characteristic that distinguished the strains according to the type of infection has been identified. In fact, regardless of the type of infection, isolates of California showed greater genomic similarity and phylogenetic proximity with each other than with the rest of the strains of biovar *equi*. The CIP 52.97 and 1/06-A strains were related less to the remaining strains of the biovar. The vast majority of singletons detected in the pangenome of *C*. *pseudotuberculosis* biovar *equi* were uncharacterized proteins and more detailed analysis of at least three of these proteins into the genome of strain MB14 identified probable new factors that contribute to the virulence of the species. This emphasizes the importance of characterizing proteins with unknown function in bacterial pathogens. # Methods ## Human or animal subjects and/or tissue or field sample The study was not submitted to any committee. The experiments do not require the approval of committees, as they involve the genome sequencing of previously isolated bacterial strains. ## Data collection, isolation, and cultivation The *C*. *pseudotuberculosis* strains described in this study were previously isolated from horses in the state of California, USA. Information about the animal host, date and site of isolation of the microorganism, and the clinical condition of the horses can be observed in. presents this information in summary form. Abscess samples were obtained and used for the cultivation of *C*. *pseudotuberculosis* in Brain-Heart Infusion (BHI). After isolation, the bacteria were maintained in 25% glycerol at -80°C. For extraction of the genomic DNA, the strains were also grown in BHI medium at 37°C with shaking. The extraction occurred when the optical density of the culture reached the mid-log phase of bacterial growth according to the procedures described by Pacheco and colleagues for clinical isolates of *C*. *pseudotuberculosis*. To evaluate the DNA extraction protocol, an aliquot of DNA was migrated in a 1% agarose gel. ## Sequencing, assembly and annotation of genomes The seven genomes in this study were sequenced with the Ion Torrent PGM platform (Thermo Fisher Scientific) using the chip 318 according to the manufacturer's protocol. The quality of sequenced reads was visualized with FastQC software (<http://www.bioinformatics.babraham.ac.uk/projects/fastqc/>) and, when necessary, was trimmed and filtered using a Phred score of 20. The assembly was performed with MIRA 4 software and the redundancies of the contigs were removed with the SeqMan Pro tool of the Lasergene software (DNASTAR). Most of the remaining gaps were manually closed using local blastn or using GapBlaster software, which uses a reference genome to map sequenced reads to generate sequences that closed the gap. The complete genome of *C*. *pseudotuberculosis* 316 was used as a reference. The final few contigs were ordered in software MAUVE using *C*. *pseudotuberculosis* MB11 as the reference genome because its assembly was confirmed by the *in vitro* optical map. The high-quality scaffolds were submitted to the web software Pannotator for automatic annotation using the *C*. *pseudotuberculosis* 316 as the reference genome. Subsequently, the predicted CDSs were manually cured with Artemis v.14.0.0 software to meet the gene annotation criteria of UniProt (<http://www.uniprot.org>) and to remove pseudogenes formed due to errors of the sequencing platform. The level of genomic similarity between the sequenced isolates was assessed by blastn and visualized as a circular map created with the program BRIG v.0.95. The Whole Genome Shotgun project of each genome has been deposited at GenBank under the accession numbers MCOC00000000, MCOB00000000, MCOA00000000, MCNZ00000000, MCNY00000000, MCNX00000000, and MCNW00000000. The versions described in this paper are versions MCOC01000000, MCOB01000000, MCOA01000000, MCNZ01000000, MCNY01000000, MCNX01000000, and MCNW01000000. ## Identification and analysis of GEIs and prophages The PAIs and RIs were identified with GIPSy v1.1.2 software. *C*. *glutamicum* ATCC 13032 was used as a non-pathogenic strain of reference. The complete lists of virulence factors, resistance genes, locations and composition of PAIs and RIs can be viewed in the. The nucleotide sequence of each RI was recovered with Artemis v.14.0.0 software. The RIs of each strain were grouped into a multi- fasta and compared all-against-all using Gegenees v.2.2.1 software. As a result of the comparison, a heatmap containing the percentage similarity of these islands was generated, which was subsequently used for calculating a Neighbor- Joining dendrogram with SplitsTree4 v.4.14.2 software. The sequence of the genes from each PAI was also recovered using Artemis. These sequences were compared with the genome of the strain MB20 using tblastx in CGView Server to create a circular map of the location of these PAIs. The identification of these PAIs was subsequently confirmed by blastn between the genomes of MB122 and MB336, which were observed in the ACT software. A database of pili genes was created from the genome of *C*. *pseudotuberculosis* 1002. Those genes were compared to the nucleotide sequences of strains 1002, MB11, MB14, MB30, and CIP 52.97 using blastn with a minimum coverage cutoff of 50%. Analysis was performed in the web tool Simple Synteny. The identification of prophages in the genomes was performed with the web tool PHAST (Phage Search Tool). A database containing all the genes identified in these prophages was generated to evaluate their conservation in the analyzed genomes. These genes in.fasta format were compared with the nucleotide sequences of the twelve genomes of Californian isolates generating a circular map in BRIG software. ## Phylogenomic analysis and SNP-based clustering of *Corynebacterium* The genomes of twelve reference species (RefSeq) from the *Corynebacterium* genus were recovered from the National Center for Biotechnology Information (NCBI) database in the GenBank file format. These files were used to determine the phylogenetic relationship among the isolates from California with the rest of the species from the *Corynebacterium* genus. All available genomes of biovar *ovis* of *C*. *pseudotuberculosis* were also used in the analysis. The construction of the phylogenomic tree was performed using two approaches. At first, the construction of a phylogenetic tree in the Neighbor-Joining model was performed using the core genome of the genus calculated in PGAP v.1.11. The tree was displayed and edited in the SplitsTree4 program. In the second approach, only the 19 genomes of the biovar *equi* were used. Therefore, the nucleotide sequences of these 19 genomes were compared all-against-all using blastn in the Gegenees program. For comparison, we used the protocol of high accuracy with the fragmentation of genomes in 200 bp blocks with a step size of 100 bp. The similarity between the genomes was plotted as a heatmap containing percentage identity values. These values were used to construct a Neighbor- Joining dendrogram in the SplitsTree4 program. A SNP-based genotyping analysis was performed using PGAP software. SNPs of the core genome were used to build a Maximum Likelihood tree that was subsequently analyzed in the SplitsTree software. ## Pangenome analysis of biovar *equi* Initially, the pangenome of the biovar was calculated with the PGAP V.1.11 program using the GF method with the program’s default parameters. The data from 1.Orthologs_cluster.txt file were used for the construction of a flower graph containing information about the core and singletongenes. Subsequently, the 2.CDS.variation.txt file was used for the building of the Heap's Law graph using a script in R. The comparison of virulence factors and resistance genes contained in the genomes was performed using the ACT program. A sequence database of singleton genes from strains of *C*. *pseudotuberculosis* biovar *equi* was obtained using a script in perl developed by our laboratory named getFastaFromOrthologs.pl. The script uses the 1.Orthologs_cluster.txt file to identify singletons and extract their amino acid sequences from the.pep files. These genes were classified in Cluster of Orthologous Groups (COG) using the Batch Web CD-Search Tool (<https://www.ncbi.nlm.nih.gov/Structure/bwrpsb/bwrpsb.cgi>). # Supporting Information The authors thank the Coordenação de Aperfeiçoamento de Pessoal de Nìvel Superior (CAPES) and Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq) that supported this study. [^1]: The authors have declared that no competing interests exist. [^2]: **Conceptualization:** RAB RTJR JME SJS VA AS. **Data curation:** RAB RTJR AAOV KCP LJB MVCV LCG. **Funding acquisition:** AS. **Investigation:** RAB RTJR. **Methodology:** RAB RTJR LCG SJS VA AS. **Project administration:** AS. **Resources:** JME SJS VA AS. **Software:** RAB RTJR AAOV KCP LJB MVCV LCG. **Supervision:** SJS VA AS. **Visualization:** RAB. **Writing – original draft:** RAB. **Writing – review & editing:** RTJR AAOV KCP LJB MVCV LCG JME SJS VA AS.

# Introduction The hardest biological material in the human body is tooth enamel, composed of both mineral and organic phases. Disturbances during amelogenesis impact maturation or mineralization stages, and can lead to defects in enamel translucence, known as enamel hypomineralization. Molar-incisor hypomineralization (MIH) is an example of an enamel condition, and is characterized by the appearance of lesions ranging from white to brownish coloration which present rapid caries progression and hypersensitivity. These asymmetrical lesions affect the first permanent molars, usually with the permanent incisors, and more recently MIH has been reported to affect canines too. Prevalence of MIH varies but has been consistently reported to range between 1% and 35% in all parts of the world, with most reported frequencies around 12%. Regarding the timing of these lesions’ appearance, reports have shown a variation in the chronology of crown completion and enamel formation of incisors and molars. Gleiser and Hunt documented crown completion of first permanent molars to be between 2.5 and 4.4 years. Boyde reported the chronological age of enamel completion in the upper central incisor as 4.64 years and Reid documented complete crown formation in lower central incisors to be at 3.8 years. The enamel maturation stage is particularly important for tissue development, and a very sensitive stage in terms of hypomineralized enamel dysplasia. Different systematic reviews concluded that most of the previously studied environmental factors that were believed to be risk factors for MIH did not fully explain the disease etiology. In contrast, etiological factors associated with enamel forming genes or in immune response-related genes have been reported to play a role in the condition’s onset. Over the past few years, our group has focused on studying different phenotypes that may be present in the same individuals with the hypothesis that they may be influenced by the same underlying genetic variation. Conditions such as cleft lip and palate, have shown an increased frequency of abnormal tooth sizes and morphology. Furthermore, this work has shown that the frequency of dental anomalies as a consequence of disturbances in dental development was much higher in individuals born with clefts, indicating that dental phenotypes should be considered an extended phenotype of oral clefts. Later, it was also shown that *IRF6* and *TGFA* interaction might contribute to oral clefts, and that genetic associations for MIH can be identified when studied in combination with cleft lip and palate. *IRF6* is a gene involved in the structure formation of the oral and maxillofacial regions and has been documented as responsible for not only influence amelogenesis, but also play a role in root and crown anomalies of first molars in mice. This same study showed that *irf6* conditional knockout mouse presented reduced enamel density preeruption and delayed enamel formation. *TGFA* on the other hand, is an essential cell regulator, acting during proliferation, differentiation, migration and apoptosis. *TGFA* has also been associated with orofacial clefts, and a previous study using the case-parent trio design raised the possibility that *IRF6* and *TGFA* interact and lead to orofacial clefts, supporting the finding from our group. In light of the aforementioned evidence, it was hypothesized that *IRF6* and *TGFA* interact and contribute to predisposition of MIH. Additionally, it was hypothesized that environmental factors affecting children that were 3 years of age or older could play a role in the etiology of the disease. Those factors included the presence of any respiratory issues, malnutrition, any type of food intolerance, infection of any sort, and the use of any type of medication. # Results One thousand and sixty-five individuals from both Brazilian and Turkish cohorts were analyzed. All four cohorts are described in detail in the (– Files) and a breakdown of the populations included in the study is described in. All markers were in Hardy-Weinberg equilibrium (PLINK threshold p-value bellow 0.001), with the exception of rs930655 in the sample from Curitiba (Federal University of Paraná), that was excluded from further analysis. It was also excluded rs2073487, rs642961 and rs2902345 in the sample from Rio de Janeiro, and rs1523305 and rs2902345 in the sample from Curitiba (Pontifical Catholic University of Paraná). The numbers in bold in highlight the distribution of the genotypes deviating from Hardy-Weinberg equilibrium. ## Single marker analysis No significant associations between the selected SNPs and MIH were detected in the study populations. ## Gene-gene interaction analyses The cohort from Rio de Janeiro showed a trend toward statistical evidence of interaction between *TGFA* rs1523305 and *IRF6* rs642961 (p = 0.03, OR = 1.77) and between *IRF6* rs2073487 and *TGFA* rs2902345 (p = 0.04, OR = 1.60). Significant results (p\<0.0004) were found for the cohort from Istanbul between *TGFA* rs930655 and all *IRF6* markers. ## Gene expression analysis It was investigated the localization of Tgfa and Irf6 in wild type mice at critical stages of dental development. It was observed expression of Irf6 in ameloblasts, apparently more membrane-bound than cytoplasmic or nuclear, whereas expression of Tgfa was not observed. ## Environmental factors Associations were identified between MIH and any type of medications taken at three years of age. This variable was designed to capture medication used early in life. In this analysis MIH was used as the dependent variable and medication intake as the independent variable. In the Curitiba cohort (Federal University of Paraná), out of 191 individuals who reported having taken medications, 52 were affected by MIH and 139 were not. In contrast, out of 166 individuals who reported not taking any medications, 36 were affected with MIH and 130 were not. In the cohort from Rio de Janeiro, 35 individuals reported having taken medications and were affected by MIH, 15 reported having taken medications and were not affected by MIH, 43 did not take any medications and were affected with MIH and 81 did not take medications and were not affected by MIH. In both groups it was found statistical evidence for an interaction of *IRF6* and *TGFA* genotypes and medication intake at 3 years of age. # Discussion MIH was originally defined as hypomineralization of systemic origin of one to four permanent first molars frequently associated with affected incisors. This definition influenced the work done for more than a decade that focused on the identification of an environmental etiological factor. Our group brought to light the possibility that MIH, in fact, has a multifactorial mode of inheritance, and therefore has a genetic component, which was estimated in a proportion explaining the variation in the population of at least 20% based on data from twins. Further, our group has proposed that MIH is possibly a localized and multifactorial expression of amelogenesis imperfecta. Some argue that the genetic condition affecting enamel is called amelogenesis imperfecta, and that affect all teeth and have a Mendelian mode of inheritance. On the other hand, enamel hypomineralization (enamel hypoplasia or enamel hypocalcification) would be simply caused by impaired ameloblast function and not by altered genes. Therefore, enamel hypomineralization would be a chronological disturbance that would encompass MIH. This line of thought has led to the hypothesis that the cause of the impaired ameloblast function could be identified because that would be the result of systemic or local causes and this hypothesis has hindered progress of the field for the last 20 years. MIH, like dental caries or periodontitis, fits well in the complex or multifactorial inheritance framework, and similar to cardiovascular diseases as an example, is determined by more than one gene and can be influenced by the environment. The evidence that MIH frequency varies across the world is supported by the results found in the present study. The association found in the Turkish population was different from the results obtained in the cohorts from Brazil, which indeed indicates that genetic factors contributing to the MIH phenotype may vary depending on geographic origin. Both MIH and clefts apparently have higher frequencies in the north of Europe and this frequency declines as one travels toward the Mediterranean. Interestingly enough, these differences are not clearly seen for the figures from the Americas, with Brazil and the United States with reported frequencies of 10% to 13%. Additional evidence that interferon regulatory factor 6 *(IRF6)* and transforming growth factor alpha *(TGFA)* may interact, as previously suggested in the formation of cleft lip and palate, is also supported here as for the formation of MIH in the Turkish population. It was not possible to determine the exact mechanism underlying this population-specific association. The interpretation of these results should be analyzed cautiously. When one looks at the frequency of the less common allele of rs17015215 in the Turkish sample, there are two thirds of the individuals affected by MIH in comparison to 1/7 of the individuals that served as comparison. The frequency of the less common allele of rs930655 is also low: 9/39 in the affected individuals and 9/54 in the unaffected. When using these small numbers, one obtains very high odds ratios between 25 and 40. The issue of overestimation of odds ratios is well documented in the literature. To avoid the overestimation of the strength of the association, when one wishes odds ratios to be approximations of relative risk, prevalence ratios were calculated, which are the ones presented in. Our group has previously reported an interaction of *TGFBR1* and childhood pneumonia as a possible gene-environment mechanism that increases the chance of MIH occurrence. Additionally, it was reported statistical evidence of potential interaction between genes related to tooth development and immune response. This present work reports evidence for both gene-gene (*IRF6*-*TGFA*) and gene- environment (*IRF6*-medications taken and *TGFA*-medications taken) interactions, which support the hypothesis that MIH is a complex genetic condition. In the gene expression analysis, it was observed expression of Irf6 in ameloblasts, but expression of Tgfa was not observed. These results can possibly be explained by the timing of the analysis or simply meant that *TGFA* and *IRF6* do not colocalize, and hence, have no direct joint effect of enamel disturbances. The variable medication intake at 3 years of age was created to capture children with history of medication intake from birth and past 3 years and to serve as a surrogate for children more prone to infections and/or illnesses. The 52 children that were included in these analyses had a perfect correlation of medication used before 1 year of age and medication used at 3 years of age. Three years of age was the cutoff chosen with the assumption that mothers would more likely remember that phase since their toddler was able to speak approximately 200 words and being mostly understood. Knowing that enamel development of first permanent molars is completed around 4 years of age, a possible mechanism underlying MIH is an interference in enamel formation due to an organic stress in response to illness or infection. This organic stress could happen at any time between birth and past 4 years of age in children with a particular genetic background that includes hypomorphic alleles of *IRF6*, *TGFA*, and other genes such as *TGFBR1*, and *AQP5*. It is likely that disturbances of first permanent molar development may happen after the first year of life, since mineralization of these teeth is not complete until later. Previous work suggested that MIH is associated with medication intake although this finding was not confirmed by others. The use of medications, which is a surrogate for illness, apparently can lead to MIH depending on genetic variation in *IRF6* and *TGFA*. To properly confirm this hypothesis, the establishment of a cohort of pregnant women that can be prospectively followed from birth to the timing when the permanent dentition is developing is suggested. This approach would allow to minimize issues of recall bias and to provide more definitive answers that the currently published literature could not. The *IRF6*-*TGFA*-medication association with MIH also supports the idea that MIH and cleft lip and palate could possibly be linked. A pleiotropic gene effect could explain this link. Our group has proposed that isolated cleft lip and palate actually is a syndrome that involves disturbances of the dentition and one line of investigation would be to test if genetic variants associated with oral clefts also associate with MIH. We have proposed this approach for isolated tooth agenesis as well, and it provided a tool for gene discovery. Tgfa and Irf6 did not show the same pattern of expression in mouse teeth and the mechanism of how they may interact to lead to MIH is yet to be determined. Among the limitations that can be listed for this present study are the small sample sizes of each independent cohort that may not have allowed the detection of small effect sizes. Depending on the study, clinical information was obtained from reviewing medical records or by the use of questionnaires, and these differences can be source of variation. The fact that detailed information about the type of medication the patients were taking was not available due to self- reporting of this information forced the inclusion of any type of medication intake in the analysis. However, these were not vitamins or other supplements. This kind of observational study that aims to recover information from many years prior suffers from potential issues related to recall bias. For these reasons, the results presented here should be taken cautiously. Another limitation that can be pointed out is that the question on medications taken was asked years after the fact at the moment of participation in the study. Depending on the cohort studied, assessments were done at different age groups (6 to 12, 6 to 10, or at 8 years of age) and one can think this may introduce some variation. In contrast, the strengths of this study include the well characterized phenotype studied here, determined by experienced dentists from different centers, and the diverse geographic populations studied. The combination of these factors increases the confidence in the results obtained. In summary, the present study provided evidence that *IRF6* and *TGFA* might interact and be involved with MIH in certain populations, although none of these genes appear to have by themselves a major role in MIH. This effect may become more likely apparent when children have to make use of medications around the age of three years. Additional studies to test potential gene-gene interactions in diverse populations are warranted in order to confirm results reported by previous studies and the results found here. Study designs aiming to obtain a more detailed information about what specific medications are involved in the development of MIH are also suggested. These studies should try minimize recall bias by having precisely defined research questions, appropriate data collection methods, well-trained interviewers, being prospective in nature, blinding for researchers and patients, and including nested case-control designs. Despite the limitations described for the present study, the results both reinforce the detrimental effects that medication intake can cause to oral health, and the link between genes associated with orofacial clefts and the MIH phenotype. # Methods ## Subjects The total study population consisted of 1,065 individuals from four populations including three Brazilian cohorts and a Turkish cohort. Salivary sample collection and DNA extraction procedures were described previously. Two groups were from Curitiba, Brazil; the first group consisted of 356 subjects (169 females and 187 males) and the study protocol was approved by the Municipal Department of Education and the Committee for Ethics in Research in Human Health Sciences of the Federal University of Paraná (UFPR; approval no. 1.613.829/2016). The second group from Curitiba consisted of 200 individuals (108 females and 92 males) and the project was approved by the Ethics and Research Committee of the Pontifical Catholic University of Paraná, under the protocol number 1.971.986. The third cohort was from Rio de Janeiro, Brazil, consisted of 174 individuals (73 females and 101 males), and the study protocol was approved by the local Ethics Committee for Research (Hospital Universitário Clementino Fraga Filho–HUCFF/UFRJ–approval protocol number 4598514.7.00005257). Finally, the Turkish population consisted of 335 subjects (164 females and 171 males) and the study was approved by the Istanbul University (approval protocol number 2006/2508) Institutional Review Board. Further oversight was obtained at the University of Pittsburgh Institutional Review Board (IRB approval protocol numbers PRO0710045 and PRO12080056). MIH was evaluated using the European Academy of Pediatric Dentistry (EAPD) criteria, and all subjects/guardians read and signed a written informed consent before their participation in the study. ## Cohort from Curitiba, Brazil (Federal University of Paraná): Eligibility criteria, calibration of the examiners and data collection The sample consisted of 356 school children, which were randomly selected from a representative population from the city of Curitiba public schools. To ensure accurate population representation, two-stage cluster sampling was performed. First, two schools from each administrative district were randomly selected. Next, two to three classrooms from each selected school were randomly chosen using the website [www.randomizer.org](http://www.randomizer.org). The children included were all eight years old and presenting four erupted first molars in the oral cavity. Subjects who had orthodontic braces, syndromes, or amelogenesis imperfecta were excluded from the study. The affected group consisted of children with at least one tooth affected by MIH and the comparison group consisted of children who had no teeth presenting hypomineralization. A structured socioeconomic data questionnaire was completed by the children’s caregivers. Four examiners, previously trained and calibrated to diagnose Molar- incisor hypomineralization, selected thirty intraoral photographs of MIH affected teeth in order to define the diagnosis. Sixty photographs of different clinical scenarios that involved manifestations of MIH were then analyzed by the examiners. After one week, the examiners independently analyzed the same photographs in a different order (duplicate examination, kappa \> 0.75). The clinical examination was performed in a school environment using artificial light, a dental mirror, a dental probe with a blunt tip, and sterile gauze. Data collection was performed from November 2016 to September 2017. ## Cohort from Curitiba, Brazil (Pontifical Catholic University of Paraná): Eligibility criteria, calibration of the examiners and data collection This sample consisted of a total of 200 individuals, enrolled in Curitiba municipal schools during 2018. Of these, 100 were affected by MIH and 100 non- affected, 108 were females and 92 were males. This cohort was recruited independently from the cohort recruited by investigators at the Federal University of Paraná, however considering the geographical proximity, there is a possibility that some individuals participated in both studies. Although this was a slight possibility, the data from each group were analyzed separately. The inclusion criteria were children between 6 and 10 years of age, affected by MIH in completely erupted teeth. As for exclusion criteria, individuals undergoing orthodontic treatment or scoring 2, 3, 4, 5 and 6 of ICDAS (International Caries Detection and Assessment System) were excluded from the study population. Two interviewers were calibrated on MIH diagnosis using photographs according to a previously published criterion. The intra-rater reliability was performed one month after to the first assessment for the diagnosis of both case and comparison groups (kappa = 1). The clinical examinations were performed by calibrated dentists in the schools, using natural light, an oral mirror and an exploratory probe when it was necessary. ## Cohort from Rio de Janeiro, Brazil: Eligibility criteria, calibration of the examiners and data collection This sample consisted of 174 individuals, from 7 to 14 (10.13±1.9) years of age, treated at the Pediatric Dentistry Clinic of the Federal University of Rio de Janeiro. Recruitment took place from July 2015 to April 2017, and divided into children with MIH (n = 78) and without MIH (n = 96). The eligibility criteria for this study included patients having all the first permanent molars erupted. The exclusion criteria consisted of children affected by congenital syndromes, enamel defects (hypoplastic lesions, fluorosis, amelogenesis imperfecta or tetracycline stain) and patients undergoing orthodontic treatment. To assess for inter and intra examiner-reliability, a theoretical exercise was previously applied for two days. Twenty clinical images of dental enamel defects including fluorosis, hypoplasia, amelogenesis imperfecta, dental caries and MIH were shown to the examiners (F.M.F.S., M.C.M). Two weeks after the first assessment, a new assessment was carried out with the examiners and the kappas were 0.88 and 0.89, respectively. The subjects’ health data were collected through health records, including the demographic data (child´s age, sex, place of birth, residence, parent’s education and income), mother’s health during pregnancy, and children’s medical history (medication intake, systemic diseases, high fever, malnutrition, asthma, bronchitis, epilepsy, and severe infections). The clinical examination was performed by a calibrated dentist (F.M.F.S). Examinations were carried out using a mirror and a probe, at the dental chair and using artificial light. ## Cohort from Istanbul, Turkey: Eligibility criteria, calibration of the examiners and data collection Eligible individuals were enrolled at the Pediatric Clinic of Istanbul University, Turkey (n = 335, ages ranging from 6 to 12 years). Subjects affected by syndromes, fluorosis, or the ones who had fixed appliances were excluded from the study. Cases were defined as subjects affected by MIH, while controls were defined as subjects with no evidence of MIH. Calibrated examiners carried out the clinical examination, with F.S. having calibrated M.B. and M.K. Exam calibrations were performed according to the following protocol: First, the calibrator presented to the examiner the criteria for MIH detection, showing pictures of several situations to be observed in the exam and discussing each of these situations in a session that lasted one to two hours. Next, the calibrator and examiner analyzed ten to twenty subjects and discussed each case. The intraexaminer agreement was assessed by a second clinical examination in 10% of the sample after two weeks, with a kappa of 1.0. M.K. pre-screened subjects, and M.B. performed the full exam. This cohort and methodology have been reported before. Clinical examinations were performed using a flashlight and intra-oral mirror and gauze was used to dry and clean the teeth prior to examination. Artificial light and dental operatory were used for all evaluations. An explorer was gently used for assessing the smoothness of tooth surfaces. ## Single marker analysis The amount of DNA and the purity of each sample was determined by spectrophotometry (NanoDrop 1000, Thermo Fisher Scientific, US). The DNA concentration was obtained by readings at 260 nm and the purity by 260/ 280 nm proportion. Five single nucleotide polymorphisms (SNPs) in *IRF6* (rs2073487, rs2013162, rs17015215, rs861019, rs642961) and four in *TGFA* were genotyped. PCR reactions were carried out using Taqman chemistry in 3.0 μl reaction volumes in an ABI PRISM Sequence Detection System 7900 (Applied Biosystems, Foster City, CA, USA). Genotyping calls were analyzed using SDS software version 1.7 (Applied Biosystems). PCR reactions were repeated twice when necessary and allele frequencies were calculated. Genotypes are available as (– Files). ## Gene-gene interactions Gene-gene interaction analyses were performed using logistic regression as implemented in PLINK. In order to test the interaction between *IRF6* and *TGFA*, the presence of MIH was considered a dependent variable and the SNPs were considered independent variables. A file containing all *IRF6* genotypes was ran against a covariate file containing the *TGFA* genotypes organized by copies of alleles, and this analysis was repeated with a *TGFA* genotype file against the *IRF6* covariate file. ## Immunofluorescence Expression of Tgfa and Irf6 proteins was performed on paraffin sections from heads of wild type mice at E16.5. Maintenance and handling of mice were approved by the Animal Care Unit at the University of Iowa. Tissues were deparaffinized and rehydrated in a series of ethanol dilutions. Slides were boiled for 20 min in antigen unmasking solution (Vector Laboratories, H-3300). Sections were blocked with 20% donkey serum (sigma, D9663A), then incubated overnight at 4°C with the following primary antibodies: polyclonal goat anti-Tgfa (1:50, R&D, AF-239-SP) and polyclonal rabbit anti-Irf6 (1:50, sigma SAB2102995). After rinsing in PBS, sections were incubated with secondary antibodies conjugated to Alexa Fluorophore 488 or 555 (Molecular Probes, Invitrogen, CA). The nuclei were counterstained with DAPI in PBS (1:10000). The images were taken using a ZEISS 700 confocal microscopy. ## Gene-environment interactions Tests for potential associations between MIH and factors that affected the children of 3 years of age or older in the cohorts from Curitiba (Federal University of Paraná) and Rio de Janeiro were carried out. The variables included the presence of respiratory issues, malnutrition, any type of food intolerance, infection of any sort and any type of medication intake between 3 years of age and the time the sample was collected. This information was part of the questionnaires obtained for both MIH affected and unaffected children. Mothers responded the questions and all participants were asked the same questions, which included medical history of the child (diseases and medications taken) and details of the child’s diet. Vitamins and other supplements were not counted as medications. The variable “medications taken at 3 years of age” was designed due to the evidence that cuspal enamel formation of first permanent molars is not completed until past 3 years of age. These data suggest that the first molar susceptibility window for MIH goes beyond the first year of age. The suggestion that first permanent molar mineralization is completed by one year of age likely limits the ability of unveiling associations. We have complied with the STROBE guidelines in this study. ## Statistical analysis All SNPs in all samples were tested for deviation from Hardy–Weinberg equilibrium using chi-square. In the single-marker analysis, association tests were performed comparing genotypes to phenotype between affected individuals and their respective comparison groups as implemented in the PLINK software. Both odds ratios and 95% confidence intervals were calculated. In order to account for multiple testing, Bonferroni correction was performed and p-values of 0.001 (0.05/36) or below were considered significant. In the gene-gene interaction analysis, p-values below 0.0004 (0.05/108) were considered statistically significant. In the gene-environment interaction analysis the covariates were dichotomous and analyses were performed using logistic regression as also implemented in PLINK. The power necessary to detect significant associations between a gene marker and the phenotype in each cohort was calculated using a Genetic Power Calculator tool. The frequency of the high-risk allele in the population was defined as equal to 0.2, the prevalence of MIH as 0.12, D’ as 1.0 (amount of linkage disequilibrium), the respective number of cases of each cohort and the case- comparison group ratio. When imputing the effect size in the heterozygote form as 1.3 and the effect size of homozygous form as 5.0, it was obtained 69%, 57%, 51%, and 79% power in the cohort from Curitiba (Federal University of Paraná), Curitiba (Pontifical Catholic University), Rio de Janeiro and Istanbul, respectively. Power will decrease with lower effect sizes for both forms, and will increase to 82%, 70%, 64%, and 89%, respectively if the effect size for the homozygous form is 6.0. # Supporting information This paper is based in part on a thesis submitted to the graduate faculty, Federal University of Rio de Janeiro, in partial fulfillment of the requirements for the PhD degree (for F.M.F.S.) and on a thesis submitted to the graduate faculty, Pontifical Catholic University of Paraná, in partial fulfillment of the requirements for the MS degree (for E.G.C.). 10.1371/journal.pone.0241898.r001 Decision Letter 0 Cray Jr. JJ Academic Editor 2020 JJ Cray Jr. This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. # Transfer Alert This paper was transferred from another journal. As a result, its full editorial history (including decision letters, peer reviews and author responses) may not be present. 9 Sep 2020 PONE-D-20-24540 Gene-environment interaction in molar-incisor hypomineralization PLOS ONE Dear Dr. Vieira, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. The expert reviewers have identified several major issues. Of specific concern is the description of variables, potential missing data, inclusion criteria, approach to analysis and data interpretation. Please submit your revised manuscript by Oct 19 2020 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <plosone@plos.org>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript: A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols> We look forward to receiving your revised manuscript. Kind regards, JJ Cray Jr., Ph.D. Academic Editor PLOS ONE Journal Requirements: When submitting your revision, we need you to address these additional requirements. 1\. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at <https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main _body.pdf> and <https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_titl e_authors_affiliations.pdf> \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: No Reviewer \#2: Yes Reviewer \#3: Partly Reviewer \#4: No \*\*\*\*\*\*\*\*\*\* 2\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes Reviewer \#2: Yes Reviewer \#3: I Don't Know Reviewer \#4: No \*\*\*\*\*\*\*\*\*\* 3\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes Reviewer \#2: Yes Reviewer \#3: No Reviewer \#4: Yes \*\*\*\*\*\*\*\*\*\* 4\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: No Reviewer \#3: Yes Reviewer \#4: Yes \*\*\*\*\*\*\*\*\*\* 5\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: In this study, the authors analyzed causes of Molar Incisor Hypomineralization (MIH). The term “MIH” came into use 20 years ago, and causes of MIH are unknown. This reviewer thinks that the topic is very important for dentists. However, the following points should be clarified prior to further consideration of publication of the manuscript. MIH is defined as a hypomineralization of systemic origin of one to four permanent first molars frequently associated with affected incisors by Weerheijm (2001). Ref: Molar incisor hypomineralisation (MIH) Eur J Paediatr Dent. 2003;4(3):114-20. The authors described that these enamel defects usually occur when there are disturbances during the mineralization or maturation stage of amelogenesis. The first molar is finished mineralization of crown by 3 years old. However, the authors concluded that environmental factors affecting children that were 3 years of age or older were also hypothesized to play a role in the disease etiology. The authors also described that genes interacted and contributed to predisposition of MIH. If genetic factors cause hypomineralization, the disease is not MIH but amelogenesis imperfecta. Genetic abnormalities cause hypoplasia in all teeth rather than locally. Reviewer \#2: The English language needs to be professionally revised. There are straightforward grammar issues. For example paragraph 4 in introduction “Risk factors for MIH did not fully explained” … and paragraph 5 in introduction “has been documented as responsible for not only affect amelogenesis…”. “The chance of the less common allele of TGFA rs930655 in addition to the IRF6 marker allele to increase the chance …” Reviewer \#3: This manuscript deals with an interesting topic and currently one of the forefront problems in the field of paediatric dentistry, i.e. MIH, focusing on its aetiology. However, there are some issues with this manuscript. In comments, only methodology and results of the study are addressed. There is no information on how children were selected. Data on children’s age is missing. Who performed dental examinations? Were the dentists who performed dental examination calibrated? Regarding MIH diagnostic criteria; description of the dental examination results related to MIH affected teeth is essential (e.g. which teeth are MIH-affected, what was MIH severity). Text describing environmental factors is very vague. How was the data on environmental factors obtained (e.g. medical history, a questioner)? Who answered the questions about environmental factors (e.g., children, parents, teachers..)? Were all participants asked the same questions? On which environmental factors that could be associated with MIH were participants asked? Why do the authors proceed from a hypothesis that environmental factors affecting 3 year-old children or older play a role in the aetiology of MIH? In MIH, the insult to the ameloblasts is likely to occur either prenatally or in the first year of life (Mangum and Farah). Since the aim of the study was to clarify the aetiology of MIH, it is not clear why various potentially harmful aetiological factors would be considered as a single potential cause of MIH. The results of the study did not show significant association between the selected SNPs and MIH. The study also does not provide solid evidence for the IRF6 and TGFA interaction, nor that the development of MIH is associated with any of these genes. Moreover, as the authors have already stated themselves, the size of each cohort is small for such research. Figure 1: A description of each of the six images is missing. Reviewer \#4: Thank you for the opportunity to review this manuscript, ‘Gene- environment interaction in Molar Incisor Hypomineralisation’. I congratulate the authors for undertaking very broad and comprehensive approach to investigating genetic interactions in MIH. However, due to two major concerns with this paper, I cannot support publication of this manuscript. My main concerns are: \(1\) The choice of genetic variants – these appear to have been measured due to their relevance in a different condition (cleft lip and palate) which the authors suggest may be linked. However, adopting this candidate-based approach and then, by stratifying the data when significant results were not obtained so that a ‘statistically significant’ could be reported, seems to misrepresent the real outcome of the study. I suggest there are potentially more biological plausible interactions that could have been investigated that could be supported by existing evidence relating potential genetic and environmental risk factors. I note the authors do acknowledge some of the limitations however these major limitations do not seem to be considered when reporting the study conclusions. \(2\) The observational component is particularly weak – participant recall of ‘medication use’ is prone to many biases and does not contribute to the existing evidence base regarding MIH. In order to investigate a complex causal relationship such as this, the authors needed to develop a detailed analysis plan with consideration of potential confounders etc. \*\*\*\*\*\*\*\*\*\* 6\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No Reviewer \#2: **Yes: **Azza Tagelsir Ahmed Reviewer \#3: No Reviewer \#4: No \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <figures@plos.org>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0241898.r002 Author response to Decision Letter 0 11 Sep 2020 September 9, 2020 To: The Editor, PLoS One Dear Dr. Cray Jr.: after carefully reading yours and the reviewers’ comments, we decided to substantially revise and resubmit our work entitled “Gene-environment interaction in molar-incisor hypomineralization.” We believe we have addressed all the concerns and are hoping it is suitable for publication in your journal. This work explores genomic variants associated with cleft lip and palate as potentially associated with MIH. We are looking forward to seeing our revised work being well received. Below are point-by-point answers for the critiques and all changes are marked in yellow. Sincerely, Alexandre R. Vieira, D.D.S., M.S., PhD Professor of Oral Biology, Pediatric Dentistry and Human Genetics Director of Clinical Research and Director of Student Research Department of Oral Biology University of Pittsburgh School of Dental Medicine 412 Salk Pavilion Pittsburgh, PA 15213 Office: 412-383-8972 FAX: 412-624-3080 E-mail: <arv11@pitt.edu> Point-by-point response to reviewers: Reviewer \#1: In this study, the authors analyzed causes of Molar Incisor Hypomineralization (MIH). The term “MIH” came into use 20 years ago, and causes of MIH are unknown. This reviewer thinks that the topic is very important for dentists. However, the following points should be clarified prior to further consideration of publication of the manuscript. MIH is defined as a hypomineralization of systemic origin of one to four permanent first molars frequently associated with affected incisors by Weerheijm (2001). Ref: Molar incisor hypomineralisation (MIH) Eur J Paediatr Dent. 2003;4(3):114-20. The authors described that these enamel defects usually occur when there are disturbances during the mineralization or maturation stage of amelogenesis. The first molar is finished mineralization of crown by 3 years old. However, the authors concluded that environmental factors affecting children that were 3 years of age or older were also hypothesized to play a role in the disease etiology. The authors also described that genes interacted and contributed to predisposition of MIH. If genetic factors cause hypomineralization, the disease is not MIH but amelogenesis imperfecta. Genetic abnormalities cause hypoplasia in all teeth rather than locally. RESPONSE: The definition cited above from Karin Weerheijm (2001) that the condition is o systemic origin does not exclude the possibility that it has a genetic component. We have suggested this for 8 years now and more and more scientists have surrendered to the idea at this point. We indeed believe MIH is possibly an extension of amelogenesis imperfecta, a localized one, at least in some instances. We added the opening paragraph of the Discussion section to address this reviewer’s concern. Further, we had originally included discussion in the paper regarding the time of mineralization of molars crowns and the variable “infection at 3 years of age.” It was on the original third paragraph of the Discussion section. Reviewer \#2: The English language needs to be professionally revised. There are straightforward grammar issues. For example paragraph 4 in introduction “Risk factors for MIH did not fully explained” … and paragraph 5 in introduction “has been documented as responsible for not only affect amelogenesis…”. “The chance of the less common allele of TGFA rs930655 in addition to the IRF6 marker allele to increase the chance …” RESPONSE: We carefully revised the text for grammar and style and made the above pointed out corrections. Reviewer \#3: This manuscript deals with an interesting topic and currently one of the forefront problems in the field of paediatric dentistry, i.e. MIH, focusing on its aetiology. However, there are some issues with this manuscript. In comments, only methodology and results of the study are addressed. There is no information on how children were selected. Data on children’s age is missing. RESPONSE: This information was in the supplemental file. We moved if to the text. Who performed dental examinations? Were the dentists who performed dental examination calibrated? RESPONSE: This information was in the supplemental file. We moved if to the text. Regarding MIH diagnostic criteria; description of the dental examination results related to MIH affected teeth is essential (e.g. which teeth are MIH-affected, what was MIH severity). RESPONSE: This information was in the supplemental file. We moved if to the text. The original cohorts had originally assigned subjects are affected or not affected. Text describing environmental factors is very vague. How was the data on environmental factors obtained (e.g. medical history, a questioner)? Who answered the questions about environmental factors (e.g., children, parents, teachers..)? Were all participants asked the same questions? On which environmental factors that could be associated with MIH were participants asked? RESPONSE: We added the information as requested. How the data was originally obtained (questionnaire) was original described in the methods. Why do the authors proceed from a hypothesis that environmental factors affecting 3 year-old children or older play a role in the aetiology of MIH? In MIH, the insult to the ameloblasts is likely to occur either prenatally or in the first year of life (Mangum and Farah). RESPONSE: We had originally added some Discussion regarding the variable of infection at 3 years of age. It is quite possible the disruption of amelogenesis happens later as well, not necessarily during the secretion phase only, but during the mineralization phase as well. Since the aim of the study was to clarify the aetiology of MIH, it is not clear why various potentially harmful aetiological factors would be considered as a single potential cause of MIH. RESPONSE: It is not the case we considered the various risk factors as a single potential cause. It is just that our analysis suggest one particular one associated. The results of the study did not show significant association between the selected SNPs and MIH. The study also does not provide solid evidence for the IRF6 and TGFA interaction, nor that the development of MIH is associated with any of these genes. Moreover, as the authors have already stated themselves, the size of each cohort is small for such research. RESPONSE: This is correct and it is clearly stated in the Results and Discussion sections. Suggestive results only come from the combined analyses of different factors. Figure 1: A description of each of the six images is missing. RESPONSE We added the description as requested. Reviewer \#4: Thank you for the opportunity to review this manuscript, ‘Gene- environment interaction in Molar Incisor Hypomineralisation’. I congratulate the authors for undertaking very broad and comprehensive approach to investigating genetic interactions in MIH. However, due to two major concerns with this paper, I cannot support publication of this manuscript. My main concerns are: \(1\) The choice of genetic variants – these appear to have been measured due to their relevance in a different condition (cleft lip and palate) which the authors suggest may be linked. However, adopting this candidate-based approach and then, by stratifying the data when significant results were not obtained so that a ‘statistically significant’ could be reported, seems to misrepresent the real outcome of the study. I suggest there are potentially more biological plausible interactions that could have been investigated that could be supported by existing evidence relating potential genetic and environmental risk factors. I note the authors do acknowledge some of the limitations however these major limitations do not seem to be considered when reporting the study conclusions. RESPONSE: We revised the conclusions to address this concern. \(2\) The observational component is particularly weak – participant recall of ‘medication use’ is prone to many biases and does not contribute to the existing evidence base regarding MIH. In order to investigate a complex causal relationship such as this, the authors needed to develop a detailed analysis plan with consideration of potential confounders etc. RESPONSE: This is an issue inherent to any observational study that includes recalling events many years prior. We added text in the last paragraph of the discussion to acknowledge this. 10.1371/journal.pone.0241898.r003 Decision Letter 1 Cray Jr. JJ Academic Editor 2020 JJ Cray Jr. This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 22 Sep 2020 PONE-D-20-24540R1 Gene-environment interaction in molar-incisor hypomineralization PLOS ONE Dear Dr. Vieira, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. Although the manuscript was given an overhaul, the same themes concerning major issues were again expressed. Please submit your revised manuscript by Nov 06 2020 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <plosone@plos.org>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript: A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols> We look forward to receiving your revised manuscript. Kind regards, JJ Cray Jr., Ph.D. Academic Editor PLOS ONE \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation. Reviewer \#1: (No Response) Reviewer \#2: (No Response) \*\*\*\*\*\*\*\*\*\* 2\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: No Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 3\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: No Reviewer \#2: N/A \*\*\*\*\*\*\*\*\*\* 4\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 5\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: No \*\*\*\*\*\*\*\*\*\* 6\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: This author commented two problems of this manuscript. I couldn't get satisfying answers from authors. The causes of disorders in amelogenesis can be divided into those caused by genes (genetic factors) and those not caused by genes (systemic causes, local causes). The former is hereditary disorder (amelogenesis imperfecta), which affects all teeth. The latter is enamel hypomineralization (enamel hypoplasia or enamel hypocalcification) caused by impaired ameloblast function. Enamel hypomineralization is chronological disturbance. MIH is included in enamel hypomineralization. MIH is a local enamel hypomineralization not amelogenesis imperfecta. Authors responded to my comments as that “MIH is possibility an extension of amelogenesis imperfecta”. The authors also have suggested that genetic abnormality is a one of the factors of MIH. I cannot believe this hypothesis. The disorder is not hypomineralization of enamel but amelogenesis imperfecta. “amelogenesis imperfecta” and “enamel hypomineralization” are often misunderstood as a single disorder and expressed in a term. However, they are actually different disorders. The former means hereditary disorders while the latter means nonhereditary congenital disorders considered as chronological disturbances. If enamel hypomineralization is caused by genetic abnormalities, enamel hypomineralization (MIH) is recognized in all teeth. This reviewer also commented on the timing of enamel mineralization. MIH is defined hypomineralization of enamel affecting affects one or more permanent first molars with or without permanent incisor involvement. Calcification of the first molar begins at birth, and crown formation (mineralization of enamel) completed at 30-36 months. The same opinion was found in reviewer \#3. Reviewer \#3 Why do the authors proceed from a hypothesis that environmental factors affecting 3 year-old children or older play a role in the aetiology of MIH? In MIH, the insult to the ameloblasts is likely to occur either prenatally or in the first year of life. The authors responded that “It is quite possible the disruption of amelogenesis happens later as well, not necessarily during the secretion phase only, but during the mineralization phase as well.” This reviewer thinks that the authors don’t consider mineralization phase of the first molar. Environmental factors affecting children that were 3 years of age or older never concern to MIH. This factor affects permanent teeth except for the first molars. Reviewer \#2: After reviewing the revised submission and the response to the reviewer comments, I would like to inform the authors that NOT all my concerns were addressed. only the English language point was addressed!! Please refer to the attached point by point review document (8 points in total) and provide point by point explanations and amendments within the body of the manuscript. \*\*\*\*\*\*\*\*\*\* 7\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No Reviewer \#2: **Yes: **Azza Tagelsir Ahmed \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <figures@plos.org>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0241898.r004 Author response to Decision Letter 1 22 Sep 2020 September 22, 2020 To: The Editor, PLoS One Dear Dr. Cray Jr.: after carefully reading yours and the reviewers’ comments, we decided to revise again and resubmit our work entitled “Gene-environment interaction in molar- incisor hypomineralization.” We believe we have addressed all the concerns and are hoping it is suitable for publication in your journal. The two most salient comments from one of the reviewers, the idea that MIH does not have a genetic component and the timing of mineralization of the first permanent first molar being at 1 year of age appeared to us not supported by the current knowledge. We addressed these concerns and added more discussion and believe those are a matter of opinion and we urge that PLoS let the readers react and decide how to interpret our work. Hopefully our work will motivate a change in direction for the field with less of an emphasis on association studies of elusive environmental factors and better designed gene-environmental model studies. This work explores genomic variants associated with cleft lip and palate as potentially associated with MIH. We are looking forward to seeing our revised work being well received. Below are point-by-point answers for the critiques and all changes are marked in yellow. Sincerely, Alexandre R. Vieira, D.D.S., M.S., PhD Professor of Oral Biology, Pediatric Dentistry and Human Genetics Director of Clinical Research and Director of Student Research Department of Oral Biology University of Pittsburgh School of Dental Medicine 412 Salk Pavilion Pittsburgh, PA 15213 Office: 412-383-8972 FAX: 412-624-3080 E-mail: <arv11@pitt.edu> Point-by-point response to reviewers: Reviewer \#1: This author commented two problems of this manuscript. I couldn't get satisfying answers from authors. The causes of disorders in amelogenesis can be divided into those caused by genes (genetic factors) and those not caused by genes (systemic causes, local causes). The former is hereditary disorder (amelogenesis imperfecta), which affects all teeth. The latter is enamel hypomineralization (enamel hypoplasia or enamel hypocalcification) caused by impaired ameloblast function. Enamel hypomineralization is chronological disturbance. MIH is included in enamel hypomineralization. MIH is a local enamel hypomineralization not amelogenesis imperfecta. Authors responded to my comments as that “MIH is possibility an extension of amelogenesis imperfecta”. The authors also have suggested that genetic abnormality is a one of the factors of MIH. I cannot believe this hypothesis. The disorder is not hypomineralization of enamel but amelogenesis imperfecta. “amelogenesis imperfecta” and “enamel hypomineralization” are often misunderstood as a single disorder and expressed in a term. However, they are actually different disorders. The former means hereditary disorders while the latter means nonhereditary congenital disorders considered as chronological disturbances. If enamel hypomineralization is caused by genetic abnormalities, enamel hypomineralization (MIH) is recognized in all teeth. RESPONSE: The assumption that a genetic condition affecting teeth would have to be present in al teeth is simply incorrect. We understand that is a misconception in the field and we added more discussion to the matter to address this concern. It is not a matter of opinion anymore, MIH fits well a condition with a complex mode of inheritance. Genetic conditions that have mendelian forms of inheritance are actually not the most typical examples of disease in human. Most common diseases in humans, such as cardiovascular diseases, diabetes, cancer (and dental caries, periodontitis, MIH) have complex modes of inheritance. This reviewer also commented on the timing of enamel mineralization. MIH is defined hypomineralization of enamel affecting affects one or more permanent first molars with or without permanent incisor involvement. Calcification of the first molar begins at birth, and crown formation (mineralization of enamel) completed at 30-36 months. The same opinion was found in reviewer \#3. RESPONSE: As we discussed originally, most of the children included in the analysis had medications since their first year of life. We described evidence (original reference 19) that the mineralization of first permanent molars can be still occurring pass 36 months and up to 48 months. We added more discussion to highlight this concern. Reviewer \#3 Why do the authors proceed from a hypothesis that environmental factors affecting 3 year-old children or older play a role in the aetiology of MIH? In MIH, the insult to the ameloblasts is likely to occur either prenatally or in the first year of life. The authors responded that “It is quite possible the disruption of amelogenesis happens later as well, not necessarily during the secretion phase only, but during the mineralization phase as well.” This reviewer thinks that the authors don’t consider mineralization phase of the first molar. Environmental factors affecting children that were 3 years of age or older never concern to MIH. This factor affects permanent teeth except for the first molars. RESPONSE: As we discussed originally, most of the children included in the analysis had medications since their first year of life. The impression that first molars are somehow immune from mineralization disturbances after the first year of life, knowing that mineralization of these teeth continues into 36 to 48 months is probably not current. We added more discussion to highlight this concern. Reviewer \#2: After reviewing the revised submission and the response to the reviewer comments, I would like to inform the authors that NOT all my concerns were addressed. only the English language point was addressed!! Please refer to the attached point by point review document (8 points in total) and provide point by point explanations and amendments within the body of the manuscript. Throughout the manuscript 1\. English language needs to be professionally revised. There are straightforward grammar issues. For example, paragraph 4 in introduction “Risk factors for MIH did not fully explained” … and paragraph 5 in introduction “has been documented as responsible for not only affect amelogenesis…”. “The chance of the less common allele of TGFA rs930655 in addition to the IRF6 marker allele to increase the chance …” RESPONSE: We made these corrections in the past version. 2\. A disorder is used commonly to refer to a group of conditions. The authors need to logically justify why they are using the word “disorder” to refer to MIH in many areas of the manuscript? Otherwise, the word disorder needs to be substituted throughout. RESPONSE: We revised the text as requested. Introduction 3. First paragraph, line 7: Did the authors omit the preposition “with” mistakenly in the sentence “usually permanent incisors”…??? This part of the sentence needs to be rephrased as the involvement of incisors is not always necessary to identify a case of MIH! RESPONSE: We revised the sentence as suggested. 4\. Third paragraph, line 17: “When full maturation is not complete at eruption, a post eruption maturation occurs through mineral ions from the saliva …” All newly eruptive enamel undergoes post eruptive mineralization regardless of the mineralization status of enamel. Please explain why you introduced “post- eruptive mineralization” here? Does it serve any purpose related to the objective of the study? if yes, what? RESPONSE: We deleted the sentence as suggested Methods 5. Were the subjects’ medication history collected at 3 years, after 3 years of age, or from birth to 3 years? There are many contradictory and confusing statements. In the beginning of the results section under environmental factors “any type of medications taken after three years of age (Table 4)…”, At the end of the same paragraph “..In both groups we found statistical evidence for an interaction of IRF6 and TGFA genotypes and medication intake at 3 years of age (Table 4)…” Other statements in the discussion section: “The variable medication intake after 3 years of age was created to capture children with history of medication intake from birth and past 3 years….” “another limitation we can point out is that we included only medication taken after 3 years of age…” In the method section: “..any type of medication intake between 3 years of age and the time the sample was collected…” These statements are very confusing and need to be explained? the authors need to be consistent when stating the one (or many) time point(s) where history of medications intake was collected. Please amend and maintain consistency throughout the manuscript. RESPONSE: We revised the text to make it consistent as suggested. The variable was created to captured use of medication from birth and passed 3 years of age. 6\. Referring to the previous point and acknowledging that the mineralization of the first permanent molar takes place from birth up to 3-4 years, why did the authors set up the cut off age to after 3 years to collect medication history? please give detailed explanation. RESPONSE: 3 years of age is likely easier to remember because the child has achieved a vocabulary of 200 words and most pf what she says can be understood. We added a comment in the discussion to address this concern. 7\. What was the reason behind the limited availability of the medication’s intake history to two cohorts out of the four cohorts? RESPONSE: This variable was not included in these original studies. 8\. Referring to the supplemental document, the subjects’ health data were collected from health records in the Rio de Janeiro, Brazil cohort, while structured data questionnaires were used to collect the same data from the Federal University of Paraná cohort. Explain the following issue: a. What do the authors think about the non- homogenous method of collecting medication history between cohorts (questionnaire in one cohort and medical records in another cohort)? this needs to be emphasized in the limitation paragraph in the discussion. How would the previous limitation affect the results of this study? Discussion § Paragraph 1, line 4: the author wanted to emphasize the geographical variations of MIH estimates, however, the generalizability of the declining trend between north and south MIH estimates needs to be addressed with caution!, first of all because the study includes cohorts from south America (Brazil) and the south European/Mediterranean regions. Second, this “claimed” trend is actually reversed when the south American MIH prevalence estimates is compared with the recently published north American MIH estimates (10% - 13%). Thus, the statement needs to be rephrased to reflect the populations studied! RESPONSE: We added the suggested limitation on how the data were collected depending on the cohort. The best estimated frequency is Brazil is 13% as well and aligns of the figures from North America. We added a comment in the discussion to acknowledge this reviewer‘s comment. 10.1371/journal.pone.0241898.r005 Decision Letter 2 Cray Jr. JJ Academic Editor 2020 JJ Cray Jr. This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 21 Oct 2020 PONE-D-20-24540R2 Gene-environment interaction in molar-incisor hypomineralization PLOS ONE Dear Dr. Vieira, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. There is one particular concern that should be addressed about the patient medication history. Also if the authors could do another run through for grammar and readibility. Please submit your revised manuscript by Dec 05 2020 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <plosone@plos.org>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript: A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols> We look forward to receiving your revised manuscript. Kind regards, JJ Cray Jr., Ph.D. Academic Editor PLOS ONE \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation. Reviewer \#2: (No Response) \*\*\*\*\*\*\*\*\*\* 2\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#2: Partly \*\*\*\*\*\*\*\*\*\* 3\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#2: N/A \*\*\*\*\*\*\*\*\*\* 4\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 5\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#2: No \*\*\*\*\*\*\*\*\*\* 6\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#2: 1. The English language still needs to be revised. There are many grammar mistakes. \_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_ \_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_ \_\_\_\_ 2\. The authors response to point 6 where the reviewer is questioning the cut off of 3 years to collect medication history is not satisfying nor convincing. How is the child's vocabulary status be a valid reason to choose this cut off age, when we know the the medication history is collected through direct questioning of the parent/caregiver? \_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_ \_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_ \_\_\_\_\_ 3\. Many sentences need to be re-written in a third person narrative This sentence is one example but this style needs to be adopted through the manuscript. "The fact that we did not always have detailed information about the type of medication the patients were taking due to self-reporting of this information forced us to include any type of medication intake in the analysis. However, these were not vitamins or other supplements. But we are aware that this kind of observational study that aims to recover information from many years prior suffers from potential issues related to recall bias. For these reasons, we ask that our results are taken cautiously.” \*\*\*\*\*\*\*\*\*\* 7\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#2: **Yes: **Azza Tagelsir Ahmed \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <figures@plos.org>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0241898.r006 Author response to Decision Letter 2 21 Oct 2020 Point-by-point response to reviewer 2: Reviewer \#2: 1. The English language still needs to be revised. There are many grammar mistakes. RESPONSE: We carefully revised the text for grammar and typos. We did not find many grammar mistakes but corrected the ones identified. 2\. The authors response to point 6 where the reviewer is questioning the cut off of 3 years to collect medication history is not satisfying nor convincing. How is the child's vocabulary status be a valid reason to choose this cut off age, when we know the the medication history is collected through direct questioning of the parent/caregiver? RESPONSE: We added more detail in the methods to address this concern. Evidence of cuspal enamel formation of first permanent molars can be detected past 3 years of age. The comment of the vocabulary was in regard to recall bias, not enamel formation. 3\. Many sentences need to be re-written in a third person narrative This sentence is one example but this style needs to be adopted through the manuscript. "The fact that we did not always have detailed information about the type of medication the patients were taking due to self-reporting of this information forced us to include any type of medication intake in the analysis. However, these were not vitamins or other supplements. But we are aware that this kind of observational study that aims to recover information from many years prior suffers from potential issues related to recall bias. For these reasons, we ask that our results are taken cautiously.” RESPONSE: We made the changes to adequate to the style the reviewer is requesting. 10.1371/journal.pone.0241898.r007 Decision Letter 3 Cray Jr. JJ Academic Editor 2020 JJ Cray Jr. This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 23 Oct 2020 Gene-environment interaction in molar-incisor hypomineralization PONE-D-20-24540R3 Dear Dr. Vieira, We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements. Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication. An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at <http://www.editorialmanager.com/pone/>, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to- date. If you have any billing related questions, please contact our Author Billing department directly at <authorbilling@plos.org>. If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact <onepress@plos.org>. Kind regards, JJ Cray Jr., Ph.D. Academic Editor PLOS ONE Additional Editor Comments (optional): Reviewers' comments: 10.1371/journal.pone.0241898.r008 Acceptance letter Cray Jr. JJ Academic Editor 2020 JJ Cray Jr. This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 26 Oct 2020 PONE-D-20-24540R3 Gene-environment interaction in molar-incisor hypomineralization Dear Dr. Vieira: I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department. If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact <onepress@plos.org>. If we can help with anything else, please email us at <plosone@plos.org>. Thank you for submitting your work to PLOS ONE and supporting open access. Kind regards, PLOS ONE Editorial Office Staff on behalf of Dr. JJ Cray Jr. Academic Editor PLOS ONE [^1]: The authors have declared that no competing interests exist.

# Introduction Except for the infection with *S. enterica* serovars Gallinarum or Pullorum, chicken infection with the other remaining *S. enterica* serovars is usually not associated with any obvious clinical signs. Despite the absence of gross clinical signs, chickens respond to oral infection by an inflammatory response associated with heterophil and monocyte/macrophage infiltration into the cecal mucosa. The scope of this response is age dependent and is more obvious in chickens up to 2 weeks of age than in adult birds. In agreement with this, proinflammatory cytokines such as IL1β, IL6, IL17 and IL22, together with IFNγ and iNOS are induced in the cecum after infection, either by epithelial cells, resident phagocytes, or infiltrating phagocytes or lymphocytes. A similar cytokine gene expression can be recorded also in the spleen, although the induction rates in the spleen after oral infection are usually lower than those observed in the cecum. The low response of splenic leukocytes to *S. enterica* infection can be overcome by intravenous infection. The chicken response to intravenous infection with *S. enterica* is characterized by splenomegaly associated with macrophage and heterophil infiltration and Th1 and Th17 cytokine signaling, similar to the response in the cecum after oral infection. Another puzzling phenomenon is that the immune response of naive or vaccinated chickens to *S. enterica* infection is the same in terms of a qualitative response. So far the only described differences are mainly in quantitative expression of the immune response – the vaccinated chickens respond to *S. enterica* infection by lower cellular infiltrates and lower proinflammatory cytokine signaling than the naive chickens. This conclusion is valid for both the cecum after oral infection and the spleen after intravenous infection. However, there is at least one difference between the oral and intravenous challenge; namely the production of anti-LPS antibodies. Orally infected chickens produce quite low anti-LPS antibodies whilst intravenous challenge leads to an extremely high antibody production which, unlike the oral challenge, is independent of previous contact with the antigen, i.e. the vaccination status. The reason for a high and rapid antibody production is rather unclear since B-lymphocytes and antibody production are considered as dispensable for the chicken's defense against *S. enterica* infection. In the search for markers for the protection of vaccinated chickens against *S*. Enteritidis infection we used the model of intravenous infection. We hypothesized that if the spleen sizes differ between the vaccinated and infected, naive and infected and non-infected chickens, , there must be significant differences in gene expressions among these 3 groups of chickens. We therefore purified mRNA from the spleens of intravenously infected chickens and subjected it to transcriptome characterization by 454 pyrosequencing. This approach resulted in an unbiased identification of chicken genes which are induced in response to systemic *S. enterica* infection. In addition, we have shown that some of the newly identified genes were induced also in the cecum of orally infected chickens. However, chickens which had been vaccinated prior to the challenge did not induce these genes in the cecum after oral challenge which in turn can be used as a marker of vaccine efficacy and specific immunity to *S*. Enteritidis. # Results ## Expression in the spleen Pyrosequencing resulted in 140,827 reads when cDNA from the spleen of the non- infected chicken was sequenced, 100,971 reads from the spleen of the chicken after *S*. Enteritidis infection and 53,762 reads from the spleen of the chicken which had been vaccinated prior to the infection. Average read size was 426 bp. Considering 1,000 bp as an average gene size and 20–23,000 genes forming chicken genome, total chicken transcriptome represents approx. 20–23 Mb of mRNA sequence. We therefore achieved approx. 1× coverage for the transcriptome of the spleen of vaccinated chicken, 2× coverage for the infected spleen transcriptome and 3× coverage for the non-infected spleen transcriptome. This means that we identified only the highly expressed genes and many low level expressed genes, despite their differential expression in the spleen of infected or non-infected chickens, might remain undetected. Combining all 3 samples in the *de novo* assembly resulted in the identification of 8,844 isotigs which were subjected to Blast2GO analysis. After the analysis, the number of expressed genes decreased to 6,633 transcripts because some of the isotigs were identical to different parts of the same genes. After applying all the quality selective criteria, 23,663 reads from the spleen of the non-infected chicken, 21,442 reads from the spleen of the infected chicken and 18,536 reads from the spleen of the vaccinated and infected chicken were finally included in the quantification of expression (the majority of the excluded transcripts comprised of rRNA, polyA sequences or repeated sequences). For 99 and 78 genes we predicted that these might be down- or upregulated in the spleen after i.v. *S*. Enteritidis infection, respectively. Similar results were observed also when we performed BLASTX analysis of all individual reads against chicken genome only (data not shown). When gene ontology classification on cellular processes was retrieved for these transcripts, the downregulated transcripts were classified as involved in transcription, translation, signal transduction, phosphorylation, cell adhesion and differentiation. The transcripts upregulated after the infection were classified as associated with proteolysis, cellular transport, acute-phase response, response to LPS, innate immune and inflammatory response, lipid metabolic processes, angiogenesis and apoptosis. As can be seen in, the suppression of gene expression was always lower in the vaccinated chicken than in the non-vaccinated chicken after the challenge, and processes like gene expression or translation were not affected in the vaccinated chicken at all (the same or even higher number of transcripts in the vaccinated and infected chicken as in the non-infected control). Similarly, except for cellular transport, the cellular processes upregulated after *S*. Enteritidis challenge were always expressed at a higher level in the non-vaccinated chicken when compared with the chicken which was vaccinated prior to the infection. In the next step we designed real time PCRs to verify the expression of all 78 genes predicted as upregulated after i.v. infection with *S*. Enteritidis (we did not elaborate on genes suppressed after *S*. Enteritidis in this study any further). Significant upregulation, either in the non-vaccinated or vaccinated chickens, was confirmed in 40 of them. When we compared the expression of individual genes in the spleens of naive or vaccinated chickens after i.v. challenge, 18 genes were expressed differently although the difference only rarely exceeded a factor of 2. ## Expression in sorted splenic leukocytes Some of the genes upregulated after the infection in the above mentioned experiments were previously reported to be expressed by different leukocyte subpopulations. Furthermore, we have shown that splenomegaly is associated with macrophage and heterophil infiltration after intravenous infection which may influence the total transcription in the spleen. In the next experiment we therefore used cDNA from our previous study to test which leukocyte subpopulations, if any, were responsible for the expression of genes associated with the chicken's response to *S*. Enteritidis. Five different groups of genes in relation to their basal expression in particular leukocyte subpopulations prior to infection and to their expression profile after the infection were identified as follows: i) genes similarly expressed in all leukocyte subpopulations, ii) genes preferentially expressed in B-lymphocytes, iii) genes constitutively expressed in macrophages, iv) genes inducible in macrophages and v) genes coding for immunoglobulins ( and for total data set see). Genes that were equally expressed in macrophages, B-lymphocytes, CD4, CD8 and γδ T-lymphocytes comprised of LMNB1, PRDX1, SLC35B1, TRAM1, and a transcript of an unknown function (NAv3). The expression of SLC35B1 and TRAM1 slightly decreased after *S*. Enteritidis infection in all subpopulations, independent of the previous vaccination status but without reaching statistical significance. Expression of the remaining genes hardly changed in sorted leukocyte subpopulations after the infection. Prior to the infection, B-lymphocytes were the major producers of PRDX4, SEC11C, ERLEC1 and TXNDC5 and all immunoglobulins. The expression of PRDX4, SEC11C, ERlEC1 and TXNDC5 decreased in expression in B-lymphocytes after *S*. Enteritidis infection suggesting a suppression of common B-lymphocyte function and specialization of B-lymphocyte towards immunoglobulin expression after the infection. Immunoglobulins are described separately due to their unexpected expression profiles (see below). Genes transcribed constitutively in macrophages were the most numerous and included AH221, ANG, ANXA2, ASAH, ASS, C1QA, C1QB, C1QG, CCDC86, CTSB, CTSC, CTSD, CTSS, EXFABP, FN1, FTH1, GSTA, HMOX1, IRG1, MD1, MGST1, OTFB and VIM. Out of these genes, the basal expression of EXFABP was as high as the expression of macrophage inducible genes after their induction. Upregulation of these genes in the spleen without their induction in their main producers indicates that upregulation in the spleen was caused mainly by infiltration of macrophages with their characteristic expression profile. Macrophage-inducible genes included genes coding for avidin, serum amyloid A and trappin-6 (AVD, SAA, TRAP6). Average upregulations of these genes in macrophages were 46×, 23× and 61×, respectively. We consider these genes as macrophage- inducible despite the fact that none of the comparisons have come out as significant. However, the statistical non-significance was caused mainly by a great variation in the induction rate in macrophages after the infection resulting in a high standard deviation and non-significance when only 3 samples were compared. The last group was formed by the transcripts coding for the constant part of immunoglobulins. Prior to infection, these genes were transcribed the most in B-lymphocytes. However, the transcription of all 4 immunoglobulin genes did not increase after the infection in B-lymphocytes and, instead, there was a significant increase in the abundance of mRNA coding for the constant part of immunoglobulins in both T-lymphocytes and macrophages in response to the infection. Consequently, the transcription of the constant parts of immunoglobulins in T-lymphocytes and macrophages after the infection reached nearly the same level as in B-lymphocytes. ## Expression of newly identified genes in the cecum Since intravenous infection can be considered as a rather artificial route of infection, in the next step we verified our results using cDNAs from the cecum of chickens orally infected on the day of hatching and sacrificed 3 days later. Out of the 40 genes identified as upregulated in the spleen, 15 genes were significantly induced also in the cecum. These included all 4 immunoglobulin genes, and TRAP6, AVD, SAA, AH221, EXFABP, IRG1, C1QA, ANG, LMNB1, OTFB and PRDX1. One of our aims was to find markers for the protection of vaccinated chickens. In the last verification experiment we therefore used cecal cDNA from 46-day-old chickens, vaccinated and infected, together with appropriate controls. The highest upregulations in naive and orally infected newly hatched chickens were observed for TRAP6, EXFABP, SAA, IRG1, AVD, AH221, C1QG, ANG, C1QA and C1QB. The response of 46-day-old chickens, except for the expression of immunoglobulins, was similar to the response of 4-day-old chickens. In 42-day-old chickens we could also test the response of chickens which were protected against challenge by previous oral vaccination. The vaccinated and orally challenged chickens responded to the infection only by a significant increase in the transcription of C1QG and MD1, though upregulations of these 2 genes were lower than 2 fold. Unlike the non-vaccinated chickens, genes such as TRAP6, EXFABP, SAA, IRG1, AVD, AH221 or ANG were not significantly induced in the vaccinated chickens and can be therefore used as new markers of vaccination status, in addition to culture detection of the challenge strain or cytokine (e.g. IL-8, IL-17 or IFNγ) targeted real-time PCR. Genes encoding immunoglobulins were not induced in the cecum of 42-day-old chickens after the infection with *S*. Enteritidis. However, when the expression levels in 4- and 46-day-old chickens were compared, the basal expression of immunoglobulins in the cecum in the older chickens, though no longer inducible, reached a considerably higher expression level than in the 4-day-old chickens. This also means that a gradual increase of immunoglobulin gene transcription must have occurred sometime between day 4 and day 46 and this process was accelerated by *S*. Enteritidis infection in the young chickens. ## Avidin in the chicken response to S. Enteritidis infection Since avidin was the most inducible gene in the spleen after *S*. Enteritidis infection, in the last experiments we tested its potential role in the defense against *S*. Enteritidis infection. First we tested the direct antibacterial effect of avidin on *S*. Enteritidis but avidin did not affect *S*. Enteritidis growth in LB broth up to 2.5 mg/ml concentration (data not shown). Next we tested whether avidin may influence phagocytosis and/or invasion of *S*. Enteritidis into the HD11 macrophage-like cell line. When the HD11 cells were pretreated with avidin prior to *S*. Enteritidis infection, a higher adhesion but lower invasion of *S*. Enteritidis was recorded when compared with the adhesion and invasion into avidin non-treated cells. On the other hand, pretreatment of *S*. Enteritidis prior to the adhesion and invasion assay did not result in any difference from the assay performed in the absence of avidin. Finally we tested whether avidin may protect chickens against *S*. Enteritidis challenge *in vivo*. The chickens were intravenously administered avidin to reach 3 µg of avidin per gram of body weight and, 8 hours later, half of the chickens were intravenously challenged with *S*. Enteritidis. Four days later when the chickens were sacrificed, we did not record any differences in total bacterial load in the liver or spleen (not shown). Similarly, we did not record any differences in the composition of splenic leukocytes (B-lymphocytes, CD4, CD8 and γδ T-lymphocytes, and macrophages) after avidin treatment determined by flow cytometry (data not shown). The only significant differences as a result of avidin administration were the differences in IgM and PRDX1 expressions in the spleen of avidin treated and *S*. Enteritidis infected chickens compared with the chickens infected by *S*. Enteritidis without avidin treatment. The differences were numerically quite low; a 2 fold increase in the transcription for IgM and around 1.5 fold increase in PRDX1 in avidin pretreated chickens. We therefore concluded that avidin likely has functions different from a direct effect on the immune response and chicken resistance to *S*. Enteritidis. # Discussion First of all we have to stress that designations and functions of the majority of the genes identified as responding to *S*. Enteritidis infection in this study are based mostly on their sequence similarities to different GenBank entries, rather than their proven function in chickens. The majority of these genes have never been associated with *S*. Enteritidis and chickens although some of these transcripts were described as responsive to *Salmonella* in other experimental animals or were characterized as LPS inducible or as belonging among acute phase proteins. This is true mainly for genes coding for serum amyloid A, avidin, immune responsive gene 1 or extracellular fatty acid binding protein,. The main motif of the immune response to the i.v. infection with *S*. Enteritidis was LPS inactivation, which was also supported by MD1 induction leading to a decrease in TLR4-LPS responsiveness in HEK293 cells or increased proliferation of B-lymphocytes. In agreement with this observation, three classes of heavy immunoglobulin chains (IgM, IgG and IgA) together with a λ light chain were induced after *S*. Enteritidis infection. The rapid increase in immunoglobulin mRNA in both the vaccinated and non-vaccinated chickens 4 days post infection indicates that the antibody response might be a T-cell- independent response to the LPS consistent with our previous reports on the rapid increase of anti-LPS antibodies in the chicken serum after intravenous infection. Furthermore, the spleen also responded by an increased transcription of all 3 subunits of the C1q complement complex which binds to the conserved domains of IgG and IgM in a complex with antigen. Several types of cathepsin proteases were induced, as well as TRAP6, a protease inhibitor, likely protecting host tissues against activity of its own proteases released during inflammation. Six genes could be characterized as having a detoxification function (glutathione S-transferase α class, microsomal glutathione S-transferase 1, peroxiredoxin 1, peroxiredoxin 4, thioredoxin domain-containing protein 5, endoplasmic reticulum lectin 1), most of them dismutating reactive oxygen species. Oxidative burst by phagocytes is decreased also by SAA binding of LPS. Finally, the restoration of damaged tissues during the initial immune response by angiogenesis, for example, was induced as early as 4 days post infection as documented by an increased expression of angiogenin, annexin a2, fibronectin and ferritin heavy chain. Most of the genes identified in this study were associated with macrophages or B-lymphocytes whilst T-lymphocytes were not involved in the response to i.v. infection with *S*. Enteritidis to such an extent that would affect transcription on a level of the whole spleen. Interestingly, not all genes inducible on the level of total spleen were inducible also in the sorted leukocytes. There are two, mutually non-exclusive explanations for this observation. First, there might be other cells in the spleen, which express the genes not inducible in the sorted leukocytes tested in this study. Our unpublished data show that heterophils are responsible for a high expression of serum amyloid A, trappin-6 or extracellular fatty acid binding protein. Secondly, the observed upregulation in the spleen but not in the sorted leukocytes can be explained by an extensive infiltration of particular cell types such as macrophages with their own constitutive gene expression profile. Their numerical increase, despite the absence of induction, then leads to changes in the net gene transcription of the whole spleen. In the case of avidin, its expression was induced in sorted macrophages approx. 46×, and macrophages increased from 0.5% to 10%, i.e. 20× after the infection as determined by flow cytometry (see ref. and data not shown). The combination of the induction and infiltration, i.e. 46×20 = 920, is in agreement with the total increase in avidin expression in the spleen determined by real-time PCR to be 1008×. Quite unexpected profiles were observed for the expression of immunoglobulins. Whereas these genes were constitutively expressed in B-lymphocytes, they were inducible in all T-lymphocyte subpopulations and even macrophages. We excluded that this could be caused by contamination of sorted T-lymphocytes and macrophages by B-lymphocytes since in such a case, the contamination with B-lymphocytes should influence the expression of immunoglobulins also in T-lymphocytes and macrophages from the non-infected chickens. We also excluded the possibility that surface markers used for macrophage and T-lymphocyte sorting might be present on the surface of clonally expanding B-lymphocytes as there would have to be B-lymphocytes positive for CD4, CD8, TcR1 and macrophage surface markers. Finally, we also consider as unlikely the hypothesis that the real-time PCR detected target sequences are similar to but different from the immunoglobulin transcripts since the same results were obtained for 4 independent targets and the transcription of the light chain was equivalent to the sum of the expression of transcripts for heavy chains, as one would expect. Finally, comparing the expression in different tissue samples, IgY and IgA dominated over IgM in the cecum whilst IgY and IgM dominated over IgA in the spleen, which confirmed the correctness of the real-time PCR results. This means the increase in the expression of immunoglobulins in T-lymphocytes and macrophages is likely correct whereas its biological function remains unknown. However, several older papers described that T-lymphocytes and macrophages were capable of transcription of immunoglobulin genes though the function might be different from immunoglobulin secretion. Since intravenous infection is quite an artificial mode of infection, we also tested the expression of the newly identified genes in the cecum of orally infected chickens. AVD, SAA, TRAP6, AH221, EXFABP and IRG1 were also highly upregulated in the cecum after oral infection, both in 4- and 46-day-old chickens. Although some of these proteins, e.g. avidin and serum amyloid A, have been known for a long time, their biological function is poorly understood. AH221, also known as CCLi10 or predicted C-C motif chemokine 3, is produced by macrophages as a chemotactic protein during inflammation. The function of chicken IRG1 (immune responsive gene 1) is even less clear (predicted chicken IRG1 was 74% identical and 83% similar to murine immune responsive gene 1 at amino acid level). IRG1 is induced by LPS or *Mycobacterium tuberculosis* in murine bone marrow derived macrophages independent of TLR2 or TLR4 sensing of pathogen-associated molecular patterns but the biological relevance of this is unknown. On the other hand, although trappin-6 has never been studied in chickens and its identification in this study was based only on sequence similarities (42% identical and 58% similar to bovine trappin-6 at amino acid level), its likely function is the protection of the host's extracellular proteins from degradation by its own proteases such as neutrophil elastase or proteinase 3. We have shown that trappin was expressed by macrophages and our unpublished data show that it is also highly transcribed in heterophils. This can serve as additional, though indirect, evidence that the trappin 6-like transcript codes for a functional protease inhibitor. Chicken macrophages and heterophils therefore likely release extracellular proteases and in parallel also protease inhibitors protecting their own tissues from proteolytic degradation. Chicken EXFABP was first characterized as a protein capable of binding unsaturated fatty acids with an unknown role in chondrocyte development. EXFABP also stimulates cell proliferation and its suppression results in apoptosis. Interestingly, recent reports showed that quail lipocalin Q83 and chicken EXFABP, which share 88% similarity, have dual binding capacities and besides the fatty acid binding capability, they can also bind bacterial siderophores. Consequently, chicken EXFABP inhibited the growth of *E. coli* in iron-limited media *in vitro* and this likely affects the multiplication of gram negative bacteria after LPS sensing also *in vivo*. Interestingly, since *Salmonella* produces glycosylated enterochelin resistant to the activity of Lcn2 in mice and likely also EXFABP in chickens, it may obtain a growth advantage over the rest of the cecal microbiota. The contribution of serum amyloid A and avidin, although known for a long time, to the defense against *S*. Enteritidis is unclear. Serum amyloid has been shown to bind LPS and cholesterol. Avidin, besides biotin binding, was shown to block chondrocyte proliferation without any effect on their differentiation. This means that these two proteins together with EXFABP may decrease LPS concentration and the associated inflammatory response, suppress cell proliferation during the inflammatory response and provide host cells with fatty acids, cholesterol and biotin. This might be consistent with our results showing a decrease in the phagocytosis of avidin treated HD-11 cells *in vitro* and the absence of any direct antimicrobial activities of chickens administered with avidin prior to *S*. Enteritidis infection. Taken together, 6 transcripts (AVD, SAA, TRAP6, AH221, EXFABP and IRG1) therefore as central to the control of the chicken response to *S*. Enteritidis infection, both in the spleen during the systemic presence of *S*. Enteritidis and in the cecum during a cecum-localized infection. Moreover, as the transcription of these genes increased in the cecum of naive chickens after *S*. Enteritidis infection but essentially did not change in the vaccinated chickens, these genes may be used as new markers for chicken response to *Salmonella* infection. # Materials and Methods ## Ethics Statement The handling of animals in the study was performed in accordance with current Czech legislation (Animal protection and welfare Act No. 246/1992 Coll. of the Government of the Czech Republic). The specific experiments were approved by the Ethics Committee of the Veterinary Research Institute (permit number 48/2010) followed by the Committee for Animal Welfare of the Ministry of Agriculture of the Czech Republic (permit number MZe 1226). ## Experimental animals, sample collection and pyrosequencing Three samples of spleen from ISA Brown chickens were used for RNA isolation. The first spleen originated from a 46-day-old, non-infected chicken, the second one from a non-vaccinated chicken infected intravenously with *S*. Enteritidis on day 42 of life, and the third one from a chicken which was orally vaccinated on day 1 and day 21 of life with *S*. Enteritidis 147 ΔSPI1 mutant, and intravenously challenged on day 42 of life with wild type *S*. Enteritidis. Both infected chickens were sacrificed 4 days after the challenge. Approximately 30 mg of spleen was collected from each chicken and stored in RNALater (Qiagen) at −70°C for RNA isolation. Total RNA was isolated with RNeasy Mini Kit (Qiagen) followed by mRNA Isolation Kit (Roche) to enrich the total RNA for mRNA species. cDNA libraries from the three spleen samples were prepared with the GS Rapid Library Preparation Kit (Roche) and approx. 2 molecules of cDNA per bead were used in emulsion PCR. All steps of the cDNA library preparation for sequencing were performed with the GS Junior Titanium series kits according to the manufacturer's instructions (Roche). The pyrosequencing was performed with the GS Junior 454 sequencer (Roche) and each of the 3 samples was sequenced in a separate sequencing run. ## Data analysis In the first step we used the De Novo Assembler software provided with the GS Junior. This software was used to assemble the chicken spleen transcriptome using the sff files generated as an output of sequencing of each of the 3 samples. Out of this analysis we took the 454Isotig.fna file which contained sequences of all transcripts assembled and identified after merging the data from all 3 splenic samples. The data present in this file were used in two independent analyses. Firstly, the 454Isotig.fna file was uploaded into Blast2GO software to associate each transcript with a gene designation and gene ontology classification according to the GeneBank. In a second independent analysis we used the 454Isotig.fna file as a reference file using GS Reference Mapper software provided with GS Junior and analyzed the sff sequencing files of each of the 3 samples. Data from the ReadStatus.txt output file after this analysis allowed us to determine the number of reads in each of the 3 sequenced samples matching different splenic transcripts present in the 454Isotig.fna reference file. Only reads longer than 100 nt and matching the reference transcripts by more than 60% of their sequence were included for the quantification of gene expression. These values were arbitrarily selected but effectively excluded short repetitive sequences or sequences with a polyA motif. The last negative selection was applied using chicken rRNA gene sequences as an exclusion filter. Transcripts predicted as being downregulated after *S*. Enteritidis infection included those for which we identified at least 10 independent reads in the transcriptome of non-infected spleen and when the fold downregulation was 3-fold or higher. A slightly less stringent selection criteria were applied for the transcripts predicted as being upregulated after *S*. Enteritidis infection since these were subjected to verification by real-time PCR. The tentatively upregulated transcripts were chosen as those which were present either in the infected or vaccinated and infected spleen at 10 or more reads and the calculated induction was at least twofold. In addition, for further analysis we also included the transcripts which were detected once or not at all in the transcriptome of the non-infected spleen but were recorded in the transcriptome of either the infected or vaccinated and infected spleen at least 8 times. ## Real-time PCR verification of the pyrosequencing data Real-time PCR was used for the verification of the pyrosequencing data. Primers for the quantification of expression real-time PCR were designed using Primer3 software (for primer sequences see). First we used the same cDNAs as in the pyrosequencing reactions followed by an additional 2 spleen samples for each group of chickens available from our previous study. After such screening we reduced the number of tested genes to those in which the real-time PCR confirmed the results from pyrosequencing which meant at least a twofold induction and statistical significance or threefold average upregulation without reaching statistical significance – the latter case, however, happened only once. Using the reduced set of real-time PCRs, we finally determined the gene expression in i) sorted splenic leukocyte subpopulations using already available cDNA, ii) the cecum of 4-day-old orally infected chickens using cDNA from our recent paper, and iii) the cecum of 46-day-old orally infected chickens using cDNA from another recent paper of ours. Real-time PCR was performed in 3 µl volumes in 384-well microplates using QuantiTect SYBR Green PCR Master Mix (Qiagen) and a Nanodrop II Stage pipetting station (Innovadyne) for PCR mix dispensing. The amplification of PCR products and signal detection were performed using a LightCycler II (Roche) with an initial denaturation at 95°C for 15 min followed by 40 cycles of 95°C for 20 s, 60°C for 30 s and 72°C for 30 s. Each sample was subjected to real-time PCR in duplicate and the mean Ct value of the duplicates was used for subsequent analysis. The Ct values of the genes of interest were normalized (ΔCt) to an average Ct value of three house-keeping genes (GAPDH, TBP and UB) and the relative expression of each gene of interest was calculated as 2^−ΔCt. These expression levels were used for data analysis and are presented in the tables and figures as average ± SD. As an alternative, the fold upregulation calculated as a ratio of averages of infected to non-infected samples are shown. However, also in this case, the significance of these upregulations was calculated by comparing the expression levels, i.e. the 2^−ΔCt values determined for the individual samples. ## Biological testing of the transcriptome data Since avidin was identified as the gene with the highest induction after *S*. Enteritidis infection, we tested whether it has a direct role in the chicken defense against *S*. Enteritidis. First we tested whether avidin might be of direct antimicrobial effect to *S*. Enteritidis. Twofold serial dilutions of avidin in LB broth were prepared in 96-well microplates starting with a 2.5 mg/ml avidin concentration. A fresh 18-hour-old culture of *S*. Enteritidis was inoculated into the wells of the microplates and the growth of *S*. Enteritidis was visually inspected after a 24 hour incubation at 37°C. The influence of avidin and its ligand biotin on the invasiveness of *S*. Enteritidis into the chicken macrophage-like cell line HD11 was tested using standard gentamicin protection. The HD11 cell line was pre-treated for 8 and 24 hours with avidin (20 µg/ml) or avidin and biotin (biotin was used at 50 ng/ml), prior to the addition of *S*. Enteritidis at multiplicity of infection of 10. In parallel, *S*. Enteritidis was exposed to avidin or avidin and biotin for 6 hours prior to to the addition to HD11 cell line. The last experimental group consisted of *S*. Enteritidis added to the HD11 cells at the same time as avidin or mixture of both. Controls included the adhesion and invasion assay performed in the absence of avidin or biotin. Adhesion was determined 1 hour after addition of *S*. Enteritidis to the HD11 cell line and the invasion after an additional hour of incubation in the presence of 100 μg/ml gentamicin. In all the cases, the numbers of intracellular (or adherent) bacteria were determined after lysis of the cell line with 1% Triton X-100 and plating tenfold serial dilutions on LB agar plates. In the last experiment we tested the protective effect of avidin *in vivo*. Four groups of 8-week-old chickens were included in this experiment, each consisting of 6 chickens. The first group was used as a non-treated control. Chickens in group 2 were given 0.1 ml of 20 mg/ml avidin intravenously into the wing vein to reach an avidin concentration of 3 µg per gram of body weight. These chickens were sacrificed 8 hours later. Chickens in group 3 were intravenously administered avidin into the left wing vein like the chickens in group 2 and 8 hours later, they were intravenously infected into right wing vein with 10⁷ CFU of *S*. Enteritidis in 0.1 ml PBS. Chickens in group 4 were given only *S*. Enteritidis like the chickens in group 3. All infected chickens in groups 3 and 4 were sacrificed 4 days later. During the post mortem analysis, *S*. Enteritidis counts in the spleen were determined in all infected chickens. Spleen samples were placed into RNALater and stored at −70°C. RNA purification, reverse transcription into cDNA and real-time PCR were performed as described previously. Finally, in 3 chickens of all 4 groups, the cellular composition of splenic leukocytes was characterized by flow cytometry exactly as described earlier. ## Reproducibility and statistics 454 pyrosequencing was considered as an initial screening and was performed on individual samples from the spleen of a non-infected chicken, an infected chicken and a vaccinated and infected chicken. An initial verification of the 454 screening by real-time PCR was done with 3 chicken samples in each group. However, when performing the experiments with avidin administration, an additional 6 spleen samples after i.v. challenge and 9 from the non-infected chickens were analyzed with the same results. Data for expression in the cecum were obtained from 9 infected and 10 non- infected 4-day-old chickens and from 6 non-infected, 6 infected and 6 vaccinated and infected 46-day-old chickens. The adhesion and invasion assays were performed in duplicate and the experiment was performed on two independent occasions with similar results. The last experiment with avidin administration was repeated on two independent occasions with 3 chickens in each group, except for the non-infected chickens for which we included two different control groups in the repeated experiment, the first one not being treated at all and the second one treated intravenously with sterile water used for dissolving the avidin. This means that the data were calculated for 6 chickens in each group treated with avidin or *S*. Enteritidis and 9 controls without any avidin or *S*. Enteritidis inoculation. Statistical significance was calculated using GraphPad Prism software (GraphPad Software, Inc.), either by ANOVA followed by a post-hoc Tuckey's test or a t-test as indicated in the text. Differences with p\<0.05 were considered as significant. # Supporting Information Authors would like to thank Peter Eggenhuizen for his English language corrections. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: IR MM. Performed the experiments: MM JR LV JV HS HH FS. Analyzed the data: IR MM JV. Contributed reagents/materials/analysis tools: MM JR LV JV HS HH FS IR. Wrote the paper: IR MM.

# Introduction Tuberculous meningitis (TBM) is the most feared presentation of extra-pulmonary tuberculosis (TB) because about a third of all patients die from disease. We and others have previously shown that adjunctive dexamethasone, administered with anti-tuberculosis drugs, improved the outcome of adults with TBM, but the mechanism underlying this effect is not understood. Dexamethasone did not have any significant effect on cerebrospinal (CSF) white cell infiltration or cytokine expression in 93 patients recruited to the clinical trial. A subset of patients had serial brain magnetic resonance imaging, which suggested dexamethasone might decrease the incidence of hydrocephalus and infarction. In children steroids have been shown to reduce CSF protein and cause a more rapid normalization in CSF glucose over placebo. However, no study has demonstrated the mechanism by which dexamethasone reduced case-fatality from TBM. We hypothesized that dexamethasone improved outcome from TBM by altering the intra-cerebral expression of MMP and the tissue inhibitors of MMPs (TIMPs). MMPs are important mediators of extracellular matrix degradation and are implicated not only in inflammatory central nervous system (CNS) diseases such as multiple sclerosis, HIV dementia and Alzheimer's disease but also in TB. The blood brain barrier (BBB) is rich in type IV collagen, a substrate of MMP-9 (gelatinase B), and its breakdown is a key initial step in the pathophysiology of CNS leukocyte influx. We and others have shown that CSF concentrations of MMP-9 are elevated in all forms of meningitis and CSF MMP-9 concentrations (corrected for CSF white cell count) were significantly associated with fatal TBM and the extent of cerebral tissue damage. We found that IFN-γ synergistically increases MMP-9 secretion from astrocytes, the most abundant CNS cell and a key component of the BBB. Mouse models of pyogenic bacterial meningitis demonstrate MMP-9∶TIMP-1 ratios are important predictors of tissue destruction, although MMP-9 may also have a significant role in host defense. Our cellular studies on TB-infected macrophages implicate MMP-1 (collagenase-1), -3 (stromelysin-1), -7 (matrilysin) and -10 (stromelysin-2) as critical in the host response to TB. In this study, we measured serial CSF concentrations of a number of MMPs/TIMPs identified as key in TB in a sub-set of adults with TBM recruited to a randomized, placebo-controlled trial of adjunctive dexamethasone. Our aim was to investigate the relationship between dexamethasone treatment, CSF MMP/TIMP expression, and clinical outcome. # Results ## Comparison of baseline variables We have compared the baseline clinical features of those included in the MMP study (n = 37) with the rest of the proven HIV uninfected patients recruited to the controlled trial (n = 400). Comparison of the patients who received placebo or dexamethasone in the MMP study revealed only CSF opening pressure was significantly different between the groups. The patients were well-balanced with regard to the most important prognostic variables (MRC grade and coma score). Comparison of the patients in the MMP study with the rest of the HIV uninfected patients recruited to the trial revealed some important similarities and differences. Clinical assessments of disease severity (by MRC grade and Glasgow coma score) were not significantly different between the two study groups. However, the patients not included in the MMP study were significantly older and lighter, had lower numbers of white cells in the CSF, and had worse outcomes. 185 CSF specimens were taken from the 37 patients included in the MMP study. 141 (76%) specimens were available for analysis of longitudinal changes in MMP/TIMP concentrations: 66/94 (70%) were from placebo and 75/91 (82%) from dexamethasone group patients (a mean of 3.8 (+/−1.2) samples per patient). ## Baseline CSF MMP and TIMP concentrations Baseline pre-treatment concentrations of MMPs and TIMPs are summarized in and are similar between the two groups with the exception of TIMP-2, which was significantly higher in the patients give placebo. MMP-2, -3 and -9 concentrations were detectable in all 30 pre-treatment CSF samples analyzed and MMP-8 concentrations were detectable in 29 (97%). CSF concentrations of MMP-1 and -7 were only detectable in 9 (30%) patients and MMP-10 in 10 (33%) patients. All 30 patients had detectable TIMP-1 and -2 and 29 (97%) patients had detectable CSF TIMP-4. ## Dexamethasone specifically reduces CSF MMP-9 concentration at day 5 of TBM treatment 22 patients had both a pre-treatment and at least one paired ‘early’ follow up sample available for analysis, taken a median 5 days (range 3–8) into anti- tuberculosis treatment, equally spilt between placebo and dexamethasone groups. Only CSF MMP-9 concentrations were significantly lower in the early follow-up samples of patients given dexamethasone compared to placebo (p = 0.01), confirmed by repeated measures ANOVA with CSF white cell count as a covariate (p = 0.002). 3/11 (27%) patients in the placebo group had a fall in their CSF MMP-9 concentration compared with 9/11 (82%) of the dexamethasone group between pre-treatment and the early follow-up sample (p = 0.03). However, we could not find a significant relationship between those patients whose CSF MMP-9 concentration fell and outcome; 1/12 (8%) patients with a fall in CSF MMP-9 died and 1/10 (10%) died in the non-faller group, the data were similar for death or disability (data not shown). There was a similar but less marked effect of dexamethasone on MMP-8 concentrations. A trend for lower CSF MMP-8 concentrations in the dexamethasone arm (p = 0.08) was not significant in the repeated measures ANOVA. TIMPs are the natural inhibitors of MMPs but we were unable to find any effect of dexamethasone on changes in CSF TIMP concentration in the same early follow-up (day 5) samples. MMP-9 concentrations at later time points were not statistically different between the two groups. At day 30 CSF MMP-9 concentration was 252.6 ng/ml (IQR 154.6–298.6) in the placebo group and 249.9 ng/ml (IQR 67.5–406.6) in the dexamethasone group. At day 60 CSF MMP-9 concentration was 281.6 ng/ml (IQR 94.1–345.0) in the placebo group and 181.9 ng/ml (IQR 101.8–300.7) in the dexamethasone group. Day 270 concentrations for all CSF MMP/TIMP and their statistical change from baseline are summarized in. Of note MMP-3, -8, -9 and TIMP-1 fell and TIMP-4 rose significantly over the course of 9 months anti- tuberculosis treatment. No other significant relationships were found at any time points for any other MMPs and treatment group. MMP/TIMP concentrations at all time points for both groups have been summarized in supplementary. ## The relationship between CSF WCC and MMP-9 concentrations Neutrophils play a key but possibly under-recognized role in host defense to mycobacteria and release preformed MMP-9 from granzymes to cross the BBB. CSF MMP-9 concentration and neutrophil count were strongly correlated on analysis of all 141 samples (R² = 0.19, p\<0.001). There was a significant correlation between MMP-9 concentration and CSF WCC (R² = 0.31, P = 0.004) in all 29 available early follow-up (day 5) samples and a stronger correlation with CSF neutrophil count (R² = 0.52, p\<0.001) which was dexamethasone independent. CSF neutrophil count was not independently related to outcome. ## The relationship between MMP-9 and other CSF inflammatory parameters BBB disruption is potentially related to MMP activity. There were significant correlations between CSF total protein (the only CSF analyte to fall significantly due to dexamethasone) and MMP-9 (R² = 0.18, p\<0.001) concentrations but we could find no relationship between the albumin index (a marker of BBB permeability) and CSF MMP/TIMP concentration. At the day 5 time point MMP-8 and -9 were closely correlated (R² = 0.67, p\<0.001, n = 29) but there was only a weak relationship between MMP-9 and its specific inhibitor TIMP-1 (R² = 0.13, p\<0.05, n = 29). There were no other relationships between MMP-9 and any of the other MMP/TIMP concentrations at this time point. We investigated the relationship between CSF concentrations of TNF-α, IFN-γ, IL-6, IL-8, IL-10 (data not presented) and MMP-9 in all 141 specimens and found only IL-10 and MMP-9 to be correlated (R² = 0.22, p\<0.001). Concentrations of IL-1β and IL-12p70 levels were below the level of detection in all but 6 and 8 specimens respectively so were not analyzed. Specific examination of pre-treatment and early follow-up samples did not reveal any other significant relationships. ## Analysis of all baseline clinical and laboratory variables and relationship to baseline clinical severity and treatment outcome measurements We investigated the relationship between increasing severity of coma (i.e. GCS of \<15 or not on admission to hospital) and clinical and laboratory variables including CSF MMP/TIMP concentrations by logistic regression. In univariate analysis only gender and CSF chloride concentration went forward to entry into the final model, but neither remained significant. We then performed univariate analysis to look for relationships between combined poor outcome (death and disability) and clinical and laboratory findings. Gender, baseline GCS, cranial nerve palsy and hemiparesis on presentation went forward into multivariate analysis where only hemiparesis remained significant in the final model (OR = 0.05, p = 0.04 \[95% CI 0.003, 0.92\]). We found no other significant relationships between pre-treatment MMP/TIMP concentrations (n = 30) and mortality, cranial nerve palsy, hemiparesis and paraparesis on admission by multivariate analysis, although there was a significant relationship between lower pre-treatment MMP-2 concentrations in CSF and hemiparesis (OR = 0, p = 0.02 \[95% CI 0, 0.24\]) on univariate analysis. There was no relationship between duration of symptoms and MMP/TIMP concentrations. # Discussion In this study we have investigated the influence of dexamethasone on the CSF secretion of MMPs/TIMPs in (n = 37) adults with TBM. Despite the small number of patients we found dexamethasone significantly decreased CSF MMP-9 concentrations early in the course of TBM treatment and thus may represent one mechanism by which corticosteroids improve outcome. As dexamethasone did not influence the CSF concentrations of other MMPs this suggests a specific mode of action. Our previous investigations failed to demonstrate a significant effect of dexamethasone on serial CSF cytokines or chemokines, although CSF protein fell faster in the dexamethasone treated group and there was a trend toward lower IFN-γ concentrations. These data are consistent with our *in vitro* data that indicate IFN-γ influences intra-cerebral MMP-9 secretion. MMP-9 is quantitatively the most significant MMP released by macrophages and concentrations are increased in lungs and pleural fluid of TB patients. Serum MMP-9 concentrations may be raised and associated with pulmonary disease severity. *In vitro*, MMP-9 is secreted from monocyte cell lines in response to stimulation by a number of TB components such as lipoarabinomannan (LAM), a process dependent upon the transcription factor AP-1. Previous studies suggest MMP-9 may be an important molecule in TBM pathogenesis. A study of adults with TBM, viral, and bacterial meningitis found increased CSF MMP-9 concentrations were associated with more severe disease and death in all cases. A study of TBM alone has replicated this finding. Other cellular work demonstrates that *M.tb*-infected monocytes drive a network involving IFN-γ, TNF-α, IL-1β and NFκB resulting in increased astrocyte MMP-9 secretion without altered TIMP-1/2 secretion indicating a role for glial cells in the development of the matrix degrading phenotype present in CNS TB. However, murine studies suggest MMP-9 is also important in host defense to *M. tuberculosis*. Macrophages from mice resistant to *M. tuberculosis* produce more MMP-9 mRNA in response to intracellular infection than susceptible mice. Non-specific MMP inhibition blocked early dissemination of *M. tuberculosis* and MMP-9-knockout mice had reduced pulmonary recruitment of macrophages and poorer granuloma formation. Therefore *in vivo* a balance may exist between the MMP-9 necessary for host- defense and excess causing the unwanted effects opposed by dexamethasone. Our data suggest that the majority of early CSF MMP-9 is released from neutrophils as there was a strong correlation between both baseline values. Other studies have reported strong correlations between MMP-9 and neutrophils in viral meningitis and with total white cells in TBM,. Neutrophils release large amounts of preformed MMP-9 in response to many inflammatory stimuli, which may explain the lack of correlation observed between CSF MMP-9 and the pro- inflammatory cytokines measured. Additionally there was a trend in our data to suggest that MMP-8, a collagenase principally derived from neutrophils and important in bacterial meningitis, might also be decreased by dexamethasone. The role of the neutrophil in the pathogenesis of TB is not well understood. Peripheral neutrophil counts have been inversely related to risk of TB in contacts of infectious cases and neutrophil depletion *ex vivo* impaired the ability of whole blood to restrict mycobacterial growth via depletion of secreted anti-microbial peptides. A report that neutrophil derived anti- microbial peptides are released in response to *M. tuberculosis* infection but not expressed in the tuberculous granuloma is consistent with the proposed early role for neutrophils in TB. In addition, adults with TBM and a high proportion of neutrophils in the CSF have an increased chance of survival. Given the strong correlation between CSF neutrophils and MMP-9 concentrations we investigated the influence of dexamethasone on CSF neutrophil numbers, but did not observe any differences between the treatment groups at any time point. However, dexamethasone may affect function rather than absolute numbers. Corticosteroids can rapidly prevent neutrophil degranulation independently of effects on protein synthesis. The restricted release of neutrophil-derived proteins in dexamethasone treated patients may have contributed to the significant reduction in CSF protein observed in the dexamethasone group in the published study involving a larger number of patients from the parent trial. The study has a number of limitations. First, we only studied a small sub-group of patients recruited to the controlled trial. In addition we only studied time points at 0 and 5 days that lead to our conclusions, further work must delineate more clearly when dexamethasone is having this early effect. We could only study patients from one centre, with bacteriologically confirmed disease, in whom multiple serial CSF specimens were available. Therefore because of its small size the study lacked power to detect small effects of dexamethasone and to link effects with outcome. In addition, the patients selected may not be representative of all the patients in the trial. However, the baseline clinical variables most strongly associated with outcome – disease severity grade and coma score - were well matched, both within the patients included in the MMP study and when these patients were compared with the rest of the trial patients. The worse outcomes in patients not included in the MMP study may be partly explained by their older age and lower numbers of white cells in the CSF, both independent risk factors for death from TBM. Also, we used ELISAs that detected both active and pro-MMP-9 and thus could not differentiate between biologically active and inactive or degraded MMP-9. Differences in the activity of MMP-9 (assessed by zymography) have been reported in experimental pyogenic meningitis where treatment with antibiotics alone increased CSF MMP-9 activity, but dexamethasone co-administration suppressed this effect. In summary, this investigation has shown that dexamethasone may influence outcome from TBM by reducing MMP-9 secretion. However, future research needs to delineate the roles of neutrophils and their MMPs, specifically MMP-8 and -9, in the early stages of TBM treatment more clearly. Targeting neutrophil-derived, MMP-driven, inflammatory responses early in TBM may be a potential therapeutic strategy. # Methods ## Study participants and setting We studied a subset of patients from a previously reported randomized, double- blind, placebo-controlled trial of adjunctive dexamethasone for the treatment of TBM in 545 Vietnamese adults. To be eligible for selection for the current study patients had to have been recruited at the Hospital for Tropical Diseases (HTD), Ho Chi Minh City, have a microbiologically confirmed diagnosis of TBM (*Mycobacterium tuberculosis* cultured from the CSF), negative HIV-1 & 2 antibodies, and at least two serial CSF specimens taken within the first 60 days of treatment (when dexamethasone was stopped) available for analysis, unless they had died within this period. Disease severity at admission was defined by the Glasgow coma scale (GCS). Assessment of GCS was performed by summation of clinical assessment of eye opening (awarded 1–4 points), best verbal response (1–5 points) and best motor function (1–6 points) where lower values indicate more severe neurological dysfunction. Disease severity was also assessed using the United Kingdom Medical Research Council grade (UK MRC) for TBM : patients in grade I had a GCS of 15 (fully conscious) and no focal neurological signs; patients in grade II had a GCS of between 11 and 14 and/or focal neurological signs; and patients in grade III had severe coma with a GCS of less than 11. The ethical and scientific committees of Imperial College, Oxford University and the Hospital for Tropical Diseases approved the study and written, informed consent to participate in the study was obtained from the patients or their relatives. All clinical investigation was conducted according to the principles expressed in the Declaration of Helsinki. ## Treatment and specimen collection Adults were allocated randomly to start dexamethasone sodium phosphate or placebo as soon as possible after the start of four drug anti-tuberculosis therapy as previously described. Briefly, adults with grade II or III disease received intra-venous study drug for 4 weeks (week 1 0.4 mg/kg/day); week 2 0.3 mg/kg/day, week 3 0.2 mg/kg/day, week 4 0.1 mg/kg/day), then 4 weeks of tablets starting at 4 mg total dose per day, reducing by 1 mg per week until zero. Those with grade I disease received 2 weeks of intra-venous study drug (week 1 0.3 mg/kg/day, week 2 0.2 mg/kg/day) then 4 weeks of tablets (week 3 0.1 mg/kg/day; then 3 mg total per day reducing by 1 mg per week until zero). Serial CSF samples (frozen at −70°C) and routine clinical investigations were collected as part of normal clinical care pre-treatment and on study days 3, 7, 30, 60 and 270. ## Assessment of the clinical and CSF inflammatory response to treatment The primary outcome of the trial was death or severe disability 9 months after randomization and was assessed blind to the treatment allocation. Likewise, treatment allocation was unknown to those assessing conventional and experimental CSF inflammatory responses including MMPs/TIMPs. CSF white cell count, protein, glucose and lactate concentrations were determined by standard methods. Blood brain barrier integrity was assessed by measurement of paired CSF and plasma albumin concentrations by standard methods. The albumin index was calculated using the formula \[albumin*_csf*\]/\[albumin*_plasma*\]. The concentrations of CSF cytokines (IFN-γ, TNF-α, and the interleukins -1β, -6, -8, -10, and 12p70) were determined by cytometric bead array assay (BD Biosciences) as described. ## CSF MMP and TIMP concentration measurement MMP-1, -2, -3, -7, -8, -9 and -10 and TIMP-1, -2 and -4 were measured by ELISA (R&D, Abingdon, UK) according to the manufacturer's instructions. The lower level of sensitivity for the ELISAs were 156 pg/ml for MMP-1 and -7, 78 pg/ml for MMP-2 and -10, and 31 pg/ml for MMP-3, -8, -9, TIMP-1, -2 and -4. Samples with high MMP/TIMP concentrations were diluted as appropriate. Values below the level of detection were assigned an arbitrary value halfway between 0 and the lower level of sensitivity. ## Statistical analysis The analysis followed a pre-defined plan, unless otherwise stated as exploratory analysis. Continuous variables were compared by Student's t-test if normally distributed and Mann-Whitney U test if non-normally distributed. Where more than two groups were compared the Kruskal-Wallis test was used for non-parametric data. Categorical data were compared using Fisher's exact test or the Χ² test. All p-values were two sided and a value of \<0.05 was taken as significant. The changes in MMP/TIMP concentration between pre-treatment and follow-up samples were calculated using repeated measures ANOVA with dexamethasone as a covariate. Baseline MMPs and TIMPs were analyzed by univariate analysis to examine relationships between presenting clinical features and the specific outcomes of death and combined poor outcome (defined in the original trial as death or severe disability at 9 month follow-up). Clinical and laboratory features with P\<0.15 were then entered into a multivariate model using forward and backward likelihood ratio logistic regression. All analyses were performed using SPSS version 15.0 (SPSS Corp, Chicago, IL, US). # Supporting Information The authors wish to thank the doctors and nurses at The Hospital for Tropical Diseases who cared for the patients and allowed collection of samples for this study to occur. [^1]: Conceived and designed the experiments: JAG JF JSF GET. Performed the experiments: JAG CTHT. Analyzed the data: JAG GET. Contributed reagents/materials/analysis tools: MTHN PHN SXD NDTH CVL HTT. Wrote the paper: JAG CTHT JF HTT JSF GET. Recruited patients to the original trial: CTHT MTHN PHN SXD NDTH CVL HTT. Collected samples for the original trial: MTHN PHN SXD NDTH CVL HTT. [^2]: The authors have declared that no competing interests exist.

# Introduction *PRKCZ* encodes a protein belonging to the atypical subclass of the protein kinase C family of serine/threonine kinases that has been implicated in the regulation of cellular transformation and carcinogenesis. PRKCZ has previously been observed to be involved in multiple signal transduction pathways, including activation of the ERK/MAPK cascade, p70 ribosomal S6 kinase signalling cascade, transcription factor NF-κB, as well as regulation of cell polarity. The regulation of these pathways may explain some of the mechanisms by which PRKCZ can promote human cancers. Indeed, the roles of PRKCZ in various cancer types have been examined in recent years. For example, it was reported that *PRKCZ* expression level is two fold higher in glioblastoma cell lines compared with normal astrocytes. Subsequent studies showed that this high level of expression is correlated with increased proliferation of glioblastoma cells, while reduced expression is correlated with inhibition of migration and invasion. The involvement of activated PRKCZ in epidermal growth factor (EGF)-induced chemotaxis has also been examined in lung and breast cancer, and it was shown that PRKCZ is able to elicit a migration response of these cells by acting as a downstream mediator in the phosphatidylinositol 3-kinase (PI3K)/AKT pathway. Additionally, PRKCZ participates in cell polarity pathways, and studies have illustrated that loss of cell polarity, which results in tissue disorganization, may contribute to cancer development. It has also been observed that PRKCZ is mislocalized in a subset of ovarian cancers, and it was suggested that this mislocalization may reflect a role for apical-basal loosening, thus disrupting cell-cell adhesion, as well as increasing cell growth; however, additional evidence supporting the role of PRKCZ in ovarian cancer remains limited. In the present study, we tested the hypothesis that PRKCZ plays a role in ovarian cancer cell viability, proliferation and migration. We detected an increase in cell proliferation in SKOV3 cells when PRKCZ was over-expressed. Moreover, SKOV3 cells exhibited a decrease in cell migration when endogenous PRKCZ expression was down-regulated by small-interference RNA (siRNA). Our data further illustrate that up-regulation of PRKCZ leads to expression alterations of IGF1R and ITGB3 in SKOV3 and OVCAR3 cell lines, suggesting that PRKCZ may participate in ovarian cancer progression by modulating the expression of other important signalling molecules. # Materials and Methods ## Cell Culture Ovarian cancer cell lines SKOV3 and OVCAR3 were purchased from American Type Culture Collection (Manassas, VA). SKOV3 cells were maintained in McCoy’s medium supplemented with 10% FBS. OVCAR3 cells were maintained in RPMI-1640 medium supplemented with 20% FBS and 0.01 mg/ml bovine insulin. Cells were incubated at 37°C in a humidified atmosphere of 5% CO₂ and 95% air. ## *PRKCZ* Expression Vector & Generation of Stable Clones PCR conditions to amplify human *PRKCZ* in a 25 μL reaction volume were as follows: 2.5 μL of 10X Platinum HiFidelity Buffer (Invitrogen), 1.5 μL of 10 mM dNTPs (Invitrogen), 1.0 μL of 50 mM MgSO₄ (Invitrogen), 0.3 μL of 30 μM EcoRI-tagged forward primer (`5’-TCATGAATTCACTATGCCCAGCAGGACC-3’`), 0.3 μL of 30 μM SalI-tagged reverse primer (`5’-CATAGTCGACACACCGACTCCTCGGT-3’`), 0.5 μL of Platinum HiFidelity *Taq* Polymerase (5U/μL, Invitrogen), 17.9 μL of ddH₂O, and 1 μL (50 ng) of pooled human cDNA (derived from 13 human cell lines: NTERA-2, Hs578T, HepG2, Ht1080, SW872, T45D, MCF-12A, SKOV3, Fetal Normal Muscle Cells, Colo-205, MOLT-4, RPMI 8226, and SK-MEL-28). Thermal cycling parameters were as follows: initial incubation for 2 minutes at 94°C; 40 cycles of 30 seconds at 94°C, 30 seconds at 73°C, 2 minutes at 72°C. PCR products were resolved by 1.0% agarose gel electrophoresis, visualized under UV, and gel extracted and purified according to the manufacturer’s protocol (Qiagen). Subsequently, they were transferred to pEGFP-N2 (N-terminal GFP tag) expression vector (Clontech). Correct *PRKCZ* sequence within vector was confirmed by sequencing. Each cell line was transfected with the plasmid vectors PRKCZ-pEGFP or vector controls, using Fugene 6 Transfection Reagent (Roche). Following transfection, cells were cultured with G418 sulfate (800 μg/ml and 500 μg/ml for SKOV3 and OVCAR3, respectively). Surviving colonies were individually selected and maintained in G418 sulfate-containing medium. ## Quantitative Real-Time PCR Primer pairs for genes of interest were designed individually by using Primer3 input software (Whitehead Institute, Howard Hughes Medical Institute, NIH). *PRKCZ*-forward: 5’-GGCCACAGACTGGATTTTCT-3’, *PRKCZ*-reverse: `5’- CTCGCTGGTGAACTGTGTGT-3’`; *IGF1R*-forward: `5’-GGTGGAGAACGACCATATCC-3’`, *IGF1R*-reverse: `5’-GCCAGCGCACAATGTAGTAA-3’`; *ITGB3*-forward: `5’-ATGGGACACAGCCAACAACC-3’`, *ITGB3*-reverse: `5’-GTGGCACAGGCTGATAATGA-3’`; *TIMP1*-forward: `5’-TGACCAAGATGTATAAAGGG-3’`, *TIMP1*-reverse: `5’-GTTGTGGGACCTGTGGAAGT-3’`. Quantitative real-time RT-PCR was performed on an ABI Prism 7000 Sequence Detector (Applied Biosystems) using SYBR Green PCR Master Mix (Applied Biosystems). Each of the 20 μL PCR reactions contained 1 μL (50 ng) of cDNA and 0.45 μM of each of the primers. The thermal cycles for PCR reaction were as follow: initial denaturation for 10 minutes at 95°C, followed by 40 cycles of 95°C for 15 seconds, and annealing extension at 60°C for 1 minute. The housekeeping gene *HPRT1* (forward: `5’- ATGGTCAAGGTCGCAAGCTTG-3’`, reverse: `5’- CAAATCCAACAAAGTCTGGCT-3’`) was used to normalize gene expression values. Reference cDNA consisted of a pool of 13 cell lines was used to generate standard curve to quantify cDNA levels of samples. ## Western Blotting Cells were washed three times with cold phosphate-buffered saline (PBS), lysed with NETN lysis buffer (20mM Tris-HCl, pH 7.5; 150mM NaCl; 1 mM EDTA, pH 8.0; 0.5% Nonidet P-40; 1 mM PMSF; 1x protease and phosphatase inhibitors) on ice for 10 minutes and centrifuged for 8 minutes at 12,000 rpm to separate lysates from cell debris. Protein concentrations were determined with BCA Protein Assay Kit (Pierce). Equal amounts of protein from cell lines were loaded and separated on 8–10% SDS-PAGE. Proteins were transferred to Hybond ECL nitrocellulose (Amersham) and blotted using anti-PRKCZ, anti-IGF1R, anti-ITGB3, anti-IRS2 (Cell Signalling), and anti-pIRS2 (Abcam) antibodies at 1:1000 dilutions. Secondary conjugates, HRP-Donkey anti-mouse or HRP-Donkey anti-rabbit (Jacksons Immunochemicals) were incubated for 1 hour at a 1:5000 dilution. Protein bands were visualized by chemiluminescence using ECL detection system (Amersham). ## Cell Viability Assays In brief, cells were seeded in 96-well plates at a concentration of 1000 cells/well with a final volume of 100 μL of culture media and were incubated at 37°C, with or without myristoylated pseudosubstrate peptide (40 μM), a PRKCZ inhibitor. After each incubation period, 10 μL of the MTT (3-(4,5-Dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide) labelling reagent (Roche) were added to each well at a final concentration of 0.5 mg/mL. Cells were incubated for an additional 4 hour period, followed by addition of 100 μL of solubilization solution (10% SDS in 0.01 M HCl). Plates were allowed to stand overnight at 37°C and the spectrophotometrical absorbance of the samples was measured using a microplate (ELISA) reader at a wavelength of 570 nm with background subtraction at 630 nm. Each assay was performed in triplicates. ## BrdU Proliferation Assay Approximately 1 x 10⁵ cells were plated onto coverslips within 6-well plates and allowed to grow to \~60% confluency overnight. On the next day, cells were incubated in 10uM of BrdU for 6 hours, washed with PBS, then fixed with 4% paraformaldehyde, denatured with 2M HCl, and neutralized with 0.1 M sodium borate. Cells were then incubated with mouse anti-BrdU (1:200, Dako) for 1 hour, washed, followed by incubation with Alexa Fluro 647 anti-mouse (Molecular Probes) for an additional 30 minutes, and counterstained with DAPI. BrdU incorporation was observed and counted in five fields of view per well through microscopy. Each cell line was performed in triplicates per experiment and the ratio of BrdU-positive cells to total cell number was calculated. ## Scratch Wound Healing Assay Cells were plated into 6-well plates and cultured to confluence. Cells were rinsed with PBS and serum starved overnight in 0.5% FBS media at 37°C and 5% CO₂. The next day, three separate scratches were introduced through the monolayer of cells in each of the wells using sterile 200 μL plastic pipette tips. Cells were then rinsed gently with PBS to remove cellular debris and replaced with fresh culture media supplemented with 0.5% FBS. The wounded cells were allowed to incubate at 37°C and representative fields were photographed with an inverted-phase microscope at different time intervals. ## siRNA Transfections Knockdown of *PRKCZ*, *IGF1R*, and *ITGB3* expression in ovarian cancer cell lines was achieved by transfection of siRNAs (Ambion). siRNAs targeting of these genes was performed with Dharmafect-4 transfection reagent (Dharmacon). In brief, cells were seeded in 12-well or 6-well plates at densities of 1 x 10⁵ or 2 x 10⁵ cells/well, respectively. Cells were then treated with siRNA transfection mixtures following the manufacturer’s protocol. Scrambled siRNA (Ambion) was used as a control. Additional controls included mock-treated cells that received transfection reagent without siRNA, as well as untreated cells that received only fresh media. Cells were harvested after 48 or 72 hours for RNA and protein extraction, respectively. ## Statistical Analyses All data were represented as means ± the standard deviation (SD) of the mean. Statistical calculations were performed with Microsoft Excel analysis tools. Differences between groups were analyzed by student *t*-test. *P* values of \< 0.05 were considered statistically significant. # Results ## PRKCZ plays a role in cell viability in SKOV3 ovarian cancer cells To address the potential roles of PRKCZ in ovarian tumourigenesis, including cell viability, proliferation, cell migration, as well as relevant downstream signalling pathways, we performed several *in vitro* functional assays using ovarian cancer cells. The endogenous PRKCZ transcript and protein levels were low for both SKOV3 and OVCAR3 compared to THP-1 cells, a human acute monocytic leukemia cell line that expresses high endogenous levels of PRKCZ (Fig). PRKCZ has previously been demonstrated to be involved in cell survival in various cell types. To assess whether it has the same effect on ovarian cancer cells, stable cell lines over-expressing PRKCZ were generated (Fig –), and MTT cell viability assays were performed using randomly selected clones. It was observed that over-expression of PRKCZ significantly correlated with an increase in the viability of SKOV3 cells, and this effect was abolished by the addition of a myristoylated pseudosubstrate peptide that targets PRKCZ. In contrast, over-expression of PRKCZ did not have an effect on OVCAR cell lines when compared to parental and empty-vector control cells. This result suggests that PRKCZ can enhance cell viability in a subset of ovarian cancer cells. To investigate whether the increased cell viability seen in SKOV3 cells that over-express PRKCZ was due to an increase in cell proliferation, BrdU cell proliferation assays were performed. SKOV3 cells over-expressing PRKCZ displayed a higher percentage of cells with BrdU incorporation, indicating that the rate of proliferation in these cells is higher than parental and empty vector controls. ## Knockdown of PRKCZ inhibits the cell migration of SKOV3 ovarian cancer cells The ability of cancer cells to migrate is one of the key processes in cancer progression. To determine whether PRKCZ can affect the migratory properties of ovarian cancer cells, scratch wound healing migration assays were performed. No migration differences were observed when PRKCZ-transfected cells were compared with controls (data not shown). To examine if the endogenous levels of PRKCZ in parental cells are sufficient for their migration, wound healing assays were repeated with the same ovarian cell lines that have been subjected to *PRKCZ* siRNA knockdown. It was observed that knocking down the endogenous level of *PRKCZ* can in fact decrease the migration rate of SKOV3 cells (Fig and). This phenotype was not observed in OVCAR3 cells (data not shown). ## IGF1R and ITGB3 are Potential Downstream Effectors of PRKCZ IGF1R and ITGB3 participate in important cellular signaling pathways involved in cancer development and both have previously been implicated as prognostic factors in ovarian cancer; since PRKCZ plays a role in the regulation of various cellular signal pathways, we sought to determine whether PRKCZ can affect the regulation of IGF1R and ITGB3 in ovarian cancer. To examine if there is a direct molecular relationship between PRKCZ, IGF1R and ITGB3, the expressions of IGF1R and ITGB3 in parental and PRKCZ-expressing ovarian cancer cells were compared at the transcriptional and protein levels. Whereas the transcript level of *IGF1R* was not altered in SKOV3 cells, an increase in PRKCZ protein expression correlated with an increased level of IGF1R protein, suggesting that PRKCZ may participate in IGF1R translation or protein stability (Fig). Additionally, the expression level of phosphorylated IRS-2 (insulin-receptor substrate-2), a known downstream target for IGF-IR, was also increased in SKOV3 cells, confirming that PRKCZ is involved in the activation of IGF1R signalling pathway in this particular cell line. Interestingly, in contrast to SKOV3 cells, expression of *IGF1R* was decreased in OVCAR3 cells that over-express PRKCZ, at both the transcript and protein levels (Fig). In addition to IGF1R, mRNA and protein expression of ITGB3 were significantly decreased in both SKOV3 cells and OVCAR3 cells that over-express PRKCZ, compared to parental and empty-vector controls (Fig –). The concurrent expression alterations observed for IGF1R and ITGB3 in PRKCZ- expressing cells prompted the question of whether these changes are occurring within the same biological pathway. Therefore, we performed *IGF1R* siRNA gene knockdown experiments in SKOV3 cells to determine if reducing *IGF1R* expression would have an effect on *ITGB3* gene expression. Results from quantitative real- time PCR analysis indicated that *IGF1R* knockdown can lead to derepression of *ITGB3* mRNA expression in cells over-expressing PRKCZ. To further address whether activation of the IGF1-signalling pathway plays a role in *ITGB3* gene regulation, the transcript level of *ITGB3* was examined after SKOV3 cells were stimulated with IGF1, a known ligand for IGF1R. Interestingly, similar to the *IGF1R* siRNA knockdown experiment, *ITGB3* expression was derepressed in PRKCZ- expressing cells when stimulated with IGF1. Moreover, to examine if there is a negative feedback mechanism that regulates expression of *IGF1R*, we induced SKOV3 cells with IGF1 and examined their *IGF1R* mRNA expression. Upon stimulation with IGF1, the transcript expression of *IGF1R* was decreased in all SKOV3 cells, as observed by quantitative RT-PCR. This suggests that IGF1 stimulation in SKOV3 cells may in fact be able to repress the transcript expression of *IGF1R*. ## TIMP-1 as a Potential Downstream Effector in IGF1 Signalling Based on our observations that both ITGB3 mRNA and protein levels are decreased in PRKCZ-expressing cells in two ovarian cancer cell lines (SKOV3 and OVCAR3), we sought to identify potential downstream players within this signalling pathway that may play role in ovarian cancer. Transcription of *TIMP1* (TIMP metallopeptidase inhibitor 1) has previously been shown to be up-regulated by ITGB3 in the ovarian cancer cell line MDAH 2774 and thus may be a candidate target. Therefore, the mRNA level of *TIMP1* in SKOV3 and OVCAR3 cells was examined. Interestingly, *TIMP1* expression was decreased in PRKCZ clones of both of these cell lines, which correlated with the expression of *ITGB3* (Fig). To further examine if the decrease in *TIMP1* expression was directly related to the decreased level of *ITGB3*, SKOV3 parental cells were subjected to *ITGB3* knockdown. No difference in *TIMP1* expression was observed between these cells (Fig), suggesting that the decreased level of *TIMP1* in PRKCZ-expressing cells was ITGB3-independent. In addition to *ITGB3*, we examined if induction of IGF1 signalling has an effect on *TIMP1* expression. Upon IGF1 stimulation, SKOV3 parental and empty- vector controls exhibited a 2-fold decrease in *TIMP1* mRNA, but no further decrease in *TIMP1* expression was observed in PRKCZ clones. This is an interesting finding, as IGF1 and PRKCZ signalling pathways may converge through their potential roles in the regulation of *TIMP1* expression. ## Effects of IGF and ITGB3 Signalling on Cell Migration in SKOV3 Cells The lack of migratory changes in PRKCZ-expressing SKOV3 cells as observed from the migration experiments as described above may be due to lack of stimulation. Since an increase in IGF1R expression was observed in cells over-expressing PRKCZ, similar migration assays were repeated with the addition of IGF1. An increase in migration was observed in parental and empty-vector controls upon stimulation with IGF1 in the wound healing assay, illustrating that the IGF1 signalling pathway is involved in migration of SKOV3 cells. However, this effect was not observed in PRKCZ-expressing cells. The lack of response in PRKCZ- expressing cells may perhaps be due to negative feedback exerted by the over- expression of IGF1 receptor in these cells. Matrigel migration assays also did not show an increased level of invasion in any of cells upon stimulation of IGF1 (data not shown). Lastly, to evaluate if ITGB3 plays a role in ovarian cancer migration, scratch wound healing assays were performed with SKOV3 cells treated with *ITGB3* siRNA. No difference in migration rate was observed between siRNA-treated and control cells (Fig). # Discussion Human tumourigenesis is a multistep process in which cells can acquire properties, through genetic alterations, that allow them to transform to a higher malignant derivative. In the present study, we aimed to determine if alteration in PRKCZ expression can drive such processes in ovarian cancer, including cell viability, proliferation, and cell migration. PRKCZ has previously been implicated to be involved in various cancer cell types, including glioblastoma, prostate cancer, squamous cell carcinomas of the head and neck, squamous cervical cancer and soft tissue sarcomas. Our current study revealed that PRKCZ also plays a role in proliferation in a subset of ovarian cancer cell lines, as it was observed that it affects cell viability in SKOV3 cells, but not OVCAR3 cells. The ability of cancer cells to migrate during cancer progression is associated with the acquisition of abnormal motile behaviour resulting from various molecular alterations. Aberrant expression of PRKCZ is an example of such alteration, as its role in migration has previously been demonstrated in breast cancer, head and neck tumour cells, and pancreatic cancer. Comparing migration of parental cell lines and PRKCZ-expressing counterparts indicated that over- expressing PRKCZ alone is not sufficient to exert increased migratory properties in the two ovarian cancer cell lines tested. Nevertheless, we demonstrated that siRNA knockdown of *PRKCZ* expression in SKOV3 parental cells can decrease its rate of migration. Our observation suggests that the endogenous level of PRKCZ is sufficient for cell motility in this particular ovarian cancer cell line. In addition to PRKCZ, it is also important to note the potential roles of PRKCI (protein kinase C iota) in ovarian cancer. Similar to PRKCZ, PRKCI belongs to the atypical protein kinase C family group, and it has been implicated in the establishment of cell polarity, motility, proliferation, and survival of cancer cells. Interestingly, the *PRKCI* gene has been shown to be amplified and over- expressed in serous epithelial ovarian cancers, and an increase in its DNA copy number is associated with a decrease in progression-free survival for the disease. Since PRKCZ and PRKCI are highly homologous to one another, sharing \~70% overall amino acid sequence identity, it is possible that these two proteins function redundantly. Indeed, it has previously been demonstrated that disruption of either PRKCZ or PRKCI expression can inhibit tight junction formation in cultured epithelial cells, suggesting that these two proteins have an overlapping role in establishment of cell polarity. For that reason, it may also be important to further investigate if PRKCI can contribute to the phenotype that we have observed in PRKCZ-expressing cells. Additionally, since siRNA knockdown of *PRKCZ* did not have an effect in OVCAR3 cells, it may be of interest to determine if these cell lines are more dependent on the activity of PRKCI. PRKCZ is involved in various cell signalling pathways, thus may alter multiple downstream targets. To this end, we sought to identify some relevant molecular players that may be affected by PRKCZ expression. A large body of evidence has supported the importance of IGF1R expression in ovarian cancer, and that increased expression of IGF1R is associated with aggressiveness, as well as drug resistance of the disease. Our data indicate that over-expressing PRKCZ in certain ovarian cancer cell lines can alter the expression of IGF1R, and the type of expression alteration is cell-line dependent. Specifically, while no *IGF1R* transcript level alterations were observed in SKOV3 cells, the level of protein expression was found to be increased in cells that over-express PRKCZ, which may be explained by post- transcriptional processes such as protein translation, post-translation modification and decrease in protein degradation; however, the exact mechanism involved remains to be investigated. On the other hand, both gene and protein expression levels of IGF1R were found to be decreased in the OVCAR3 cell line when PRKCZ is over-expressed, suggesting that regulation of *IGF1R* by PRKCZ in this particular cell line may be occurring at the transcript level. These results suggest that regulation of IGF1R by PRKCZ may also be occurring within different biological pathways and may be dependent on other molecular characteristics specific to each of the cell lines, once again illustrating the heterogeneity of this disease, and further investigations are required to determine the exact mechanisms responsible for the dual effect PRKCZ has on IGF1R expression. The role of ITGB3 in ovarian cancer has been implicated in a study by Kaur et al. in which over-expression of ITGB3 in SKOV3ip1 cells (cell line generated from ascites developed in nu/nu mouse by administering an intraperitoneal injection of SKOV3 cells) was found to be associated with decreased invasion, protease expression, as well as colony formation. These observations were consistent with their subsequent *in vivo* experiments, which showed that tumours expressing ITGB3 were less aggressive compared to those that do not express this protein. Moreover, upon examination of ITGB3 expression in ovarian tissue of patients with invasive ovarian cancer, the same group found that patients with high ITGB3 expression had a significantly better prognosis, a finding that is consistent with other recent studies. Interestingly, our assessment of ITGB3 expression in PRKCZ-expressing SKOV3 and OVCAR3 ovarian cancer cells showed that ITGB3 is down-regulated in the presence of PRKCZ, as its gene and protein expressions were both decreased compared to controls. Future IHC studies should reveal whether this correlation occurs in ovarian tumour specimens. The concomitant altered expression of IGF1R and ITGB3 in PRKCZ-expressing cells led to the question of whether these genes are activated in the same signalling pathway. Indeed, *IGF1R* siRNA knockdown in SKOV3 cells revealed that *ITGB3* transcription may be dependent on expression of IGF1R; however, the effects of IGF1R on OVCAR3 cells may differ since its expression is decreased in PRKCZ- expressing cells. Nevertheless, results from this study suggest one possible mechanism by which ITGB3 expression may be altered, and as a consequence, a more aggressive phenotype of ovarian cancer cells is developed. Given that PRKCZ expression correlates with the expression of both IGF1R and ITGB3, we further examined if alteration of these genes can affect the migration phenotype of ovarian cancer cells that over-express PRKCZ. Interestingly, scratch wound healing migration assays showed that upon IGF1 stimulation, SKOV3 parental control cells, but not PRKCZ-expressing cells, displayed an increase in cellular motility, which was contrary to what was expected. The lack of response in PRKCZ-expressing cells may perhaps be due to a negative feedback mechanism exerted by the over-expression of IGF1 receptor in these cells, thus hindering the cells’ ability to respond to IGF1 signalling. Additionally, unlike results obtained from scratch wound assays, IGF1 stimulation had no effects on any of these cells in matrigel migration assays, suggesting that while IGF1 signalling may be involved in cell migration, it may be insufficient for increasing the invasive properties of these cells, because the cells are incapable of breaking down the matrigel matrix to cross the barrier. Based on these observations, we propose the following model by which PRKCZ may participate during tumour progression in a subset of ovarian cancer. In cells with normal expression of PRKCZ, the expressions of *IGF1R*, *ITGB3*, and *TIMP1* are in equilibrium. These same cells, when stimulated with IGF1, can decrease the expression of *TIMP1*. When PRKCZ is deregulated and over- expressed, it increases the translation or stability of IGF1R, thus enhancing IGF1 signalling, leading to a repression of *ITGB3* expression, which ultimately can lead to changes in cellular processes that can enhance the aggressiveness of a tumour cell (eg. impaired apoptosis, increased proliferation). When *IGF1R* gene expression is decreased (eg. via siRNA knockdown) in these PRKCZ-expressing cells, rescue of *ITGB3* expression occurs. Additionally, when cells are stimulated with IGF1, the overall expression of IGF1R decreases due to a negative feedback mechanism that leads to suppression of *IGF1R* transcription and rescues the repression of *ITGB3*; however, there may be another yet to be identified pathway downstream of IGF1 signalling that can lead to derepression of *ITGB3*. One possible pathway may be PI3K/AKT, as IGF1 is a potent activator of this signalling cascade. The results from our study suggest the potential roles of PRKCZ in ovarian cancer development; however, further investigations using an animal model are needed to elucidate the roles of PRKCZ and its molecular partners in ovarian cancer. A better understanding of the interaction of these molecules may be useful in development of therapeutics for the subset of ovarian cancer patients who display expression alterations of these genes. We would like to thank our colleagues at the Lunenfeld-Tanenbaum Institute and the University of Toronto for their valuable experimental advice and reagents, including Dr. Theodore Brown, Dr. Alicia Tone, and Dr. Katherine Sodek. We would also like to thank everyone from the Andrulis laboratory for their helpful advice and discussions, especially Dr. Nalan Gokgoz, Dr. Christopher Huggins, and Dr. Winnie Lo. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: KKYS ILA. Performed the experiments: KKYS. Analyzed the data: KKYS ILA. Contributed reagents/materials/analysis tools: KKYS ILA. Wrote the paper: KKYS ILA.

# Introduction Functional magnetic resonance data (fMRI) of the human brain acquired in a task- absent (“resting state”) condition has attracted increasing interest in recent years. Due to the absence of an experimental paradigm, analysis procedures based on an activation model are not applicable. New types of techniques have been developed focusing on functional connectivity rather than task activation. For instance, correlation of time series between a pre-specified seed region and all other voxels of the brain is robust and conceptually clear. However, it can only be successfully applied if some prior knowledge exists for identifying seed regions. Another widely used technique is based on independent component analysis (ICA) whose primary advantage is its freedom from hypotheses preceding the analysis and the need for selecting seed regions. However, the number of independent components is difficult to specify and assumptions must be made about what constitutes a valid network. For comprehensive reviews of the above and related methods see. More recently however, graph-based methods have been proposed for the analysis of functional and structural magnetic resonance data of the human brain. Their main feature is that they take brain regions as nodes in a graph. Some of these methods have also been applied to the analysis of resting state fMRI data. Given the small world properties of the human brain, graph-based methods provide a valuable tool for elucidating network structures. In the present study, we focus on a particular type of graph-based method that identifies nodes which play central roles within the network structure. Such nodes are characterized by a measure called “node centrality”. Node centrality is a key concept in social network analysis of which several competing definitions exist and some of which have been applied to fMRI data analysis in the past. Here we discuss several of these approaches - in particular “betweenness centrality”, “degree centrality” and “eigenvector centrality”. Sporns et al. for instance advocate a combination of various graph measures including degree, betweenness centrality and closeness centrality. Thus far, centrality measures have been applied to a pre-selected set of nodes consisting of at most several hundred elements. Here, we propose to apply this measure to all voxels in a region of interest covering the entire cerebrum thereby avoiding any selection bias. However, due to computational complexity, closeness and betweenness centrality measures are not suited for compiling brain maps with thousands of voxels. Therefore in this study, we will focus primarily on ‘eigenvector centrality’. To our knowledge, eigenvector centrality has not yet been used in the context of fMRI data analysis. Eigenvector centrality specifically weights nodes based on their degree of connection within the network. It does so by counting both the number and the quality of connections so that a node with few connections to some high-ranking other nodes may outrank one with a larger number of mediocre contacts. Google's “PageRank” algorithm is a variant of eigenvector centrality. Both the human brain and the world wide web exhibit small world properties suggesting that an algorithm that is effective as part of a search engine may also be effective in analyzing network properties of the human brain. Eigenvector centrality can be used on a variety of different similarity metrics. Here, we present applications based on linear correlations and on spectral coherences between times series. This latter approach allows us to draw conclusions about connectivity patterns in different spectral bands. The motivation for choosing spectral measures came from Salvador et al. who have emphasized the importance of investigating interregional dependencies in the frequency domain rather than in the time domain. We propose to use eigenvector centrality as a mapping tool for the entire brain or parts of it. Such maps can be subjected to statistical tests to detect groupwise differences in centrality between experimental states. For abbreviation, we will call this method ECM (Eigenvector Centrality Mapping). # Materials and Methods Several definitions of node centrality exist - each having a slightly different interpretation. Common to all of these definitions is that they are based on a symmetric matrix containing pairwise similarity measures. Let be such an similarity matrix where entries contain a pairwise similarity measure between time series in voxels and. The number of voxels is determined by user-specified regions of interest (ROI) to which all subsequent analysis steps are restricted. In the experiments reported in this study, the ROI covered the entire brain excluding the cerebellum and consisted of voxels. The matrix is symmetric so that each voxel can be viewed as a node in an undirected weighted graph in which similarity values correspond to weights along the edges of the graph. In graph-based applications, these weights represent distances between nodes and are therefore non-negative. As a result, centrality measures are generally also defined to be non-negative. See Bonacich for a discussion of this point. ## Degree centrality The simplest centrality measure is called “degree centrality”. The degree of a node is defined asThus, a node has a high degree if it has strong connections to many other nodes in the graph. ## Eigenvector centrality Eigenvector centrality was first introduced by Bonacich, and a later variant of it is a central part of Google's PageRank algorithm. Much like degree centrality, it favours nodes that have high correlations with many other nodes. However, in contrast to degree centrality it specifically favours nodes that are connected to nodes that are themselves central within the network. Thus it takes into account the entire pattern of the network. As before let denote an similarity matrix. Then the eigenvector centrality of node is defined as the -th entry in the normalized eigenvector belonging to the largest eigenvalue of. Note that with this definition fulfils the characteristics described above. To see why let be the largest eigenvalue and the corresponding eigenvector, thenwith proportionality factor so that is proportional to the sum of similarity scores of all nodes connected to it. Uniqueness of this definition is ensured by the Perron-Frobenius theorem which states that any square matrix with strictly positive entries has a unique largest real eigenvalue with strictly positive components. This is also true for irreducible square matrices with non-negative entries. An irreducible matrix has at least one non-zero off-diagonal element in each row and column. Since we assume that represents distances between nodes we have. In the present context, we may assume that is irreducible because fMRI time series are almost never entirely dissimilar so that a sufficient number of non-zero entries in exist. Thus, an eigenvector belonging to the normalized largest eigenvalue exists and its entries provide a centrality measure for each node which is uniquely defined and non-negative. Note that symmetric matrices with negative entries may have several largest eigenvalues that are not distinct so that the requirement of non-negativity is essential for ensuring the uniqueness of this definition (see for an example). Eigenvector centrality is related to principal components analysis (PCA) in that both methods are based on eigenvector decompositions of similarity matrices. However, PCA differs from eigenvector centrality in that it only allows linear correlations as a similarity metric. But linear correlations may be negative so that the first principal component is not uniquely defined because of possible multiplicities of eigenvalues. In our experiments we used linear correlations which were re-scaled to be non- negative and also a spectral coherence metric which is non-negative by definition (see below). Other similarity metrics such as mutual information or wavelet transform coherence (WTC) might be used for eigenvector centrality mapping (ECM) as well. Many algorithms for computing eigenvectors of symmetric matrices are known. In the present context, it suffices to find the eigenvector belonging to the largest eigenvalue. For this special case, the power iteration method is one of the most efficient, and was used in our experiments. ## Betweenness centrality The betweenness centrality of some node is defined as:where is the number of shortest geodesic paths from to, and is the number of shortest geodesic paths from to that pass through node. This is normalized by dividing through the number of pairs of nodes not including, which is. Betweennnesss centrality is computationally expensive. For weighted graphs, its complexity is which can be reduced for unweighted graphs to where is the number of edges (non-zero correlations) making it computationally impracticable for large values of or. Note that generally, correlations are thresholded at some user-defined level prior to applying betweenness centrality. It was used e.g. by He et al. for analyzing spontaneous fluctuations in a network consisting of 90 regions of interest. We tested betweenness centrality on a region of interest containing 17,398 voxels that covered parts of the left hemisphere of one subject. The computation took 26 hours using 4 parallel 2.6 GHz processors for a single data set. Application to a region of interest with full brain coverage was not feasible. ## Linear correlation Linear correlation has been proposed as a metric for analysing functional connectivity. A high positive correlation between two fMRI time series indicates a strong similarity, a high negative a strong dissimilarity. Note that this measure is quite agnostic about any form of causal influence between brain regions. It is defined as follows. Let and be time series of length in two voxels and. Their correlation is defined aswhere denote the sample mean and the standard deviations. Because the similarity matrix should be positive, linear correlations between time series must be re-scaled accordingly. We propose to usewhere denotes the correlation between two time series and the corresponding scaled version. Note however that strong negative correlations may indicate some form of inverse coupling. Therefore, an alternative way to handle negative correlations might be to take absolute values instead of the approach proposed here. ## Spectral coherence Salvador et al. have noted that interregional dependencies can be more readily observed in the frequency domain than in the time domain. Therefore, we have also used frequency based similarity metrics for ECM. Specifically, we employ spectral coherence for this purpose. It has been previously applied to fMRI data analysis. In the following, we give a brief overview. For more information see for instance. Let denote real-valued stationary time series in two voxels, and let be some frequency of interest. We assume that they are normalized to zero mean. Their cross-correlation function evaluated at lag is defined asand the corresponding cross-spectral density is:Analogously, the auto-spectral density of a single time course isSeveral choices for the weighting factors exist. Among the most common ones are Parzen or Tukey windows. Here, we used the Tukey window which is defined as:where is the number of lags to compute the autocorrelation for. As a rule of thumb, should be chosen to be in the range where is the length of the time series. In the data presented below the time series length was, and we used throughout. Since is not necessarily symmetric the cross-spectrum is generally a complex function. The real part of is known as the cospectrum denoted as and the imaginary part as the quadrature spectrum. The spectral coherence between at frequency is defined as: Note that information about phase lags is not included in the above measure. Frequency-dependent phase coherence can be computed using the above definitions as follows ## Experiment 1 Functional MRI/EPI data were acquired of 35 normal volunteers on a 3T MRI scanner (Siemens Tim Trio) using TR = 2.3 sec, TE = 30ms, 3×3 in-plane resolution, 3 mm slice thickness, 1 mm gap between slices. Each scanning session began with a task-absent (“resting state”) scan lasting 7.6 minutes during which subjects were asked to fixate a fixation cross. A second resting state scan with the same acquisition parameters followed about 10 minutes later within the same scanning session. In between these two scans, subjects were scanned in another task absent condition using sagittal instead of axial slices. Data from this scan were not used for the present study. All data sets were initially fieldmap corrected using the software system Lipsia. Data preprocessing then continued using FSL, and consisted of motion correction, bandpass filtering (SPECS), and spatial smoothing (SPECS). Finally, preprocessed data sets were registered into standard MNI152 (Montreal Neurological Institute) brain space using FSL's nonlinear registration software FNIRT, and resampled to an isotropic voxel grid with a resolution of 3×3×3. We manually defined a region of interest containing about 52,000 voxels covering the entire cerebrum to which subsequent ECM analysis was applied. ## Experiment 2 Functional MRI/EPI data were acquired of 22 normal volunteers on a 3T MRI scanner (Siemens Tim Trio) using TR = 2.3 sec, TE = 30ms, 3×3 in-plane resolution, 3 mm slice thickness, 1 mm gap between slices. The study was approved by the ethics committee of the University of Leipzig. All subjects have written informed consent. The subjects were asked to attend two scanning sessions, in one of which they were asked to refrain from eating after 6 pm of the previous day. During both sessions, we first acquired resting state data for 6.5 minutes during which subjects were asked to fixate a fixation cross. During the following 34 minutes they were shown pictures of food and tools that they were asked to respond to by button presses. Finally, another 6.5 minutes of resting state data were acquired. In the present study, we only analyzed the initial resting state data acquired before visual stimulation began, and ignored the rest of the experiment. Data processing was done using the software system Lipsia. All data sets were initially corrected for motion and slicetime offsets. A baseline correction was applied using a highpass filter with a cutoff frequency of 1/90 Hz, and a spatial smoothing with a Gaussian filter of fwhm = 8 mm was used. All data sets were initially registered to an AC/PC coordinate system where the data were resampled to an isotropic voxel grid with a resolution of 3×3×3. We manually defined a mask containing 40,000 voxels covering the entire brain while excluding the cerebellum and parts of CSF. ## Data processing For both experiments, we computed pairwise similarity matrices between time series of any two voxels inside the mask using scaled linear correlation and for experiment 2 also spectral coherence, and applied the ECM algorithm to these matrices. The resulting centrality maps were then transformed as described by van Albada et al. in order to ensure that they obey a Gaussian normal distribution as required for subsequent statistical tests. The results were corrected for multiple comparisons using cluster-size and cluster-value thresholds obtained by Monte-Carlo simulations, using a significance level of. Clusters in the resulting maps were obtained using an initial z-value threshold of 2.33. The Monte Carlo simulation determines the size and peak value a cluster must have in order to be considered statistically significant. Thus, a cluster with only a moderately high peak value might be considered significant if it is large enough. On the other hand, a cluster with a very high peak value might be significant even if it is rather small. Computation times for ECM were about 20 minutes per dataset on a 2.6 GHz Opteron processor. About 6 GByte of computer memory are needed to store an matrix with voxels to cover the cerebrum at resolution. # Results ## Experiment 1 shows group averages of eigenvector centrality in the two scans. A network of hubs including sensorimotor areas of the marginal ramus of the cingulate and mid-cingulate, thalamus, primary visual cortex, insula and operculum are common to both. shows results of a paired t-test contrasting the two scans. During the first scan, eigenvector centrality scores were significantly higher in left and right thalamus and in the cerebellum. During the second scan, eigenvector centrality was larger in posterior cingulate, medial frontal and right opercular cortices, and medial frontal areas. For comparison, we additionally computed another centrality map - this time using degree centrality instead of eigenvector centrality. Note that during the second scan, degree centrality was larger almost everywhere in the brain. ## Experiment 2 shows group averages of ECM based on scaled linear correlation. The overall ECM pattern looked very similar although we found statistically significant differences in precuneus when contrasting hungry against sated state. Recent literature postulated that posterior midline cortex, comprising precuneus and posterior cingulate cortex constitutes the core hub within the default network of the human brain with strong connections to ventral medial prefrontal cortex/anterior cingulate cortex, inferior lateral parietal cortex, and the hippocampi. This particular portion of the precuneus, located in the anterior section adjacent to the marginal ramus of the cingulate sulcus, has been implicated in self-related processing (in contrast to the episodic memory- related role of the posterior precuneus). Although the enhanced centrality of precuneus during the hungry state indicates a changed core hub of the default network, we found no other areas with significantly changed centrality values across the brain. Nonetheless, the increased centrality of anterior precuneus during the hungry state is consistent with the proposed self-related functionality of this region. We next employed spectral coherence to investigate frequency based similarity metrics because of their known advantages in the observation of interregional dependencies. We found strong effects of frequency in ECM across spectral bands as shown in. In frequency bands 1/10 Hz up to 1/20 Hz, ECM was significantly larger in precuneus, the striatum and several more areas. Very low frequency bands (1/25 Hz, 1/30 Hz, 1/35 Hz) dominate at the temporal poles and mediodorsal frontal areas. and show differences between the sated and the hungry state across various spectral bands based on spectral coherence. In particular, differences appear in the anterior precuneus at 1/20 Hz and 1/30 Hz, and in the ventral striatum at 1/30 Hz. # Discussion We propose eigenvector centrality as a new method for analyzing fMRI data. It is parameter-free, computationally fast and does not depend on prior assumptions. In contrast to previous studies using centrality measures, we have applied them here to a large region of interest consisting of thousands of voxels. Under those circumstances, betweenness centrality becomes computationally intractable. The computational speed allowed us to obtain whole brain centrality maps and use them in a manner similar to contrast maps obtained in standard regression analyses. In the first experiment, we found significant differences between ECMs of two resting state scans following each other within the same session. In particular, left and right thalamus had higher eigenvector centrality scores during the first scan. Thalamus has been implicated in mediating attention and arousal in humans, suggesting that subjects' attention and/or arousal may have declined with time spent in the scanner. We also found higher centrality in the cerebellum during the first scan. The cerebellum is involved in the coordination of voluntary motor movement and muscle tone. Perhaps the mental effort of remaining motionless for a prolonged period of time may have played a role in this context. On the other hand, posterior cingulate and anterior medial frontal cortex appeared stronger in the second scan - regions that are associated with the “default mode network”. A possible explanation might be that subjects were more relaxed and more “at rest” during the second scan so that the typical “default mode” pattern emerged more clearly. For comparison, we also computed degree centrality and found that during a second resting state scan, degree centrality increased almost everywhere indicating a general increase in correlations across the brain. This may be due to a global physiological influence such as respiration or heart rate. Eigenvector centrality on the other hand did not show such a global effect. Rather it highlighted specific regions that were differentially affected by the prolonged duration of the experiment. For the second experiment, we additionally used frequency instead of time based similarity metrics with the known advantages in the detection of interregional dependencies, we identified regions with significant changes in their centrality scores that match well with previous findings from experiments addressing paradigms related to food and eating in hungry and sated state. We found the ventral striatum as the most prominent region within the network (in the 1/30 Hz band, see) which is well known as a key region implicated in reward, e.g. such as consummatory food, and displays functional connectivity throughout the prefrontal and motor cortex. The spectral coherence measure assumes that the coupling between fMRI time series is stationary over time. This assumption may sometimes be unrealistic. In such cases, the wavelet transform coherence (WTC) might be better suited because it describes coherence and phase lag between two time series as a function of both time and frequency. It has recently been used for analyzing resting state fMRI data. For the present work, we have only used spectral coherence but not phase coherence. However, it might be advantageous to include phase coherence and use it in conjunction with spectral coherence. We plan to explore that possibility in future work. In both experiments, we found high centrality values in cortical and subcortical areas, but also in white matter regions. This agrees with results found by Mezer et al. who reported clusters of similar BOLD fluctuations not only in the cortical and subcortical regions, but also within the white matter. The origin of such effects is still unclear and remains the object of future research. It should be noted that low frequency fluctuations may also be caused by aliasing effects (undersampling) so that the actual sources of these signals need not be in that same low frequency range. Nonetheless, recent studies have confirmed that oscillations - even at very low frequencies - appear robust and reliable, so that these results are not unexpected. It remains to be shown whether these findings indicate the existence of natural frequencies at which specific networks operate. Such natural frequencies have recently been postulated by Rosanova et al. for the human corticothalamic circuits based on EEG and TMS data. Our findings suggest that analogous patterns might exist at much lower frequencies observable in fMRI even though the exact nature of these connectivity patterns remains to be investigated. In this context, it may be interesting to use alternative frequency-dependent similarity metrics as described e.g. in Salvador et al.. The initial analyses presented in this study demonstrate that eigenvector centrality is a computationally efficient tool for capturing intrinsic neural architecture on a voxel-wise level. The independence of centrality approaches from a priori hypotheses, makes it a valuable methodological addition to the “model-free” analytic toolbox. # Supporting Information We thank Karsten Müller, Franziska Busse and Stefan Kabisch for their support in acquiring the data for this work. [^1]: Conceived and designed the experiments: DSM AH BP JL HS MS AV RT. Performed the experiments: DSM AH BP JL HS MS AV RT. Analyzed the data: GL DG. Contributed reagents/materials/analysis tools: GL DG. Wrote the paper: GL. [^2]: The authors have declared that no competing interests exist.

# Introduction Mosquito-borne diseases are a major health problem worldwide, and cause important morbidity and mortality in tropical areas. To a lesser extend, the European population is also exposed to a variety of mosquito-borne pathogens. Outbreaks of mosquito-borne diseases with significant human health implications occurred on a mass scale in Europe in the last century. These outbreaks included the following: Dengue virus in Greece in 1928, West Nile virus in Camargue in 1962 and in Romania in 1996, Sindbis virus in Finland in 2002 and more recently, Chikungunya virus in Italy. Numerous indexes can provide a comprehensive understanding of the potential for mosquito-borne disease transmission and the dynamics of these diseases in human populations, such as the vectorial capacity, the basic reproductive rate (R0) and the entomological inoculation rate (EIR). These indicators of disease transmission levels can also quantify the impact of vector-control strategies. These indexes mainly depend on entomological parameters that can be measured in the field, including the human-biting rate (HBR) (average number of bites per individual per day received from a mosquito species). This parameter is currently estimated by entomological methods, such as human landing catches, or other strategies based on attractant traps (e.g., light traps, carbon-dioxide traps, odour-baited traps). These entomological methods have proven efficacy for monitoring the density of mosquitoes relative to the density of the human population – but the HBR has been shown to vary within small geographic areas, meaning that the results of local catches cannot be extrapolated to larger areas. These methods are not adapted to consider differences found within a population, which include differential attractiveness to mosquitoes or other environmental and socioeconomic factors that could induce important variations in individual exposure to vector bites. Moreover, entomological methods can be labour intensive, expensive and difficult to implement when mosquito numbers are low or because of logistical constraints. In addition, the deliberate exposure of human volunteers to vectors has raised some ethical issues related to human landing catches, which remains the most reliable method to estimate host/vector contacts. The use of immunologically based techniques to estimate individual exposure to arthropod vector bites, such as those from mosquitoes, ticks, sand flies and *Glossina*, has been described in several studies –. The saliva of hematophagous arthropods contains a complex mixture of biologically active proteins. These proteins may modify hemostatic responses and induce both cellular immunity and the production of specific antibodies. As described previously, mosquito salivary gland extracts can induce an IgG antibody response in individuals living in endemic areas – and in travellers transiently exposed to vectors in tropical areas, suggesting that salivary proteins can potentially be used as immunological markers to evaluate individual exposure to mosquito bites. Mosquito densities and species diversity can be influenced by the surrounding landscape, even in restricted areas. The Mediterranean coast of southern France presents areas with distinct demographic and ecological conditions, ranging from large wetland areas in the Rhone River delta (Camargue) to highly urbanised environments (city of Marseille). These contrasting landscapes mirror the density and geographical spread of some mosquito species, notably *Aedes caspius*. *Ae. caspius* is a Paleartic species that has demonstrated the ability to transmit the Rift Valley fever virus and the Chikungunya virus in the laboratory. This mosquito species was also suspected to be involved in the 1993 Rift Valley fever outbreak in Egypt. Despite its low vector competence for these viruses, *Ae. caspius* should be considered a potential vector in wetland areas due to its high anthropophily and its abundance. *Aedes caspius* is well adapted to swampy environments: it tolerates varying levels of salinity in larval breeding sites, and its larval development is linked to the alternating dry and wet seasons in areas where its eggs are laid. After abundant rainfalls events, a massive, synchronous adult population emerges and becomes a nuisance. In Camargue, *Ae. caspius* is active from March to November. *Ae. aegypti* and *Ae. albopictus* mosquitoes, both vectors of arboviruses (*e.g.*, yellow fever, dengue or chikungunya viruses), were not endemic in the study area at the time of the present work. To assess whether exposure to different densities and/or species of mosquitoes throughout the year could influence the antibody response against mosquito salivary gland extracts, we tested the IgG response to *Ae. caspius* salivary gland extracts of individuals living in three southern French areas (Camargue, Fos-sur-mer and Marseille) with distinct ecological environments at three time points (February 2007, September 2007 and January 2008). We concomitantly evaluated the IgG responses to salivary gland extracts from *Culex pipiens, Aedes albopictus* and *Aedes aegypti* as controls. The temporal and spatial evolution of IgG responses according to mosquito species will be discussed. # Results ## Kinetics of IgG antibody responses against Ae. caspius salivary gland extracts (AecSGE) from individuals living in distinct ecological environments First, we determined whether mosquito density, linked to the ecological environment and season, could influence the IgG responses against mosquito salivary gland extracts. Thus, the IgG responses against *AecSGE* were assessed in individuals living in Camargue, Fos-sur-mer or Marseille at three time points: February 2007 (T1) and January 2008 (T3), which corresponded to periods outside of the *Ae. caspius* exposure peak, and September 2007 (T2), which corresponded to the *Ae. caspius* exposure peak period. Independent of the sampling time (i.e., T1, T2 or T3), the IgG antibody responses against *Aec*SGE were significantly different among the sites (Kruskal-Wallis test, *p*\<0.0001). Additionally, independent of the site, the IgG response against *Aec*SGE increased significantly from T1 to T2 (Wilcoxon matched pairs test, *p*\<0.0001, p\<0.0018 and p\<0.0001 for the Camargue, Fos-sur-mer and Marseille sites, respectively) and decreased significantly from T2 to T3 (Wilcoxon matched pairs test, *p*\<0.0001 for the three sites). These variations gradually decreased in individuals living at the Camargue site (*i.e.*, the mean change in ΔT2-T1OD was +0.26, with a 95% confident interval (95% CI) from +0.14 to 0.38) compared to those living in the city of Marseille (*i.e.*, +0.05 \[0.01 to 0.09\]), with intermediate variation observed for those living in Fos-sur-mer (*i.e.*, +0.13 \[0.07 to 0.18\]). The mean changes in ΔT2-T1OD between the Camargue and Fos- sur-mer as well as the Camargue and Marseille sites were significantly different (*p* = 0.019 and *p*\<0.0001, respectively, Mann-Whitney test). No significant differences (Mann-Whitney test, ns) were observed when comparing the mean change in ΔT2-T1OD between the Fos-sur-mer and Marseille sites. In contrast, outside of the period of peak exposure to mosquito bites (*i.e.*, T1 and T3) at the three sites, the antibody response against *Aec*SGE returned to the background level at each site (Camargue, +0.04 \[−0.06 to 0.14\]; Fos-sur-mer, +0.09 \[−0.13 to −0.04\]; Marseille, +0.01 \[−0.02 to 0.04\]). Variations in the IgG antibody responses detected between T3 and T1 were considered not to differ (mean ΔT3-T1OD \<0.1). The cut-off value for seropositivity (mean aOD ±3 standard deviation) was defined as 1.03 based on the IgG reactivity of sera from individuals living in Marseille who were not exposed to *Ae. caspius*. Individuals showing aOD values above this cut-off level were classified seropositive. Seroprevalence significantly increased among individuals living at the Camargue study site from the T1 time point to the end of the exposure peak (T2), with values increasing from 29% to 54% (chi-squared test, *p*\<0.0249). In addition, seroprevalence significantly declined after the exposure peak and returned to the baseline level (T3 = 29%, chi-squared test, *p*\<0.0249). Although an increase in seroprevalence was observed from T1 to T2 and a decrease was observed from T2 to T3 for individuals living at the Fos-sur-mer study site, these variations were not significant (chi-squared test). No significant change in seroprevalence was observed between T1 and T3 at the Fos-sur-mer and Camargue sites. Additionally, no significant differences were observed when comparing seroprevalence between the two study sites independently of the sampling time (chi-squared test). ## Kinetics of IgG antibody responses to Cx. pipiens salivary gland extracts (CxpSGE) in individuals living in distinct ecological environments The IgG responses to *CxpSGE* were assessed using the same sera as that used for the *AecSGE* assay. High inter-individual heterogeneity in the antibody responses was observed at all time points and for all sites. Independent of the sampling time (i.e., T1, T2 or T3), no significant difference was observed in the IgG antibody responses to *Cxp*SGE between the sites (Kruskal-Wallis test). With respect to the kinetics, despite statistically significant variation being detected in the IgG antibody response to *Cxp*SGE between the T2 and T3 time points for the Camargue and Fos-sur-mer sites (Wilcoxon matched pairs test, *p*\<0.0001 and p = 0.0028 for the Camargue and Fos-sur-mer sites, respectively), these variations were weak. The IgG antibody variations for the Camargue and Fos-sur-mer sites were considered not to differ (below than 0.1 ΔOD), indicating global stability of IgG responses to *Cxp*SGE throughout the year. Conversely, in the city of Marseille, the variations in the IgG responses observed from T1 to T2 and T2 to T3 were significant (Wilcoxon matched pairs test, *p* = 0.016/74 and p\<0.0001 for ΔT2-T1OD and ΔT3-T2OD, respectively) and relevant (ΔT2-T1OD +0.1 \[0.01 to 0.19\]; ΔT3-T2OD −0.15 \[−0.21 to 0.08\]). ## Kinetics of antibody responses to Ae. albopictus (AealSGE) and Ae. aegypti salivary gland extracts (AeaeSGE) in individuals living in distinct ecological environments To estimate the specificity of the IgG response to *AecSGE*, the same sera were assessed for IgGs against salivary gland extracts of two mosquitoes from the *Aedes* genus (*i.e.*, *Ae. albopictus* and *Ae. aegypti*) that were not endemic in the study area until 2008. Independent of the sampling time (*i.e.*, T1, T2 or T3), no significant difference was observed when the IgG antibody responses to *AealSGE* or *AeaeSGE* were compared between sites (Kruskal-Wallis test). With regard to the kinetics analysis, despite statistically significant variations being detected in the IgG antibody response to *AealSGE* or *AeaeSGE* between some of the time points at the three sites, these variations were below 0.1 ΔOD and were considered to not differ. ## Correlation of the IgG response between mosquito species To estimate the level of cross-reactivity of the IgG response to the salivary gland extracts between two mosquito species at T2 (September 2007) at the three sites, a Spearman's rank correlation coefficient (rho) test was used, and the corresponding *p*-values were determined. Significant positive correlation coefficients (rho\>0.42; *p*\<0.0083) were obtained when the mosquitoes of the *Aedes* genus were compared at the three sites, mainly for IgG responses against non-prevalent mosquitoes (*i.e.*, *albopictus* and *aegypti*). Conversely, no significant correlation was observed between the IgG responses against *CxpSGE* and those against the three other *Aedes* species at the three sites, except for the Fos-sur-mer site, where a significant positive correlation coefficient (rho = 0.43; *p* = 0.0079) was obtained when the IgG responses against *CxpSGE* and *AecSGE* were compared. # Discussion Numerous studies have reported that mosquitoes' salivary components can induce an antibody response in humans under natural conditions. Here, we analysed human antibody responses against *AecSGE* according to spatial (environment) and temporal (seasons) variations in the level of *Ae. caspius* mosquito exposure. The specificity of the IgG response was also estimated at the genus and species levels. The Mediterranean coast of southern France includes areas with distinct demographic and ecological conditions that greatly influence the dispersion and composition of the mosquito fauna. Thus, three sites were selected in this area on the basis of environmental patterns influencing *Ae. caspius* density: the Camargue area, the town of Fos-sur-mer and the city of Marseille. In the Camargue wetlands, *Ae. caspius* is well adapted to the rural and swampy environment where it encounters favourable climatic and biotope conditions. The city of Marseille presents an urban habitat that is more suitable for *Cx. pipiens* mosquitoes than rural mosquitoes, such as *Ae. caspius*. Fos-sur-mer, a mid-sized town located between the Camargue area and Marseille, has an intermediary environment and is also exposed to *Ae. caspius* mosquitoes. To confirm the geographic distribution and the different densities of *Ae. caspius* among these three sites, collection of mosquitoes was conducted in July 2007 using carbon dioxide dry ice traps. The collected specimens indicated a decreasing mosquito density gradient from Camargue to Marseille. *Cx. pipiens* and *Ae. caspius* were the most abundant mosquitoes at Camargue (31% and 29%, respectively), as previously described. *Ae. caspius* mosquitoes were captured at the Fos-sur-mer site (21%), but none were found in Marseille. Thus, Camargue, Fos-sur-mer and Marseille were considered sites with high, medium and very low levels of exposure to *Ae. caspius* bites, respectively. In the present study, we showed that the IgG antibody responses to *AecSGE* evolved in accordance with the *Ae. caspius* density, which is influenced by both seasonal changes and the ecological environment. The variations in the IgG antibody levels against *AecSGE* between the peak exposure at T2 and the T1 baseline level were approximately 2- and 4-fold higher in Camargue than in Fos- sur-mer and Marseille, respectively. This positive relationship between anti- salivary protein IgG levels and the seasonal variation of human exposure to mosquito bites has previously been reported. Moreover, the average IgG response level against *AecSGE* observed at T3 returned to T1 baseline levels only four months after the exposure peak (T2), suggesting a short-lived IgG response. A decrease in the IgG response against salivary gland extracts after a period of non-exposure has been described for outdoor workers exposed to ticks and for travellers transiently exposed to *An. gambiae* and *Ae. aegypti* mosquitoes. The transient anti-saliva IgG response and its relationship with mosquito density may be useful for assessing mosquito exposure and could thus, provide new immunological tools to evaluate anti-vector strategies or to monitor vector populations. Recently, Drame and colleagues have confirmed the potential of *An. gambiae* saliva for use as an immunological exposure marker to assess the risk of malaria transmission and the efficiency of antivectorial strategies in a malaria-endemic area. It is worth pointing out that baseline IgG levels against *AecSGE* were significantly and pertinently higher at Camargue and Fos-sur-mer than at Marseille. Repeated seasonal exposure to *Ae. caspius* seemed to favour maintaining a high baseline IgG level throughout the year. To test this hypothesis, a comparison of the kinetics of the IgG response against *AecSGE*, collected outside the *Ae. caspius* exposure peak (cold season) between individuals who had lived for a long period (*i.e.*, at least 5 years) in Camargue and newcomers could be performed. Further studies may also analyse the kinetics of the antibody response in children born in Camargue. Collectively, these data suggest that to determine the prevalence of seroreactivity against *AecSGE*, several parameters should be considered, including mosquito density and the environment, historical mosquito exposure (time spent in a particular area) and individual behaviour (*e.g.*, outdoor/indoor activities, use of mosquito nets or repellents). To evaluate the specificity of this IgG antibody response, the same sera were first tested against *CxpSGE*. The mosquito collection performed in July 2007 indicated that *Cx. pipiens* was present at all sites, with decreasing densities observed from Camargue to Marseille. In contrast to the IgG response observed to *AecSGE*, the IgG levels against *CxpSGE* were considered not to differ spatially and temporally, with the exception of the levels from Marseille. Individuals from Marseille presented a significant and pertinent increase in the IgG levels against *CxpSGE* after the peak of exposure. These results indicated that unlike the responses against *Ae. caspius*, the IgG responses against *CxpSGE* appeared to not be associated with the decreasing density of *Cx. pipiens* from Camargue to Marseille. This phenomenon could be attributed to a distinct *Cx. pipiens* behaviour that could occur between sites. The temperate species *Culex pipiens* Linné can effectively be divided into two different biological forms: *Culex pipiens pipiens* (*Cx. p. pipiens*) and *Culex pipiens molestus* (*Cx. p. molestus*). These two subspecies are relatively morphologically similar but exhibit different physiological and behavioural traits. In contrast to the rural *Cx. p. pipiens*, the urban *Cx. p. molestus* is anthropophilic, breeds in underground urban habitats, is able to lay its first batch of eggs without a blood meal, does not hibernate and can mate in confined spaces. In Camargue, *Culex pipiens* L. mosquitoes were found at a high density in bird-baited traps compared to horse or human baited-traps, suggesting that the non-anthropophilic form of *Culex pipiens* L. dominates in this rural area. This species is a moderately efficient laboratory West Nile virus vector, but in southern France, it is considered to be a main vector in the dry areas and a secondary vector in the wetlands such as Camargue. Although few entomological data are available for Marseille, the anthropophilic *Cx. p. molestus* form is very likely to occur there, which could represent an explanation for the pertinent IgG increase observed during the warm season. Nevertheless, continuous exposure or an insufficiently long period of non- exposure to *Cx. pipiens* bites throughout the year could limit the IgG baseline feedback, potentially explaining the moderate seasonal variations of IgG that have been observed. Collectively, these data showed that IgG antibody responses against *AecSGE* and *CxpSGE* evolved differently according to site and season, suggesting a specificity of the serological response against *AecSGE*. Cross-reactivity was evaluated using correlation tests between the IgG levels against *AecSGE* and *CxpSGE* at the exposure peak. Differences were detected in the results according to the location. For Camargue and Marseille, the absence of a significant correlation between the IgG levels against *AecSGE* and *CxpSGE* corresponded to the low antigen cross-reactivity between these two species, which belong to different genera. Conversely, for Fos-sur-mer, a significant positive correlation was detected between these two species. Because both species are present at this site, the correlation could be due more to dual exposure to *Ae. caspius* and *Cx. pipiens* than to antigen cross-reactivity. Thus, our results support a genus-specific IgG response. It is of note that the IgG response heterogeneity (*i.e.*, between individuals from the same area) observed for *Cx. pipiens* and, to a lesser extent, for *Ae. caspius* might reflect heterogeneous exposure to mosquito bites due to individual behaviours (*e.g.*, outdoor/indoor activities, use of mosquito nets or repellents). Additionally, these two mosquito species exhibit distinct circadian biting activities (*e.g.*, clearly diurnal and nocturnal biting activities are observed for *Ae. caspius* and *Cx. pipiens* mosquitoes, respectively), which could further increase this inter-individual heterogeneity. Finally, intra-genus specificity was estimated using SGE from two *Aedes* species that were not endemic at the three sites during the time of the study. The very low IgG levels observed against *AealSGE* and *AeaeSGE*, independent of the site and timing, indicated that the IgG responses to *AecSGE* were specific at the species level. Nevertheless, the significant correlation coefficients obtained when the levels of IgG against mosquitoes from the *Aedes* genus were compared suggest cross-reactivity. These correlations were not attributed to dual exposure (because these mosquitoes were not present in the study area), but to the presence of shared salivary antigens between different *Aedes* species. Collectively, these data showed that the IgG antibody response against *AecSGE* may be related to seasonal and geographical variations in *Ae. caspius* density. The pertinent increase and transient IgG response at the peak of exposure appears to be species-specific, and these results strongly suggest that human antibody responses may be used to assess the individual level of exposure to mosquito bites. Nevertheless, other parameters should be considered, including historical individual exposure, which could influence the baseline IgG level. Further studies are needed to characterise specific *Aec*SG antigens, for instance, using an immunoproteomic approach, as described previously. This step of identifying the antigenic protein repertoire is necessary to determine the diversity and specificity of this repertoire. Salivary proteins of different arthropod species can share sequence similarities and cross-reacting antigens, resulting in the need to select species-specific antigens. Recombinant forms of selected salivary gland antigen candidates could be used for the development of a more sensitive and specific immunological test to accurately assess individual exposure to mosquito bites. Thus, specific immune responses against mosquito saliva antigens could be used in control and surveillance programs to assess the efficiency of anti-mosquito strategies, to estimate exposure levels and to identify new infestation areas. This strategy could be extended to other mosquito species that are involved in the transmission of infectious diseases and could represent a tool for estimating the risk of vector-borne disease transmission. Collectively, these data confirm that human antibody responses may be used to assess individual exposure to mosquito bites from particular species and estimate the level of these pests. # Materials and Methods ## Ethics statement All participants gave their written informed consent to participate in the study, and the Marseille-2 Ethical Committee approved the protocol (N°2006-A00581-50). Mosquito larval sampling was carried out in non-privately owned areas and non-protected areas outside the boundaries of the regional nature park of Camargue. The field study did not involve endangered or protected species. No specific permissions were required for the described field studies. ## Study sites The study was conducted in the Provence-Alpes-Côte d'Azur (PACA) area in southeastern France. Three study sites in PACA were chosen: (i) Camargue, a large wetland area of 150,000 hectares located inside the Rhone River delta, principally covered with pools of water, marshes and irrigated fields, where the human population is distributed between towns, hamlets and isolated houses, ; (ii) Fos-sur-mer, a town with 14,000 inhabitants (population density, 151 inhabitants/km²) with a mixed residential and agricultural landscape, located approximately 15 km from the border of the Camargue area; and (iii) the city of Marseille, a dry, an urban area with approximately 852 400 inhabitants and located approximately 30 km from Fos-sur-mer. ## Studied population Volunteers were recruited from Camargue (n = 41, 54% male, mean age ± SD: 45.7±11.3, Caucasian), Fos-sur-mer (n = 26, 42% male, mean age± SD: 51.5±11, Caucasian) and Marseille (n = 38, 47% male, mean age± SD: 40.3±12.2, Caucasian). For each individual, blood samples were collected by venous puncture at three different time points: February 2007, September 2007 and January 2008. Sera were obtained through centrifugation of the blood samples and were stored at −20°C. Eligible participants were individuals who did not travel to countries or areas that are endemic for *Ae. aegypti* and *Ae. albopictus* mosquitoes in the six months prior to and during the study. ## Mosquitoes and salivary gland extraction Adult female *Ae. caspius*, *Cx pipiens*, *Ae. aegytpi* and *Ae. albopictus* mosquitoes were used in this study. *Ae. caspius* and *Cx. pipiens* species were collected at the larval stage in the field in Camargue from August to September 2009, and the mosquitoes were reared in an insectarium. The *Ae. albopictus* mosquito colony came from the Alpes-Maritimes area and was bred in a laboratory at the Entente Interdépartementale pour la Démoustication (EID) Méditerranée (Cagnes-sur-Mer). *Ae. aegypti* mosquitoes came from the Bora-Bora reference colony, which was bred in a laboratory at the Institut de Recherche pour le Développement (Montpellier). All of these mosquitoes were maintained under identical standard conditions of 26°C and 60% humidity. All mosquitoes consumed no blood meals and were maintained on a diet of a 10% syrup solution. Salivary glands from 5- to 8-day old adult female mosquitoes were dissected on ice in phosphate-buffered saline (PBS) under a stereomicroscope. The salivary glands were pooled by species in a microcentrifuge tube and were then stored frozen at −20°C until needed. At that time, the salivary glands were disrupted by ultrasonication (Vibracell 72412, Bioblock Scientific, Illkirch, France) for 5 min on ice at maximum amplitude. Salivary gland homogenates were then centrifuged for 15 min at 16,100× g and the protein concentration of the supernatant was determined in duplicate by the Lowry method (DC Protein assay Kit, Bio-Rad) according to the manufacturer's instructions. Salivary gland proteins were then suspended in 0.1 M (pH 9.6) bicarbonate buffer to obtain a protein concentration of 1 µg/µL. ## ELISA The sera were tested by ELISA for the presence of IgG antibodies that bind to salivary gland proteins. To optimise the working conditions of the ELISA tests, a checkerboard titration was performed to establish salivary gland protein extracts and serum conditions. Based on the results of this procedure, Maxisorp Microtiter Immunoplates (Nunc, Denmark) were coated with 2 µg/ml (50 µl/well) of either *Ae. caspius*, *Cx. pipiens*, *Ae. albopictus* or *Ae. aegypti* salivary gland extracts diluted in 0.1 M bicarbonate buffer (pH 9.6) overnight at 4°C. Three washes were performed with 250 µL of PBS (pH 7.4, Sigma Co., USA) plus 0.05% Tween-20 (Sigma Co., USA) between each incubation. The plates were blocked for 2 h at 37°C with 200 µL of blocking solution buffer consisting of PBS, 0.05% Tween and 5% skimmed milk (Beckton, Dickinson Bioscience, USA). Serum diluted 1∶50 in blocking buffer was added (50 µl/well) to the plates, and they were incubated at 37°C for 1 h. Subsequently, 50 µl of horse radish peroxidase (HRP)-conjugated rabbit anti-human IgG (1∶10,000, Invitrogen, Rockville, USA) diluted in the blocking buffer were added, followed by incubation for 1 h at 37°C. Enzyme activity was detected by incubation with 50 µl of tetramethylbenzidine substrate (KPL, USA) for 10 min at room temperature. The reaction was stopped using 50 µl of 1 M H₂SO₄. The optical density (OD) at 450 nm was determined with a microplate reader (Versa Max® Turnable Multiplate Reader, Molecular Devices, UK). Each serum sample was tested in duplicate and in control wells without salivary gland extracts. To improve the consistency of the results, sera from different study sites were randomly arranged on each plate, and the samples collected at different time points for each individual were tested on the same plate. A pool of 5 sera collected in September 2007 from individuals living in Camargue (selected based on ELISA optimisation tests) was used as a positive control on all plates coated with salivary gland extracts from *Ae. caspius* and *Cx. Pipiens*. A pool of 5 sera from individuals living in inter-tropical areas, kindly provided by Dr. F. Remoué, was used as a positive control against salivary gland extracts from *Ae. aegypti* and *Ae. albopictus*. Only plates presenting inter-assay variations in absorbance values of positive controls lower than 20% were included in the analysis. The levels of IgG antibodies were expressed as the adjusted OD (aOD), which was calculated for each serum sample as the mean OD value for wells with salivary gland extracts minus the OD value of the control wells, i.e., without salivary gland extracts. Individual variations in IgG antibody responses were assessed according to the OD differences (ΔOD) between pairs of sera collected throughout the year. To consider pertinent ΔOD between pairs of sera, an arbitrary threshold of 0.1 ΔOD was defined. The mean aOD at the three time points for individuals not exposed to *Ae. caspius* living in Marseille plus 3 standard deviations was used as the cut-off value for seropositivity. ## Statistical analyses After verifying that the values in each group did not assume a Gaussian distribution, the Kruskal-Wallis, Mann-Whitney and Wilcoxon matched-pairs signed-rank tests and Spearman's rank correlation coefficient were computed when appropriate with STATA version 9.0 (Stata-Corp, USA). The frequencies were compared by the chi-squared test. All differences were considered significant at *p*\<0.05. However, for multiple tests, a Bonferroni correction was applied and *p*-value significance was then indicated. # Supporting Information The authors gratefully acknowledge D. Fontenille, M.N. Lacroix from IRD UR016 and G. Lacour from EID Méditerranée for access to mosquitoes bred in the insectary and D. Caire, N. Bonton and G. l'Ambert (EID Méditerranée) for access to wild mosquito larvae in the Camargue area. The authors are indebted to S. Bourdon for technical support and F. Gardella and T. Coffinet who conducted the entomological survey. The authors also acknowledge all individuals for their participation. [^1]: Conceived and designed the experiments: LA CR EO-P TF FP. Performed the experiments: AF AP. Analyzed the data: AF LA CR. Contributed reagents/materials/analysis tools: ID FR. Wrote the paper: AF LA. [^2]: The authors have declared that no competing interests exist.

# Introduction Earth is currently experiencing biodiversity losses of a magnitude described by many as constituting a sixth mass extinction. Conservation of threatened cryptic species often hinges on our ability to identify where they occur, so that critical populations can be monitored and managed appropriately. Field surveys involving the capture of individuals by live traps is a common method employed by wildlife researchers to establish the occurrence of small terrestrial mammal species and to monitor populations (e.g., their abundance, survival and recruitment) over time. These direct sampling methods provide immediate and generally unambiguous species identifications and enable additional information to be collected from individuals, such as genetic material and their sex, age, body mass and condition. However, recently, the proliferation of commercial wildlife camera traps has led to a sharp increase in the use of camera traps for small-medium sized mammal occurrence surveys and monitoring. Camera traps are generally defined as remotely triggered cameras that automatically take images and / or videos of passing animals. Remotely deployed cameras eliminate the need for researchers to trap and physically handle wild animals (thus avoiding often severe regulatory constraints) and can be deployed for long periods (up to several months), thereby reducing operational costs, time and effort. Although camera traps may be seen and / or heard by animals to some degree, they still provide an opportunity to detect and monitor rare and / or trap-shy species that may otherwise be missed or under-detected by direct census methods, as well as collect potentially valuable information about behaviour and activity. When deployed in the field, camera traps are generally mounted horizontally on trees, oriented passively outward or toward one or more bait holders. Such a setup has proved highly successful for detecting a wide variety of medium-large mammals, such as cats, ungulates and foxes. For detecting smaller mammal species, having the camera positioned closer to the subject is advantageous. In such cases, whether the best camera orientation is horizontal or vertical is an open question. De Bondi et al. demonstrated that small-medium sized mammals could be successfully detected by vertically orienting camera lenses toward a bait holder placed a standard height above ground. This alternative method has since been used to successfully detect the critically endangered central rock- rat (*Zyzomys pedunculatus*) and invasive black rat (*Rattus rattus*). However, the accurate identification of small mammals from camera traps is still challenging, especially where morphologically similar looking species co-exist. If camera trapping is going to continue to be used for small threatened species surveys and monitoring, then further ‘proof of concept’ evidence is required in challenging environments to ensure that individual species can be consistently detected and identified where they occur, and that efforts to reduce field time and cost do not compromise estimates of species occurrence and persistence. The black-tailed dusky antechinus, *Antechinus arktos*, is one of 15 species of *Antechinus*, a genus of small (16–170 g) carnivorous marsupials endemic to Australia. Antechinus are predominantly nocturnal insectivores renowned for their semelparous reproductive system, which features a short, promiscuous mating period, concluded by the abrupt death of all males in a population. Collectively, members of the genus occur in coastal / near coastal forest across all of Australia’s states and mainland territories. The geographic distributions of several species, including *A*. *arktos*, are severely limited. The species is known only from three sites located a maximum 8 km (straight line distance) apart within cloud forest at the summit of the Tweed Shield Volcano caldera (900–1200 m elevation), which straddles the border of Queensland (Qld) and New South Wales (NSW) in mid-eastern Australia. Therefore, *A*. *arktos* has been classified as Endangered in both states and is currently being considered for federal threatened species listing. Currently, the most important conservation priorities for *A*. *arktos* are to ensure the continued persistence of the three known populations and locate and protect previously unknown populations, should they exist. However, *Antechinus arktos* exemplifies the challenges associated with detecting and monitoring small, elusive / rare mammals. Recently, a two-year mark-recapture study of *A*. *arktos* was undertaken using Elliott (metal box) traps. This method proved effective at capturing *A*. *arktos*; however, trap success (number of captures divided by number of trap nights) was consistently low throughout its limited range, especially outside their short pre-breeding / breeding period, never exceeding 3.5% (or 7 individuals per 200 trap nights). Additionally, *A*. *arktos* co-occurs with much larger populations of the brown antechinus (*Antechinus stuartii)*, bush rat (*Rattus fuscipes*) and fawn-footed melomys (*Melomys cervinipes*), which may restrict access of *A*. *arktos* to live traps. Thus, to be 95% confident of detecting the species at other sites if they are present requires 600 trap nights conducted within this short enhanced detection-timing window. Deploying such a large number of traps at remote sites limits the number of sites that can be surveyed. The camera trapping approach described by De Bondi et al. may provide an alternative means of detecting *A*. *arktos* for occurrence surveys and monitoring programs. However, first camera trapping trials within known *A*. *arktos* habitat must be conducted. In contrast to many previous camera trapping studies, *A*. *arktos* co-occurs with a morphologically similar congener *A*. *stuartii* and other highly abundant small mammal populations, which may confound accurate identification. Therefore, using *A*. *arktos* as a model, we aimed to: 1) assess the utility of infrared digital camera traps for detecting and distinguishing *A*. *arktos* from other co-occurring small mammals; 2) identify factors that influence temporal variation in detection probability for each species; 3) examine diel activity patterns and their extent of overlap among species’ and 4) provide recommendations concerning the applicability of camera trapping as a survey method for *A*. *arktos* and other similar small mammal species. # Methods All aspects of the study were carried out with approval from the Queensland University of Technology ethics department (approval number: 1400000005) and the Queensland Parks and Wildlife service (approval number: WITK14454114). ## Study site Our study was conducted at Best of All Lookout (28.2415°S, 153.2640°E) within Springbrook National Park, \~100 km south of Brisbane, Queensland, Australia. Located at the rim of the Tweed Shield Volcano caldera, Springbrook National Park has major remnants of UNESCO World Heritage listed ‘Gondwana Rainforests of Australia’ that provides important refugia for many relict and endemic species of flora and fauna. Mean elevation of the site is 950 m and consists of complex notophyll vine forest (Regional Ecosystem 12.8.5) and simple microphyll fern forest (Regional Ecosystem 12.8.6). The study area was centred in a steep headwater gully containing the cloud-stripping stream lily, *Helmholtzia glaberrima*, and a small stand of Antarctic beech, *Lophozonia moorei*. ## Camera-trapping design Camera trapping was conducted between 10 August and 14 October 2016, which encompassed the breeding period from pre-mating to post-mating male die-off of the two local *Antechinus* spp. Eleven Ltl Acorn^® infrared digital cameras (Ltl-5310 series) were deployed at the study site and left in the same position for five consecutive deployments (ranging from 11–16 days in duration). Placement of the cameras utilized a pre-existing live-trapping grid consisting of 4 (200 m long and 20 m apart) parallel transects oriented down slope. Cameras were randomly assigned to stations set \~50 m apart along each transect. We recognise this is a small area to deploy camera traps; however, *A*. *arktos* has a very limited known distribution and may occur patchily even within sites it is known to occur. Setting camera traps within the live-trapping grid, where the species has previously been captured, allowed us to be confident that *A*. *arktos* was present, and thus detectable, rather than absent due to unsuitable habitat. Cameras were mounted on wooden stakes that were hammered into the ground, with the lens positioned 35 cm above the ground surface and directed downward towards a bait container. To limit the occurrence of false positive camera detections, vegetation and leaf litter were cleared from within the camera’s infrared sensor zone. Several layers of cream-coloured masking tape were placed over the cameras’ LED lights to reduce the intensity of the flash when the cameras were triggered and avoid over-illuminating or possibly startling the target species. A seven-day field trial conducted in July 2016 at the study site was used to optimize camera settings and quantify the operational periods of the cameras. Based on information collected during this field trial, we configured all cameras to record a single photograph (JPEG format, 5 megapixels) when triggered, immediately followed by a 20-s video (AVI format, 640 by 480 pixels per frame), followed by a 10-min interval, 24 hrs per day, as some antechinus species are known to be at least partially diurnal. We considered a 10-min minimum interval to the next possible recording event of the cameras to be an acceptable compromise between detecting the same individuals repeatedly (which reduces battery life, as well as available storage of recorded media) and the likelihood of failing to record individuals that only infrequently visited the cameras’ monitoring areas. To increase the probability of detecting target species (particularly *A*. *arktos*), we used a bait mixture of peanut butter, oats and bacon, which is commonly used in live trapping studies and has proven to be a superior bait type for antechinus, together with a single application (2-s spray) of FeralMone^™, a generic carnivore attractant (Animal Control Technologies, Somerton, VIC, Australia; Jesse Rowland pers.comm.). Bait containers comprised 60 X 75 mm PVC vent cowls that were secured to the ground with tent pegs. These devices permitted target species to see and smell the bait mixture but prevented them from directly accessing or removing it. Bait (including FeralMone), camera batteries and storage media (SD cards) were retrieved and replaced after each successive deployment. Animal recordings were identified to species level based on body size, body and head shape, tail length, ear morphology and behaviour (particularly movements captured on video). Reference photographs of *A*. *arktos* and *A*. *stuartii* taken during live trapping aided in identification. If identification was still uncertain after viewing both image and video files the animal was classified as an “unknown mammal”, “unknown antechinus” or “unknown murid” for the purposes of categorisation and the record was not used in subsequent analyses. Because successful and consistent identification between the two *Antechinus* spp. was an important element of our study, all antechinus observations were independently assessed by two researchers (ELG and AMB) and only individuals positively identified by both researchers were categorised to species level. Camera trap recordings were transcribed into separate binary response variables, one for each species per day of study, with a ‘1’ indicating species presence and a ‘0’ indicating species absence. Due to a high number of missing values (caused by the uneven deployment lengths ranging from 11–16 days) only the data from the first 11 days of each deployment were used for subsequent analyses, which was the minimum length of time that any of the cameras was set. We consider trimming deployment periods in the manner described above to be the best means of addressing missing observations, as plots of the capture proportions of each target species expressed as a function of the number of days of camera deployments clearly demonstrated that there were no marked changes in capture rates of any target species after day 11. Prior to statistical analysis we developed a list of explanatory variables that might reasonably influence detection probabilities, and conducted exploratory data analysis on both the response and explanatory variables to test for outlier observations and evaluate the extent of collinearity among independent variables, as suggested by Zuur et al.. ## Data analyses The data are repeated occurrence observations of the target species at fixed sites, with one level of spatial structure (Camera trap ID) and two levels of temporal structure (day of deployment nested within deployment). ‘Deployment’ can be considered to be both a study design and seasonal variable that reflects the breeding stage and survival of the two antechinus species over time. Camera deployments \#1 and \#2 occurred prior to antechinus breeding periods, deployment \#3 coincided with breeding, and camera deployments \#4 and \#5 took place post-breeding. ‘Day of deployment’ reflects the duration of bait effectiveness and possible neophobic responses of species to the cameras within each deployment. Rainfall and moon phase were also included as possible explanatory variables due to their reported influence on the activity of other small mammal species. We used generalized linear mixed models (GLMMs) to determine which variables best accounted for the detection probabilities of the target species. Because the response variable was binary, it was fitted to a binary distribution through a log-link function. Camera trap ID was included as a random effect nested within ‘deployment’ and ‘day of deployment’ to account for pseudoreplication (repeated measures) of observations over both time scales. Significance of fixed effects was assessed by computing Wald statistics. The ‘best’ (minimal adequate) model was determined by fitting the full statistical model and then excluding non-significant terms using a stepwise backward selection process recommended by Crawley. To examine the effects of individual explanatory variables on detection probability we plotted the fitted relationships from the minimal adequate model of each species. Following Rendall et al. we also used estimated daily detection probabilities from the minimal adequate model for each species to calculate the cumulative detection probability for each day since deployment using the formula: $$P = \ 1\ –\ \left( {1\ –p_{1}} \right)\ *\ \left( {1\ –p_{2}} \right)\ *\ \left( {1\ –p_{3}} \right)\ldots\left( {1\ –p_{n}} \right)$$ Where *P* is the cumulative nightly detection probability, *p*₁ is the detection probability for night one, and _n is the total number of survey nights per deployment. GLMMs were fitted using the R package lme4. ## Activity patterns and overlap Date and time records for each photo and video pair were used to investigate and compare the diel activity patterns of each target species. Because none of the target species had natural features or markings that allowed for individual identification within a species and because continued presence of species (after each 10-min interval) at a camera trap represented continued foraging activity, all capture records were included in the analyses as suggested by Carver et al.. In addition, multiple records of individuals (e.g., two *R*. *fuscipes* individuals observed simultaneously at the same camera trap) were treated as separate events. All analyses of activity patterns were performed using the Overlap package in R studio version 3.1.1 using code adapted from Meredith and Ridout. Probability density functions of activity for each species were estimated non-parametrically using kernel density estimates. Then, to compare the activity times of each pair of sympatric species, the degree of overlap between the two estimated densities were measured. Various measures of overlap have been proposed; however, we used the ‘coefficient of overlapping (Δ)’ recommended by Ridout and Linkie. This is a quantitative measure ranging from 0 (signifying no overlap) to 1 (signifying complete overlap). There are three alternative means of estimating the coefficient of overlapping, labelled Δ₁, Δ₄ and Δ₅. Here, however, we used only Δ₁ and Δ₄, which were recommended for sample sizes less than 50 and sample sizes greater than 75, respectively. Confidence intervals for coefficients of overlapping were obtained as percentile intervals from a recommended 10 000 bootstrap samples. # Results ## General findings In total, 8 273 JPEG and AVI video pairs were recorded over 725 camera trap nights. Of these pairs, 3 207 (38.8%) were deemed to be ‘false triggers’, with 49.9% caused by flies active during daylight hours. Additionally, 334 (4%) of observations could not be identified to species. This was overwhelmingly due to poor image quality, with less than six antechinus image and video pairs (across all cameras; 0.1%) unable to be identified to species level due to disagreement between researchers. The remaining 5 168 image and video pairs represented fauna from 10 different taxonomic groups, including our four target species: *A*. *arktos*, *A*. *stuartii*, *R*. *fuscipes* and *M*. *cervinipes*. Non-target animals included: northern brown bandicoot (*Isoodon macrourus*), possums (*Trichosurus spp*.), short-beaked echidna (*Tachyglossus aculeatus*), feral cat (*Felis catus*), macropod, and various rainforest birds (not identified). *R*. *fuscipes* was detected most frequently, constituting 56% of all mammal observations, with trap success (number of detections divided by number of camera trap nights) ranging from 77–98% between deployments. Next most numerous was *M*. *cervinipes* (15.2% of all observations; trap success ranging from 33–65%), *A*. *stuartii* (13.2% of all observations; trap success ranging from 15–59%), *I*. *macrourus* (3.9% of all observations; trap success ranging from 7–35%) and finally *A*. *arktos* (2.1% of all observations; trap success ranging from 3–21%). The two antechinus species could easily be distinguished from the Muridae species by their smaller body sizes and pronounced, pointed snouts. Muridae could also be easily identified to species level due to marked differences in species facial features and ear morphology (*M*. *cervinipes* has a shorter face, while *R*. *fuscipes* has conspicuously rounded ears;). Distinguishing between the two antechinus species was more challenging. Generally, the much larger body size and more rounded rump of *A*. *arktos* was sufficient for definitive identification. However, very large *A*. *stuartii* males are similar in body size to small *A*. *arktos* females. In such cases, video footage was often crucial in order to confidently assign antechinus to species level. The behaviour and movements of the two antechinus species were particularly diagnostic. *A*. *stuartii* typically exhibited rapid stop-start movements and regularly climbed the wooden stakes supporting the cameras; in contrast, *A*. *arktos* tended to move in a much slower, ‘shuffling’ gait, and always remained on the ground. ## Temporal changes in detection probabilities GLMMs of detection probability were fitted for each of the four target species. The minimal adequate detection model for *A*. *arktos*, *A*. *stuartii* and *M*. *cervinipes* included the fixed effects of ‘deployment’ and ‘days since deployment’. The most parsimonious detection model for *R*. *fuscipes*, however, was the null, with detection probability consistently high (close to 1) irrespective of effects from any of the explanatory variables. *A*. *arktos* and *A*. *stuartii* both had significant quartic relationships with deployment number; detection probabilities were lower during camera deployments \#2, \#4 and \#5 compared with deployments \#1 and \#3. In comparison, *M*. *cervinipes* had a strong negative linear relationship with deployment number. Within individual camera deployments, there was a strong negative linear relationship between detection probabilities and ‘days since deployment’ for *A*. *arktos*, *A*. *stuartii* and *M*. *cervinipes*; the slopes of these relationships differed markedly between species. *A*. *arktos* decreased from an average detection probability of 0.2 on the first day of each deployment to 0.06 on day 11, *A*. *stuartii* from 0.54 to 0.25 and *M*. *cervinipes* from 0.82 to 0.21. Cumulative detection probability curves show that detection probability reached 95% after just one night for *R*. *fuscipes*, two nights for *M*. *cervinipes* and five nights for *A*. *stuartii*. However, a cumulative detection probability of 95% was never achieved for *A*. *arktos*. After 11 nights the cumulative detection probability of *A*. *arktos* was just 76%. ## Activity patterns and overlap Periods of activity were strongly coincident for all four target species, as indicated by values of overlap coefficients being \>0.7. *A*. *arktos* was predominantly nocturnal (91% of observations were recorded during darkness), with a primary peak in activity between 18:00 and 19:00 hours, followed by smaller spikes of activity from 21:00 to 22:00 and at 03:00 hours, respectively. Similarly, *A*. *stuartii* had a primary peak of activity between 18:00 and 19:00 hours, with smaller spikes from 21:00 to 22:00 and at 02:00 hours. However, *A*. *stuartii* displayed stronger diurnal activity, with 115 detection events (20% of all captures) recorded during daylight hours. The majority of this daytime activity (86 captures) occurred during the species’ breeding period within deployment \#3 from late August to mid-September. In comparison, *R*. *fuscipes* and *M*. *cervinipes* displayed a unimodal, peak in activity between 18:00 and 19:00 hours and were strongly nocturnal, with 97% and 96% of captures recorded during darkness, respectively. # Discussion ## Effectiveness of the camera trapping design Our study confirms that remote infrared digital camera traps can be successfully used to detect and differentiate small, closely related, morphologically similar mammal species, including the endangered black-tailed dusky antechinus. Previous studies have highlighted the difficulties in distinguishing between small co- occurring mammalian species (including Muridae and antechinus) from images recorded by horizontally oriented camera traps. However, using a modified vertical mounting design established by De Bondi et al. and extended here we conclude we could confidently identify most detections of small mammals to species. The vertical cameras captured dorsal perspectives of animals against a cleared ground surface, allowing critically diagnostic features such as body and head shape, tail length and shape, and ear morphology to be easily distinguished. The standardized dimensions of the bait holders and fixed distance of the cameras to the ground also provided a means to accurately estimate body size. Additionally, the 20-s video recordings corresponding to each still image allowed multiple angles of each individual to be viewed and its behaviour closely observed. We found that video recordings of behaviour (especially close-ups) were particularly useful and often allowed us to discriminate antechinus species with confidence. Although our findings are specific to the taxa and study area, the method could be applied to other small mammals such as civets, martens, shrews and other rodents. However, despite the general success of this approach, we caution that prior live trapping at the site and / or a certain degree of familiarity with the target species was essential when attempting to taxonomically classify recorded individuals, even from paired monochrome images and video clips. We recommend that live trapping and field-based observations be used in concert with camera trapping, especially at new sites that have not been previously assessed. The present study also used two independent experts to examine all antechinus images and video, to improve accuracy of identification and ensure that the reviewer of the camera footage did not get complacent with identifications. Disagreement between independent experts resulted in the removal of just five possible antechinus detections from 665 paired images and video footage. The inclusion of a third independent reviewer may have allowed for a majority vote in these cases where the two experts could not agree. We advise that future studies employ as many independent reviewers as possible. Our infrared cameras also recorded a large number of false triggers compared with other studies. This was likely due to the high infrared motion sensitivity of our particular camera model even at its lowest sensitivity. Most false triggers were of large flies moving near the bait stations during daylight hours. This issue can easily be avoided by configuring cameras to operate only during the night. However, in our case this would have meant foregoing collection of important data on the diurnal activity of *A*. *stuartii*. The wide detection zone of our particular camera model (Ltl Acorn^®) may have also resulted in a number of false positive images, caused by animals triggering a camera trap outside of the cameras field of view. In some instances, the trigger speed (time taken to take a photograph after it has detected heat/motion) of the cameras may also have been too slow for rapidly moving animals, resulting in out-of-focus or only partial images, which impeded classification to species. Use of ‘white-flash’ cameras may circumvent these problems. Unlike infrared illumination, white flash provides the ability to take colour images at night and thus enable unique pelage attributes of *A*. *arktos* (i.e., fuscous black hindfeet and tail, orange eye ring and cheek patch) to be used as an additional diagnostic feature in separating them from the more uniformly brown *A*. *stuartii*. ## Factors influencing detection probability Changes to detection probability with time invariably differs from one species to another, depending on their level of attraction to baited camera traps and other key biological factors such as population density and activity. Therefore, unsurprisingly *R*. *fuscipes*, which occurs at high population density at our and other sites where it is found and has a strong attraction to peanut butter and oat bait, had a near constant detection rate between and within camera deployments. In contrast, results of our GLMMs show that both deployment number and days since deployment are important factors influencing the detection probabilities of *A*. *arktos*, *A*. *stuartii* and *M*. *cervinipes*. Both *A*. *arktos* and *A*. *stuartii* exhibited marked differences in detection probabilities between camera deployments: probabilities were higher during camera deployments \#1 and \#3 (pre- and during breeding) than deployments \#4 and \#5 (post-breeding). This finding is consistent with previous live-trapping studies that found *Antechinus* spp. were significantly more trappable just prior to and during their breeding periods (owing to increased activity and movement during this time) and lowest post-breeding following the period of male die-off, when only pregnant females remain in the population. However, such low detection rates of both *Antechinus* spp. in our study during camera deployment \#2 (when males were still alive), was unexpected. This period during the study experienced the highest total rainfall of any camera deployment (76 mm), which may have reduced the extent of activity in the antechinus. Moreover, periods of rainfall occurring early in the camera deployment may have degraded the olfactory attractiveness of our baits, similar to the reduced toxicity level of ‘1080’ bait after rainfall. In contrast, we found that *M*. *cervinipes* exhibited a strong negative linear relationship of detection probability with respect to camera deployment number, suggesting either: 1) natural declines in either activity periods or population size occur between deployments; and / or 2) a persistent learned decrease of attraction to the bait, because the baits were inaccessible and thus offered no food reward, leading *M*. *cervinipes* to forage elsewhere for better opportunities. Optimal GLMMs for *A*. *arktos*, *A*. *stuartii* and *M*. *cervinipes* also identified a significant negative linear relationship between detection probability and days since camera deployment. Distinct peaks in detection rates on the first night of each camera deployment, when baits were fresh and most effective, suggest that the attractiveness of baits declines markedly over time. Remote cameras have a distinct advantage over alternate survey methods such as live trapping because of the former’s longer operational periods without the need of human intervention. Ideally, with remote cameras such as those used in the present study, several weeks or even months of data can be collected with only two visits to a site: one to deploy the cameras and the second to retrieve or reset them. We noted that the initial high peak of interest in the camera trap baits (and lure combination) only lasted a single evening and was followed by progressively lower detection rates until the bait was replaced. To remain optimally effective, in our study baits would have needed to be refreshed every two-to-three days. For species with high initial detection rates over this time such as *R*. *fuscipes* (0.96–0.98), *M*. *cervinipes* (0.72–0.82) and *A*. *stuartii* (0.48–0.54), we found that a single camera deployment would be sufficient to achieve 95% cumulative detection probability if the species was present. Conversely, for rarer species such as *A*. *arktos* (detection rate 0.16–0.20), at least 14 survey days (five visits or successive three-day baiting deployments) would be necessary to achieve 95% cumulative detection probability if present. This extra effort would markedly add to the cost of an occurrence survey or long-term monitoring project for *A*. *arktos* and similarly rare species. This latter finding has general significance for other camera trapping studies and will be further discussed below. ## Diel activity and overlap Information about the date and time that our cameras’ images and videos were recorded allowed us to examine the activity patterns of the four target species. Both *R*. *fuscipes* and *M*. *cervinipes* were found to be completely nocturnal, displaying a unimodal peak in periods of activity, consistent with previous studies of these species. *A*. *arktos* and *A*. *stuartii* also exhibited primary peaks in activity, although smaller activity peaks were evident later in the night. Ours is the first study of activity patterns in *A*. *arktos*, a species which was discovered only very recently. We found *A*. *arktos* to be primarily nocturnal, an activity pattern consistent with most of its congeners but notably at odds with some other members of the Dusky antechinus species complex. In contrast, *A*. *stuartii* showed greater diurnal activity during its breeding season (74% of all daytime observations occurred during this period). Such a shift in the timing of activity periods has also been reported for island populations of *A*. *minimus* and may be necessary to maximize reproductive success when competition among males for mating opportunities within populations is strong. In general, we noted strong overlap of the activity periods in all four target species, with each displaying a peak in activity during the same two-hour period following sunset. In south-east Queensland, *R*. *fuscipes*, *M*. *cervinipes* and several *Antechinus* spp. are often sympatric and a certain level of interspecific competition is believed to occur. The patterns of activity observed for the target species in our study suggest that diel temporal partitioning is not a mechanism used to promote coexistence. Rather, differences in microhabitat use and diet are likely to be the principal factors limiting competition among these species. However, although we found no evidence of temporal partitioning in our study, a degree of avoidance at proximity may still occur. On several occasions (six and two times, respectively) video footage clearly showed *A*. *stuartii* fleeing from *R*. *fuscipes* and *A*. *arktos* to avoid confrontation. Dickman suggested that evasive action by a subordinate before an encounter with a dominant species could be due to early detection, either by sound or smell. Such a behavioural strategy could occur between the larger *R*. *fuscipes* and *A*. *arktos* and the smaller *A*. *stuartii*. ## Applicability of camera traps as a survey method for rare or elusive species Our study demonstrates that infrared digital camera traps can be used to detect and identify *A*. *arktos* and other small, morphologically similar mammals at a rate comparable to live-trapping. Although our study was not specifically designed to compare survey methods, three nights of live trapping were conducted in August 2016 (four days prior to the first camera trap deployment) and so some general comments are warranted. Live trapping (600 trap nights) captured 41 *A*. *stuartii*, 7 *A*. *arktos*, 77 *R*. *fuscipes* and 58 *M*. *cervinipes* (including recaptures). In comparison, the first three nights of camera trapping (deployment \#1, 33 trap nights) recorded 68 *A*. *stuartii*, 15 *A*. *arktos*, 228 *R*. *fuscipes* and 117 *M*. *cervinipes*. Because live traps are limited to one capture per night, but cameras can record multiple individuals (and recaptures) per night, we expected that camera traps would record more species at more trap stations than live traps. Future studies will aim to formally test the relative efficacy of live versus camera trapping for this species. Nevertheless, such a high number of camera recordings from multiple stations highlights the potential use of this technique for other small mammal surveys and monitoring in the future. It is particularly relevant for areas inaccessible to large numbers of live traps or of high conservation value, where minimal disturbance to the target species is preferable. However, to achieve consistently high detectability of *A*. *arktos*, our work suggests baits need to be refreshed every two-to-three days, making the remote cameras more maintenance demanding than might be assumed. This limitation may not only apply to rare small mammals. Strongly declining rates of detection were also recorded in our study for *A*. *stuartii* and *M*. *cervinipes* and have been previously documented for other small mammals. Examining fifteen camera trapping studies conducted on small-medium mammals over the last 10 years, we found that 80% deployed cameras longer than 1 week and 67% deployed cameras for at least 2 weeks before rebaiting / collection. This is an important finding for camera trapping generally, indicating that such routine camera deployment lengths may not be optimal when targeting some small mammals. For management of threatened mammals, the problem of declining bait effectiveness is more acute because on any given deployment, probability of detection is already low. It is uncertain if other attractants that have been used in camera trap studies, such as linseed oil, truffle oil or vanilla essence may enhance the time length of optimal attractiveness. In any case, we recommend including the time since camera deployment as a covariate in future surveys of small mammals, especially when cameras are operational for extended periods. In cases of declining detection for rare fauna, a standard three night live trapping survey (where species can be identified at point of capture and other ecological information obtained) may be a more practical detection method than a camera trapping survey, which may require multiple, successive three-day baiting deployments. # Supporting information We would like to thank Aila Keto (Australian Rainforest Conservation Society) and the rangers at Springbrook National Park (QPWS) who provided logistical support and accommodation. We would also like to thank Todd Landers for his generous help during initial field trials. Finally, we thank the residents of Springbrook for their interest and all the volunteers who ventured out with us to help collect this data. [^1]: The authors have declared that no competing interests exist.

# Introduction Humans have transformed most of the earth's terrestrial biosphere into highly modified biomes, resulting in the loss and decline of many species. To counter this, much focus has been on the establishment of formal conservation reserves. While these are a critical component of biodiversity conservation worldwide, reserve networks rarely provide comprehensive, adequate and representative coverage of ecosystems. It is often the case that those natural ecosystems associated with agricultural land are the most poorly represented in conservation reserve networks. Australian box gum woodlands are a classic example of this. Despite comparatively low levels of reservation in agricultural landscapes, many species associated with such ecosystems still persist there. Such species have been shown to use the relictual and semi-natural habitats of these landscapes. These habitats or countryside elements (‘elements’) often have very different and distinct characteristics, each offering a different range of resources for wildlife. The role that countryside elements and matrix habitats play in biodiversity conservation is increasingly appreciated. This has been demonstrated by work conducted in Australia, Central and South America , and West Africa. However, to date there have been few works comparing different kinds of ‘elements’ and quantifying their relative usage for a single species. In this study, we examined the relative value of four kinds of wooded countryside elements in the conservation of a threatened arboreal marsupial, the squirrel glider *Petaurus norfolcensis*, in agricultural landscapes of south-eastern Australia. These ‘elements’ were: *scattered trees*, *tree plantings*, *linear roadside remnants* and small *native vegetation patches*. These four broad kinds of ‘elements’ can be found in many agricultural landscapes around the world. While their use as wildlife habitat has been reported, there are few unequivocal examples of their relative value in the conservation of different species or taxa,. To achieve this we address the following questions: 1. Is the use of an ‘element’ for denning and feeding proportional to their availability? If not, is there evidence of preferential usage of some kinds of ‘elements’? 2. Does home range size depend on the proportional availability of an ‘element’? Differences in home range size within species have often been attributed to habitat quality, with smaller home range indicating better quality habitat. If an ‘element’ offers a significantly higher or lower quality habitat relative to others, it would be expected that their availability would have a significant influence on home range size. Our work is the first to attempt to quantify the relative contribution of different countryside elements for the squirrel glider and is one of the few that attempt to compare the ecological value of countryside elements, using empirical data. An understanding of the value of such ‘elements’ for biodiversity is essential for better integration of biodiversity conservation and broader management of agricultural landscapes. Without such data, managers are in danger of undervaluing certain ‘elements’. This may, in turn, lead to the loss of critical habitats and further threaten associated species in these landscapes. # Methods ## The squirrel glider The squirrel glider is a nocturnal, arboreal, gliding marsupial in the Family Petauridae. It is a medium-sized possum weighing between 190–300 g and which feeds on invertebrates, insect exudates, sap of trees and shrubs, and pollen and nectar. The species is listed as threatened in three of the four Australian states in which it is found (Victoria – *Flora and Fauna Guarantee Act 1988*; South Australia – *National Parks and Wildlife Act 1972*; New South Wales – *Threatened Species Conservation Act 1995*; and Queensland – no formal listing). ## Study area Our investigation encompassed five study areas within the south-west slopes of New South Wales, Australia. The region is the most extensively and intensively disturbed of the 13 botanical regions of NSW, with an estimated 85% of the original cover of native vegetation removed in the past 200 years. The five study areas were located in heavily modified agricultural landscapes, used predominantly for livestock grazing and dryland cropping. Study areas were approximately 3 km×3 km. Woody vegetation occurred primarily as relictual scattered paddock trees, native vegetation plantings and remnant temperate *Eucalyptus* woodlands on private lands, road reserves and travelling stock reserves. ## Ethics statement We conducted trapping and radio tracking under The Australian National University Animal Ethics Committee protocol number C.RE.39.05. Squirrel Gliders are a native species and therefore protected. Relevant permits to handle the animals were obtained from New South Wales Government agencies. Animals were captured in wire mesh cage traps covered with non-transparent, heavy-duty plastic sleeves to minimise stress on animals and protect from cold and wet weather. Qualified wildlife veterinarians anaesthetised captured animals using isofluorane gas delivered via a portable gas anaesthesia machine. Isofluorane anaesthesia ensures recovery of animals within minutes, which is considerably faster than injectable agents, thus minimising holding time and stress. Anaesthetising animals ensured that accurate measurements of body size and reproductive status could be made without undue stress to the animals. It also enabled the veterinarians to properly fit radio-collars and microchip each animal. Gliders not fitted with collars were ear tagged (hamster ear tags, Sieper & Co., Sydney, Australia) in each ear. When animals had recovered from the anaesthetic, they were released at the exact point of capture. Land accessed was a mix of privately owned farmland, local government managed road reserves, and travelling stock reserves managed by the Livestock Health and Pest Authority and relevant access permissions were obtained from the respective land managers. ## Radio-tracking We captured gliders using drop-door, wire mesh cage traps (170 mm×200 mm×500 mm) over a three night period at each site in March 2005. We fitted 32 gliders with a single stage brass loop radio transmitter, weighing 4.5 grams (Sirtrack, New Zealand). When selecting which gliders were to be collared, we preferred adult gliders and attempted to achieve an equal sex ratio and an equal spatial coverage of animals within and between sites. We tracked gliders to their diurnal denning site at least twice a week and to a nocturnal location at least 1–3 times every 14 days, over a 4–5 month period. For each fix, we recorded the countryside element in which the glider was located. ## Home range We derived home range estimates by using both diurnal and nocturnal fixes. The woody vegetation was scattered or represented in irregularly shaped patches, surrounded by cleared agricultural matrix. Common parametric approaches such as the minimum convex polygon method were therefore unsuitable. This is because they would have (inappropriately) included large areas of unused cleared agricultural land – as has been found by Martin et al. and van der Ree and Bennett. We estimated home range by using the non-parametric, grid cell method. The size of the grid cell is arbitrary and can have a major influence on either underestimating or over-estimating the den range size. Three sized cells were tested 40×40 m, 50×50 m and 60×60 m and compared to the minimum convex polygon (MCP) method (where MCP could be appropriately used, i.e. for some animals in areas dominated by remnant vegetation patches or large clusters of scattered trees). We calculated the minimum convex polygon using Home Range extension within Arc View GIS (ESRI, California, USA). We selected a 50×50 m grid cell as it aligned more closely with the commonly used MCP method. We connected disjointed cells by including cells that were crossed by the most direct line joining to consecutive locations, taking into consideration gap-crossing ability (\[i.e. gaps in canopy \<70 m; see). We estimated the home range size of each squirrel glider on 95% of fixes. This was done to give an objective, repeatable method of comparison of normal home range. We deleted 5% of fixes at the extremities of each animal's range to reduce the influence of exploratory movements or outlying fixes outside the ‘normal’ home range. ## Countryside elements Within our study area, we recognised four categories of countryside elements that contained woody vegetation. ### (1) Linear roadside remnants These were linear strips of remnant vegetation along roads. Remnant roadside vegetation is a major feature across agricultural landscapes providing a network of remnant vegetation corridors across what are otherwise generally heavily cleared landscapes. The width of the roadside reserves in this study ranged from 40–60 m. Vegetation along road reserves is subject to high levels of disturbance such as road construction and maintenance. However, grazing pressure by domestic livestock is often low and irregular. These areas regularly contain regeneration of overstorey species. Native understorey species are generally present but their dominance and diversity varies. ### (2) Native vegetation patches These were patches of remnant vegetation where the understorey was dominated by a diverse array of native plants. In this study, these areas were mostly on travelling stock reserves, and some patches of remnant native vegetation on freehold land. The travelling stock route network was established more than 150 years ago to facilitate the movement of domestic livestock between properties and to markets. The network is made up of travelling stock routes (which today are often incorporated into the road reserves) and holding paddocks which are generally referred to as travelling stock reserves (TSRs). Because of their reservation for these purposes, these areas generally escaped clearing and continuous high-intensity livestock grazing. The *native vegetation patches* varied in size with the largest patches occurring on TSRs (≈100 ha) and the smallest patches on freehold land (\>5 ha). ### (3) Scattered trees These were scattered, (mostly large, old) relictual trees remaining on land used for grazing or cropping. They include dead and living trees, are often widely spaced (more widely spaced than expected in their natural state) and contain a simple understorey, generally dominated by introduced grasses and forbs, with a low diversity of native plant species. ### (4) Tree plantings These were Australian native vegetation plantings, generally containing dense stands of trees (predominately *Eucalyptus*) and shrubs (such as *Acacia* and *Melaleuca*). The species composition was predominantly locally indigenous species, but often included species naturally found outside the region. *Tree plantings* vary in their shape and size as they were planted for various purposes such as shelterbelts or to reduce rising water tables. The level of grazing by livestock within them also varied. The vegetation was assigned to the countryside element of best fit, using the above descriptions as a guide. Where elements were adjacent to each other, boundaries were defined by differences in vegetation structure and composition, and/or management practices. We calculated the area of each countryside element available to an individual squirrel glider by measuring the total area of woody vegetation attributed to that ‘element’, within a 1000 m radius of the centre point of all fixes for each individual glider. We used a 1000 m radius, as 2000 m is approximately the maximum home range length that has been reported for our study species. We measured the area of woody vegetation using geographical information systems software (ArcGIS 9.2-esri) to draw polygons over the canopy of woody vegetation interpreted from satellite imagery (spot 5-Astrium). We deemed that woody vegetation isolated by a gap distance of greater than 70 metres was unavailable to gliders. ## Data analyses For each squirrel glider, our data consisted of a count of the number of fixes in each ‘element’, classified by circadian time - day or night. Associated data were the total area of woody vegetation in each ‘element’, available within a 1000 m radius of the centre point of all fixes for each individual glider. From these data, we can obtain the expected number of fixes by apportioning total fixes according to the relative area of each ‘element’. Thus, for day and night data separately, we have a 2-way contingency table cross-classified by animal ID and ‘element’ type, where the cells are the observed count of the number of fixes and a concomitant variable is the expected number of fixes based on the relative availability of each of the ‘elements’. ## The Model Considered as a two-way contingency table, our data can be modelled bywhere *y_ij* are the expected frequencies, *u* is the grand mean, *f_ij* are the ‘expected’ frequencies derived from the percentage are occupied by *‘element’_j animal_i* and *‘element’_j* are constants to account for the marginal distribution of the two-way animal by ‘element’ table. The above model is a particular case of a class of models commonly known as log- linear models, and which are often used in the analyses of multiway contingency tables. The log-linear model belongs to the class of generalised linear models. These models can be fitted by the use of maximum likelihood methods available in many statistical software packages, such as GENSTAT Version 15. The goodness-of- fit of the model and the significance of individual terms can be assessed by examining an analysis of deviance. We completed further linear regression analysis to quantify relationships between home range size and total woody vegetation and each of the countryside elements. We avoided statistical issues associated with intrinsic collinearity of countryside elements by considering each of the ‘elements’ separately. # Results ## Radio-tracking We captured 52 individual gliders and fitted 32 with radio-transmitting collars. The numbers of fixes for individuals varied as signals were lost for some animals throughout the study. Over a five month period, we tracked gliders to 1027 independent locations (655 diurnal and 372 nocturnal locations). We tracked individual gliders to an average of 21±1.16 (mean ± s.e.) diurnal locations and to 12±0.6 (mean ± s.e.) nocturnal locations. ## Usage of countryside elements We found that gliders used all four wooded countryside elements nocturnally and all but one element (*tree plantings*) for diurnal denning. Some individual gliders exclusively used one category of ‘element’, *scattered trees*, *linear roadside remnants* or *native vegetation*. The fit of the model as defined above is summarised by the analysis of deviance. As can be seen from, 64% (367/576) and 71% of the residual deviance was accounted for by the ‘expected’ count based on the percentage of available ‘elements’ for each animal, for night and day use respectively. The final residual deviance, based on 92 degrees of freedom after fitting the terms for animal, ‘element’ and expected frequency was 209 and 286, for night and day, respectively. This suggested that there remains non-random unexplained variation. A further breakdown of the components of this residual variation revealed evidence of preferential selection of *native vegetation patches* and *scattered trees* nocturnally, and a strong preference to *scattered tree*s for diurnal use, as is shown in. ## Relationship between home range size and availability of CSEs Home ranges varied from 2.5 to 12 ha and averaged 4.9±0.45 ha (mean ±1 s.e.). Home range size increased with addition fixes, but plateaued towards the end of our study. We found no significant relationship between total woody vegetation cover and home range size. However, there were significant relationships between home range size and area of woody vegetation within three ‘elements’. We found a significant (p\<0.001) negative relationship between home range size and the available area of *scattered trees* and significant (p = 0.009 and p = 0.022) positive relationships with the available area of *native vegetation patches* and *linear roadside remnants*. # Discussion The contribution that countryside elements make towards biodiversity conservation is increasingly recognised in many ecosystems around the world. The challenge for conservationists is to recognise different ‘elements’ and to understand the role each play in the conservation of different species. Our work is one of the few studies that quantify the relative value of different countryside elements for a single species. We provide empirical evidence of the disproportionate value of *scattered trees* in agricultural landscapes. Our work also highlights the need to examine the roles and values of countryside elements in biodiversity conservation, particular for species such as the squirrel glider which has a distribution largely confined to highly modified agricultural landscapes. Across our five study sites, gliders relied entirely on countryside elements (*scattered trees*, *linear remnant vegetation*, *native vegetation patches* and *tree plantings*) located within road reserves, traveling stock reserves and freehold land. Our study demonstrates that the squirrel glider will use all of these four wooded ‘elements’, with availability being a key factor in determining usage. In agricultural landscapes, roadside vegetation provides important habitat for many species of wildlife, particularly in facilitating migration and dispersal,. We found that the squirrel glider would commonly use *linear roadside remnants*, with some individuals relying entirely on this ‘element’. The ability of *linear roadside remnants* to support squirrel glider populations has been previously reported. While it is clear this kind of countryside element is of conservation importance to this species, we found evidence that suggests *linear roadside remnants* may provide inferior quality habitat compared *to native vegetation patches* and *scattered trees*. Larger home range sizes associated with increased area *of linear roadside remnants* and the under-utilisation of this ‘element’ is evidence of this. This may be explained by structural differences in the vegetation between the ‘elements’, such as a *linear roadside remnants* containing a higher proportion of small regrowth trees (which have been shown to offer poorer quality habitat than large trees). Issues associated with the geometry of linear habitats also may explain our findings. Lindenmayer et al. and Recher et al. highlight the problems for species inhabiting linear habitats, such as disrupted social behaviour and additional expenditure of energy in obtaining food. While *native vegetation patches* in agricultural landscapes are often small and highly fragmented, they have been shown to be important habitat for many species. In our study, the majority (over 90%) of *native vegetation patches* occurred on traveling stock reserves. These reserves themselves have been shown to have high conservation value (higher than remnant vegetation on private land, particularly for arboreal marsupials). We found that *native vegetation patches* were heavily used by gliders, with a number of individuals using this ‘element’ exclusively. The use of *native vegetation patches* was higher than predicted for both diurnal and nocturnal activity. While *native vegetation patches* were preferentially used, gliders in areas dominated by *native vegetation patches* had larger home ranges. This result may suggest that while quality habitat exists within *native vegetation patches*, it is often dispersed among poorer habitats, resulting in gliders having to cover larger areas to gather resources. *Tree plantings* have been shown to play an important role in the conservation of some species in agricultural landscapes, particularly birds. For arboreal marsupial conservation, *tree plantings* may have a more limited role, at least in the short to medium term. Our data indicated that *tree plantings* were used less than predicted based on availability. When compared to other treed ‘elements’, plantings would seem to be of minor importance to the species. However, *tree plantings* were nevertheless still used and should not be discounted as their value may increase over time as trees mature, given squirrel gliders’ preference for large trees. *Tree plantings* are the only countryside element examined here that can be readily introduced into these highly modified landscape. Our data suggests that for *tree plantings* to be used, they must be located in association with other countryside elements containing remnant trees. *Scattered trees* are widely recognised as a critically important countryside element in many agricultural landscapes globally and are considered a keystone structure in such landscapes. Numerous studies have highlighted the ecological value of *scattered trees* for various taxa of wildlife. Fischer et al. have shown that *scattered trees* have a disproportionate value for birds and bats. Our work provides further evidence of this. The home ranges of gliders were significantly smaller with increasing coverage of *scattered trees*, an opposite pattern to all other ‘elements’. We also found that *scattered trees* were used at higher rates than predicted, particularly as diurnal den sites. Both results indicate that scattered trees have a comparatively higher habitat value for the squirrel glider than the other three kinds of countryside elements we examined. *Scattered trees* are often the oldest living structures in disturbed landscapes. This was the case in our study, with *scattered trees* represented by predominately large mature relictual *Eucalyptus* trees. These trees generally had well developed hollows and a large canopy, both kinds of resources have been shown to be preferentially sought by the squirrel glider. Despite this, *scattered trees* are arguably under the greatest threat of all countryside elements and populations of such trees are rapidly diminishing resource in many agricultural landscapes. It is estimated that tens of millions of *scattered trees* will be lost in grazed landscapes in Australia over the next 50 years, due to factors such as natural attrition, clearing, and a lack of regeneration – a problem common to many agricultural landscapes globally. Current conservation strategies offer little protection for *scattered trees*, and they are cleared for cropping and irrigation. Without a concerted shift in conservation policy and awareness in general of the value of *scattered trees*, there is a real risk of losing this critical countryside element. # Conclusions Agricultural landscapes are dynamic. Changing technologies, land uses and attitudes continually transform these systems. There are significant threats to the integrity and extent of various kinds of countryside elements within agricultural landscapes. Therefore, understanding the contribution they make to biodiversity conservation is essential. In the case of the squirrel glider, we found that four kinds of countryside elements are important, but their relative values vary. We suggest that these countryside elements need to be better integrated into landscape management. We thank the Hume Livestock Health and Pest Authority (formally Hume and Wagga Wagga Rural Lands Protection Boards), Wagga Wagga City and Greater Hume Shire (formerly Culcairn Shire Council) Councils, and private landholders for allowing us access to their properties, Cody Keys for assistance with trapping and radio tracking, and Damian Michael, Chris MacGregor, Ben MacDonald, Nicki Munro, and Jake Gillen for assistance with trapping. We thank our two wildlife veterinarians Karen Viggers and Arianne Lowe. We also thank Paul Adams (Academic Editor) and the two anonymous reviewers for their thoughtful contributions. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: MJC DBL RBC. Performed the experiments: MJC. Analyzed the data: MJC DBL RBC. Contributed to the writing of the manuscript: MJC DBL RBC.

# Introduction The Portuguese oak (*Quercus faginea* Lam., identified by A. Fabião) is a species from *Fagaceae* family, native to the Western Iberian Peninsula and the North African countries of Morocco, Tunisia, and Algeria. It is a medium-sized deciduous or semi-evergreen tree growing to a height of 20 m and a diameter of 80 cm. This species occupied once (XV-XVI centuries) much of the Portuguese territory and the wood was valued and intensively exploited for naval construction. However, the large stands of *Q*. *faginea* trees are now reduced to a few scattered stands, especially in mixed stands with other oaks. This decline occurred mostly in the last three decades of the twentieth century as a consequence of land use change from forest to agriculture and to a preferential reforestation with other species e.g. *Q*. *suber*, *Q*. *rotundifolia*, *Eucalyptus globulus* and *Pinus pinaster*. Today the oak forests in Portugal are mostly of *Q*. *suber* (23% of the total forest area), and *Q*. *rotundifolia* (11%) while the other oak stands occupy about 2% of the forest area with *Q*. *faginea* restricted to only a few scattered stands. The wood of *Q*. *faginea* is not presently used in a significant extent. However, the wood potential and the environmental and cultural importance of the species are acknowledged. General descriptions of *Q*. *faginea* wood refer good aesthetic appearance, white yellowish sapwood and brown yellowish heartwood, high density, and considerable mechanical strength. Recent research efforts have been made to increase knowledge on growth and wood characteristics with the objective of contributing to value *Q*. *faginea* for high-quality end-uses, thereby strengthening the species sustainability. Results already obtained include ring analysis and heartwood development, wood anatomy and density and their relationship. The enological potential of *Q*. *faginea* heartwood was also recently considered, showing a balanced tannins content, mainly of hydrolysable tannins such as ellagitannins, of volatile phenols and phenolic acids, in levels similar to or greater than those shown by the American (*Q*. *alba*) or French (*Q*. *petraea*) oaks. A chemical characterization of *Q*. *faginea* has not been made, namely considering the differences between heartwood and sapwood. In fact the transformation of sapwood to heartwood is accompanied by accumulation of solvent extractible compounds (the so-called extractives) and their proportion and properties may have a significant impact on the wood utilization. For instance, the heartwood in oaks shows high concentrations of phenolic extractibles (up to 10%), that may have multiple biological effects, including antioxidant properties, and has small amounts of lipids that vary according to the species and other factors such as geographical origin and tree. In this study, the summative chemical composition of the heartwood and sapwood of twenty *Q*. *faginea* trees from two sites in Portugal was evaluated, and the content and composition of lipophilic and polar extractible compounds were determined. The underlying rationale is an evaluation of the chemical-related wood quality of *Q*. *faginea* and of its variability for value wood products that could enhance the exploitation of this endogenous forest species, also taking into account the potential relevance of its wood extractives. # Material and methods The study was carried out in two locations: one stand in the northeast of Portugal (site 1), near Macedo de Cavaleiros and the other stand in the centre of Portugal (site 2), near Vimeiro. For the first site, the authority who issued the permission was Instituto da Conservação da Natureza e das Florestas ICNF, and for site 2 the private owner was asked and gave permission. ## Site characterization and sampling A total of twenty *Q*. *faginea* trees were randomly selected and harvested from two naturally regenerated and unmanaged stands (10 trees per stand) within the natural geographic distribution area of the species: one stand with 34–60 year old trees in the northeast of Portugal (site 1), near Macedo de Cavaleiros (41° 30′ N, 7° 01′ W; 554 m altitude); the other stand with 112–150 year old trees in the centre of Portugal (site 2), near Vimeiro (39° 29′ N, 9° 01′ W; 100 m altitude). The climate is of the Mediterranean type with Atlantic influence with a mean annual temperature of 12°C and 15°C, and annual precipitation of 700 mm and 890 mm at site 1 and site 2, respectively. Soils are classified as leptosols at site 1 and cambisols at site 2. The tree characteristics are summarised in. The chemical determinations were performed on a stem disc collected at breast height with approximately 5 cm thickness. The sapwood and heartwood were separated with a chisel and the samples were ground in a knife mill (Retsch SM 200) and sieved (Retsch ISO 9001). The 40–60 mesh fractions were kept for analysis. ## Chemical analysis Chemical summative analysis included determination of ash, soluble extractible compounds in dichloromethane, ethanol and water, Klason and acid-soluble lignin, and the monomeric composition of polysaccharides. All determinations were made with duplicate samples. The ash content was determined by incinerating 2.0 g of the sample at 525°C overnight and weighing the residue, reported as percent of the original samples (TAPPI 211 om-93). The soluble extractible compounds were determined with procedures adapted from Tappi 204 cm-97, in a soxhlet system with dichloromethane, ethanol and water during 6 h, 16 h and 16 h respectively. The extractible compounds solubilised by each solvent were determined by mass difference of the solid residue after drying at 105°C and reported as percent of the original sample (TAPPI T204 om-88). The lignin content was analysed from the extracted samples by acid hydrolysis with 72% sulphuric acid. Klason lignin was determined as the mass of the solid residue after drying at 105°C (TAPPI T 222 om-02). The acid-soluble lignin was determined by analysing the UV absorbance at 206 nm using a UV/VIS spectrophotometer (TAPPI Useful Method UM 250). The remaining acid solution was kept for sugar analysis. The polysaccharides were calculated based on the amount of the neutral sugar monomers released by total hydrolysis, after derivatization as alditol acetates and separated by high-performance anion exchange chromatography (Aa Dionex ICS-3000 system equipped with an electrochemical detector). The separation was performed with Aminotrap plus CarboPac SA10/Carbopac PA10 anion-exchange columns. The mobile phase was an aqueous 2 nM NaOH 135 solution at a flow rate of 1.0 ml/min at 25°C. ## Composition of dichloromethane extracts The lipophilic extractible compounds that were solubilized from the wood samples with dichloromethane were recovered as a solid residue after evaporation of the solvent and dried overnight under vacuum at room temperature. Aliquots (2 mg) of each sample were taken and derivatized as described below. They were dissolved in 100 μL of pyridine and the compounds with hydroxyl and carboxyl groups were trimethylsilylated into trimethylsilyl (TMS) ethers and esters, respectively, by adding 100 μl of bis(trimethylsily)-trifluoroacetamide (BSTFA). The reaction mixture was heated at 60°C for 30 min in an oven. The derivatized extracts were immediately analyzed by GC—MS by injection in a GC—MS Agilent 5973 MSD with the following GC conditions: Zebron 7HG-G015-02 column (30 m, 0.25 mm; ID, 0.1 μm film thickness), flow 1 ml/min, injector 280°C, oven temperature program, 100°C (1 min), rate of 10°C/min up to 150°C, rate of 4°C/min up to 300°C, rate of 5°C/min up to 370°C, rate of 8°C/min up to 380°C (5 min). The MS source was kept at 220°C and the electron impact mass spectra (EIMS) taken at 70 eV of energy. Compounds were identified as TMS derivatives by comparing their mass spectra with a GC—MS spectral library (Wiley, NIST), and by comparing their fragmentation profiles with published data. The full scan of the chromatogram was used in order to find all possible compounds. For semi-quantitative analysis the area of peaks in the total ion chromatograms of the GC—MS analysis was integrated and their relative proportions expressed as area proportion of the total chromatogram area. Each aliquot was injected in triplicate and results presented by mean (only standard deviation inferior to 5% was considered). ## Composition of ethanol-water extracts Extracts were prepared using approximately 1 g of the wood samples and ethanol/water (50/50, v/v), with a 1:10 (m/v) solid-liquid ratio for 60 min at 50°C using an ultrasonic bath. After filtration the supernatant extract was used to determine the contents in total phenolics, and condensed and hydrolysable tannins. Each assay was performed at least three times and at least three independent replicates were prepared for each standard and sample. The total phenolic content was determined spectrophotometrically by the Folin—Ciocalteu method using gallic acid as standard. An aliquot (100 μL) of the extract was mixed with 4 ml of the Folin—Ciocalteu reagent and after 6 min, 4 ml of a 7% Na₂CO₃ solution was added. After 15 min of incubation in a bath at 45°C, absorbance was read at 760 nm versus a prepared blank. A calibration curve was built using gallic acid as a standard (0–150 μg/ml). The total phenolic content was expressed as mg of gallic acid equivalents (GAE) per g of extract. Condensed tannins were determined by the vanillin-H₂SO₄ method, and the results expressed as mg of (+)-catechin equivalents. An aliquot (1.0 ml) of the extract was mixed with 2.5 ml of 1.0% (m/v) vanillin in absolute methanol and then with 2.5 ml of 25% (v/v) sulphuric acid in absolute methanol for vanillin reaction with the polyphenols in the extract. The blank solution was prepared with the same procedure without vanillin. Absorbances were recorded at 500 nm after 15 min. The tannin content was calculated from a calibration curve using catechin as standard, and expressed as mg of catechin equivalents (CE) per g of the extract. Hydrolyzable tannins were determined by the method of Willis and Allen. One ml of the extract and 5 ml of 2.5% KIO₃ were added into a vial and vortexed for 10 s. Absorbance of the red coloured mixture was determined at 550 nm versus the prepared water blank. Six concentrations of tannic acid solutions (500–2000 mg/L) were used for calibration. The final results were expressed as mg tannic acid equivalent (TAE) per g of the extract. ## Antioxidant activity of ethanol-water extracts Two methods were used to determine the antioxidant properties of these samples: ferric reducing/antioxidant power (FRAP), which measures the sample’s ferric reducing power, and 2,2-diphenyl-1-picryhydrazyl (DPPH), which measures the free radical scavenging capacity. The DPPH assay was performed using 2,2-diphenyl-1-picrylhydrazyl hydrate (DPPH) and was expressed in terms of: a) the amount of extract required to reduce 50% of the DPPH concentration (IC₅₀); and b) Trolox equivalents on a dry extract base. First, different dilutions of the initial extract and of a Trolox solution (0.2 mg/ml) in methanol were prepared. An aliquot of 100 μL of each methanolic solution of extract and of Trolox were added to 3.9 ml of a DPPH methanolic solution (24 μg/ml). The blank sample consisted of 100 μl of methanol added to 3.9 ml of DPPH solution. After 30 min incubation at room temperature in the dark, the absorbance was measured at 515 nm. The radical scavenging activity of each sample was calculated by the DPPH inhibition percentage as follows: I % = \[(Abs₀-Abs₁)/Abs₀\]×100, where Abs₀ was the absorbance of the blank and Abs_*1* was the absorbance in the presence of the extract at different concentrations. The IC₅₀ inhibiting concentration, which represents the concentration of a sample necessary to sequester 50% of the DPPH radicals, was obtained by plotting the inhibition percentage against the extract concentration. The scavenging effect on the DPPH radical of the extract was also expressed as the Trolox equivalent antioxidant capacity (TEAC) calculated from the calibration curve with the Trolox solution concentrations and the percentage of scavenging effect on the DPPH radical. The ferric reducing antioxidant power (FRAP) assay was followed as described by. The FRAP reagent was generated by mixing 300 mM sodium acetate buffer (pH 3.6), 10 mM (tripyridyl triazine) TPTZ solution and 20.0 mM FeCl₃.6H₂O solution in a ratio of 10:1:1 in volume. A sample (100 μl) of extract or standard was then added to 3 ml of FRAP reagent and the reaction mixture was incubated at 37°C for 30 min. The absorbance was measured at 593 nm in comparison with a blank. Aqueous solutions of known Trolox concentrations in the range of 0–0.5 Mmol/L were used for the calibration curve, and the results were expressed as Mmol Trolox equivalents/ g dry mass. ## Statistical analysis All results were expressed as mean and standard deviation (SD). The significance of differences (*p* ≤ 0.05) among the corresponding mean values was determined using one-way analysis of variance (ANOVA) followed by Duncan’s multiple- comparison test, and the Student’s t-test, Shapiro-Wilk normality test was also made using *Sigmaplot*^® statistical software, version 11.0. # Results and discussion The heartwood is clearly distinct in the wood cross-section, showing a rich brown colour and well defined border to the sapwood. The studied trees differ between sites by age, diameter and heartwood proportion as shown in : site 2 has the older and larger trees, therefore with a higher heartwood content (73.1% vs. 37.1% at site 2 and site 1). ## Chemical composition The chemical composition of the heartwood and sapwood of the *Q*. *faginea* trees from the two sites is given in Tables and regarding the monosaccharide composition. There were considerable chemical differences between heartwood and sapwood as regards solvent extractible compounds content. Heartwood contained significantly higher (F (1, 18) = 142.72; *p \<* 0.001) amounts of extractible compounds than sapwood: approximately two times more with on average 19.0% and 9.5% in heartwood and sapwood respectively. Polar compounds extracted by ethanol and water, which include mainly phenolics and polyphenolics, corresponded to most of the total content (93% and 96% of the total extractible compounds in sapwood and heartwood respectively). However, there was a difference between the proportion of the ethanol and water soluble compounds in heartwood ethanol-soluble compounds represented the largest proportion of total extractives (65%) while in sapwood it was the water-soluble extractibles that had the highest proportion (50% of the total extractives). The amount of dichloromethane soluble extractibles was very small (below 1%) similar in heartwood and sapwood, but statistically different (F (3, 36) = 6.60; *p* = 0.001). Comparative studies of heartwood—sapwood solvent extractible compounds in other oak species have shown in general a heartwood enrichment in these non-structural compounds, although to a variable extent depending on species. Seikel et al. reported that *Q*. *rubra* contained 6.2% and 8.2% total extractibles in sapwood and heartwood, respectively, which included sixteen phenolic compounds. Hernández and Salazar reported the chemical composition of sapwood and heartwood of four oak species: respectively 6.1% and 7.2% in *Q*. *coccolobilofia*, 8.3% and 8.9% in *Q*. *durifolia*, 9.2% and 11.6% in *Q*. *rugose*, and 8.2% and 10.6% in *Q*. *oleoides*. Ruiz-Aquino et al. found no differences between sapwood and heartwood solvent extractible compounds content: respectively 5.3% and 5.4% in *Q*. *laurina*, and 8.2% and 8.9% in Q. *crassifolia*. The higher content in solvent extractible compounds of heartwood in relation to sapwood is considered a heartwood specific feature. However the magnitude of the difference between sapwood and heartwood differs greatly between species e.g. respectively 3.5% and 5.7% or 2.4% and 3.8% in *Eucalyptus globulus*, 4.0–4.2% and 7.4–9.5% in *Acacia melanoxylon*, 9.2% and 10.0% in *Tectona grandis*. The heartwood contained slightly less lignin than sapwood (23.1% and 25.5% in relation to wood) but when calculated in relation to extractive-free wood the lignin content is the same: 28.6% in heartwood and 28.1% in sapwood. The lignin content reported for *Q*. *faginea* heartwood and sapwood is similar to the values reported by Ruiz-Aquino et al. for *Q*. *laurina* (25.1% in sapwood and 25.5% in heartwood) and *Q*. *crassifolia* (24.9% in sapwood and 25.2% in heartwood), and by Lourenço et al. for *Q*. *suber* sapwood (23.8%). The carbohydrate composition was similar in heartwood and sapwood. The major monosaccharide was glucose (60.8% and 59.6% of the total monomeric sugars in heartwood in sapwood respectively) while xylose was the dominating non- cellulosic sugar (31.6% and 33.8% of total neutral monosaccharides in heartwood ad sapwood respectively) and arabinose and galactose represented about 2.9% of the total content of neutral sugars. This sugar composition agrees with the general structure of hardwood hemicelluloses. A similar monosaccharide composition was reported for *Q*. *suber* sapwood. Overall there was little chemical variation between the *Q*. *faginea* trees and the sapwood and heartwood chemical composition was not influenced by the sites (and therefore by the different ages of trees) e.g. *p* = 0.131 for heartwood solvent extractible compounds, *p* = 0.153 for extractive-free heartwood lignin content. ## Composition of dichloromethane extract The identified lipophilic extractible compounds in *Q*. *faginea* sapwood and heartwood are reported in. The main constituents are saturated alkanoic acids (15.7% and 25.8% in sapwood and heartwood, respectively). The chain lengths vary from C10 to C26, with hexadecanoic acid as the major compound found, representing 35.5% and 41.1% of all fatty acids present in sapwood and heartwood, respectively. Both saturated and substituted alkanoic acids are substantially more abundant in heartwood than in sapwood as previously reported by Ekman, although there is significant less content of substituted fatty acids, probably due to side reactions that convert them into the saturated form. Sterols were also abundant in both wood tissues, varying from 10.2% to 13.0%. Among them, β-sitosterol constitutes 6.6% to 12.1% of the identified compounds. Fatty acids and sterols are responsible for the reduction of wood durability since they are nutrients for fungi attack. However, antifungal capacity was found for some fatty acids, as reported by Salem et al.. Triterpenes were also identified in both wood tissues, being higher in heartwood; arjunic acid was the major constituent (2.4% and 12.5% in sapwood and heartwood, respectively). This can be explained by the tree defence mechanism against external aggressors as fungi, accumulating triterpenes, sterols, flavonoids and other secondary metabolites during heartwood formation. Authors found a significant positive relationship between the compositions of the lipophilic extractibles on both sapwood and heartwood tissues (p\<0.05) concerning alkanols, alkanoic acids, sterols and triterpenes, from both sites. The same is true for all the classes of compounds from both wood tissues from both sites, except for saturated and substituted diacids. The major difference between sapwood and heartwood lipophilic extracts relies on their aromatics content, reaching in average for the two sites 22.8% of all compounds for sapwood, against 5.3% in heartwood. ## Ethanol-water extracts shows the average and standard deviation values for the yield and composition of ethanol-water (1:1) extracts regarding total phenolics, condensed and hydrolysable tannins. The results show a clear difference between heartwood and sapwood. The obtained extraction yield was high (on average 14.7% and 6.0% for heartwood and sapwood); these values are under the determined total solvent extractible compounds content due to the less intensive extraction conditions. However the obtained extract yields were higher in comparison to the value of 8.9% reported for *Q*. *faginea* heartwood extraction with 50% methanol-water or the values of 8.3%, 7.8% and 7.9% for *Q*. *pyrenaica*, *Q*. *robur* and *Q*.*petraea* respectively. The major part of solvent extractibles is made up of phenolic substances that represent about 56% of the heartwood extract and 32% of the sapwood extract, corresponding to about 8% of heartwood (86.2 and 77.4 mg GAE /g of wood) and 2.5% of sapwood (27.1 and 11.9 mg GAE /g of wood). It is clear that there is an enrichment in phenolic compounds in the heartwood in comparison to sapwood. The values obtained here for the phenolic content of *Q*. *faginea* heartwood extracts are similar to those reported in the literature (no reports on sapwood extract composition were found). Cadahía et al. and Fernandez de Simón et al. reported for *Q*. *faginea* heartwood methanolic extracts (50% methanol-water) total phenolics of 50.5 and 89.2 mg GAE/g wood, respectively, and 42.6 and 56.6 mg GAE/g of wood for *Q robur*, 53.7 and 70.3 mg GAE/g of wood for *Q*. *petraea*, and 44.8 and 56.6 mg GAE/g of wood for *Q*. *pyrenaica* heartwood. Alañón et al. obtained for methanol extracts values of 41.5 mg GAE/g of wood for *Q*. *pyrenaica*, 72.6, 48.9 and 33.3 mg GAE/g of wood for *Q*. *robur*, *Q*. *petraea* and *Q*. *alba* respectively. The polyphenolic composition of the aqueous ethanolic extracts of *Q*. *faginea* heartwood and sapwood showed a similar profile, with a higher content in hydrolysable tannins (ellagitannins) and very small amounts of condensed tannins. This is in accordance with the literature. When expressing the amounts of hydrolysable tannins as mg tannic acid/g of wood, there is a difference between extracts due to the different extraction yields with higher values for the heartwood (on average 23 mg tannic acid/g of heartwood) in relation to sapwood (12.2 mg of tannic acid/g of sapwood). These values are higher than those reported by Fernandez de Simón et al. and Cadahía et al. who showed for *Q*. *faginea* an ellagitannin content of 4.0 and 8.5 mg ellagic acid/g wood. The values are also higher when compared to the 5.1–5.9 mg ellagic acid/g heartwood of *Q*. *pyrenaica*, 8.5–12.1 mg ellagic acid/g heartwood of *Q*.*petraea* and 7.6–8.2 mg ellagic acid/g heartwood of *Q*. *robur*. However, Castro-Vázquez et al. report higher values for *Q*. *pyrenaica* with 18.2–23.0 mg ellagic acid/g wood. The site did not influence the content of total phenolics of *Q*. *faginea* wood (F (1, 18) = 3.18; *p* = 0.080). The same was reported for *Q*. *pyrenaica* heartwood. However for *Q*. *alba* and *Q*. *robur*, Miller et al. referred that species seem to be more important than the geographical origin to explain the differences in the wood phenolic composition. ## Antioxidant properties of ethanol-water extracts The antioxidant characteristics of *Q*. *faginea* heartwood and sapwood extracts were evaluated by measurement of the scavenging capacity against a given radical (DPPH assay) and by the ferric reducing antioxidant power (FRAP). The results are summarised in. Both heartwood and sapwood extracts exhibited high antioxidant activities (IC₅₀ of 2.6 μg/ml and 3.3 μg/ml respectively) when compared to Trolox (IC₅₀ 3.8 μg/ml) which is used as an antioxidant standard compound. Ellagitannins should be the main compounds responsible for the antioxidant capacity of the extracts since they have a higher ability to donate a hydrogen atom and to support the unpaired electron as compared to low molecular weight phenolic compounds. The antioxidant activity expressed as mg of Trolox equivalents/g of wood (112.7 mg Trolox equivalents/g of heartwood vs. 84.8 mg Trolox equivalents/g of sapwood) are similar to those reported for wood extracts of other oaks: 65.3 mg Trolox equivalents /g of wood and 112.9 mg Trolox equivalents /g of wood for American and French oaks respectively. The heartwood and sapwood ethanolic extracts present FRAP activity with a mean value of 2.8 mMol Trolox/g extracts of heartwood and 2.1 mMol Trolox/g extracts of sapwood. The reducing ability of the extracts showed an increasing trend with the polyphenol content, as expected. Phenolic compounds can act as reducing agents, hydrogen donors, singlet oxygen quenchers or metal chelators due to their redox properties and consequently antioxidant capacity. The heartwood extract, with the highest concentration, exhibited the highest reducing capability (0.4 mMol Trolox equivalents/g of heartwood), while the sapwood extract possessed the lowest ferric reducing abilities (0.15 mMol Trolox equivalents/g of sapwood). The present results were compared to the values of other extracts from wood of *Quercus* spp. Alañón et al. evaluated the FRAP of aqueous methanolic extract of *Q*. *pyrenaica*, *Q*. *robur*, *Q*. *petraea* and *Q*. *alba* in mMol of Trolox equivalents/g of wood as 0.54, 0.82, 0.59 and 0.39 respectively. In methanol extracts of American and French oaks, the FRAP values were 0.29 and 0.45 mMol Trolox equivalents/g of wood, respectively. Overall the chemical features of the polar extracts of heartwood and sapwood of *Q*. *faginea* allow considering their potential use for the maturation and aging of alcoholic beverages, as previously suggested. Also the use of extracts as a source of compounds with antioxidant properties may be envisaged as additives in the various applications for which such characteristics may be important. # Conclusions The heartwood and sapwood of *Q*. *faginea* are clearly outsingled with a well- established border. Chemically, heartwood and sapwood differ considerably in their extractives content, namely in polar extractives e.g. compounds soluble in ethanol and water, which are approximately two-fold in heartwood in relation to sapwood. The ethanol-water extracts are rich in phenolics, especially the heartwood extracts. The content of ellagitannins is high and of condensed tannins is low and similar in both heartwood and sapwood extracts. The antioxidant properties of *Q*. *faginea* heartwood and sapwood extracts are excellent and above those of standard antioxidant compounds. Lipophilic extracts from *Q*. *faginea* heartwood and sapwood are mainly constituted by saturated alkanoic acids and sterols. Aromatic compounds were also identified in considerable extent only in the sapwood lipophilic extracts. The variability between trees is low and no differences between trees of different ages from two sites were found. *Q*. *faginea* shows a very good potential for cooperage and other applications for which a source of compounds with antioxidant properties is desirable. We thank José L. Louzada for providing the samples from Site 1 and Sofia Knapic for project management. We also thank Joaquina Silva and Lidia Silva for help with the chemical analysis. [^1]: The authors have declared that no competing interests exist. [^2]: **Conceptualization:** IM VS JF HP. **Data curation:** IM VS JF. **Formal analysis:** IM JF HP. **Funding acquisition:** HP. **Investigation:** IM JF. **Methodology:** IM JF. **Project administration:** IM HP. **Resources:** IM VS JF HP. **Supervision:** HP. **Validation:** IM JF. **Visualization:** IM VS JF HP. **Writing – original draft:** IM VS JF HP. **Writing – review & editing:** IM VS JF HP.

# Introduction Papillary renal cell carcinoma (PRCC) accounts for approximately 15% of renal cancers and is the most common type of renal cell carcinoma after clear cell renal cell carcinoma (ccRCC). Its pathological classification is controversial, and its clinical behaviour is highly variable. Not only can PRCC be divided into histological type 1 and type 2, studies in recent years have shown that type 2 PRCC is heterogeneous; it can be divided into further subtypes according to tumour-related molecular biomarkers, and these subtypes are associated with different clinical processes and prognoses. For example, subgroup C2c contains type 2 PRCC tumours with the CpG island methylator phenotype (CIMP), and the overall survival rate is the lowest in this group. Because the specific molecular pathogenesis of PRCC is poorly understood, there has been no effective targeted therapy for papillary carcinoma in the past. There is accumulating evidence indicating that abnormal gene expression and mutation are related to the development of PRCC. Linehan et al. found through clinical studies that the occurrence of PRCC type 2 was closely related to mutations in the chromatin modification genes SETD2, BAP1 and PBRM1, while mutations in the structural domain of the tyrosine kinase MET were related to type 1 PRCC. In hereditary leiomyomatosis and renal cell carcinoma (HLRCC) (which is an aggressive form of type 2 PRCC), mutation of Fumarate Hydratase (FH) located on chromosome 1 is relatively common. However, there is currently no standard treatment for the disease, and mortality rates for PRCC patients remain high. Therefore, understanding the exact molecular mechanisms involved in the carcinogenesis, metastasis, and recurrence of PRCC and formulating effective diagnosis and treatment strategies are of vital importance to improving the survival rate of patients. Microarray technology and bioinformatic analysis, important and widely used methods for screening and identification of differentially expressed genes (DEGs) related to disease development, have helped us to explore a wide variety of DEGs involved in PRCC canceration and progression. However, the false positive rate in single microarray analysis is relatively high, and it is difficult to obtain reliable results. Therefore, in this study, 3 mRNA microarray datasets from the Gene Expression Omnibus (GEO) were downloaded and analysed to identify DEGs between PRCC and normal tissues. Subsequently, Gene Ontology (GO) analysis, Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis and protein-protein interaction (PPI) network construction were used for analysis to help us understand the key genes and pathways involved in carcinogenesis and progression. From this analysis, a total of 214 DEGs and 17 hub genes were selected that may be potential molecular targets and biomarkers of PRCC. # Materials and methods ## Acquisition of microarray data The GEO (<http://www.ncbi.nlm.nih.gov/geo>) database of the National Center for Biotechnology Information (NCBI) is a public functional genomics database that is used to store gene expression datasets, platform information and original series. Three gene expression datasets (GSE7023, GSE48352 and GSE15641) were downloaded from GEO. GSE7023 is based on the GPL4866 platform (Affymetrix GeneChip Human Genome U133 Plus 2.0 Array) and includes 35 PRCC samples and 12 normal samples. GSE48352 is based on the GPL16311 platform (\[HG-U133_Plus_2\] Affymetrix Human Genome U133 Plus 2.0 Array) and includes 24 PRCC samples and 8 normal samples. GSE15641 was based on the GPL96 platform (\[HG-U133A\] Affymetrix Human Genome U133A Array) and includes 11 PRCC samples and 23 normal samples. ## Screening of DEGs Differential expression analysis was performed for each dataset using the GEO2R online analysis tool provided with the GEO database. The screening criteria for DEGs between PRCC and non-cancerous samples were \|log₂FC (fold change) \|\> 1 and FDR adjusted P value \<0.05. Based on the annotation data in the platform, the probes were converted into the corresponding gene symbol. Probe sets without corresponding gene symbols and duplicate data were removed. Then, the intersection of three datasets was taken to determine the common DEGs, and the online tool Venn Diagram (<http://bioinformatics.psb.ugent.be/webtools/Venn/>) was used to draw the Venn diagram of the DEGs. ## KEGG and GO enrichment analyses The Database for Annotation, Visualization and Integrated Discovery (DAVID; <https://david.ncifcrf.gov/home.jsp>) (version 6.8) was used to conduct GO enrichment analysis of the DEGs. GO includes three categories: biological pathway, cellular component and molecular function. KOBAS 3.0 (<http://kobas.cbi.pku.edu.cn/kobas3/annotate/>) was used for KEGG pathway enrichment analysis. P\<0.05 was considered statistically significant. ## PPI network construction and module analysis The Search Tool for the Retrieval of Interacting Genes online database (STRING; [https://string-db.org](https://string-db.org/) (version 11.0)) was used to construct the DEGs PPI network, and the analysis results were visualized with Cytoscape (version 3.6.1); each interaction with a combined score \>0.4 was considered statistically significant. The Molecular Complex Detection (MCODE) (version 1.5.1) plug-in in Cytoscape was used to identify the most important modules in the PPI network. The selection criteria were as follows: degree cutoff = 2, node score cutoff = 0.2, K-core = 2 and max depth = 100. The genes in the module were then subjected to GO and KEGG enrichment analyses using DAVID and KOBAS 3.0, respectively. ## Selection and analysis of hub genes Through the degree algorithm, 17 genes identified to have scores greater than 15 by cytoHubba in Cytoscape were identified as hub genes. These hub genes were analysed and visualized with the Biological Network Gene Oncology (BiNGO) tool (version 3.0.4). The heatmap of the hub genes was constructed using the UCSC cancer genomics browser (<http://xena.ucsc.edu/>). The associations of the hub genes with overall survival and recurrence-free survival were analysed using Kaplan-Meier Plotter (<https://kmplot.com/analysis/>). The online database Oncomine ([https://www.oncomine.org](https://www.oncomine.org/)) was used to analyse heatmaps of PECAM1 and PLAU gene expression in clinical PRCC samples and normal tissues. The online database UALCAN (<http://ualcan.path.uab.edu/>) was applied to analyse the expression of these two genes based on tissue type (PRCC tissue versus normal tissue), stage, subtype and race. # Results ## Identification of DEGs in PRCC After the microarray results were normalized, DEGs in the three datasets were identified (906 in GSE7023, 823 in GSE48352, and 2051 in GSE15641). The overlap between the three datasets contained 214 genes, as shown in the Venn diagram. The set of 214 overlapping genes contained 9 upregulated genes and 205 downregulated genes between PRCC and normal tissues. ## GO and KEGG enrichment analyses of the DEGs GO and KEGG enrichment analyses of the DEGs were performed using DAVID and KOBAS 3.0, respectively. The GO analysis results indicated that in terms of biological process, the DEGs mainly participated in excretion, kidney development, angiogenesis, negative regulation of growth, epithelial cell differentiation and oxidation reduction processes. In terms of molecular function, DEGs were mainly associated with receptor binding, heparin binding, prostaglandin E receptor activity, calcium ion binding, sodium-independent organic anion transmembrane transporter activity and transcriptional activator activity. In the cellular component category, the DEGs were mainly enriched in the terms extracellular exosome, apical plasma membrane, integral component of plasma membrane, extracellular region, and cell surface. KEGG pathway analysis showed that the DEGs were mainly enriched in the following pathways: metabolic pathways, pathways in cancer, complement and coagulation cascades, proteoglycans in cancer and PPAR signaling pathway. ## PPI network construction and module analysis The PPI network of the DEGs was constructed with STRING and Cytoscape, and the most significant module, which consisted of 10 nodes and 43 edges, was identified. DAVID and KOBAS 3.0 were used to analyse the genes in this module by functional enrichment. The results showed that the genes in this module were mainly enriched in angiogenesis, cell adhesion, platelet degranulation, leukocyte transendothelial migration, Pathways in cancer, PI3K-Akt signaling pathway and so on. ## Selection and analysis of hub genes The 17 genes whose degree scores were greater than 15 were defined as the hub genes and included ALB, EGF, KDR, CXCL12, REN, PLG, PECAM1, KNG1, CDH5, AQP2, C3, THY1, WT1, MGAM, PLAU, AGTR1, and DCN. The results of biological process analysis of the hub genes are shown in. Heatmap analysis showed that the hub genes could distinguish PRCC samples from normal samples. The associations of the hub genes with overall survival and recurrence-free survival were analysed using Kaplan-Meier Plotter. PRCC patients with changes in EGF, KDR, CXCL12, REN, PECAM1, CDH5, THY1, WT1, PLAU, and DCN showed poorer overall survival and recurrence-free survival. However, patients with altered KNG1 expression in PRCC tissues had poor overall survival but no significant difference in recurrence- free survival. Among these genes related to significant differences in overall survival and recurrence-free survival, PECAM1 and PLAU were identified as seed genes in the significant modules screened by MCODE, suggesting that they might play a crucial role in the carcinogenesis and progression of PRCC. Oncomine analysis of the carcinogenesis of PRCC and normal tissues showed that PECAM1 and PLAU were downregulated in PRCC tissues in different datasets. UALCAN was used to analyse the expression of these two genes based on tissue type (PRCC tissue versus normal tissue), stage, subtype and race. As shown in , PECAM1 and PLAU were downregulated in PRCC tissue compared with normal kidney tissue, and this finding was consistent with the UCSC and Oncomine analysis results. Moreover, the expression of both genes in PRCC carcinoma tissues was related to stage, subtype and race. Notably, PECAM1 was most significantly downregulated in stage 1 Caucasian patients with type 1 PRCC, while PLAU was most significantly downregulated in Asian patients with stage 4 CIMP-type PRCC. # Discussion The pathological classification of papillary renal cell carcinoma is controversial, and its morbidity and mortality are increasing annually. However, the potential molecular pathogenesis of PRCC is poorly understood. Therefore, in this paper, three mRNA datasets from PRCC tissues and normal tissues were analysed. A total of 214 DEGs were identified, specifically, 205 downregulated genes and 9 upregulated genes. GO enrichment analysis showed that the DEGs were mainly enriched in excretion, renal development, angiogenesis, negative regulation of growth, epithelial cell differentiation and other processes and were located in exosomes and other sites. By PPI network analysis, 17 hub genes were identified. Survival analysis of patients stratified by the expression levels of the hub genes showed that changes in EGF, KDR, CXCL12, REN, PECAM1, CDH5, THY1, WT1, PLAU and DCN were associated with overall survival and recurrence-free survival. These results suggest that these genes may play an important role in the carcinogenesis, progression and recurrence of PRCC. EGF can activate several pro-oncogenic intracellular pathways leading to tumour cell proliferation and angiogenesis. KDR is a specific receptor for VEGF, which promotes angiogenesis. CXCL12 has been shown to be a stimulatory chemokine related to tumour neovascularization and to be involved in promoting the migration and invasion of cancer cells. The renin-angiotensin system regulates angiogenesis, cell differentiation and proliferation, opening a new avenue for understanding the occurrence of renal carcinoma. PECAM1 encodes a protein associated with angiogenesis and extracellular circulation that is involved in tumour growth and spread. CDH5 is abnormally expressed in a variety of human cancers and plays an important role in angiogenesis. THY-1 is a renal differentiation marker and is useful in the characterization of tumours of renal origin. The WT1 gene, which encodes a transcription factor with four zinc finger DNA-binding motifs, shows variable stage-specific expression in the developing kidney. PLAU encodes a serine protease involved in degradation of the extracellular matrix and possibly in tumour cell migration and proliferation. DCN is an important component of the ECM and functions as a tumour suppressor in RCC that can inhibit the proliferation and metastasis of RCC cells. The genes analysed above may provide a new direction for further studies on the pathogenesis of PRCC. KEGG pathway analysis showed that the DEGs were mainly concentrated in metabolic pathways, pathways in cancer, complement and coagulation cascades, proteoglycans in cancer and the PPAR signalling pathway. Studies have shown that metabolic reprogramming of PRCC is associated with upregulation of many proteins and enzymes involved in the glycolytic pathway. PPARα antagonism inhibits several pivotal cancer-relevant metabolic pathways, leading to reduced tumour growth in RCC. Further study of these potential pathways will contribute to a deeper understanding of PRCC. PECAM1 and PLAU were identified as seed genes in the significant module determined by MCODE. Studies have shown that PECAM1 expression is significantly elevated in patients with ccRCC in the early stage of the disease. However, our analysis showed that PECAM1 expression was downregulated in patients with stage 1 type 1 PRCC. Therefore, PECAM1 may be an early diagnostic indicator to distinguish ccRCC from PRCC. UALCAN analysis showed that low expression of PECAM1 and PLAU in PRCC tissues was associated with stage, subtype, and race. Notably, PECAM1 was most significantly downregulated in Caucasian patients with stage 1 type 1 PRCC, while PLAU was most significantly downregulated in Asian patients with stage 4 CIMP-type PRCC. These results suggest that PECAM1 and PLAU play critical roles in the typing and the efficient diagnosis and treatment of PRCC. In summary, a bioinformatics approach was used in this study to analyse the DEGs involved in the occurrence and development of PRCC. Due to the current limited understanding of these DEGs, their specific biological functions and molecular regulatory mechanisms still need to be confirmed by further studies. # Supporting information [^1]: The authors have declared that no competing interests exist. [^2]: ‡ These authors share first authorship on this work.

# Introduction Transplantation of genetically incongruous organ generates the immune response, which may eventually result in the destruction of the grafted tissue. Because currently used immunosuppressants induce impairment of the recipient's immune system, a major goal in transplantation is to prevent rejection by inducing tolerance while avoiding global immunosuppression. The administration of MHC class I allochimeric molecule \[α1h1/u\]-RT1.Aa that contains donor-type (Wistar Furth, WF; RT1u) immunogenic epitopes displayed on recipient-type (ACI, RT1a) sequences and produced by the alteration of the immunodominant determinant in the α1-helical region of class I MHC RT1.Aa to that of RT1.A1 and RT1.Au sequences induced indefinite survival of heterotopic cardiac allografts in rats when administered in conjunction with sub-therapeutic dose of cyclosporine (CsA). The results of several studies indicate that the soluble MHC class I proteins either directly inhibit T cell functions by receptor blockade or induce apoptosis of activated CD8+ T cells. Alterantively, the MHC molecules may modulate CD4+ T cells responses via phagocytosis and indirect presentation by antigen-presenting cells. However, the cellular and molecular mechanisms underlying the immunosuppressive function of allochimeric \[α1h1/u\]-RT1.Aa molecule and the mechanism(s) by which this immunosuppressant regulates gene expressions of host's T cells and induce allograft tolerance remains largely unknown. In vascularized transplantation models, the spleen contributes to graft rejection by generating alloantigen reactive T cells. It is well established that immune response relies on the ability of T cells to move, scan and to form the immunological synapse with the antigen presenting cells (APCs). Interaction of T cell with APC involves: active migration towards the APCs, the adhesive contact required to scan the surface of APC, and the polarization and redistribution of cytoskeleton which allows the close apposition of cell membranes necessary for T cell receptor (TCR) interaction with major histocompatibility complex \[MHC; 14–25\]. All these functions require polarized cytoskeleton and proper segregation of membrane, and adhesion and intracellular signaling proteins. To find out which genes and molecular pathways are affected by the allochimeric molecule treatment, we have analyzed, using microarray and quantitative reverse transcription polymerase chain reaction (qRT PCR), the quantitative and temporal patterns of gene expression profile of splenic T cells in rats treated with allochimeric \[α1h1/u\]-RT1.Aa molecule in conjunction with sub-therapeutic dose of CsA in comparison to sub-therapeutic dose of CsA-treated and untreated rats at post-transplantation day 1, 3 and 7. # Results ## Overview of Gene Expression Profiles Gene expression profiles were generated from ACI host splenic T cells at 1, 3 or 7 days of post transplantation. Control transplanted animals received no treatment. Experimental transplanted animals were treated with sub-therapeutic dose CsA in conjunction with allochimeric peptide or with sub-therapeutic dose of CsA (see for details). ## Principal Components Analysis Principal component analysis (PCA) is a standard tool used in modern data analysis, which simplifies (reduces) complex (multidimensional) and confusing data sets to lower dimensions (less complex). The goal of principal component analysis is to filter out the irrelevant “noise” in existing data set and to identify the most meaningful basis to present a data set. The principal components analysis (PCA) was performed using Partek Genomics Suite. PCA transforms the data to a new coordinate system such that the greatest variance by any projection of the data comes to lie on the first coordinate (called the first principal component), the second greatest variance on the second coordinate, and so on. Those characteristics of the data set that contribute most to its variance are retained. When the T Cell microarray data were analyzed by PCA, 50% of the variation in samples was revealed in the first two principal components. The type of treatment appeared to be the most significant effect and the treatment time was another significant variable. The type of treatment appeared to be the most significant effect since each treatment could be seen as distinct groups. Both the untreated (Green) and the cyclosporine treated (Red) samples seemed to form similar cluster shapes. For example, the distance between each of the green ovoids is less than the distance between green (untreated) and red ovoids (cyclosporine treated). This suggests a greater variation due to treatment than time after treatment for these two conditions. Cyclosporine plus peptide (Blue) sample, on the other hand, was much more spread out. This suggests another variable such as time after transplantation has an important role in this treatment condition. Treatment time did appear to have some effect on the PCA for all treatment conditions since the ovoids representing 7 days treatment samples were always to the left of the 1 or 3 day treatments. ## Gene Changes Due to Treatment Differential gene expression was determined in Partek Genomics Suite with the criteria described in. Since post transplantation time also had an influence on the gene expression profiles, we examined the effect of treatment on gene expression for each day separately. Treatment with CsA and CsA + peptide were compared with the corresponding untreated control for that day. Gene changes were divided into those that were unique to each treatment and those that were common to both the CsA and CsA + peptide treatments. Overall, more than 80% of these changes were genes down regulated compared to the untreated control groups, which is expected considering the immunosupressive function of CsA and CsA+peptide. The number of gene expression changes associated with sub- therapeutic doses of CsA was greatest at day 1 and quickly diminished by days 3 and 7. On the other hand, gene expression changes unique to CsA + peptide not only persisted but increased from 286 changes at day 1 to 683 changes at day 7 post treatment. Lists of individual gene changes for each day can be found in the supplementary tables. Since we were primarily interested in the genes involved in early events of immune response associated with the immunosuppressive function of allochimeric molecule, we focused our analysis on genes differentially expressed due to CsA + peptide treatment (row 3). The increasing number of genes meeting the criteria for differential expression suggested a persistent immunsupression in the CsA + peptide treated animals. The increased number of changes also suggested that there might be early and additional late immunosuppressive effects of peptide treatment. The overlap of CsA + peptide gene changes within the given days can bee seen in. There is surprisingly little overlap in these changes. Only 32 genes were differentially expressed at both 1 and 3 days post-treatment with CsA + peptide. When day 3 and 7 post- teatment with CsA+peptide were compared, the 171 genes common for both these days were differentially regulated. Only one gene was differentially regulated at all three time points. That gene was identified as an EST (AA800192) and was down regulated approximately 2 fold at all three days. This EST has sequence homologies with endogenous retroviral mRNA, LTR-repeat sequences and MHC class I in rat. Considering the fact that LTR repeats play a role not only in specific DNA rearrangement but also act as a MHC I gene promoter cis-acting activation response element – this finding is especially interesting. ## Early Gene Changes Associated with CsA and CsA + Peptide Treatments The sub-therapeutic dose of CsA used in this study is not effective in preventing heart allograft rejection in rat model. The same dose of CsA is, however, much more effective when administered in combination with the allochimeric molecule. The numbers of gene changes shown in is consistent with this idea. Initially (day 1), the CsA treatment downregulates a significant number of genes, which are probably the early responders, induced shortly after transplantation. Although this down regulation does not persist in CsA treatment, many gene changes do persist in the CsA + peptide treatment. Our overall assessment is that majority of the changes are not really persistent but rather can be divided into early and late changes. The lists of differentially expressed genes may provide some insight into the molecular mechanisms of action of these two immunosuppressants. Most interestingly, among the genes down regulated at 1 day and 3 days post transplantation in CsA + peptide treatment were genes involved in actin filament polymerization and positive regulation of protein polymerization (RhoA and Rac1; 29–33), cell adhesion, and cell junction assembly and maintenance (Catna1, Vcam, and CD9; 17, 23, 24, 34–36), vacuolar transport, vacuole organization and biogenesis (Rhob, Cln8, ATP6v1b2; 37, 38–40) and MAPK pathway genes \[Spred1 and Dusp6; 41, 42\] involved in tubulin cytoskeleton reorganization and interaction between actin and microtubule cytoskeleton. The down regulation of these genes was also confirmed independently by qRT PCR method. All these genes are directly or indirectly involved in the pathways participating in T cell polarization, scanning ability and the formation of the immunological synapse. To identify additional functional related patterns from the differentially expressed gene lists, we performed a pathway analysis. This was done by comparison of our gene lists with canonical KEGG pathways using “Pathway Express”. Pathway Express ranks the probability of a pathway being significantly affected with an impact factor \[described in 43\]. Based on an impact factor of three or greater we found 13 KEGG pathways to be affected from the CsA treated differential expressed gene list at day 1. In the CsA + peptide treatment at day 1 the additional 6 pathways were found to be affected. These additional pathways also appeared to be consistent with the pathways involved in T cell polarization, migration and immunological synapse formation and functions. ## Comparison of 1 and 7 Day CsA + Peptide Treatments Interestingly, pathway analysis revealed much more similarity between 1 day and 7 days after treatment with CsA + peptide. Approximately one third of the significant pathways were identical. These included NK and B Cell activation pathways along with several of the associated signaling pathways, Toll like receptor, VEGF, and MAPK. In addition, pathways involved in cell adhesion remained significantly effected. Some of the new pathways appearing at day 7 were also related to Cell adhesion, Adherens junction, Focal adhesion and regulation of actin cytoskeleton. Majority of these pathways are involved in T cell polarity, migration and the formation of immunological synapse with antigen presenting cells. # Discussion Our study of T cell expression profile showed that the expression levels of many genes either directly or indirectly involved in the molecular pathways responsible for T cell polarization, movement, scanning ability and the formation of the immunological synapse are down regulated during CsA + allochimeric molecule treatment when compared to CsA treatment alone or to the untreated control. The generation of immune response requires several highly specialized molecular and cellular events, which take place in sequential manner. First there is a migration of T cells toward the antigen presenting cells (APCs), followed by an initial adhesive contact between T cell and APC which allows for the scanning of APC surface by T cell, and finally the formation of the immunological synapse i. e. the close apposition of T cell and APC membranes necessary for the interaction between the T cell receptors (TCR) and major histocompatibility complex (MHC). The supramolecular spatial organization and highly dynamic clustering of the molecules in the immunological synapse allows T cell receptors to response to the antigen and to facilitate polarized secretion of cytotoxic granules and cytokines. All these events involve highly orchestrated polarization and rearrangements of cytoskeleton, membranes, vacuolar/endosomal compartments and cell adhesion and signaling molecules. Our study showed that the following genes, directly or indirectly involved in all of the above events, are down regulated in T cells originating from allochimeric MHC molecule treated animals. Rac1, RhoA and RhoB genes belong to the Rho family of GTPases. Rho GTPases cycle between an inactive, GDP-bound, and an active, GTP-bound state, and act like “molecular switches” controlling the dynamics of cell proliferation, movement, cytoskeleton and TCR signaling. Engagement of the TCR has been shown to lead to the activation of Rac1 and Rac2. Consistent with a key role for Rac proteins in TCR mediated signaling, CD4+ T cells from Rac2−/− mice are defective in TCR mediated proliferation and IL-2 and IFN-γ production. TCR stimulation leads to the activation of RhoA, and blocking RhoA activation has been shown to decrease production of IL-2. In addition, it was shown that inhibition of ROCKs, key effectors of RhoA, caused a decrease in T cell proliferation and production of IL-2 and IFN-γ, and resulted in prolonged survival of cardiac allograft. In the light of these data it is not surprising that immunosuppressive function of allochimeric molecule used in our study relies on the down regulation of genes belonging to Rho GTPase cascade. The signaling Rac/Rho cascade triggered by TCR engagement is also tightly intertwined with the remodeling of cytoskeleton essential for decreased T cell rigidity that enables closer contacts between the T cell and the APC and the formation of immunological synapse. Rac and Rho are also involved in the integrin-mediated adhesion between T cells and APCs, control of actin polymerization at the T cell/ APC interface, and in proper clustering of lipid rafts and TCR at the immunological synapse. In addition, Rho GTPases are involved in the acquisition of a polarized morphology of T cells i.e. in the processes that are critical for the movement and migration of T cells. Unlike Rac and RhoA, the RhoB is an immediate early response gene, which has a short half-life, is associated with the plasma membrane and endosomal compartments, and similar to Cln8 (Ceroid-lipofuscinosis, neuronal 8) and ATP6v1b2 (ATPase, hydrogen-transporting, lysosomal, V1 subunit B, isoform 2) genes, it is involved in the trafficking of receptors and signaling molecules through the vesicular/endocytic pathway. Considering all of these data we believe that the immunosuppressive function of allochimeric molecule may depend on the impairment of T cell motility and scanning ability, T cells adhesion with APCs, and the proper TCR clustering, and possibly also the formation of immunological synapse. Another genes down regulated in allochimeric molecule treatment are Catna1, Vcam and CD9, which are involved in intercellular adhesion, interaction with cytoskeleton, signaling and regulation of inflammatory response. Vcam-1 (vascular cell adhesion molecule-1 also known as CD106) is a membrane protein and an integrin ligand, which mediates leukocyte-endothelial cell adhesion and signal transduction, and is an important factor in the initiation of an inflammatory response. Indeed, conditional Vcam-1 mutant mice have an impaired immune response , and the over expression of Vcam-1 is associated with several chronic inflammatory diseases. In addition, VCAM-1 is linked to actin cytoskeleton in immunological synapse. Catna1 (catenin alfa 1; cadherin- associated protein; Ctnna1) is an intracellular protein that associates with cadherins and cytoskeleton and is required for the formation and maintenance of functional intercellular adhesion complex. CD9 is a member of the tetraspanin superfamily (which also includes MHC molecules). Tetraspanins are known to be involved in cell adhesion, motility, and cell differentiation and activation. CD9, which is expressed on most mature and naïve T cells, associates with various cell surface molecules including co-stimulatory CD5 molecule. We suggest that the down regulation of these three genes in allochimeric treated animals maybe responsible for the impairment in T cell/APC adhesion, T cell motility and immune response. Allochimeric molecule treatment also causes a down regulation of Mitogen- activated protein kinases (MAPK) pathway genes Dusp6 and Spred1. The Dusp6 (Dual specificity phosphatase 6) is a phosphatase, which inactivates members of the mitogen-activated protein (MAP) kinase superfamily (MAPK/ERK, SAPK/JNK), which are associated with cellular proliferation and differentiation. Recent studies on T cells indicate that Dusp5 and DuspP6 are induced by IL-2. A novel gene Spred-1 (Sproutyrelated Ena/VASP homology 1-domain-containing protein-1) is involved in the inhibition of the Ras/Raf-1/ERK pathway impairing the growth- factor-mediated activation of ERK1/2 kinases. Studies of Spred1 expression in T cells showed that Spred-1 is highly up regulated in the tumor-infiltrating CD8⁺ T cells and that TGF-β modulates the infiltrating function of CD8⁺ T cells via TCR signaling and Spred1 expression. In conclusion, the T cells gene expression changes described in our studies indicate multiple protective influences of allochimeric peptide and will provide basis for further studies aimed at understanding of pathways and mechanisms involved in the immunosuppression and heart graft survival. # Materials and Methods ## Animals Adult male inbred Wistar Furth (WF; RT1.Au) and ACI (RT1.Aa) rats (180–250 gm) were purchased from Harlan Sprague Dawely (Indianapolis, IN). Heterotopic cardiac transplants were placed intra-abdominally as described previously. There were three experimental groups: 1. Transplantation control group without any treatment, 2. Transplantation in the presence of sub-therapeutic dose of CsA (acute rejection) which consisted of 3 day course of oral cyclosporine delivered by gavage feed (CsA, 10 mg/kg/day; day 0–2). Transplantation in the presence of sub-therapeutic dose of CsA supplemented with allochimeric molecule (CsA + peptide). Allochimeric peptide \[α1h1/u\]-RT1.Aa (GenWay, San Diego CA; 1 mg/kg) was delivered through the portal vein into ACI recipients of WF hearts at the time of transplantation in addition to a 3 days course of oral cyclosporine delivered by gavage feed (CsA, 10 mg/kg/day; day 0–2). All experiments were performed according to the The Methodist Hospital Research Institute animal care and use NIH standards as set forth in the “Guide for the Care and Use of Laboratory Animals” (DHHS publication No. (NIH) 85–23 Revised 1985). The Institution also accepts as mandatory the PHS “Policy on Humane Care and Use of Laboratory Animals” and NIH “Principles for the Utilization and Care of Vertebrate Animals Used in Testing, Research and Training. ## T cells Isolation and FACS Analysis Spleens from ACI host rats were harvested at 1, 3 and 7 days post- transplantation. Cell suspension was made by passing spleen through a cell strainer using 3cc syringe. Cells were treated with lysing reagent (Becton Dickenson) to remove the red cells and then washed twice with complete media (10%Fcs/1640RPMI). T cell population was purified via a positive T cell isolation kit using magnetic anti-T cell micro beads (Miltenyl Biotech) and purity of T cells was confirmed by FACS analysis i.e. T cells were stained with BD Pharmingen (Franklin Lakes, NJ) reagents including FITC-conjugated mouse anti- rat CD3 and CD4 antibodies for 15 minutes at the room temperature and then washed three times with PBS and analyzed using FACScan flow cytometer (Becton Dickenson, San Jose, CA). ## RNA Isolation For RNA isolation isolated T cells were pelleted, immediately placed in RNA Later (Applied Biosystems/Ambion, Austin, TX) and after overnight infiltration at 4°C they were kept in RNA Later at −70°C until the isolation of total RNA. T cell pellets were homogenized using a TissueLyser (Qiagen, Valencia, CA) and RNA was isolated using the RNeasy Mini Kit (Qiagen, Valencia, CA) according to the manufacturer's recommendations. The quantity, purity and integrity of RNA were evaluated using a NanoDrop ND-1000 spectrophotometer (Nanodrop Technologies, Wilmington, DE) and an Agilent Bioanalyzer (Agilent Technologies, Palo Alto, CA). ## Quantitative RT PCR Total RNA was isolated from purified spleenic T cells using RiboPure kit (Applied Biosciences, Foster City, CA). Complementary DNA (cDNA) was made using High Capacity cDNA Reverse Transcription Kit (Applied Biosciences, Foster City, CA). The cDNA was used to determine the expression levels of house keeping actin β (ACTβ) gene and gene of interest. Total RNA (1000 ng) was reverse transcribed (RT) and PCR amplified using the High Capacity cDNA Reverse Transcription Kits (Applied Biosystems) according to the manufacturer's protocol. The RT reaction consisted of a 10 min incubation at 25°C, 120 min incubation at 37°C, followed by a 5 min 85°C termination step, and the resulting complementary DNA (cDNA) was stored at − 20°C. For Real time PCR amplification samples were run in duplicate and only 1 gene was analysed per reaction. The cDNA template reaction contained Assay On Demand Gene Expression primers (see below; Applied Biosystems, Foster City, CA) and TaqMan® Fast Universal PCR no AmpErase UNG master mix (Applied Biosystems, Foster City, CA). Reactions were heated to 95°C for 10 min, followed by 40 amplification cycles of 95°C for 15 s and 60°C for 1 min using Applied Biosystems 7500 Standard System. The amount of target mRNA relative to housekeping gene mRNA was expressed as fold incerase /decrease. Relative changes were measured using real time PCR in the 7500 Fast or Standard Real Time PCR System (version 1.3.1). To calculate the relative quantity (RQ) of particular gene, 2-delta delta ct method implemented in the software was used. The data are presented as the fold change (RQ values) in gene expression level normalized to endogenous reference gene. ## PCR Primers All PCR primers were purchased from SABiosciences (Frederick, MD, USA): Spred 1, cat \#PPR55168; Vcam, cat#PPR45334; RhoA, cat# PPR56555A; RhoB, cat# PPR42656A; Rac, cat \#PPR48291A; CD9, cat# PPR42666A; Catna, cat# PPR48584A; Cln8, cat# PPR42624A; ATP6v1b2, cat# PPR44008A; Dusp6, cat# PPR43415. ## Microarray The microarray hybridization and analysis were performed by Cogenics (Morrisville, NC) according to the manufacturer's protocol (Affymetrix, Santa Clara, CA). Total RNA from each sample was converted to double-stranded cDNA with the BioarrayTM Single-Round RNA Amplification and Labeling Kit (Enzo Life Sciences), Double-stranded DNA purified with Purification Kit (Enzo Life Sciences was ttranscribed in vitro using the BioarrayTM HighYieldTM RNA Transcript Labeling Kit (Enzo Life Sciences). For each sample, biotinylated cRNA spiked with bioB, bioC, bioD and cre (Hybridization Control) was hybridized to an Affymetrix Rat 230 2.0 Microarray for 16 hours at 45°C. After hybridization, washing and staining, arrays were scanned with an Affymetrix GeneChip® Scanner 3000. Quality checks and data analyses were performed using Affymetrix GeneChip Operating Software (GCOS) and Expression Console. All microarray data reported in the manuscript are described in accordance with MIAME guidelines. ## Microarray Data Analysis The processed image file of the Affymetrix Rat 230 2.0 array contains over 31,000 probe sets representing approximately 28,700 well-substantiated rat genes (Affymetrix; [www.affymetrix.com](http://www.affymetrix.com)). The probe set level data were analyzed with either Rosetta Resolver Gene Expression Data Analysis System (Rosetta, Seattle, WA) or Partek Genomics Suite, version 6.4, build 6.09.0129 (Partek, St. Louis, MO). The criteria for identification of differentially expressed transcripts was a log ratio p-value\<0.001, an absolute fold change greater than 1.5, and a log (10) intensity measurement \>−1.6. Enrichment of biological pathways for differentially expressed genes was determined using Pathway Express. The microarray data used in this study were deposited in the National Center for Biotechnology Information (NCBI) Gene Expression Omnibus (GEO) with the accession number GSE15074 ([www.ncbi.nlm.nih.gov/projects/geo](http://www.ncbi.nlm.nih.gov/projects/geo)). # Supporting Information [^1]: Conceived and designed the experiments: WL MK RMG. Performed the experiments: NT TSS MG. Analyzed the data: WL EGB MK RMG. Wrote the paper: MK. Performed heart transplants in rats: YG. [^2]: Dr. Eric Bremer is employed and funded by a salary from Precision Biomarkers, Inc. Dr. Bremer had a role in data analysis.

# Introduction Cell migration is essential for normal embryonic development, wound healing, and immunity but can be devastating in tumor invasion and metastasis. Netrins are secreted, laminin-related proteins that direct cell and axon migration during neural development (reviewed by). Netrin-1 and netrin receptors DCC, the DCC paralogue neogenin, and UNC5 proteins, are also expressed in many adult tissues, but their function in mature tissues is poorly understood. Netrin-1 is widely expressed by neurons and glia in the adult CNS. Reduced expression of netrin-1 has been documented in brain tumors, including glioblastoma, however, a role for netrins regulating brain tumor cell migration has not been established. Although substantial evidence suggests an anti-oncogenic role for DCC, how disruption of netrin signaling might contribute to malignancy is poorly understood. In colorectal cancer, allelic deletion involving chromosome 18q21 occurs in \>70% of tumors and the *dcc* gene was first identified as a putative tumor suppressor from this chromosomal deletion. *Dcc* expression is reduced in many cancers, including most high-grade gliomas and loss of DCC correlates with the development of highly invasive glioblastoma multiformae. Furthermore, ectopic expression of *dcc* in transformed epithelial cells reduced tumorigenicity, and expression of DCC antisense RNA in transformed fibroblasts resulted in an increased growth rate, anchorage independence, and tumorigenicity when the cells were transplanted into nude mice. No increased incidence of tumor formation has been detected in conventional DCC knockout mice, however, conclusions drawn from this study were complicated by the possibility that tumors may not have had time to develop due to the early post-natal lethality of DCC knockouts. Unc5 homologue netrin receptors signal chemorepulsion, and co- expression of DCC often facilitates UNC5 function (reviewed by). Four UNC5s, UNC5A-D, are expressed in mammals. Altered expression of UNC5A, B, C, and D has been detected in various cancers and tumor cell lines. Here we investigated the possibility that netrins and netrin receptors influence tumor cell migration. Using human glioblastoma cell lines, we provide evidence that DCC is required for chemoattraction to netrin-1 and slows the rate of spontaneous cell migration. Our findings support a role for netrins as autocrine inhibitors of cell motility that regulate focal adhesions (FA). # Results ## Glioblastoma cells express netrin and netrin receptors To determine if netrins regulate glioblastoma cell migration, we first characterized netrin and netrin receptor expression in human astrocytoma cell lines U87, U343, and U373, and in cultures of astrocytes isolated from newborn rat cortex. Western blot analysis using an antibody that binds netrin-1 and netrin-3 detected a ∼75 kDa band corresponding to full-length netrin in conditioned medium collected from all cells tested. The DCC_IN monoclonal antibody detected a ∼185 kDa band corresponding to DCC in astrocyte and U87 cell lysates. In contrast, DCC was not detected in lysates of U343 or U373 cells. The DCC homologue neogenin was expressed by astrocytes and was detected in all glioblastoma cell lysates. RT-PCR revealed *dcc* expression by U87 cells but not U343 or U373 cells, and *neogenin* and *unc5* expression by all three cell types. U87 and U343 cells express *unc5b and c*, and U373 cells express *unc5b, c*, and *d*. *Netrin-1* expression was detected in U343 and U373 cells, and *netrin-3* expression in U87 cells. Netrin-1 and netrin-3 are essentially functionally equivalent: both bind DCC and UNC5 proteins and evoke chemoattractant or chemorepellent responses from responsive cells. We then sought to determine if netrin might exert an autocrine influence on cell migration. We first assessed the relative motility of the three cell lines using a transfilter chemotaxis assay as described. Briefly, cells were cultured on the upper surface of a porous membrane and allowed to migrate across. Following migration, cells remaining on the upper surface of the membrane were scraped off, and the cells that migrated to the underside were fixed, stained, and counted. While this assay is often employed to assess the migration of cells in response to a putative attractant or repellent cue, here we used it in the absence of added factors to compare the relative rates of spontaneous migration of the three glioblastoma lines. U343 and U373 cells, lacking DCC, migrated significantly faster than DCC-expressing U87 cells. Notably, the U343 cells, which were originally derived from a grade IV glioblastoma multiformae, migrated significantly faster than either the U87 or U373 cells, both of which were originally derived from less aggressive grade III astrocytomas. ## Autocrine netrin inhibits U87 cell motility We hypothesized that DCC and netrin expressed by U87 cells might exert a kinetic influence on the rate of cell movement, independent of netrin's influence on directional migration. We therefore tested the effect of disrupting DCC and netrin function on the spontaneous rate of U87 cell migration. The rate of spontaneous migration was not affected by addition of a DCC function-blocking antibody (DCC_FB). In order to disrupt autocrine netrin function, netrin function-blocking antibody (Net_FB) was added to both the top and bottom compartments of the migration chamber. This resulted in an approximately 10-fold increase in spontaneous migration across the filter relative to the number of cells migrating in either medium alone (Control) or in the presence of a control IgG. ## Autocrine netrin-1 inhibits migration of U373, but not U343, glioblastoma cells Netrin's capacity to inhibit U87 cell motility in a DCC-independent manner led us to determine if a similar mechanism was active in U343 or U373 cells, which do not express DCC. The addition of netrin function-blocking antibody to both the top and bottom compartments of the transfilter assay significantly increased the rate of U373 migration , indicating that endogenous netrin-1 inhibits the rate of U373 migration. Unlike U87 and U373 cells, blocking netrin function did not alter the rate of U343 cell migration. Although U343 cells express neogenin and UNC5 homologue netrin receptors, the absence of an increase in the rate of migration may be the result of mechanisms that restrain inappropriate cell motility being more severely disrupted in these cells. ## Netrin-1 is a chemotropic attractant for U87 glioblastoma cells Transfilter migration assays were then used to determine if DCC-expressing U87 cells can respond to a gradient of netrin-1 as a chemoattractant. Addition of 100 ng/ml netrin-1 to the bottom compartment of the migration chamber (NB) produced a significant increase in the number of U87 cells that migrated across the membrane relative to control (medium alone:, 16 hr assay;, 48 hr assay). In contrast, when netrin-1 was added to both the top and bottom compartments (NTB), migration was not significantly different from control. This indicates that U87 cells respond to a gradient of netrin-1 as a chemotropic attractant. When challenged with a gradient of netrin-1 whilst in the presence of the DCC_FB antibody in the top and bottom wells (NB DCC_FB), U87 cell migration was not significantly different from control, indicating that the tropic response of U87 cells to netrin-1 requires DCC. Neither U343 nor U373 cells, which do not express DCC, altered their migration in response to a gradient of netrin-1, despite expressing neogenin and UNC5 netrin receptors. These findings suggest that although these receptors may be sufficient to mediate autocrine inhibition of migration, they are insufficient for these cells to generate a chemotropic response to a gradient of netrin-1. ## Chemoattractant response of DCC-expressing U343 and U373 cells to a gradient of netrin-1 To further investigate the contribution of DCC to the regulation of cell motility, we reintroduced the *dcc* gene back into U343 and U373 cells by transfection with a cDNA encoding a DCC-GFP chimera (pDCC-GFP, described by). The proportion of cells expressing DCC-GFP was increased by passaging the cells with Geneticin selection, such that the vast majority of cells seeded in the migration assays expressed DCC. Expression of DCC by U343 and U373 cells was confirmed by western blot. Unlike the parental U343 and U373 cell lines, DCC- GFP-expressing U343D and U373D cells migrated up a gradient of netrin-1, indicating that ectopic expression of DCC now rendered these cells competent to generate a chemotropic response to netrin-1. Like DCC-expressing U87 cells, the gain-of-function migration towards netrin-1 exhibited characteristics of chemotropic attraction, as the cells only responded to a gradient. Uniform presentation of exogenous netrin-1 resulted in migration that was not significantly different from control. The DCC_FB antibody blocked the chemoattractant response of U343D and U373D cells, indicating that DCC is required for chemoattraction to netrin-1. Consistent with the slow migration of DCC-expressing U87 cells, the number of DCC-transfected U343 and U373 cells that migrated under control conditions was substantially reduced relative to that of the parental cells. These findings suggest that DCC expression decreases the motility of these cells; however, application of the DCC function-blocking antibody (DCC_FB) did not increase the rate of migration, as was also found for the U87 cells. In constrast, DCC_FB completely blocked the chemoattractant migratory response of the U87 cells, and the DCC-transfected U343 and U373 cells to a gradient of netrin-1. These findings are consistent with DCC activating a mechanism that slows non-directional cell migration; however, this mechanism can be differentiated from DCC-dependent chemoattraction due to its insensitivity to DCC_FB. ## Chemoattraction to netrin-1 is converted to repulsion by laminin-1 Laminin-1 exerts a neuromodulatory influence that converts the response of *Xenopus* retinal ganglion cell growth cones to netrin-1 from attraction to repulsion. We therefore investigated the possibility that laminin-1 might influence the migratory response of U87 cells to a gradient of netrin-1. When U87 cells were challenged with an ascending gradient of laminin-1 (LB), the number of cells that migrated across the membrane increased. In the presence of a uniform concentration of laminin (LTB), U87 migration was not significantly different from control, indicating that a gradient of laminin-1, like netrin-1, is a chemoattractant for these cells. Interestingly, the combination of an ascending gradient of netrin-1 and a uniform concentration of laminin-1 (LTB NB) dramatically reduced the number of U87 cells that migrated across the membrane to the extent that it was significantly less than control, suggesting that laminin-1 converted the reponse to netrin-1 from attraction to repulsion. Consistent with this, confronting cells with a descending netrin-1 gradient in the presence of a uniform concentration of laminin-1 (LTB NT) resulted in an increase in migration relative to control. Importantly, this finding provides strong evidence that laminin-1 does not influence the response to a gradient of netrin-1 by arresting cell motility. When the cells were simultaneously exposed to uniform concentrations of netrin-1 and laminin-1, (LTB NTB), fewer cells migrated across the membrane, indicating that the combined action of netrin-1 and laminin-1 exerts a non-directional effect that inhibits U87 cell motility. These results are consistent with laminin-1 switching netrin-1 from an attractant to a repellent for U87 cells, as previously described for the axons of *Xenopus* retinal ganglion cells. Addition of DCC_FB antibody in the presence of a uniform concentration of laminin-1 and either an increasing gradient (LTB NB DCC_FB) or uniform concentration (LTB NTB DCC_FB) of netrin-1, did not significantly affect migration compared to control, indicating that the laminin-induced repellent response to netrin-1 requires DCC. This is consistent with a requirement for DCC in chemorepellent responses to netrin-1, documented in many cell types, including glial precursor cells. ## Netrin-1 and DCC do not affect U87, U343, and U373 cell proliferation or survival DCC and UNC5 homologues have been proposed to function as dependence receptors, activating apoptosis in the absence of netrin-1. This raises the possibility that the effects described above may be due to an influence on cell survival and not motility. Thus, we examined the consequences of manipulating netrin function on the survival of U87, U343 and U373 cells. No significant change in cell number, or activation of caspase-3, an indicator of apoptosis, was detected following 16 hrs treatment with exogenous netrin-1, laminin-1, or both; nor following disruption of netrin or DCC function using blocking antibodies. Further testing, by blocking netrin and DCC function for 48 hrs, again resulted in no detectable increase in caspase-3 activation. In contrast, staurosporine, applied as a positive control, activated caspase-3 and caused extensive cell death. These findings are consistent with previous analyses of glial precursor cells, indicating that netrin-1 and DCC do not regulate oligodendroctye precursor survival either *in vitro* or *in vivo*, and they support the conclusion that the results of transfilter assays reflect changes in cell migration and not effects on cell survival or proliferation. ## Endogenous netrin promotes the maturation of focal complexes into focal adhesions Cell migration requires the formation of transient adhesive contacts with the extracellular matrix (ECM). Initial contacts occur at the leading edge of lamellipodia where integrins bind ECM ligands and recruit proteins such as vinculin and paxillin to form immature adhesive contacts called focal complexes (FC) (reviewed by). The transition from FC to FA is marked by consolidation of the adhesive contact, an increase in size, and the recruitment of additional proteins including zyxin. The effect of disrupting netrin function on adhesive complex formation in glioblastoma cells was investigated by examining the distribution of paxillin, which is present in both FAs and FCs, and zyxin, which is present in FAs but not FCs. The influence of netrin on FC formation was quantified by subtracting the distribution of zyxin immunoreactivity from paxillin immunoreactivity to create images representing regions of paxillin, but not zyxin localization. Using the ‘paxillin minus zyxin’ images, the density of FCs present in each lamellipodium was calculated. Exposure of U87 cells to an isotype control antibody (RbIgG), DCC_FB, or netrin-1, resulted in no change relative to control. In contrast, application of Net_FB resulted in increased FC density. A similar increase was observed when netrin function was inhibited in U373 cells, but not U343 cells, in which FC density was high in all conditions examined ( and, data not shown). To determine if inhibiting endogenous netrin function influences FA density, images depicting regions of paxillin and zyxin colocalization were generated. From the ‘paxillin and zyxin’ images, the density of FAs in each lamellipodium was calculated. In U87 and U373 cells, addition of netrin function-blocking antibody resulted in decreased FA density. In all other conditions analyzed for U87 and U373 cells and in all conditions analyzed for U343 cells, no change in FA density was observed. Notably, the increase in FC density and corresponding decrease in FAs correlates precisely with the changes in motility evoked by disrupting netrin function and measured using the microchemotaxis assay. The absence of an effect of the DCC function-blocking antibody provides evidence that DCC function is not essential for this non-directional effect of netrin on motility, consistent with the lack of an effect of disrupting DCC function on the rate of cell migration. That the addition of exogenous netrin-1 protein did not influence FC or FA density suggests that the relatively high level of netrin protein secreted by the cells is sufficient to saturate the inhibitory response in the conditions described here. These data are consistent with a mechanism in which autocrine secretion of netrin promotes the maturation of FCs into FAs, and that the accumulation of these adhesive structures restrains cell movement. ## Netrin and netrin receptors are localized to focal adhesions, but not focal complexes We next investigated the possibility that netrin and netrin receptors might be localized to FCs or FAs and thereby directly influence their maturation. U87, U343, and U373 cells were labeled with antibodies against paxillin and one of DCC, netrin, or UNC5 proteins. U87 cells were also labeled with anti-DCC and anti-zyxin. In U87 cells netrin , DCC and UNC5 immunoreactivity colocalized with large paxillin-positive foci characteristic of FAs (white arrowhead), but not smaller paxillin-positive structures characteristic of FCs (black arrowhead). In U343 and U373 cells that lack DCC expression, netrin and UNC5 immunoreactivity was similarly localized to FAs but not FCs. Consistent with localization to FAs, DCC and zyxin immunoreactivity colocalized in U87 cells. Colocalization with markers of FAs is consistent with netrins and netrin receptors regulating cell- substrate adhesion and motility. # Discussion Here we provide evidence that secreted netrins can function as autocrine inhibitors of cell motility. Our findings support the conclusion that DCC is required for cells to migrate directionally in response to a gradient of netrin-1. Ectopic DCC expression conferred on U343 and U373 cells the capacity to respond to a gradient of netrin-1. DCC expression also slowed the rate of spontaneous migration in these cells, consistent with DCC restraining cell movement. The glioblastoma-derived cell lines tested express either *netrin-1* (U343, U373) or *netrin-3* (U87). Disrupting endogenous netrin-1 or netrin-3 function dramatically increased the rate of U87 and U373 cell movement. U87 cells express DCC while U373 cells do not, indicating that in addition to DCC slowing cell migration, netrins must influence the motility of these cells through a DCC-independent mechanism. Unc5 homologue netrin receptors are required for axonal growth cone repulsion and collapse induced by netrin-1, and co-expression of DCC often facilitates UNC5 function (reviewed by). Our findings support the hypothesis that UNC5 proteins, in collaboration with DCC, underlie the netrin-mediated inhibition of motility described here; however the role of UNC5 homologues in these cells remains to be tested directly. Consistent with increasing the rate of cell motility, disrupting endogenous netrin function increased the number of lammelipodial FCs, immature adhesive contacts that are associated with cell movement. Netrin, DCC, and UNC5 immunoreactivity was co-localized with FA but not FC markers, suggesting that netrin may act at the nascent FA itself to promote the maturation of FCs to FAs. ## Netrin, focal adhesions, and cell motility Netrin-1 signaling through DCC directs the organization of F-actin, regulating the activation of RhoGTPases, PAK1, MAPK, FAK, and Src family kinases. FAK and Src are also activated downstream of UNC5 proteins in response to netrin. FAs are sites of interaction for many proteins. Our evidence indicates that netrin and netrin receptors are localized to FAs. We hypothesize that netrins may contribute to restricting cell movement by promoting FA maturation. Numerous proteins present in FAs have been implicated in signaling downstream of netrin: FAK, Src, the Ena/VASP proteins, Rho-family GTPases Cdc42, Rac, and RhoA and the GEF Trio. FAK is activated by autophosphorylation that creates a binding site for Src-family kinases. Association with FAK initiates a FAK-Src signaling complex. Extensive tyrosine phosphorylation is a key signaling event observed in focal adhesions, as it is thought to create ‘docking’ sites for recruitment of SH2 domain-containing proteins required for further signaling events (reviewed by). FAK and Src regulate the phosphorylation of UNC5B on multiple tyrosine residues upon netrin binding, and that following these phosphorylation events, Src associates directly with UNC5B via its SH2 domain. Interestingly, this is enhanced by, but does not absolutely require, DCC function, perhaps reflecting that co-expression of DCC can facilitate UNC5 function (reviewed by). Notably, FAK is required for the maturation of adhesive complexes, and, together with Src, is essential for the normal turnover of FAs (reviewed by). The findings we present here provide a foundation for investigating the role of netrin-1 in the formation of focal adhesions. ## An emerging role for netrin in adhesion and tissue morphogenesis Netrin-1 and netrin-3 are secreted proteins, which raises the question of how they may contribute to anchoring a cell to either the substrate or another cell. The majority of netrin-1 protein in the CNS is not freely diffusible, but is bound to cell surfaces and extracellular matrix (reviewed by). DCC binding immobilized netrin-1 mediates cell-substrate adhesion, consistent with a role for netrin mediating cell-matrix interactions. Key roles for netrins and netrin receptors have been identified during tissue morphogenesis (reviewed by), including development of the mammary epithelium, pancreas, lung, lymphangiogenesis, and during angiogenesis. Furthermore, overexpression of netrin-1 by cells in the intestinal epithelium of mice led to the formation of focal hyperplasias and adenomas. These authors concluded that the phenotype induced was due to netrin-1 reducing cell death; however, our findings raise the possibility that disruption of appropriate cell-cell interactions as a result of netrin-1 overexpression may contribute to the disorganization of normal epithelial structure. ## How might secreted netrins influence tumor cell migration in vivo? Our findings suggest that loss of netrin function may lead to disruption of appropriate cell-cell and cell-matrix interactions. We have provided evidence that in the presence of laminin-1, netrin-1 becomes a repellent for U87 cell migration, and that this requires DCC. Importantly, the combined action of netrin-1 and laminin-1 may influence glioblastoma cell migration *in vivo*. Laminin-1 is restricted to basement membranes and capillary walls in developing and mature CNS. Although deregulated cell migration makes an important contribution to the dissemination of tumor cells within the brain, metastasis of brain tumor cells outside the CNS is rare. Glioma cells are attracted to endothelial capillaries *in vitro* and glioblastoma cells migrate in close association with capillary walls as they disseminate within the brain. Laminin-1 may facilitate this as it promotes glioma cell migration. Based on our evidence that laminin-1 biases cells to respond to netrin-1 as a repellent (see also), the basal lamina may inhibit the migration of glioma cells expressing netrin-1 and DCC. In contrast, in the absence of netrin function, our findings predict that deregulation of this inhibition of migration will lead to laminin-1 in the basal lamina of blood vessels becoming a permissive substrate that promotes tumor cell migration and dissemination to other brain regions. Correlated loss of DCC expression with tumor progression suggests that netrin and DCC may play an important role in tissue maintenance in adulthood. We propose that appropriate cellular organization may be stabilized by autocrine and paracrine actions of netrin. Our findings suggest that loss of effective netrin signaling may disinhibit a mechanism that normally restrains cell migration. In the absence of netrin-mediated inhibition, local cues such as laminin, are predicted to become potent promoters of migration. Numerous cell types expressing both netrin and netrin receptors *in vivo* have been described. We provide evidence that autocrine expression of netrin can restrain cell migration, and promote the maturation of focal complexes into focal adhesions. These findings identify a novel netrin function that may contribute to the formation and maintenance of tissue organization, and identify netrin and its receptors as potential therapeutic targets to inhibit tumor cell migration and dispersion. # Materials and Methods ## Cells and cell culture Human glioblastoma cell lines, U87, U343, U373 (ATCC, Rockville, MD) and astrocytes isolated from newborn mouse brain were grown as monolayer cultures in DMEM (Invitrogen, Burlington, ON), 10% heat-inactivated fetal bovine serum (FBS, Invitrogen), 1% glutamax-1 (Invitrogen) and 1% penicillin/streptomycin. All procedures using animals were performed in accordance with the Canadian Council on Animal Care guidelines for the use of animals in research, and were approved by the animal care review board of the Montreal Neurological Institute (approval ID \# 4330). ## Antibodies, conditioned media, cell lysates, western blotting, and PCR Antibodies against the following were used: cleaved caspase-3 (Asp175, mouse, Cell Signaling Technology, Beverly, MA); DCC (DCC_IN, mouse, G97-449; BD Biosciences Pharmingen, San Jose, CA; DCC_GT, goat, A-20; Santa Cruz Biotechnology, Santa Cruz, CA; function-blocking, DCC_FB, mouse, AF5; Calbiochem, La Jolla, CA); netrin-1 and 3 (PN2, rabbit); netrin function- blocking (Net_FB, PN3, rabbit); neogenin (rabbit, Santa Cruz Biotechnology); paxillin (mouse, BD Biosciences Pharmingen); Unc5C (rabbit, provided by Dr. Tony Pawson, Mount Sinai Hospital, Toronto, ON; isotype control rabbit IgGs (RbIgG; Invitrogen); and zyxin (rabbit, Abcam, Cambridge, MA). Cultures were grown to 80% confluence and conditioned media collected following 48 hrs in serum-free DMEM. For lysates, cells were grown to 80% confluence, rinsed with PBS and lysed in 1 ml of hot sample buffer (60 mM Tris/HCl, pH 6.8, 2% SDS, 10% glycerol, 100 mM DTT). For western blot analysis of cleaved caspase-3, cells were cultured at a density of 120,000 cells/well in a 12-well tissue culture plate. Nitrocellulose immunoblots were probed with DCC_IN (0.5 µg/ml), PN2 anti-netrin (4 µg/ml), anti-cleaved caspase-3 (1∶1000), or anti-neogenin (0.4 µg/ml). After washing, membranes were incubated with HRP-coupled secondary antibodies and immunoreactivity visualized using chemiluminescence (NEN, MA). PCR was carried out using standard methods. Total RNA was isolated from glioblastoma cells using Trizol (Life Technologies, MD) and cDNA prepared using SuperScript II reverse transcriptase (Invitrogen). Primers were annealed at 60°C to amplify *netrin-1* (527 bp), *netrin-3* (429 bp) and *dcc* (434 bp), 58°C for *neogenin* (545 bp) and *unc5C* (530 bp), and 55°C for *unc5A* (215 bp), *unc5B* (350 bp), and *unc5D* (324 bp). human *netrin-1* F: 5′GCCGCCACTGCCATTACTGC 3′ R: 5′GAGGGGCTTGATTTTGGGACACTT 3′, human *netrin-3* F: 5′CCGCTGGGCTTCTTCTCC 3′ R: 5′GCAGCGGCCGCAGTCAGG3′, human *dcc* F: 5′CAAGTGCCCCGCCTCAGAACG 3′ R: 5′GCTCCCAACGCCATAACCGATAAT 3′, human *neogenin* F: 5′ TGGCCCAGCACCTAACCT 3′ R: 5′TTGCCGGGCCTGTACCATTGATTG 3′ human *unc5a* F: 5′ TCGTCAAGAACAAGCCAGTG 3′ R: 5′ GCACTGGCACCAGTATTC 3′ human *unc5b* F: 5′ TCCAGCTGCATACCACTCTG 3′ R: 5′ AGCCACCAGCATCTCACTCT 3′ human *unc5c* F: 5′ GCCAGCAAGTGGAAGAACTC 3′ R: 5′ CACACTCTGCCCTTCACAGA 3′ human *unc5d* F: 5′ ATATTCCCCCATTCCTCTGG 3′ R: 5′ TAGCACAAATCCGCTGTCGTCTG 3′ ## Transfilter chemotaxis assay Cells were plated at a density of 4×10⁵ cells/ml on polycarbonate transwell culture inserts (6.5 mm diameter with 8 µm pore size, Corning). 100 µl of cell suspension was applied to the upper surface of the filter, and the filters placed in the wells of a 24-well plate over 600 µl of medium. DMEM with 0.2% BSA, 100 U/ml penicillin, 100 µg/ml streptomycin, and 2 mM glutamax was the base medium used for all assay conditions. Following migration, cells on the upper side of the filter were scraped off, and the cells attached to the lower side of the filter fixed with 4% paraformaldehyde (PFA)/0.1% glutaraldehyde (30 min, 4°C). Filters were rinsed with PBS, and cell nuclei stained with Hoechst dye. Four transwells were used per condition. Four images of each filter were captured using a 10 X objective and nuclei counted using Northern Eclipse software (Empix Imaging, TO). Where pooled results are presented, the value ‘percent migration vs. control' for a given trial represents the number of cells migrated in that condition expressed as a percentage of the mean number of cells migrating in control conditions. Recombinant netrin-1 protein was purified as described and used at a concentration of 100 ng/ml. Laminin-1 was used at 10 µg/ml (BD Biosciences, Bedford, MA). Net_FB and rabbit isotype control IgG (used as a control) were added at a concentration of 25 µg/ml. DCC_FB was added at a concentration of 10 µg/ml. Statistical significance was calculated using Student's t-test and error bars represent S.E.M. ## Plasmids and transfection U343 and U373 cells were transfected using lipofectamine (Invitrogen) with expression constructs encoding either green fluorescent protein (GFP) alone or DCC tagged at its C-terminus with GFP. Seventy-two hrs after transfection, the medium was changed to selection medium containing Geneticin (Invitrogen). ## Confocal image analysis 10⁴ cells were plated per well in chamber slides (Fisher) coated with 20 µg/ml poly-D-lysine (Sigma) at 4°C overnight, washed with Hanks buffered salt solution (Invitrogen) and allowed to dry. Cells were fixed in 4% PFA, 4% sucrose in PBS, and permeabilized with 0.25% Triton X-100 in PBS. Blocking was performed in 3% heat-inactivated normal goat serum, 2% BSA, and 0.125% Triton X-100 in PBS. Cells were then incubated with anti-paxillin and anti-zyxin, anti-paxillin and one of anti-netrin PN2, anti-UNC5, or anti-DCC_GT, or anti-zyxin and anti-DCC_GT diluted in blocking solution. Primary antibodies were detected with secondary antibodies coupled to Alexa 546 or Alexa 488 (Molecular Probes). For imaging adhesive complexes, single confocal optical slices through the base of lamellipodia were collected. Identical settings were used for each condition examined for a given cell line. The outermost region of individual lamellipodial protrusions (excluding regions of paxillin or zyxin immunoreactivity contiguous with adhesive structures in the cell body) was outlined using Image J software. Mask images were then generated representing either the regions staining with both paxillin and zyxin (using the ‘AND’ function) or representing the difference between the paxillin and zyxin images (the zyxin signal subtracted from the paxillin image). Images were adjusted to eliminate signal below a minimum value that was held constant across all images for each cell line. Signal in the ‘AND’ image corresponds to the area of each lamellipodium occupied by mature FAs that contain both paxillin and zyxin. To quantify FCs, the subtracted image representing paxillin staining without zyxin was filtered to exclude structures smaller than 3 pixels or larger than 40 pixels. The number of individual adhesions was then counted and the density of adhesions within each lamellipodium calculated. Netrin-1 was used at 100 ng/ml, Net_FB and rabbit preimmune IgG (as a control) at 25 µg/ml, and DCC_FB at 10 µg/ml. ## Analysis of cell number and apoptosis To investigate changes in cell survival or proliferation, cells were plated at a density of 30,000 cells per well in 8-well chamber slides (Fisher), allowed to settle for 2 hrs, treated as described for 16 hrs, fixed and stained with Alexa-488 conjugated phalloidin and Hoechst, and the number of live cells per 20 X field counted. To measure levels of apoptosis in these cultures, 120,000 cells were cultured in each well of a 12-well tissue culture plate, allowed to settle for 2 hrs, and treated as described for either 16 or 48 hrs. In all cases, the base medium used was DMEM with 2% FBS, penicillin/streptomycin, and glutamax-1. Statistical significance was calculated using Student's t-test and error bars represent S.E.M. We thank Phil Barker, Adriana Di Polo, Morag Park, Katherine Horn, Sonia Rodrigues, Simon Moore, Veronica Miron, and Sathy Rajasekharan for comments on the manuscript. [^1]: Conceived and designed the experiments: AAJ MD TLL TEK. Performed the experiments: AAJ MD TLL NM MS. Analyzed the data: AAJ MD TLL NM TEK. Wrote the paper: AAJ TEK. [^2]: Current address: Edinburgh Centre for Multiple Sclerosis Research, Scottish Centre for Regenerative Medicine, The University of Edinburgh, Edinburgh, United Kingdom [^3]: The authors have declared that no competing interests exist. This research was funded in part by a research grant from Biochem Pharma Inc. that partially supported the salary of Dr. Durko, who contributed to this project as a post-doctoral fellow. This does not alter the authors' adherence to all of the PloS One policies on sharing data and materials.

# Introduction Cattle breeding is a historically successful economic activity all over the world. Production of meat, milk, and leather has been present in humanity’s history since Classical Antiquity generating business and income for millions of people. Nowadays, the worldwide demand for food and animal sources is the leading reason scientists and producers have studied ways to improve chain production aiming for higher production and, as possible, fewer outlays. The Brazilian Pampa is a grassland-dominated environment characterized by an intense agriculture activity based on livestock in association with crop rotation using rice, soybeans, corn, and wheat. These natural grasslands are dominated by about 450 gramineous species and nearly 200 leguminous species. Such characteristics of the Biome give it a natural aptitude for raising ruminants, an activity introduced in the 17th century. Nevertheless, the natural grasslands from the Brazilian Pampa present high seasonality of forage production and nutritional quality throughout the year. Management strategies aiming the rumen development and the establishment of rumen microbiota and interventions to improve animal nutrition should be implemented to increase production. Interventions into the cattle microbiome might be one of those strategies. For instance, using the probiotic bacteria *Bacillus subtilis natto* benefits the development of cellulolytic microbes in the gastrointestinal tract of calves. Members of the genus *Coprococcus* have been associated with an efficient rumen microbiome capable of exhibiting increased propionate and butyrate production. A stable microbial load of Lactobacillus species has been shown to improve weight gain and immunocompetence in young calves. Still, a deeper understanding of the microbiome shifts along the gastrointestinal tract is necessary to take full advantage of these strategies. The relationship between nutrient utilization and efficiency, the gastrointestinal microbiome, and animal nutrition is one of the leading topics in cattle breeding studies. Microbes along the bovine’s gastrointestinal tract are responsible for processing complex carbohydrates (cellulose and lignin), absorption of nutrients, and production of proteins. The digestion starts in the mouth, where saliva enzymes first break down polymers into lower molecules while buffering substances keep the pH stable. Bacteria from the *Proteobacteria*, *Tenericutes*, *Firmicutes*, and *Actinobacteria* phylum are the main taxa in the cattle mouth responsible for a healthy and stable environment. Posteriorly, in the rumen and barely in the omasum, the bolus is fermented and produces volatile fatty acids (VFA), the primary food source for these animals. The rumen is composed of a diverse and stable community that includes Bacteria, Fungi, Archaea, and microeukaryotes. Finally, the cud moves to the intestines after the rumination cycle, where non-previously degraded material is processed and absorbed through villi. Different and specialized bacteria such as *Plantomyces* and *Cloroflexi* are characteristic of this environment. The knowledge about interactions between microbes has exponentially risen in the last decades with the Next-Generation Sequencing (NGS) development. This technique makes it possible to map the microbes in a biological sample without needing cultivation. Many studies were conducted, and many valuable findings were made in the rumen, such as the postulation of the core ruminal microbiota, the prospection of probiotics to increase milk quality, and the identification of putative microbial enhancers of feed quality. Nonetheless, many questions remain about how microorganisms act and respond in association with cattle, and new ones rise with any further report. For instance, we still need to know how the cattle microbiome can be effectively included as a trait in selection decisions and breeding programs or the relative contribution of environmental factors in determining the cattle’s microbiome composition. Here we performed the taxonomic identification of Bacterial and Archaeal communities from critical compartments of the gastrointestinal tract of grass feed Bradford cattle raised in natural grassland in south Brazil. We hypothesized that different gastrointestinal compartments (saliva, ruminal fluid, and feces) harbor unique microbial communities. To test this hypothesis, we analyzed 110 samples from animals from the same breed, submitted them to the same food source, and collected them during the same season. Our main target was to characterize the microbiome shifts along the gastrointestinal tract to support future works aiming at generating interventions to improve rumen development, the establishment of a healthy rumen microbiota, and maintaining and improving animal nutrition. # Materials & methods ## Experimental design We collected samples from animals raised by the Departamento de Diagnóstico e Pesquisa Agropecuária (DDPA, SEAPI-RS), located in São Gabriel, RS, Brazil, for three years (2017–2019). All sample collections were done during the autumn to minimize seasonal variation and extreme climate conditions. The Animal Welfare Committee approved animal management and research procedures of the State Foundation of Agricultural Research (FEPAGRO), registered under the number 16/13. No aggressive method was used in herding, aiming not to stress the animals. We studied sets of Bradford heifers (58 animals) randomly distributed in paddocks with natural grasslands of the Pampa Biome. To obtain representative samples that allow us to detect Bradford’s core microbiome, different animals were sampled in a 3-year experiment. The heifers were 36 months old and, by the time of sampling, weighed 343 ± 30 kg. Each animal was sampled only once. Seventeen animals were sampled in the first year of the experiment. A different group of 17 animals was sampled in the second year, and 24 animals were sampled in the third year. All animals were submitted to a continuous grazing stock method, with a forage allowance of 12 kg of dry matter/100 kg of live weight adjusted by adding or taking extra animals to the paddock as needed. We monitored the weight gain for 60 days, with the sampling occurring in the middle of this period (30 days). Based on this monitoring, we estimate the average daily weight gain (ADG) for each animal and proceed to downstream analysis. Diet aspects such as weight gain, crude protein intake, and organic matter digestibility were accompanied for 120 days after the sample collection. The native pastures included *Eryngium horridum*, *Vernonia nudiflora*, *Erianthus angustifolius*, *Eupatorium bunifolium*, *Paspalum notatum*, *Eragrostis plana*, *Axonopus affinis*, *Paspalum umbrosum*, *Desmodium incanum*, and *Paspalum plicatulum*. ## Sampling strategy and measurement of descriptive variables Samplings (n = 110) were always performed after the morning grazing cycle (approximately 9:00 AM and 12:00 AM). During this procedure, all animals were kept in a closed corral without food assessment. Saliva (n = 35) was collected with an individually sterilized syringe and stored in a 5mL tube on ice. Ruminal fluid (n = 17) was collected through an esophageal probe equipped with a vacuum system using a sterilized set of hoses and kitassato flasks as previously described. Fecal samples (n = 58) were divided into two sub-samples. One sub- sample was stored in 5mL tubes and kept on ice while transported to the laboratory, where samples were frozen at -80°C until DNA extraction. The other sub-sample was dried with forced air circulation at 55°C for 72 hours. After partially dried and grounded, feces were subjected to dry matter determination after drying at 105ºC for 12 hours. Organic matter in feces was determined by flaring crude protein (CP) at 550°C, according to the Kjeldahl method. The concentration of fecal CP in the organic matter (CPf:g/kg.OM) was used in the Eqs. (I), (II), and (III) are described below to estimate organic matter digestibility (OMD), protein concentration in the diet (PC), and daily protein consumption (CP Intake). The equations were explicitly constructed for native grasslands of Pampa Biome by using Rosa et al. methods: OMD = 0:942–38,619/fecalCP (I) PC = 1.346 \* (fecalCP) - 347.63 (II) CPintake = 8.0728 \* (fecalCP) - 347.38 (III) ## DNA extraction and amplification Microbial DNA was extracted from 0.25g of each sample (n = 110) using DNeasy PowerSoil Kit® (QIAGEN, Hilden, Germany), following the manufacturer’s instructions. DNA purity and concentration were determined using a NanoVue^TM spectrophotometer (GE Healthcare, Chicago, IL, USA). All DNA samples were stored at -20°C for downstream analyses. Amplification, sequencing, and data processing were performed as described previously, following the recommendations of the Brazilian Microbiome Project. Briefly, the V4 region from the marker gene 16S rRNA was amplified using the 515F/806R primers to amplify bacteria and archaea simultaneously. PCR products were purified with the Agencourt AMPure XP Reagent kit (Beckman Coulter, USA). The final concentration of DNA in the PCR product was quantified using the Qubit Fluorometer kit (Invitrogen), following the manufacturer’s recommendations. Finally, the PCR products were combined into equimolar concentrations and were used for library preparation with the Ion One-Touch 2 System using the Ion PGM Template OT2 400 Kit (Thermo Fisher Scientific, Waltham, MA, USA). Sequencing was performed with Ion PGM Sequencing 400 on the Ion PGM System using 318 Chips v2. The Raw sequences were deposited in the Sequence Read Archive (SRA), BioProject ID PRJNA818310. The individual accession numbers associated with each sample are available in. ## Sequencing and amplicon analysis Raw barcoded sequences were split into individual files utilizing FASTX-Toolkit. These files were imported into the R environment, where all analyses were carried out. We performed quality filtering and built an amplicon sequence variants (ASVs) table through the DADA2 package, followed by taxonomy inference using the last available SILVA database (ver. 138.1). The resulting table was converted into a phyloseq object for downstream analyses. Data were rarefied by the minimum library size and transformed by centered log-ratio to reflect the compositional nature of this type of data. Our dataset presented low variance with no need for controlling or strata. We calculated the Good’s coverage for all samples and evaluated the microbial distribution of taxa, diversity of ASVs, and distribution of phyla and genera among the three sample sources. Alpha diversity was measured by using the rarefied dataset with the microbiome package with a function called *alpha*. Boxplots summarizing the alpha diversity distribution were plotted by using ggplot2 R package The significance of numerical evaluations of alpha diversity, and abundance of phyla was tested with the Kruskal-Wallis *post hoc* Dunn test (p ≤ 0.05) from base R using the functions *kruskal*.*test* and *dunn*.*test*. Beta diversity differences among samples were measured by PERMANOVA analysis (p ≤ 0.05) using the *adonis* function from the vegan package and plotted in a PCoA plot using the Euclidean distance and the phyloseq package and the functions *ordinate* and *plot_ordination*. The statistical power of the sample number per group was estimated using the micropower R-package based on the permutational multivariate analysis of variance (PERMANOVA). The PERMANOVA powers were calculated based on the weighted Jaccard distance. The effect sizes (ω2) were calculated using matrixes of 80% powers for 58 fecal samples, 35 saliva samples, and 17 ruminal fluid samples. The R markdown file showing the amplicon analysis is provided at <https://github.com/FreitasAndy/BovinesGIT>. # Results We carried out a study with 110 samples from different compartments of the GIT of Bradford heifers (36 months, 343 ± 30 kg) distributed in paddocks with natural grasslands of the Pampa Biome, analyzing sequencing output, crude protein intake, an average daily weight gain and organic matter digestibility. With 58 fecal samples, 35 saliva samples, and 17 ruminal fluid samples, we obtained an 80% power that allowed us to detect small to medium effect sizes (Feces ω2 = 0.000132, Saliva ω2 0.0167, and ruminal fluid ω2 0.32). This number of samples provided the statistical power to answer questions about how similar or different the microbiota is from each part of the cattle’s GIT. The size of the dataset provided variability to validate the data and produce more reliable conclusions. Data analysis showed that we worked with a very homogeneous group. That means the animals presented a similar food intake and digestion capacity. On the other hand, all groups showed a wide range of weight gain. After sequence quality control, we obtained 990,505 high-quality reads associated with ASVs (a mean of 8,923 reads per sample) and 11,922 ASVs (The ASV table and associated taxonomy is presented in, the complementary metadata associated with this experiment are presented on). The coverage was higher than 99% in all samples, meaning our sequencing depth was sufficient for adequately investigating the microbiota in all samples, allowing us to include all samples in the downstream analysis. ## All GIT compartments presented different communities, but microbial communities from the rumen were less diverse Our goal was to measure the similarity among microbial communities from saliva, rumen, and feces. The ruminal fluid presented the lowest microbial richness and diversity. Microbial communities from feces and saliva had similar richness and diversity. The beta diversity showed that all GIT sampling sources presented different microbial communities (R² = 0.1; p-value = 0.001). Although saliva and feces had similar microbial richness, they also had more distinct communities between themselves when compared with ruminal samples. More than 40% of the phyla found in this study were unique from saliva samples. No phyla were uniquely found in either feces or rumen. About 60% of the microbial genera found in our experiment were unique from saliva samples. On the other hand, 12.4% of the genera found in our investigation were uniquely found in feces samples. The rumen samples had less than 1% of unique sequences. All GIT sources shared only 36.1% of phyla and 7.5% of genera, confirming that each sampling source in the cattle’s GIT harbors a specialized microbial community. ## GIT sample sources have unique microbial communities and different microbial abundance No phyla presented the same abundance in the three GIT sources (p-value ≤ 0.05). *Acidobacteria*, *Actinobacteria*, *Cloroflexi*, *Deinococcus-Thermus*, *Epsilonbacteraeota*, *Fusobacteria*, and *Synergistetes* presented decreased abundance in the rumen when compared with saliva. On the other hand, *Tenericutes* and *Fibrobacteres* presented lower abundance in saliva, higher abundance in the rumen, and lower abundance in feces. Besides, some phyla were negatively correlated. *Firmicutes* and *Proteobacteria*, for instance, presented a negative relationship irrespective of the GIT sampling site. This data showed that levels of *Firmicutes* increase over the bovines’ GIT, whereas *Proteobacteria* levels decrease over the GIT flow. Other phyla were typically dominant in a specific GIT niche. *Bacteroidetes*, *Elusimicrobia*, *Kiritimatiellaeota*, *Lenthisphaerae*, *Planctomycetes*, and *Spirochaetes* presented lower abundance or were absent in saliva when compared to rumen and feces. Finally, *Verrucomicrobia* was the only phylum with a similar low abundance in saliva and rumen but more abundant in fecal samples. Important differences in the composition of the microbial community at the genera level were also observed among fecal samples, saliva, and ruminal fluid samples. Among the most abundant genus, *Acinetobacter*, *Bibersteinia*, *Mannheimia*, *Moraxella*, *Porphyromonas*, and *Streptococcus* were enriched in saliva samples. On the other hand, feces samples were enriched by *Alistipes*, *Bacteroides*, *Prevotelaceae*\_UCG-004, *Ruminococcaceae*\_UCG-005 and *Ruminococcaceae*\_UCG-010. *Fibrobacter*, *Prevoltella*\_1, *Prevotellaceae*\_UCG-003 *Rikenellaceae*\_RC9_gut_group, and *Treponema*\_2 were more abundant in the ruminal fluid samples. ## Animals with a higher weight gain do not ingest more food We measured this on our samples by testing the relationship between the ADG and the crude protein intake (CPI), a known variable to assess diet quality, for each animal. Our results showed no association between CPI and ADG. The CPI varied from 74 to 162 g day^-1, but its variation does not follow any pattern compared with the ADG values from the same animals (p = 0.84). # Discussion Insights about how microbes correlate with cattle nutrition are crucial to selective breeding high-efficiency animals producing more and generating minor environmental damage. Identifying a detailed cattle’s GIT microbiome profile will enable us to identify novel microbial markers, as well as their interactions with modifiable environmental factors, that are informative for establishing useful strategies aiming at interventions to improve animal nutrition. Here, we provide multiple lines of evidence that the microbiota changes throughout the GIT compartments despite the interaction between rumen and mouth via rumination. Saliva, rumen, and feces samples collected from Bradford heifers (36 months old, 343 ± 30 kg at the sampling time) raised in natural grasslands of the Pampa Biome presented different microbiotas. Few taxa were shared between the different types of samples. They also showed differences in weighted diversity and total taxa count. Our findings contradict other works postulated that saliva samples could predict rumen microbes. Those papers pointed out that once ruminants pass considerable amounts of rumen material over their oral cavities by chewing the cud, oral samples may be good representations of the ruminal microbes. In this regard, targeting saliva instead of ruminal liquid might be valuable to rumen studies since the ruminal sample collection is complicated and invasive to animals. Nonetheless, although these papers have shown valuable results in different aspects, both were conducted with few samples, presented low statistical confidence, and have not considered the compositional nature of their microbiome datasets. Here, we provided a bigger sample size, more robust analysis methods, and deeper taxonomy resolution. Still, our results showed that the communities from different samples present different compositions. One of the reasons for that is the difference in many physiological aspects between the oral cavity and the rumen. Although the cud is shared between mouth and rumen, differences in pH and oxygen levels might be crucial factors contributing to these differences. Both compartments, in fact, share some taxa, but, saliva samples cannot resemble or predict ruminal microbes. For that reason, and based on our data, we do not contend this approach for future works once it could lead to misleading conclusions. Differences in microbial diversity and the presence/absence of taxa in those sources were expected based on the physical and chemical nature of the GIT organs. Saliva is the first step of interaction between food and animal. It’s susceptible to airflow, produces buffering substances, and catalyzes fiber degradation keeping an alkaline pH between 8.5 and 8.9. The rumen is an anaerobic fermentative chamber that consumes 65 percent of the starch ingested by the animal, maintaining a barely acid pH of around 5.5 to 6.8. Feces are the end of the digestion after the bolus passes through the highly acidic abomasum (pH \~ 3) and turn into a more pleasant environment for microbes in the intestines. Those differences make the rumen a more selective environment when compared with saliva and feces. Altogether these environmental restrictions lead to lower diversity. Saliva samples presented a more divergent community, possibly because most saliva microbes cannot survive in the rumen environment, despite the high interaction between rumen and mouth via rumination. Additionally, some ruminal taxa are expected to accompany the bolus through the intestines and remain in feces, which explains the closer beta diversity between them. Also, the microbial diversity observed in feces samples indicates that the intestine is a more suitable environment for establishing microorganisms. One interesting finding of our work was the negative association between the phyla *Firmicutes* and *Proteobacteria* among the different compartments of the GIT. Whereas the abundance of *Firmicutes* increases from the beginning to the end of the GIT, *Proteobacteria* does the opposite. A possible explanation for this is that *Firmicutes* can metabolize fat, which is degraded mainly in the small intestine, close to the end of digestion, where more specific phyla are needed for that. The variation of *Proteobacteria* is less clear since the *Proteobacteria* phylum comprises a broad array of microbes that can perform multiple functions. Recent studies could not explain its variations along with the GIT. Besides the phyla mentioned above, saliva samples were composed of *Acidobacteria*, *Actinobacteria*, *Cloroflexi*, and *Fusobacteria*. *Cloroflexi* and *Acidobacteria*. All are common phyla from the soil microbiome, which could be ingested during grazing. *Fusobacteria* and *Actinobacteria* are responsible for states of healthy homeostasis in animals’ GIT and members of the saliva consensus microbiota. All these phyla are aerobic or microaerophilic. Probably the main reason why none or very few bacteria from these groups can survive in the anaerobic conditions of the rumen. The rumen is a restrictive environment, with an increased abundance of phyla specialized in the anaerobic processing of organic matter. The fermentative capacity of this phyla might explain the high abundance of Bacteroidetes. The phyla include microbes from the genus *Prevotella*, one of the main bacterial genera in the rumen. Besides, the higher abundance of *Fibrobacteres*, part of the consensus ruminal microbiota, relies on the phylum’s importance for the degradation of fibers and complex carbohydrates such as cellulose. Other prominent phyla in the rumen belonged to groups with a high capacity to adapt to stressful environments like the ruminal fluid. *Elusimicrobia*, *Kiritimatiellaeota*, *Lentisphaerae*, *Patescibacteria*, and *Tenericutes* are phyla of bacteria capable of growing in extreme conditions. Most of them are typically found at high depths in the ocean, in toxic wastes, and contaminated soils. They found good conditions for survival in the ruminal environment as they can grow in anaerobic conditions. The *Acidobacteria*, *Actinobacteria*, *Cloroflexi*, and *Fusobacteria* presented a slightly higher abundance in fecal samples than in the ruminal fluid. These phyla are known to show less tolerance to anaerobic environments and belong to a group of the healthy intestinal microbiota. On the other hand, *Deinococcus-Thermus*, *Epsilonbacteraeota*, and *Synergistetes* were detected in a similar abundance registered in saliva. The mechanisms behind this similar abundance are not fully understood. Still, somehow these phyla, which are present in saliva and the intestine of animals, can survive the passage through the rumen environment. The main genera enhanced in saliva were *Moraxella*, *Porphyromonas*, and *Bibersteinia*, all of them gram-negative and oxidase-positive bacteria members of the regular oral cavity of animals. Those genera are aerotolerant anaerobes. They can tolerate, for a limited time, the presence of atmospheric oxygen, growing in the presence of oxygen concentrations in the mouth, allowing them to survive in saliva samples. On the other hand, in the rumen, there is an enrichment of bacterial genera which degrade complex carbohydrates, such as *Fibrobacter* and *Prevotellaceae* members. The pattern of anaerobic bacteria is also present in fecal samples, but with the increased abundance of *Bacteroides*, *Alistipes*, and *Ruminococcaceae* members, which are commonly found in animal and human feces but have less impact in the degradation of complex carbohydrates once these molecules are mainly degraded in previous compartments of cattle GIT. Finally, the relationship between the ruminal microbiota and the ruminant’s performance apart from the diet isn’t novel. Here we endorsed this concept by showing that Animals with a higher weight gain did not simply ingest more food. Salivary samples showed a clear microbial pattern in animals with a higher weight gain, relying on a higher abundance of microbial biomass degraders. Animals with a higher weight gain presented a higher abundance of *Lachnospiraceae XPB1014 group*, *Fibrobacter*, and *Prevotellaceae* members. # Conclusions This work hypothesized that different gastrointestinal compartments of the gastrointestinal tract (saliva, ruminal fluid, and feces) harbor unique microbial communities. The results obtained confirm our hypothesis. Few taxa were shared between the different compartments tested, and differences in microbial diversity and taxa counts were observed. Finally, our data showed that the saliva microbiome is not representative of the rumen microbiome and should not be used as an alternative, easy-to-collect sample for studies about the rumen microbiome. # Supporting information [^1]: The authors have declared that no competing interests exist.

# Introduction Flower colour is determined on the basis of flavonoids, carotenoids and betalains. Anthocyanins, a group of secondary metabolites belonging to the flavonoid family, demonstrate a wide range of orange to red and purple to blue flowers, which can attract pollinators and, importantly, protect against damage from UV irradiation. In addition, anthocyanins performs as key signals between plants and microbes. The anthocyanin biosynthesis pathway (ABP) is a branch of the flavonoid biosynthesis pathway that is derived from the phenylpropanoid biosynthesis pathway. Briefly, the ABP begins with the formation of chalcones by the chalcone synthase enzyme (CHS). Next, chalcone isomerase (CHI) converts chalcone into naringenin. Naringenin is then hydroxylated at the 3′ position of its central ring by flavanone 3-hydroxylase (F3H) to produce dihydrokaempferol (DHK). DHK can then be further hydroxylated at the 3′ position or at both the 3′ and 5′ positions of the B-ring to produce dihydroquercetin and dihydromyricetin, respectively DHK, dihydroquercetin, and dihydromyricetin generally result in the production of brick-red/orange pelargonidin-, red/pink cyanidin-, and blue/violet delphinidin-based pigments, respectively. Thus, the establishment of these three biosynthesis pathways is essential for diverse flower colours. Each species usually accumulates limited kinds of anthocyanins and exhibits limited kinds of flower colour on the basis of the expression of a specific set of biosynthetic genes, the substrate specificity of key enzymes and the temporal and spatial regulation of the biosynthetic genes. Thus, the colour range in many ornamental plants is limited by the genetic background of the species, and genetic modification (GM) is the only effective way to overcome this limitation.Many ornamental plants lack cultivars of the violet to blue flower due to the absence of delphinidin-based anthocyanins. Thus, GM of blue coloured flowers is important to horticultural breeders. There are two main strategies employed to modify flower colour using transgenic approaches. One is the control of endogenous flavonoid pigments in flowers both quantitatively and qualitatively, and the other strategy is to accumulate non-native pigments in flowers. Both strategies are effective. In rose, to out compete the endogenous DFR for substrates dihydroquercetin, *RcDFR* was targeted for downregulation via RNAi-mediated silencing. The resulting transgenic plants, which also expressed pansy *F3*'*5*'*H* and Dutch iris *DFR* genes, produced flowers that exclusively accumulated delphinidin-based anthocyanins with a concomitant colour change towards blue. In carnation, over-expression of a petunia *F3*'*5*'*H* gene in a pelargonidin producing carnation cultivar produces petals in which delphinidin derivatives contribute to about 70% of total anthocyanins. In gentian, suppression of the *F3*'*5*'*H* gene decreased delphinidin derivatives and resulted in magenta flower colours, which could be utilised as elite sources to breed gentian plants in the near future. In *Cyclamen persicum*, suppression of the endogenous *F3*'*5*'*H* transcript resulted in a hue shift from purple to red/pink in one cultivar Of these examples, we concluded that the most promising strategy to obtain blue flowers was to suppress endogenous competition enzymes of the substrate for F3′5′H and to simultaneously introduce an appropriate exogenous *F3*′*5*′*H* gene as well as downstream biosynthesis genes of the ABP pathway. Chrysanthemum (*Chrysanthemum* × *morifolium* Ramat.) is one of the most important ornamental plants in the world. The main anthocyanin biosynthesis pathway in chrysanthemum was the cyanin-based pathway, so there are no blue pigments in any of the chrysanthemum cultivars due to the lack of delphinidin- based anthocyanins and deficiency in F3′5′H. Compared to rose and carnation, molecular breeding of blue colour flower in chrysanthemum is in the nascent stage, although molecular genetic technology has been widely used to improve other aspects of chrysanthemum cultivar. To date, no studies focused on molecular breeding involving flavonoid pigments for chrysanthemum and no reports on the isolation and functional analysis of ABP genes of chrysanthmum. In the present study, we isolated seven key structural genes involved in anthocyanin biosynthesis in chrysanthemum, and suppressed a key nodal gene *F3*′*H* in chrysanthemum to block cyanidin-based anthocyanin pathway. Simultaneously, the *F3*′*5*′*H* gene from *Senecio cruentus* (*PCFH*), which produces new delphinidin derivatives in the corollas of transgenic tobacco plants, was introduced into the chrysanthemum to create flower colour-modified chrysanthemums. # Materials and Methods ## Plant materials The *in vitro* plantlets of different coloured series (white, red, pink and purple) of *Chrysanthemum × morifolium* ‘Lijin’ were germinated at 22°C under illumination in a 14 h light period during vegetative growth and 8 h light period during reproductive growth with 40% relative humidity in an artificial climate chamber in Beijing Forestry University. The white, red, pink and purple cultivars were named LW, LR, LPi and LPu, respectively, in this study. There are five stages in the development of the capitulum: S1, ray florets were not yet out of bract (0-5 mm in ray flower length); S2, ray flowers were acicular and had barely outgrew the bract (5 mm in ray flower length); S3, ray flowers clearly outgrew the bract, but the capitulum was compact (5-8 mm in ray flower length); S4, ray flowers were opening and the angle between the ray flowers and stem was over 90° (12-18 mm in ray flower length); and S5, fully opened flower and the angle between ray flowers and stem was nearly 90° (16-20 mm in ray flower length). The five different stages of ray flower development were frozen in liquid nitrogen and stored at -80°C for subsequent determination of the pigment contents and RNA isolation. ## The qualitative analysis of pigments in chrysanthemum Freeze dried tissue was used for the analysis. Samples of the ground freeze- dried petal tissue (50 mg DW) were initially extracted in 2 ml of petroleum ether and acetone (4:1), and 2 ml of methanol, acetic acid, water, trifluoroacetic acid (70:3:27:1) for 24 hours at 4°C in the dark with an absorption spectrum (220-700 nm; 400-500 nm) using a spectrophotometer (TU-1901; Beijing Puxi Co., Ltd). ## Gene isolation and bioinformatics analyses For reverse transcription polymerase chain reaction (RT-PCR) analysis, 3′-rapid amplification of the cDNA ends (RACE), and 5′-RACE, the total RNA was extracted from the five stages using TRIzol according to the manufacturer’s protocol. The first strand cDNA was synthesised using the reverse transcription (RT) system (Promega). In a preliminary study, a cDNA library of *C. morifolium* was constructed (unpublished data). To isolate the 3′ and 5′ ends of the ABP genes, primers were designed and synthesised according to the ESTs of CmCHS, CmCHI, *CmF3H*, *CmF3*′H, CmDFR, *Cm3GT* and *CmANS*. Next, 3′ RACE and 5′-RACE were performed using the SMARTTM RACE cDNA Amplification Kit (Clontech). The final full-length sequences were then amplified using TransTaq DNA Polymerase High Fidelity (TransGen). The amplified product was then purified, ligated into the pGM-T vector, and cloned into the DH5α *Escherichia coli* strain followed by sequencing of all strands. BLAST analysis was performed to confirm the homologues to other plant anthocyanin biosynthesis genes. Sequence homology searches in Genbank were performed using the BLAST (<http://www.ncbi.nlm.nih.gov>) and DNAMAN programs. Structural analysis of the deduced protein was performed using the Expasy Molecular Biology Server (<http://cn.expasy.org/tools/>). Phylogenetic trees were constructed using the neighbour-joining method and the MEGA version 4.0 software. ## RT-PCR and qRT-PCR Isolation of the total RNA from each stage of the 4 series and synthesis of the first strand cDNA were performed as previously described. The transcript levels of CmCHS, CmCHI, *CmF3H*, *CmF3*′H, CmDFR, *CmANS* and *Cm3GT* at the S0 to S4 stages were analysed using RT-PCR and quantitative real-time PCR according to the method described by Huang et al.. The primers sequence used in this study were listed in Table S1 in. *C. morifolium* β-actin (*CmActin*) was used as an internal control gene. The reactions were repeated three times. ## Transformation with *Agrobacterium tumefaciens* Full-length *PCFH* cDNA was subcloned into the binary vector pBI121 in exchange for the *GUS* structural gene to construct vector 35S-PCFH. We selected *CmF3*′*H* as the targets for RNAi. Binary vectors with RNAi-induced inverted repeat structures were constructed as described by Nishihara et al.. Approximately 500 bp of the gene was connected in sense and antisense into the pUC19 vector to construct the vector 35S-F3′Hir. The transverse thin cell layers (tTCLs) of the LPi were used as explants for the transformation experiments according to the methods described in our previous studies. The *A. tumefaciens* strain EHA105, which contained either 35S-PCFH or 35S-F3′Hir, were used to inoculate the explants (the transgenic chrysanthemum were named pPCFH and pF3′Hir, respectively). For co-transformation of 35S-PCFH and 35S-F3′Hir, which is a mixed strain treatment, 35S-PCFH and 35S-F3′Hir, which contained a single T-DNA vector in two separate *Agrobacterium* strains, were transformed into the explants (the co-transformation of 35S-PCFH and 35S-F3'Hir were named pPCFH + pF3′Hir). Bialaphos-resistant shoots were then transferred into the root-inducing medium. Plantlets were acclimatised and grown in a contained greenhouse. The flowers of the transgenic gentian plants were collected for further analysis and stored at -80°C until use. ## Flower colourimeter analysis, anthocyanin content measurement and HPLC analysis The flower colour variables were measured on all of the positive lines of pPCFH, pF3′Hir and the co-transformation lines immediately after selection. The L^\*, a^\*, and b^\* values were measured with a Konica Minolta CR-10 Chroma Meter (Minolta, Japan) on the opposite side of the medium part of each ligulate floret, all these ligulate florets were picked from the middle whorl of the capitulum. The lightness coefficient ‘L^\*’, represents brightness and darkness, the ‘a^\*’ value represents greenish and redness as the value increases from negative to positive, and ‘b^\*’ represents bluish and yellowish. Anthocyanin compounds of pPCFH, pF3′Hir and the co-transformation lines were extracted from the petals using ethanol/water/acetic acid (10:9:1). The anthocyanin concentrations were determined by measuring the absorbance at 530 nm using a spectrophotometer and these experiments were repeated three times. For HPLC analysis, anthocyanins were analysed and characterised according the methods of Lai et al, 2007, Lin and Harnly. Malvidin-3,5-*O*-glucoside (Mv3G5G, Extrasynthese, France) was used as a standard for the quantitative analysis. All of the samples were analyzed in triplicate. # Results ## Pigments in chrysanthemum Flavonols/anthocyanins and carotenoids are often coexistence in the flower petals of chrysnthemums, and their combination increases colour cultivars. To avoid the obstruction of carotenoids on the flower phenotype during the GM experiment, we first carried out a qualitative analysis of pigments in the 4 chrysanthemum cultivars to find which cultivar could only accumulate anthocyanin in their petals. The pigment types in the 4 chrysanthemum cultivars is shown in Figure S1 in. In LW, the main pigment was flavonols and carotenoids (with absorption peak at 342.8 nm / 468.00nm, 441.70nm, 417.00nm); in LR, the main pigment was flavonols, carotenoids and anthocyanins (with absorption peak at 334.5 nm, 529 nm indicated the existence of flavonols and anthocyanins, with absorption peaks at 468.00nm，441.70nm，419.00nm indicated the carotenoids); and LPi and LPu exhibited a similar pigment composition, which contained flavonols and anthocyanins (with absorption peak at 339.0nm/ 529.50nm and 340 nm/ 529.50 nm). The only type of anthocyanins in chrysanthemum was cyandin and the flavonols included apigenin, acacetin, eriodicyol, luteolin and diosmetin. From stages S0 to S4, the anthocyanin contents in the 4 chrysanthemum varieties showed an initial increased in stage S2, and then a subsequent decreased. In S0, there was no anthocyanin accumulation in the flower petals. ## Sequence analysis of full-length cDNAs of CHS, CHI, F3H, F3′H, DFR, ANS and 3GT in chrysanthemum Seven anthocyanin biosynthesis-related genes encoding predicted CHS, CHI, F3H, F3′H, DFR, ANS and 3GT were successfully isolated from the red ray floret of chrysanthemum encoding 397, 234, 356, 507, 373, 354, and 449 amino acid residues, respectively. The Genbank ID of these 7 genes were DQ521272, EU286277, DQ471438, KF313549, GU324979, EU810810 and JF433952, respectively. In CmCHS (Figure S2), similar to other plants, the four conserved amino acid residues that are essential for the CHS active sites are Cys₁₆₇, Phe₂₁₈, His₃₀₆, Asn₃₃₉. Four strictly conserved residues (Thr₅₀, Tyr₁₀₈, Met₁₁₅, Ser₁₉₂) in CmCHI, which are important for catalytic activity of the CHI enzyme (Figure S3) further support this gene identification Importantly, the amino acid residues His₂₁₆, Asp₂₁₈ and His₂₉₁ of CmF3H, which ligate ferrous iron, and Arg₃₀₁ and Ser₃₀₃, which participate in the 2-oxoglutarate binding (RXS motif), (Figure S4 in) are the same at similar positions among plant F3Hs. CmF3′H contained the conserved FxxGxRxCxG sequence (Figure S5) in which the invariant cysteine residue serves as the fifth ligand to heme iron\[–\], and the phylogenic tree of F3′Hs and F3′5′Hs demonstrated that CmF3′H was grouped into the CYP75B subfamilies. Furthermore, phylogenetic analysis showed that F3′H and F3′5′H of Asteraceae formed the same cluster and F3′H was ancestral to F3′5′H, which was accordance with the report of Tanaka and Brugliera, 2013. Moreover, F3′5′H of Asteraceae was recruited from a pre-existing F3′H precursor gene and independently evolved from other F3′5′Hs. Protein of the CmDFR gene exhibits a putative NADPH binding domain, which contains a specific conserved motif that was predicted to be related to substrate specificity and 5 strictly conserved amino acid residues in the DFR superfamily (Figure S6). This is similar to other members of the 2OG-FeII_Oxy superfamily. Furthermore, CmANS contained the active sites of His₂₃₄, His₂₈₀ and Asp₂₃₆ residues, which may coordinate iron at the catalytic centre of the iron- containing soluble oxygenases and 2-oxoglutarate-dependent enzymes. In addition, Arg₃₀₀ and Ser₃₀₂ residues were also conserved in CmANS, which may contribute to the specific binding of 2-oxoglutarate, and most likely provides a positive charge (Figure S7). For Cm3GT, The deduced amino acid sequence showed a region between residues 403 and 446 (PSPG box, underlined in Figure S8) that corresponds to the UDP-binding domain. ## The expression profiles of key structural genes The expression profiles of the 7 structure genes, with the exception of *CmCHS* and *CmCHI*, were not detected in LW, which indicated the lack of ABP in white chrysanthemum cultivars. In LPi, LR and LPu, the expression of CmCHS, CmCHI, *CmF3H* and *CmF3*'*H*, which are regarded as upstream biosynthesis genes, were mainly expressed in stage S0 to S2, while the expression levels of downstream biosynthesis genes, which contained *CmDFR*, *CmANS* and *Cm3GT*, were continuously increased from S0 to stage S4. In addition, *CmF3*'*H* showed higher transcript levels in LR and LPu, but much lower levels in LPi, made it a good candidate receptor for GM. ## Generation of transformed lines Overexpression of the *PCFH* transformants were produced from the 'LPi' cultivar using 35S-PCFH vector. Flowers from the transgenic lines did not show significant changes in colour compared with wildtype plants. Moreover, the expression of *CmF3*′*H* in transgenic chrysanthemum plants also did not change. Suppression of CmF3'*H* transformants was performed in 'LPi' cultivars using 35S-F3′Hir vector. There was a marked reduction in the endogenous *CmF3*'*H* transcript in the positive transgenic lines. In addition, anthocyanin accumulation in these transgenic flowers was significantly reduced by 20%-40% compared to wildtype plants. Furthermore, RT-PCR showed that the transcript levels of *CmF3*'*H* were significant decreased, and the transcriptional level of downstream genes were suppressed Because overexpression of *PCFH* did not produce significant phenotypic alterations in the colours of the chrysanthemum flowers, we performed a co- transformation of 35S-PCFH and 35S-F3′Hir. The flowers of the transgenic chrysanthemum plants exhibited a deeper colour compared to WT and pPCFH plant lines. The change in colour, which was grossly observed by eye, was quantified by colour measurements using a colourimeter *CIEL*^*\***a*^\**b*^\*. We observed a significant decrease in b^\* in the transgenic lines compared to wildtypes, suggesting that the colour became more blue; however, the HPLC and pigments measurement results demonstrated marked increased concentrations of anthocycanin (2.5-fold). However, the HPLC results showed that the co-transformation did not produce any new types of pigment. Compared with the pF3′Hir transgenic chrysanthemum plants, transcriptional level of endogenous *CmDFR* and *CmANS* gene recovered to the same intensity as wildtype plants. # Discussion ## Analysis of the key structural gene expression patterns in different colour chrysanthemum ray florets During the past few decades, nearly all of the enzymes involved in ABP for different flower phenotypes have been determined. The lack of activity in one or more genes of these pathways may inhibit the synthesis of anthocyanins, which results in the inability to form colourful anthocyanins. In the present study, we found that the expression of genes encoding enzymes of anthocyanin biosynthesis is similarly regulated in ray florets among the 4 chrysanthemum cultivars with different colours. However, in chrysanthemum ray florets, the red colour has also been contributed for carotenoids, or combined with red or magenta anthocyanins, which was according to our study, the pigments of LPi and LPu were cyanidian. Analysis of the expression patterns of key structural genes suggest that anthocyanin biosynthesis-related genes in chrysanthemum may be classified into two groups on the basis of their temporal expression patterns during ray floret development. The first group consists of CmCHS, CmCHI and *CmF3*’*H*, which are more highly expressed at early development stages compared to late stages. These findings indicated that a sufficient number of precursors were needed to accumulate for anthocyanin and other co-pigment biosynthesis during the early stages of inflorescence growth. The second group consists of *CmF3H*, *CmDFR* and *CmANS*, which are necessary for anthocyanin biosynthesis and are coordinately expressed throughout all of the stages of ray floret development. In addition, the gene expression profile of these four genes correlates with pigment accumulation. Similar results have also been reported in apple, litchi and *Pyrus pyrifolia*, indicating that multiple late genes determine the anthocyanin accumulation among the different genotypes. The expression of CmCHS, CmCHI, *CmF3H*, *CmF3*′H, CmDFR and *CmANS* in chrysanthemum showed that the genes responsible for anthocyanin synthesis accumulated during the early stages of capitulum development , which provided a reference for the selection of a suitable developmental stage for transgenic studies in chrysanthemum. DFR is one of the most important key enzymes for anthocyanin biosynthesis, in petunia(*Petunia hybrida*), the PhDFR proteins do not accept monohydroxylated DHK, and thus, cannot produce the corresponding monohydroxylated pelargonidin anthocyanins. F3′H catalyses the hydroxylation of flavanones, flavones and flavonols. It often competes against the introduced F3′5′H. To avoid these competitions, it is necessary to downregulate F3′H and DFR. However, because the only ABP in chrysanthemum is the cyanidin-based pathway, there is no DFR competition with F3′H. Thus, regulation of *CmF3*′*H* appeared to be important in the production of blue flower pigmentation in chrysanthemum. On the basis of the above analysis, we concluded that the most effective strategy to obtain the blue flowers by genetic modification was to suppress the activity of endogenous enzymes which produce cyanidin, and to introduce an appropriate exogenous *F3*′*5*′*H* to produce delphinidin. ## Key roles of CmF3'H in chrysanthemum It is feasible to generate white, yellow, red and blue flowers by engineering the pathway; both by overexpression of heterogenous genes and/or downregulation of endogenous genes. A growing number of potentially useful molecular tools to engineer blue or red flower colours have now become available. However, only modification of B-ring hydroxylation has been effectively demonstrated in modifying flower colour. The F3′H and F3′5′H enzymes play important roles in flower colour by determining the B-ring hydroxylation pattern of anthocyanins. In genetic breedings of the blue rose, cultivars with no F3′H activity were selected to enhance the blue hue of the transgenic petals and to achieve a high content of delphinidin. In the present study, the transcriptional levels of *F3*'*H* in LPi were much lower compared to other cultivars, which was a good candidate explant to create the blue chrysanthemum. Suppression of *CmF3*'*H* successfully changed the anthocyanin profiles and flower colour in chrysanthemum. A shift from predominantly cyanidin-derived pigments to non- pigment transgenic lines was achieved and revealed a concomitant shift in the CIEL^\*a^\*b^\* value. Previous reports have demonstrated that GM toward *F3*'*H* in other ornamental plants has successfully generated new colour flower cultivars. In *Osteospermum*, suppression of *F3*'*H* by RNAi and the introduction of gerbera DFR resulted in pelargonidin accumulation. Furthermore, downregulation of *F3*'*H* and *flavonol synthase* (*FLS*) genes as well as the overexpression of the gerbera *DFR* gene in tobacco resulted in pelargonidin production. Moreover, heterogenous expression of Gentian *F3*'*H* modulated the intensity of flower pigmentation in transgenic tobacco plants. As shown in pF3′Hir transgenic chrysanthemum plants, we observed the down- regulated expression of *CmANS* and *CmDFR*, which was important to produced the cyanidins. The same phenomenon was also observed in *I. quamoclit*, where *F3*'*H* gene expression was dramatically reduced in *I. quamoclit* and DFR was unable to catalyse dihydroquercetin, a precursor of cyanidin. In addition, expression of the introduced *F3*′*H* resulted in the suppression of the endogenous *ANS* gene and production of peonidin and quercetin in transgenic flowers, thereby resulting in pink flowers in the petunia. ## Overexpression of an exogenous F3'5' H failed to create blue but produced brighter red chrysanthemum lines Loss-of-function or reduced expression of the gene encoding the branching enzyme F3'5'H was the main reason for the loss of flux down the delphinium branch of the ABP. Specific studies demonstrated an accumulation of delphinidin-derived anthocyanins by overexpression the *F3*'*5*'*H* have been reported in the carnation and rose, while inhibition of both the *F3*'*H* and the *F3*'*5*'*H* genes modifies the colour and promotes cyanidin- and pelargonidin- based pigment accumulation in flowers of *Torenia*, *Nierembergia* and *Osteospermum*. Delphinidin-based pigments were not detected in the flowers of the transgenic chrysanthemum using HPLC, anthocyanin content measurements and CIEL^\*a^\*b^\* analysis. Co-expression of pPCFH and pF3′Hir could not successfully create blue chrysanthemums; however, the cyandin content increased 2-fold in the transgenic lines compared to WT. These results indicated that overexpression of a gene resulting in a target compound was not sufficient and that it was necessary to downregulate competing pathways to accumulate a blue compound. In our study, overexpression of exogenous *F3*'*5*'*H* results in no phenotypic change in chrysanthemum, one explanation for this absence may be due to the competition of endogenous CmF3'H with PCFH. A second reason may be that some chrysanthemum family species, which produce delphinidin-based anthocyanins, contain a CYP75B- and not a CYP75A-type F3'5'H, indicating that the chrysanthemum family may have lost the F3'5'H function during their evolution. F3'5'H function was then reacquired by gene duplication and neo- functionalisation of the CYP75B-type gene. However, our previous studies showed that overexpression of *PCFH* could produce delphinidin in tobacco, and the transgenic lines showed a blue flower phenotype. Expression of the *Campanula medium F3*'*5*'*H* gene resulted in a more efficient accumulation of delphinidin (up to 99%) compared with the petunia or Lisianthus *F3*'*5*'*H* gene in tobacco, which indicated that the gene origin is important for successful engineering. and which was accordance with the report of Gion et al., 2012 that in chrysanthemum, expression of a pansy *F3*'*5*'*H* gene under the control of rose chalcone synthase promoter resulted in transgenic chrysanthemum with flower colour change from pink to violet and accumulation of delphinidin. Similar challenges are also posed in the breedings of roses with blue flowers. The expression of *F3*'*5*'*H* genes from the petunia, gentian or butterfly pea in rose resulted in no or little delphinidin accumulation in the petals of the transgenic plants, although these genes are functional in the petunia, carnation or yeast. In contrast, the expression of the pansy (*Viola spp*) *F3*'*5*'*H* gene resulted in a significant amount of delphinidin-derived anthocyanins accumulating in the petals of transgenic rose plants. Substrate specificity is an important consideration in delphinidin production. Dihydroflavonols are reduced in corresponding 3,4-cisleucoanthocyanidins by the activity of DFR. In some plant species, such as petunia (*Petunia hybrida*) and *Cymbidium*, DFR demonstrates strict substrate specificity and cannot utilise dihydrokaempferol. Thus, these species lack pelargonidin-based anthocyanins and have no flowers of an orange/brick red colour. According to our study, there was no pelargonidin in chrysanthemum, and the 134-136 AA in CmDFR was Asn, which could only utilise DHQ. Our group has isolated five *DFRs* from cineraria, among which two utilises DHM to produce delphinidin. Thus, our strategy to produce blue chrysanthemums will focus on following two points: (1) selection of a more effective *F3*'*5*'*H*. *F3*'*5*'*H* from other species should be transformed into the *F3*'*Hi* lines to produced DHMs; and (2) suppression of the inner *DFR* in chrysanthemum and introduction of *ScDFR* to produce delphinidin in chrysanthemum. # Supporting Information [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: DS. Performed the experiments: HH H. Ke H. Keting XQ. Analyzed the data: H. Ke HH. Contributed reagents/materials/analysis tools: H. Ke H. Keting HH. Wrote the manuscript: HH H. Ke DS.

# Introduction The intestinal tract harbors trillions of commensal bacteria representing over a thousand species and encoding over one hundred and fifty fold more genes than the human genome. During the past decade, the gut microbiota was revealed as providing an important functional contribution to its host physiology and in maintenance of health. However, the understanding of these functional interactions is in its infancy. The limitation imposed by the inability to cultivate the majority of the indigenous microbial species is now bypassed with the metagenomic approach. Metagenomic libraries already allowed sequence-based explorations of the human intestinal microbiome that drew up microbial genomic and genetic diversity. Furthermore, the metagenomic approach has been used for some functional investigations of gut microbial communities, however, these applications are few compared to metagenomic studies of other environmental ecosystems. The human intestinal microbiota has been shown to participate in epithelium maturation, host nutrition and protection against pathogens and more recently in regulating gut epithelial cells proliferation, host energy metabolism and immune responses, thus reflecting the symbiotic and beneficial crosstalk between intestinal epithelial cells (IECs) and intestinal microbiota. However, there is also evidence implicating microbiota in diseases such as allergies, inflammatory bowel disease (IBD) and cancer. This evidence unmasks the unwanted face of the microbiota and stresses the importance of an accurate equilibrium between microbes and mucosal immune system. Indeed, the mucosa has the challenge to sense pathogens while being unresponsive to food antigens and commensals, in order to maintain integrity and normal function of the intestine. Traditionally, immune responses in the gut involved the gut-associated lymphoid tissue (GALT), but it is now established that IECs are also an essential component in innate immunity. Theses cells can directly sense commensal bacteria and pathogens *via* pattern recognition receptors (PRRs), including Toll-like receptors (TLRs) and Nod-like receptors (NLRs). These receptors are specialized in the recognition of conserved bacterial and viral structures and generally activate proinflammatory pathways warning the host about infections. Immune and inflammatory responses in the gut involve the transcription factor NF-κB. This DNA binding protein is the transcriptional factor of an evolutionarily conserved regulatory pathway that drives expression of a large number of genes involved in proinflammatory processes at the site of infection or tissue damage. It also controls cell survival, proliferation and differentiation gene expression induced by a wide range of noxious stimuli. Recent studies have demonstrated that NF-κB signaling is a critical element of the homeostatic immuno-inflammatory function in the gut and both deficiency in, or hyperactivation of this transcription factor are linked with pathologies such as chronic IBD or obesity. Thus, determining the factors that regulate this key pathway is of great scientific and clinical interest. In the present study, we present the development of a high throughput functional screening method of metagenomic libraries designed to explore novel NF-κB modulatory potentials within the human intestinal microbiota. A cellular tool, based on the reporter gene strategy, was obtained using the intestinal epithelial cell line HT-29, and was validated and applied for screening of a metagenomic library issued from the intestinal microbiota of Crohn's disease (CD) patients. This strategy opens a way toward identification of bacterial species and effectors involved in the interaction with IECs *via* NF-κB signaling. # Materials and Methods ## Cell culture To construct the reporter model, we selected the human colorectal carcinoma cell line HT-29 that was obtained from the American Type Culture Collection (Rockville, MD). HT-29 cells were grown in RPMI 1640 medium (Sigma) with 2 mM L-glutamine, 50 IU/mL penicillin, 50 µg/mL streptomycin and 10% heat-inactivated fetal calf serum (FCS - Lonza) in a humidified 5% CO₂ atmosphere at 37°C. The THP-1 blue™ CD14+ cells used as control for TLRs responses were obtained from Invivogen and used according to the manufacturer's instruction. ## Construction of stable NF-κB reporting HT-29 cells Stable HT-29 transfectants containing the secreted alkaline phosphatase (SEAP) reporter gene were obtained after cell transfection with the reporter plasmid pNiFty2-SEAP (Invivogen), which contains SEAP as reporter gene and zeocin as resistance antibiotic gene, using TFX-50™ (Promega) following manufacturer's instruction. Cells were cultured for 3 weeks under zeocin (50 µg/mL) selection and cloned. The HT-29/kb-seap-25 clone was selected for its response to 10 ng/ml of TNF-α after 24 h stimulation. ## Analyses of NF-κB activation For each experiment, HT-29/kb-seap-25 reporter cells were seeded at 50 000 cells per well, into 96-wells plates and incubated 24 hours before stimulation. Cells were stimulated with 10 µl of each tested substances for a final volume per well of 100 µl. SEAP in the supernatant was revealed using Quanti-Blue™ reagent (Invivogen) according to the manufacturer's protocol and quantified as OD at 655 nm. All measurements were performed using a microplate reader (Infinite 200, Tecan). ## NF-κB inhibitors NF-κB inhibitors caffeic acid phenethyl ester (CAPE), BAY 11-7082 ((E)3-\[(4-methylphenyl)sulfonyl\]-2-propenenitrile) and MG132 (ZLeu-Leu-Leu-H) were provided by Calbiochem. Cells were stimulated with a concentration range of each inhibitor in the presence or absence of 10 ng/mL of TNF-α. ## TLRs analysis ### TLRs ligands response profiles The TLR response profile was determined using the TLR1-9 agonist kit (Invivogen) according to manufacturer's instruction. Ligands and working concentrations are for TLR1/2: Pam3CSK4 (1 µg/mL); TLR2: Heat Kiled *Listeria monocytogenes* (10⁸ cells /mL); TLR3: Poly(I:C) (10 µg/mL); TLR4: *E. coli K12* LPS (10 µg/mL); TLR5: *Salmonella typhimurium* Flagellin (10 µg/mL); TLR6/2: FSL1 (1 µg/mL); TLR7: Imiquimod (1 µg/mL); TLR8: ssRNA40 (1 µg/mL); TLR9: ODN2006 (5 µM). **TLRs expression profiles by flow cytometry.** Reporter cells were washed with 2% FCS in phosphate-buffered saline (PBS) and stained with TLRs antibodies (Ab) for 30 min at 4°C. Ab for human TLR2, TLR3, TLR4, and TLR9 were obtained from eBioscience. TLR5, TLR6 and TLR7 Ab were from Imgenex and TLR8 antibody from Axxora. Isotype matched Ab were used as negative controls and all antibodies used were originally coupled with Phycoerythrin (PE). After incubation, cells were washed again with PBS+2% FCS and fixed (CellFix – BD bioscience) prior to analysis using a Becton-Dickinson FACSCalibur flow cytometer and CellQuest software (BD Biosciences). For evaluation of total (surface plus intracellular) expression, cells were fixed and permeabilized using CytoFix/CytoPerm (BD Pharmingen). For each measurement, 3×10⁴ cells were analysed. ## High throughput screening of metagenomic clones We used the HT-29/kb-seap-25 cells to screen a metagenomic library issued from the intestinal microbiota of CD patients for NF-κB modulatory capacities. The screened library consisted of 2640 *Escherichia coli* (*E. coli*) DH10B clones bearing fosmids with metagenomic inserts of 40 kb length approximately randomly selected from a global library of 25 000 clones. *E. coli* bearing a fosmid without a metagenomic insert was used as control (referred as metagenomic controls). Metagenomic clones were cultured for 24 hours in lysogeny broth (LB) medium in 96 wells plates. Then OD_{600 nm} was measured and given as a raw value. It is important to precise that this OD value corresponds to a measurement in 96 well plates and a culture volume of 150 µl. In these conditions, growth stationary phase of *E. coli* started at 0.4 OD approximately. Bacterial cells were broken with glass beads (106 µm, Sigma) in a mixer mill (15 oscillations/second, two cycles of 90 sec). The suspension was filtered by centrifugation at 4°C through a 0.2 µm microplate filter (Corning) and added to HT-29/kb-seap-25 reporter clone at 10% vol/vol to a final volume of 100 µl. Seeding of the reporter cells, addition of lysates and SEAP activity measurement were performed using a robotic pipetting workstation (Microlab Star, Hamilton). Ten selected metagenomic clones, 5 stimulatory and 5 inhibitory, were grown further in 8 independent cultures and tested to validate their effect. ## Characterization of the stimulatory clone 52B7 A preparation of supernatant and lysed pellet was done to determine the active culture fraction of 52B7. Supernatant of an overnight culture grown at 37°C was collected after filtration on a 0.2 µm filter. After centrifugation of the bacterial culture at 5000 g/10 min, pellet was resuspended in RPMI and bacterial cells were broken by shaking with glass beads in a mixer mill as described above. The 52B7 fosmid was purified using the NucleoBond extraction kit (Macherey- Nagel) and the DNA insert was sequenced (Genoscope, Evry, France). The sequence is available on GenBank (accession number HQ231916). Potential transcription units were detected using SoftBerry's software FGENESB (<http://linux1.softberry.com/berry.phtml>). Potential ORFs contained in the 40 kb insert were determined using GeneMark.hmm for prokaryotes (<http://exon.biology.gatech.edu/gmhmm2_prok.cgi>) and FGENESB. Both analyses were performed using *Bacteroides thetaiotaomicron* VPI-5482 as model organism. The output predicted protein sequences were further blasted in order to have phylogenetic and function estimation. The final phylogenetic assignment was expressed in percentage of total genes length coverage. The original host was searched using Blastn and the microbial genome database on the NCBI. To determine the potential gene(s) involved in the stimulatory effect of 52B7, fosmid transposition was performed as previously described using the EZ-Tn5™ \<KAN-2\> kit according to manufacturer instructions (Epicentre). Two hundred transposed clones of 52B7 were picked and tested for a revertant phenotype toward NF-κB activation. Three transposed clones with a revertant behaviour were validated and transposon insertion site in the metagenomic insert was determined by sequencing (Beckman Coulter Genomics) using primers homologous to the ends of the transposon (Epicentre). ## Statistical analysis Presented results were representative of a minimum of 3 independent experiments. Results were expressed as mean ± standard deviation of triplicate measurements. Data were analyzed by Student's *t* test. # Results ## Development and characterization of the NF-κB reporter system in HT-29 cells After transfection and culture of several clones under zeocin selection, we selected the best responding reporter clone, HT-29/kb-seap-25, which displayed a high signal upon TNF-α stimulation and a high ratio between unstimulated and stimulated states. To determine whether the reporter gene in the selected reporter clone reflects well the regulation of the NF-κB signaling pathway, we characterized extensively its response to known NF-κB modulating molecules, including proinflammatory cytokines, chemical inhibitors and TLRs agonists. The two proinflammatory cytokines, TNF-α and IL-1β, were able to induce expression of the reporter genes in a dose dependent manner. As expected, the induction was stronger with TNF-α than IL-1β and did not reach saturation even at the highest dose used (100 ng/mL), while IL-1β was saturated at 10 ng/mL. We examined the kinetics of activation of our NF-κB reporter systems by incubating the cells with TNF-α (10 ng/mL) at different times. Activity of the reporter clone HT-29/kb-seap-25 increased in a time dependent manner, with a maximum effect occurring after 24 hours stimulation. A weak activity was detected after 4 hours and continued to increase throughout the experiment. Three chemical inhibitors acting at different level of the NF-κB pathway were tested. Reporter cells were incubated for 24 hours with different concentrations of CAPE, BAY-117082 and MG132, and co-treated with TNF-α. NF-κB activation was reduced by 50% in cells treated with 14.2 µM of BAY 11-7082 or 22.2 µg /mL of CAPE. In the case of MG132, IC50 was 0.2 µM. No effect on the reporter system was observed when the cells were treated with the solvent (DMSO) alone (not shown). ## TLRs expression and response to TLRs ligands The TLRs have an immunological sensing role for bacterial, fungal, viral and parasitic organisms. Therefore, we first examined the presence of 8 different TLRs in the HT-29/kb-seap-25 cells. TLR5 was expressed by HT-29 cells however no TLR2 and few TLR4, TLR6 and TLR9 were detected at the cell membrane (MFI given). Interestingly, TLR3 expression was detected at cell membrane, but also in the endolysosomal compartment as revealed after permeabilization. As expected, TLR7, TLR8 and TLR9 were only found in the intracellular compartment. When we tested induction of the reporter gene by 9 different TLRs ligands, we found that TLR3, TLR4 and TLR5 ligands were active, concordantly with the presence of the cognate receptors. *S. typhimurium* flagellin was highly stimulatory, therefore correlating with TLR5 expression in these cells, while *E. coli* K12 lipopolysaccharide (LPS) and Poly-IC, ligands for TLR4 and TLR3, respectively, were also stimulatory but to a lesser degree. Despite the fact that TLR7, 8 and 9 were detected by flow cytometry in the cytoplasm of HT-29/kb- seap-25 cells, no SEAP activity was detected when the cells were stimulated with their respective agonist, imiquimod, ssRNA and CpG-ODN. Finally, the agonists were tested for control on THP-1 blue™ CD14+ which is an NF-κB reporter monocytic cell line (Invivogen). We observed that THP-1 responded to the stimulation with all tested agonists except those which receptors (TLR3, 7, 8 and 9) are located in the intracellular compartment (data not shown). ## Seeking for modulatory metagenomic clones NF-κB reporter cells were constructed in view of performing high throughput screening of NF-κB modulation capabilities within metagenomic libraries issued from human intestinal microbiota. A first screening using the HT-29/kb-seap-25 reporter clone was done on 2640 metagenomic clones from a library prepared with the fecal microbiota of CD patients. Results are represented as a plot of reporter gene activity (normalized to that of the metagenomic control) and growth level for each metagenomic clone. In a OD range between 0.4 and 0.8, growth of the metagenomic control was not correlated with NF-κB activity. A vast majority (93.5%) of the 2640 tested clones had no significant different effect on NF-κB activity as compared to the controls. Interestingly, 22 lysates (0.8%) enhanced the reporter system activity while 149 lysates (5.6%) down- activated it. We decided to validate 10 clones producing the highest enhancing or inhibitory effect (circled on). These 10 clones were cultivated as 8 independent cultures and tested in a similar way on the HT-29/kb-seap-25 cells. All of the 5 activating clones (43C8; 52B7; 90B4; 90C4 and 122B8) displayed high stimulatory effects that were significantly different from the control (p\<0.01). No inhibitory clones affected the expression of the reporter gene to an extent comparable to that of the 5 most stimulatory clones. The 5 inhibitory clones selected (31B2; 38B10; 90H7; 122A8 and 123H11) were nevertheless submitted to the validation step and we found one having a very significant and reproducible effect (122A8). This demonstrates that our screening system can successfully identify bacterial metagenomic clones able to modulate NF-κB signaling in HT-29. We also obtained a Caco-2 reporter clone (Caco-2/kb-seap-7) that was characterized similarly to HT-29, but was not used for the present screening. However, it is noteworthy that on Caco-2 reporter cells, the 10 selected lysates were without effects suggesting potential cell line specificity (data not shown). ## Toward the identification of a new bacterial stimulatory mechanism of NF-κB For further investigation, we decided to focus on the stimulatory clone 52B7. Metagenomic DNA insert of 52B7 (37,006 base pairs long) contains 43 genes and 23 transcriptional units according to GeneMark.hmm and FGENESB predictions. The total predicted genes length represents 80.8% of the total insert length. No relatives were found in NCBI bacterial genome database however, the best blastp results of predicted genes for 52B7 insert showed that it is probably issued from a *Bacteroides* genus member (98.5% of predicted genes coverage), the most closely related cultivated strain being *B. vulgatus* ATCC 8482 with 42% of predicted genes coverage. We then identified the most active fraction by testing separately supernatant and lysed pellet. We observed that the stimulatory effect was mainly present in the supernatant while only a weak activity was detected in the lysed pellet (data not shown). Moreover, we also confirmed that 52B7 supernatant significatively induced IL-8 secretion in both parental HT-29 and HT-29/kb- seap-25 reporter cells as compared with the metagenomic control. Therefore, we managed to identify the gene(s) or loci responsible for the stimulatory effect and performed mutagenesis on the fosmid by insertion of transposon EZ-TN5. Out of 192 mutants, we obtained 3 transposed clones from 52B7 of which the supernatant did not produce any more stimulation of NF-κB. Insertion regions of the transposons were determined by sequencing. Analysis of both insert annotation and sequences from the EZ-TN5 transposon revealed that in 52B7/A and 52B7/C, the transposons were inserted in the same gene encoding an ATP-binding cassette (ABC) transporter permease ( and 8^th gene). In 52B7/B transposon targeted the 38^th gene, which encodes for a putative lipoprotein with unknown function. The two targeted genes were part of two putative distinct transcription units. Each of the two targeted insert loci was therefore involved in 52B7 stimulatory effect of NF-κB in HT-29. # Discussion We have presented the construction and validation of a cell-based screening system to study NF-κB modulation in IECs and its first utilization within screening of metagenomic libraries from the human gut microbiota. Using the reporter gene strategy, we have obtained an HT-29 reporter cell clone (HT-29/kb- seap-25) that allows rapid examination of NF-κB activity regulation. A validation of the selected reporter clone was performed to check the robustness of its response to known activators or inhibitors of NF-κB pathway. We observed a dose-dependent stimulation of NF-κB in HT-29 clone upon treatment with the proinflammatory cytokines TNF-α and IL-1β. These observations are classical and well described in the literature. We tested 3 different NF-κB inhibitors, each one acting at different levels of the NF-κB pathway. Previous studies demonstrated that the proteasome inhibitor MG132 blocks TNF-α-induced degradation of IκBα and leads to the accumulation of phosphorylated IκBα. Since it inhibits proteasome activity, MG132 is not a specific NF-κB inhibitor but is widely used as such. CAPE is an active molecule purified from honeybee hives and was also described as an anti-inflammatory molecule acting at the level of DNA binding in human histiocytic cell line. BAY 11-7082 is an irreversible inhibitor of IκBα phosphorylation thereby sequestering NF-κB in an inactive state in the cytoplasm and preventing activation of downstream signaling. These three molecules were efficient in the HT-29 reporter cells. The gut epithelium is exposed to numerous microbes without developing uncontrolled inflammation. One of the proposed mechanisms explaining such epithelium tolerance towards commensals is its hypo-responsiveness to bacterial patterns involving TLRs signaling cascade. For example, Toll inhibitory protein (Tollip) was shown to inhibit IL-1R activated kinase (IRAK) activation, resulting in loss of TLR2 and TLR4 response in the gut. In addition, the luminal surface seems to be devoid of TLR2 and 4, their expression being confined to crypt cells. It is noteworthy that characterization of TLRs showed unresponsiveness of our HT-29 and Caco-2 reporter cells toward agonists of the bacterial cell wall components receptors TLR1/2, TLR2 and TLR6/2. Using mRNA and protein analysis, other laboratories, showed that both cell lines expressed very low levels of TLR2 mRNA as compared to a TLR2-ligand responsive monocyte cells THP-1. TLR1 and 6 mRNA were not expressed by Caco-2 whereas HT-29 did. In addition, it was demonstrated that these cell lines expressed increased levels of Tollip, the TLR2 signaling inhibitor, which could explain the lack of response obtained with TLR1/2, TLR2 and TLR 6/2 agonists. Challenging HT-29 cells with LPS alone resulted in a small but significant induction of NF-κB- dependent reporter gene activity which is consistent with previous results demonstrating that parental HT-29 line expresses small levels of TLR4 mRNA and protein, and high level of Tollip. By contrast, we found that HT-29 reporter cells were strongly responsive to flagellin treatment. TLR5 recognition of bacterial flagellin from both Gram-positive and Gram-negative bacteria was previously reported and was described to be constitutively expressed at the apical and basolateral side in HT-29. Nucleic acid recognition is possible *via* TLR3, TLR7, TLR8 and TLR9, which are classically located in the intracellular endosomal compartment. Following treatment with Poly-I:C, a synthetic analogue of double stranded RNA, which is recognized by TLR3, NF-κB activation was detected in HT-29 cells but not in THP-1 control reporter cells. This difference could be linked to the localization of TLR3 in HT-29. Indeed expression patterns described for this cell line suggest localization in both cytoplasm and cell membrane, a finding that we confirmed by flow cytometry. In our hands, TLR7, TLR8 and TLR9 agonists did not induce NF-κB activation in HT-29, Caco-2 nor in THP1 reporter cell lines. Validation of our reporter screening tools showed reproducible responses consistent with the literature and the reporter gene strategy proposed is robust for high throughput identification of bacterial modulatory potentials within metagenomic libraries. Metagenomic is a recent approach designed to uncover novel genes and related functions. However, this method can identify only a small proportion of genes from the natural pool due to cloning and heterologous expression limitations in *E. coli*. Nevertheless, *E. coli* is a reliable host since it was shown to be able to express up to 40% of the functional potential from randomly cloned environmental DNA. To fulfill these limitations, work is in progress to generate libraries in a Gram + recipient host. Nevertheless, the methodology presented allowed us to identify 171 metagenomic clones able to modulate NF-κB pathway. Here we presented the first results obtained on the characterization of the stimulatory metagenomic clone 52B7. The 52B7 insert is issued from a probably unknown *Bacteroides* strain and encodes number of hypothetical proteins with unknown functions. This demonstrates the importance of the functional metagenomic approach in a better understanding of the interactions between a large portion of uncultivated bacteria and their host. Interestingly, the closest relative species was *B. vulgatus*, a commensal bacteria previously found to be more abundant in CD patients than in healthy subjects. Moreover, presence of antibodies against *B. vulgatus* was reported in CD patients. However no mechanism was yet proposed to explain a possible implication of *B. vulgatus* in CD. Concerning the 52B7 insert, no proteins already known to activate NF-κB signaling (such as a TLR ligand) were detected, suggesting the involvement of potentially new stimulatory effectors. Using mutagenesis by transposition, we identified two different candidate loci responsible for the stimulatory effect of NF-κB on HT-29. The first locus, related to an efflux ABC-type transport system, includes a permease particularly targeted by the transposition experiments and the second locus highlighted includes a putative lipoprotein with unknown function. The contribution of the 2 loci on the stimulatory effect still needs to be elucidated. According to the literature, some ABC transporter proteins ensuring lipoprotein trafficking in Gram-negative bacteria were described. In addition, lipoproteins from pathogenic bacteria are known inducer of immune responses during infection, generally through interaction with TLR2. It is therefore tempting to assume a potential mechanism in which the efflux ABC transporter would drive the release of the putative lipoprotein, the latter being the bacterial effector responsible for activation of NF-κB in HT-29, potentially through a TLR2-independent manner since this receptor is absent. Future investigations will be necessary to precisely determine bacterial and eukaryotic effectors and mechanisms involved as well as their potential implication in the existing relationship between some commensals (as *B. vulgatus*) and pathogenesis of IBD. Our approach provided other excellent candidate metagenomic clones for investigation of novel host-commensal interaction mechanisms and these clones are currently being characterized using a similar strategy. In future work, the NF-κB reporter cell lines will be used in the automated screening of new metagenomic libraries in the frame of the MetaHIT project. This approach targeting NF-κB pathway is currently under development for other signaling pathways including AP1, p53, c-Myc and PPARγ, as well as for genes of interest including Thymic Stromal Lymphopoietin (TSLP) and angiopoietin-like protein 4/fasting-induced adipose factor (Angptl4/FIAF). We think that our functional metagenomic strategy will allow identification of genes that human- associated intestinal microbes may use to affect expression of genes and pathways of their host, thus opening new avenues towards a better understanding of microbiota-host interactions and potential development of new probiotics or therapeutic molecules. # Supporting Information We thank Julien Cherfils and Isabelle Cremer, INSERM UMRS 872, for the technical and material assistance they provided for the flow cytometry analysis. Thanks to Thomas Clavel for is participation in culture of metagenomic clones and lysates preparation. We are grateful to Mihai Covasa for revising the manuscript for English. [^1]: Conceived and designed the experiments: OL SDE FL JD HMB. Performed the experiments: OL. Analyzed the data: OL AC JT KG FL HMB. Contributed reagents/materials/analysis tools: KG FB. Wrote the paper: OL AC KG HMB. [^2]: OL was an employee of LibraGen and this does not alter adherence to all the PLoS ONE policies on sharing data and materials.

# Introduction Vitiligo, a cosmetic disfigurement disorder, may lead to psychological and social stigma, particularly in people with dark and intermediate skin tones. It is characterized by circumscribed milky white patches on the skin affecting about 0.06–2.28% of the world population. Based on a few dermatological outpatient records, the prevalence of vitiligo is found to be 0.5 to 2.5% in India, wherein Gujarat and Rajasthan states have the high prevalence i.e. \~8.8%. The exact etiopathology of vitiligo is not defined, however, based on extensive studies various theories such as oxidative stress, autoimmunity, and neurochemical hypothesis have been proposed to explain the underlying pathomechanisms. Autoimmunity has been strongly involved in the development of disease, as 30% of vitiligo cases are affected with at least one of the concomitant autoimmune disorders. Several studies including ours have identified critical role of CD8⁺ cytotoxic T cells in melanocyte destruction. Generation of antigenic peptides and their transport across the membrane of the endoplasmic reticulum for assembly with major histocompatibility complex (MHC) class I molecules are essential steps in antigen presentation to cytotoxic T lymphocytes. Genes within MHC class II loci along with genes involved in antigen processing and presentation i.e., proteasome subunit beta 8 *(PSMB8)* and transporter associated with antigen processing 1 *(TAP1)* have been reported to be associated with several autoimmune diseases including vitiligo. The *PSMB8*, often referred as *LMP7* encodes interferon (IFN)-γ inducible subunit of immune proteasome i.e., β5i involved in degradation of ubiquitinated intracellular proteins into peptides that are especially suited for presentation by MHC class I molecules. Whereas, *TAP1* encode a subunit of an IFN-γ inducible heterodimer which binds with peptides cleaved by the proteasome and transports them to be loaded into nascent MHC class I molecules for presentation to CD8⁺ T cells. The genome-wide association study (GWAS) on generalized vitiligo revealed that the association of *TAP1-PSMB8* seems to derive from linkage disequilibrium with major primary signals in the MHC class I and class II regions. Out of 8 different single nucleotide polymorphisms (SNPs) of *PSMB* and *TAP* gene region studied, *PSMB8* intron 6 G/T and *TAP1* exon 10 A/G were found to be significantly associated with vitiligo in the Western population. Another study showed significant association of *TAP1* exon 10 A/G polymorphism with vitiligo in Saudi population but not for *PSMB8* intron 6 G/T polymorphism. The nature of the genetic association may vary according to different ethnic backgrounds. However, despite having high prevalence of vitiligo in Gujarat, there are no reports of *PSMB8* and *TAP1* polymorphisms so far. Hence, the present study aims, (i) to investigate the association of *PSMB8* intron 6 (rs2071464) and *TAP1* exon 10 (rs1135216) polymorphisms and (ii) to estimate transcript levels of *PSMB8* and *TAP1* using a case-control approach. # Materials and methods ## Study subjects We report a case-control study including 509 ethnically age and gender matched controls and 378 patients with vitiligo from Gujarat. Unaffected individuals of age between 5 to 60 years were recruited in the study. None of the unaffected individuals had any evidence of vitiligo and any other disease. Patients with vitiligo who referred to S.S.G. Hospital at Vadodara, Gujarat, India were recruited in the study. The inclusion criteria followed were: outpatients of age between 5 to 60 years and both the parents should be Gujarati by birth. Patients with other diseases and those unwilling to participate in the study were excluded. The diagnosis of vitiligo by dermatologists was clinically based on characteristic skin depigmentation with typical localization and white color lesions on the skin, under Woods lamp. Generalized or non-segmental vitiligo (GV) was characterized by depigmented patches varying in size from a few to several centimeters in diameter, involving one or both sides of the body with a tendency towards symmetrical distribution. Whereas localized or segmental vitiligo (LV) typically has a rapidly progressive but limited course, depigmentation spreads within the segment during a period of 6–24 months and then stops; further extension is rare. Following clinical criteria to proposed by Falabella *et al*., and discussed in the Vitiligo Global Issues Consensus Conference 2012, were used for characterizing stable vitiligo (SV): (i) lack of progression of old lesions within the past 2 years; (ii) no new lesions developing within the same period. Active vitiligo (AV) was defined as the appearance of new lesions and spreading of existing lesions observed during past two-year duration. The importance of the study was explained to all participants and written consent was obtained. Informed consent in written was obtained from the next of kin, caretakers, or guardians on behalf of the minors/children enrolled in the study. The study plan and consent forms were approved by the Institutional ethical committee for human research (IECHR), Faculty of Science, The Maharaja Sayajirao University of Baroda, Vadodara, Gujarat, India (FS/IECHR/BC/RB/1). Demographic characteristics of the patients are provided in Supporting information as ‘’. ## Genomic DNA extraction Genomic DNA was extracted from PBMCs using ‘QIAamp^TM DNA Blood Kit’ (QIAGEN Inc., Valencia, CA 91355, USA) according to the manufacturer’s instructions. After extraction, concentration and purity of DNA were estimated spectrophotometrically, quality of DNA was also determined on 0.8% agarose gel electrophoresis and DNA was stored at -20°C until further analyses. ## Genotyping of *PSMB8* rs2071464 polymorphism Polymerase chain reaction- Restriction Fragment Length Polymorphism (PCR-RFLP) technique was used to genotype *PSMB8* rs2071464 polymorphism. The primers used for polymerase chain reaction are mentioned in. The reaction mixture of the total volume of 20 μL included 3 μL (100ng) of genomic DNA, 11 μL nuclease-free H₂O, 2.0 μL 10x PCR buffer, 2 μL 2 mM dNTPs (Genei^TM, Bangalore, India), 1 μL of 10 pM corresponding forward and reverse primers (Eurofins^TM, India), and 0.3 μL (3 U/μL) Taq Polymerase (Genei^TM, Bangalore, India). Amplification was performed Eppendorf Mastercycler Gradient Thermocycler (Eppendorf^TM, Germany) according to the protocol: 95°C for 10 minutes followed by 45 cycles of 95°C for 30 seconds, 58°C for 30 seconds and 72°C for 30 seconds, and 72°C for 10 minutes. The amplified products were checked by electrophoresis on a 2.0% agarose gel stained with ethidium bromide. Restriction enzyme was used for digesting the PCR product. 15 μL of the amplified products were digested with 1U of *Hha* I (Fermentas^TM, Thermo Scientific, Waltham, MA) in a total reaction volume of 20μL as per the manufacturer’s instruction. The digestion products were resolved with 50 bp DNA ladder (Novagen^TM, Perfect DNA ladder) on 3.5% agarose gel stained with ethidium bromide and visualized under E-Gel Imager (Life Technologies^TM, Carlsbad, CA). Representative gel image is shown in. More than 10% of the samples were randomly selected for confirmation and the results were 100% concordant (analysis of the chosen samples was repeated by two researchers independently). Six samples of each genotype were also confirmed by sequencing using carefully designed primers. ## Genotyping of *TAP1* rs1135216 polymorphism *TAP1* rs1135216 polymorphism was genotyped using amplification refractory mutation system-polymerase chain reaction (ARMS-PCR) method. DNA was amplified in two different PCR reactions with a generic antisense primer and one of the two allele-specific sense primers. To assess the success of PCR amplification in both the reactions, an internal control of 407 bp was amplified using a pair of primers designed from the nucleotide sequence of the human growth hormone (*HGH*). The reaction mixture of the total volume of 15 μL included 3 μL (100 ng) of genomic DNA, 4.7 μL nuclease-free H₂O, 1.5 μL 10x PCR buffer, 1.5 μL 2mM dNTPs (Genei^TM, Bangalore, India), 1 μL of 10 pM allele- specific and common primers (Eurofins^TM, India), 1 μL of 10 pM control primers (*HGH*), and 0.3 μL (3U/μL) Taq Polymerase (Genei^TM, Bangalore, India). Amplification was performed using a Mastercycler Gradient PCR (Eppendorf^TM, Germany) according to the protocol: 95°C for 10 minutes followed by 45 cycles of 95°C for 30 seconds, 61°C for 30 seconds, and 72°C for 30 seconds, and 72°C for 10 min. The PCR products were resolved on 3.5% agarose gel stained with ethidium bromide along with 50bp DNA ladder (Novagen^TM, Perfect DNA ladder) and visualized under E-Gel Imager (Life Technologies^TM, Carlsbad, CA). Two amplicons were available for each sample (one each specific for A or G allele). Representative gel image is shown in. More than 10% of the samples were randomly selected for confirmation and the results were 100% concordant (analysis of the chosen samples was repeated by two researchers independently). Six samples of each genotype were also confirmed by sequencing using carefully designed primers. ## Estimation of *PSMB8* and *TAP1* transcript levels ### RNA extraction and cDNA synthesis Total RNA from PBMCs was isolated and purified using the Ribopure-blood Kit (Ambion^TM Inc., Austin, TX, U.S.A.) following the manufacturer’s protocol. RNA integrity was verified by 1.5% agarose gel electrophoresis, RNA yield and purity was determined spectrophotometrically at 260/280 nm. RNA was treated with DNase I (Ambion^TM inc. Texas, USA) before cDNA synthesis to avoid DNA contamination. cDNA synthesis was performed using 1 μg of total RNA by RevertAid First Strand cDNA Synthesis Kit (Fermentas^TM, Vilnius, Lithuania) according to the manufacturer’s instructions in Eppendorf Mastercycler Gradient Thermocycler (Eppendorf^TM, Germany). ### Quantitative realtime PCR (qPCR) The expression of *PSMB8*, *TAP1* and Glyceraldehyde 3-phosphate dehydrogenase (*GAPDH*) transcripts were measured by qPCR using gene specific primers (Eurofins^TM, Bangalore, India) as shown in. Expression of the *GAPDH* gene was used as a reference. qPCR was performed in duplicates in 20 μl volume using LightCycler^® 480 SYBR Green I Master (Roche^TM Diagnostics GmbH, Mannheim, Germany) following the manufacturer’s instructions. The thermal cycling conditions included an initial activation step at 95°C for 10 min, followed by 45 cycles of denaturation, annealing, and extension (95°C for 10 sec, 65°C for 15 sec, 72°C for 20 sec). The fluorescence data collection was performed during the extension step. At the end of the amplification phase, a melt curve analysis was carried out to check the specificity of the products formed. The PCR cycle at which PCR amplification begins its exponential phase and product fluorescence intensity finally rises above the background and becomes visible was considered as the crossing point-PCR-cycle (C_P) or cycle threshold (C_T). The ΔC_P value was determined as the difference between the cycle threshold of target genes (*PSMB8/TAP1*) and reference gene (*GAPDH*). The difference between the two ΔC_P values (ΔC_P Controls and ΔC_P patients) was considered as ΔΔC_P to obtain the value of fold expression (2^-ΔΔCp). ## Estimation of PSMB8 protein expression ### Western blot analysis Five ml blood was drawn from healthy controls and patients with active GV and collected in EDTA vials. Red blood cells were lysed with RBC lysis buffer (0.17 M Tris/ 0.16 M NH₄Cl pH 7.2) and the remaining leukocytes were washed in PBS, and lysed in lysis buffer (1 mM EDTA, 50 mM Tris-HCl pH 7.5, 70 mM NaCl, 1% Triton, 50 mM NaF) containing 1x proteinase inhibitors (Sigma, Bangalore, India). Protein concentration was determined by Bradford assay (HiMedia Laboratories, India) and 20μg protein was loaded on 12% SDS-PAGE along with Precision Plus Protein™ Dual Color Standards (Bio-Rad, Germany). Protein was electro-blotted on PVDF membrane at 100 V for 1.5 hrs. Following the transfer, the membrane was blocked with 5% blocking buffer (5% BSA and 0.1% Tween-20 in PBS) for 1 hr at room temperature. The membrane was incubated overnight with primary antibody against LMP7/PSMB8 (ab58094). After incubation the membrane was washed four times with PBS-T (PBS containing 0.1% Tween 20) for 15 min. and incubated with a secondary anti-mouse antibody (Bangalore Genei, India) at room temperature for 1 hr. The membrane was similarly washed four times with PBS-T and protein bands on the membrane were then visualized by using Bio-Rad Clarity™ western ECL substrate (Bio-Rad, Germany) and signal was scanned using the Chemidoc™ Touch Gel Imaging System (Bio-Rad, Germany). Intensities of target proteins were normalized with that of total protein loading by staining the membrane with Ponceau. Densitometric analysis of the protein bands was calculated by ImageJ software. ## Statistical analyses Hardy-Weinberg equilibrium (HWE) was evaluated for both SNPs in patients and controls by comparing the observed and expected frequencies of the genotypes using chi-square analysis. Distribution of the genotypes and allele frequencies of polymorphisms in different groups were compared using chi-square test with 2×2 contingency tables. Major genotype/allele was used as a reference. Multiple comparisons were controlled by the Bonferroni’s method. Odds ratio (OR) with 95% confidence interval (CI) for disease susceptibility was also calculated. Haplotype and LD analysis were carried out using <http://analysis.bio-x.cn/myAnalysis.php>. For analyses of the transcript and protein levels unpaired t-test and one-way ANOVA were applied. Tukey’s multiple correction was applied for multiple testing and the p-values were adjusted. All the statistical tests were carried out using Prism 6 software (Graph Pad Software, USA). ## Bioinformatics analysis *In silico* prediction tools HaploReg v4.1 and Regulome DB were employed to predict the functional impact of non-coding polymorphism. *In silico* prediction tools SIFT, PANTHER, I-MUTANT SUITE, POLYPHEN, MUPRO were employed to predict the impact on the protein due to single amino acid variation. SNPs and GO predicts the variation effect which might terminate into a disease like a trait. The details have been described in ‘Supporting Information’ file. # Results ## *PSMB8* rs2071464 polymorphism in vitiligo Genotyping of *PSMB8* intron 6 rs2071464 SNP by PCR-RFLP using *Hha* I and subsequent sequencing results revealed that there is C\>T nucleotide change instead of previously reported G\>T change, which falls in the *Hha* I recognition/restriction site and was imputed to *PSMB8* rs2071464 SNP. The observed genotype frequencies of *PSMB8* rs2071464 SNP among the controls were in accordance (*p* = 0.071) whereas, genotype frequencies among the patients were deviated (*p* = 0.001) from HWE. When ‘C’ allele and CC genotype were used as reference group, the frequencies of the variant ‘T’ allele and homozygous ‘TT’ genotype were significantly lower in patients with vitiligo as compared to controls (49% vs. 54%, *p* = 0.031; 19% vs. 27%, *p* = 0.026 respectively) but it did not remain significant after Bonferroni’s correction. The protective role of ‘TT’ genotype in patients was suggested by OR = 0.629 (95% CI = 0.41–0.94). OR suggests that the minor allele ‘T’ might have the protective role in the disease pathogenesis. Analysis based on types of vitiligo revealed significantly lower frequency of ‘TT’ genotype (18% vs. 27%, *p* = 0.019) and ‘T’ allele (48% vs. 54%, *p* = 0.024) in patients with GV as compared to controls. No significant difference in genotype and allele frequencies between patients with LV in comparison to patients with GV or controls. Interestingly, a similar trend was observed upon analysis based on the activity of the disease. Predominantly increased frequency of the risk genotype ‘CC’ (24% vs.19%) and allele ‘C’ (53% vs. 46%) was observed in patients with AV as compared to controls. The frequency of the protective genotype ‘TT’ (18% vs. 27%, *p* = 0.005) and allele ‘T’ (47% vs. 54%, *p* = 0.007) was significantly lowered in comparison to controls. However, no significant difference in allele and genotype frequencies was observed between patients with AV and SV. ## *TAP1* rs1135216 polymorphism in Vitiligo Both, control and patient groups were following HWE (*p* = 0.663 and *p* = 0.167 respectively;). Major allele ‘A’ and ‘AA’ genotype were considered as the reference. The allele and genotype frequencies were not significantly different in patients and control. *TAP1* SNP when analyzed based on the type of vitiligo, no significant difference in genotype and allele frequencies was observed between patients with GV and LV with respect to unaffected controls. Analysis based on the activity of the disease also showed no significant difference among the genotypes as well as allele frequencies. ## Linkage disequilibrium and haplotype analyses LD analysis revealed that two polymorphisms investigated i.e., *PSMB8* rs2071464 and *TAP1* rs1135216 were in low LD association (D’ = 0.432, r² = 0.044). Haplotype evaluation of the two polymorphic sites was performed and the estimated frequencies of the haplotypes were not significantly different between patients and controls (global *p* = 0.278;). ## *PSMB8* transcript and protein levels in vitiligo Analysis of *PSMB8* transcript levels revealed a significant decrease in expression of *PSMB8* transcripts in patients as compared to controls (*p* = 0.002;) after normalization with *GAPDH* expression. The 2^-ΔΔCp analysis showed approximately 0.52-fold decrease in the expression of *PSMB8* transcript levels in patients, as compared to controls. Interestingly, analysis based on type and activity of the disease revealed that *PSMB8* transcript levels were significantly decreased in patients with GV as well as AV in comparison to controls (*p* = 0.007 and *p* = 0.006 respectively;), suggesting a role in the autoimmune basis of the disease. However, there was no significant difference in patients with LV and SV as compared to controls (*p* = 0.090 and *p* = 0.112 respectively;). Also, no significant difference in transcript levels was observed between GV vs LV and AV vs SV patients. When expression of *PSMB8* transcripts was monitored in different age at onset groups of patients, no significant difference was observed in any of the age of onset groups i.e., 21–40, 41–60 and 61–80 years when compared with 1–20 years. Gender-based analysis also showed no significant difference in *PSMB8* transcripts in both the groups (*p* = 0.396;). Furthermore, the decreased transcript expression of *PSMB8* in patients with vitiligo was confirmed at protein level by western blot analysis in PBMCs of healthy controls (n = 6) and patients with active GV (n = 7). A significant decrease (*p* = 0.0460) in expression of PSMB8 was observed in patients as compared to controls. ## Genotype—phenotype correlation for *PSMB8* rs2071464 polymorphism Further, the expression of *PSMB8* transcripts was analyzed with respect to *PSMB8* rs2071464 genotypes. Interestingly, *PSMB8* transcript levels were significantly reduced in individuals with the susceptible CC genotype when compared with CT and TT genotypes (*p* = 0.009 and *p* = 0.003, respectively;). However, no significant difference in *PSMB8* transcripts levels was observed between individuals with the CT and TT genotypes. ## *TAP1* transcript levels in vitiligo Analysis of *TAP1* transcript levels was carried out after normalization with *GAPDH* expression. No significant difference in expression of *TAP1* transcripts was observed (*p* = 0.553) between patients and controls. The 2^-ΔΔCp analysis showed approximately 1.12- fold change in expression of *TAP1* transcript in patients as compared to controls. Analysis based on type of the disease suggested no significant difference in *TAP1* transcript levels in patients with GV and LV in comparison to controls (*p* = 0.090 and *p* = 0.219 respectively;). Moreover, there was no significant difference in patients with AV and SV as compared to controls (*p* = 0.671 and *p* = 0.291 respectively;). When expression of *TAP1* transcripts was monitored in different age at onset groups of patients, no significant difference was observed in any of the age of onset groups i.e., 21–40, 41–60 and 61–80 years when compared with 1–20 years. Gender-based analysis showed no significant difference in *TAP1* transcripts in both the groups. ## Bioinformatics analyses Analysis of functional consequences of *PSMB8* rs2071464 by RegulomeDB was scored 6 and classified as having minimal binding evidence. HaploReg v4.1 predicted *PSMB8* rs2071464 could alter 7 DNA motifs. RegulomeDB revealed that the Chromatin state is altered favoring strong transcription and genic enhancer by the polymorphism in peripheral blood cells (<http://www.regulomedb.org/snp/chr6/32809075>). Analysis by HaploReg v4.1 further confirmed the enhancer chromatin state in peripheral blood and T cells due to the polymorphism (<http://archive.broadinstitute.org/mammals/haploreg/det ail_v4.1.php?query=&id=rs2071464>). *TAP1* exon 10 A\>G leads to variation in TAP1 protein from Asp to Gly at position 637. PANTHER tool showed variation Asp to Gly at position 637 is not deleterious for TAP1 function, with the score of 0.3456. POLYPHEN tool showed that the substitution does not affect the phenotype or have damaging effects on the function of TAP1 protein. I-MUTANT and MUPRO predictions revealed decreased stability of Asp637Gly variants compared to native structure, which might affect the protein function. SNPs AND GO tool revealed that the variant doesn’t show disease like trait.. # Discussion The association of MHC region has been implicated in several GWAS on vitiligo including in Indian subcontinent. Association of MHC class II region with generalized vitiligo was reported in European-derived white population by Jin *et al*.. The strong link between autoimmune diseases and MHC class II genes suggests that abnormalities in MHC class II gene products may play a crucial role in vitiligo susceptibility. Interestingly, the association of GV with SNPs in the *PSMB8-TAP1* region of the MHC has been reported to derive from LD with primary association signals in the MHC class I and class II regions. Any alterations in function or expression of *PSMB8* or *TAP1* proteins could potentially affect the antigenic repertoire expressed on the cell surface and may alter peripheral tolerance. Several studies have addressed the association of *PSMB8* and *TAP1* polymorphisms in patients with vitiligo; however, studies revealing the impact of these polymorphisms at transcript and protein levels are few. The present study suggests the association of *PSMB8* rs2071464 SNP with GV as well as with the disease activity (AV); however, *TAP1* rs1135216 SNP was not associated with vitiligo in Gujarat. Our results are in accordance with the previous study reported in Western population for *PSMB8* SNP. In contrast, two studies have found *TAP1* exon 10 SNP to be associated with vitiligo in Saudi population, and this may be due to differences in the ethnicity. Birlea *et al*., have addressed 34 SNPs spanning *TAP1-PSMB8* region in GWAS and the meta- analysis study in GV patients; however, no association was observed for *TAP1* rs1135216 and *PSMB8* rs2071627 SNPs. The *PSMB8* encodes IFN-γ inducible subunit (b5i/LMP7) of the immunoproteasome, which degrades the ubiquitin-tagged cytoplasmic proteins into peptides that are especially suited for presentation by MHC class I molecules to CD8⁺ cytotoxic T cells. Significant association of *PSMB8* rs2071464 leads us to speculate some functional consequences of this SNP in the disease pathogenesis. Intriguingly, the decreased expression was associated with the susceptible ‘C’ allele of *PSMB8* rs2071464; however, the mechanism is not yet clear. *In silico* prediction tools have predicted that *PSMB8* rs2071464 C\>T variation might alter chromatin to enhancer state and result in induced gene expression in peripheral blood cells. Recent studies have explored that several of *cis*-regulatory SNPs could affect histone modifications and change chromatin state transition from repressor to enhancer state. Our results correlate with these findings as higher expression of PSMB8 was observed in individuals having variant ‘TT’ genotype as compared to ‘CC’ genotype. A significant decrease in transcript as well as protein expression of *PSMB8* in PBMCs of patients with GV and AV is revealed in the present study. Our findings have recently been supported by the blood transcriptomics analysis of vitiligo patients which revealed significant down regulation of *PSMB8* expression in patients. In addition, another recent study has demonstrated the IFN-γ induced lower expression of *PSMB8* in PBMCs of vitiligo patients as compared to controls. Moreover, it has been observed that the down-regulation of *PSMB8* expression leads to suppression of MHC class I molecule surface expression. In addition, the IFN-γ induced immunoproteasomes have been associated with the improved processing of MHC class I antigens. It has been reported that the presentation of a majority of MHC class I epitopes was strikingly reduced in immunoproteasome-deficient mice. Moreover, Xu *et al*., have also reported a significant decrease of 26S proteasome in lesions of vitiligo patients. Thus, the decreased expression of *PSMB8* in the present study, in conjunction with the above-discussed studies advocates the possibility of reduced MHC class I molecules in the patients and indicates the crucial role of *PSMB8* in vitiligo immunopathogenesis. Autoimmune diseases are characterized by decreased expression of MHC class I on lymphocytes. The appropriate MHC class I expression is necessary for self- tolerance, and abnormalities in such expression may lead to autoimmunity. Zaiss *et al*., have reported that proteasome immuno-subunits protect against the development of CD8⁺ T-cell mediated autoimmune diseases. They showed that mice deficient for the immune-subunits β5i/LMP7 and β2i/MECL-1 develop early-stage multi-organ autoimmunity following irradiation. Several reports including ours have suggested a decreased CD4⁺⁄CD8⁺ ratio in vitiligo patients, indicating the prevalence of CD8⁺ cells in patients. Thus, a decrease in immunoproteasome levels may lead to a breakdown of self-tolerance, resulting in an increase of CD8⁺ T cells directed towards melanocytes in predisposed individuals which could not be checked upon by the insufficient numbers and functionally deficient regulatory T cells (Tregs) in patients with vitiligo. Transport of antigenic peptides across ER membrane is mediated by TAP1 and TAP2 molecules. We did not find a significant association of *TAP1* rs1132516 SNP with vitiligo, as well as there was no difference in *TAP1* transcript levels between cases and controls. The ‘G’ allele occurred predominantly in AV patients compared to SV however, it was considered non-significant due to Bonferroni’s correction. The higher frequency of ‘G’ allele in AV patients indicates its involvement in the autoimmune basis of vitiligo. The bioinformatics analysis revealed that *TAP1* rs1135216 SNP (Asp637Gly) leads to a decrease in the stability of TAP1 protein. Moreover, it has been reported that the polymorphism in *TAP1* gene product did not show any measurable change in protein function but has an influence on peptide selectivity. The binding of antigenic peptides to class I molecules depends on both length (usually 8–10 residues) and sequence. The specificity of these reactions and their biological functions are affected by the 3D conformation of the peptide, HLA complexes, compatibility of the peptide sequence with its HLA class I binding pocket etc. Interestingly, significant differences in the amino-acid signatures of the peptide-binding pockets of MHC class I α chains as well as class II β chains were observed between vitiligo patients and unaffected controls. Though *TAP1* SNP was not associated with vitiligo but the predominant presence of ‘G’ allele in combination with other SNPs in this region might affect the peptide selectivity in patients. *PSMB8* polymorphism in addition to previously reported susceptibility loci such *TNFA*, *TNFB*, *IL1B*, *IFNG*, *NALP1*, *IL4* etc. demonstrate immunogenetic predisposition in vitiligo patients from Gujarat. Overall, studies implicate a break in immunological tolerance in vitiligo. A similar type of etiopathology has been observed in alopecia areata (a common autoimmune disorder that often results in unpredictable hair loss). The melanocyte is the main autoimmune target in both the disorders. Both are IFN-γ dependent and shares common immunogenetic loci such as *AIRE*, *CTLA4*, *NALP1*, and MHC region. Surprisingly, the co-occurrence of vitiligo and alopecia areata is rare. Unequal expression of MHC class I and II might be a base for the reverse correlation between the incidents of vitiligo and alopecia areata. Hence, the genes involved in antigen processing might have a role in the breakdown of immune tolerance and precipitation of vitiligo. # Conclusion In conclusion, the association of *PSMB8* rs2071464 polymorphism with generalized and active vitiligo suggests defective antigen processing which might influence the peptide repertoire presented to the immune cells targeting melanocytes. However, further replicative studies and *in vitro* functional studies for *PSMB8* and *TAP1* are needed to delineate the role of defective antigen processing and presentation pathways in vitiligo pathogenesis. # Supporting information We thank all the subjects for their participation. We thank Dr. Y.S. Marfatia, Department of Skin and VD, The M.S. University of Baroda, Vadodara, India for his kind help in the recruitment of vitiligo patients. We thank Dr. Edwin Ostrin, MD Anderson Cancer Center, Houston, Texas, USA for providing anti-PSMB8 antibody. SDJ and MSM thanks to UGC, New Delhi for awarding SRF. MS thanks to CSIR, New Delhi for awarding SRF. [^1]: The authors have declared that no competing interests exist.

# Introduction Although biodiversity-loss rates are actively investigated to determine their relationship with habitat fragmentation, climatic and social changes, there is also a need to assess current rates against longer-term loss rates that occurred in response to past environmental change. Such assessments could be valuable for highlighting the influence of ecological stress and disturbance on biodiversity. In purely theoretical terms, the richness (*S*) of communities at any period of time can be defined as the entire number of species living in the particular area of interest. Among several indices of species richness, the rarefaction method – is often used because it is effective for standardizing the species richness of assemblages of different sizes to a common number of individuals or samples, i.e., the ‘expected taxonomic richness, *E*(*Tn*)’. This standardization is necessary due to the non-linear relationship between richness and the number of individuals recorded. Since Sanders’ original formulation and the subsequent correction by Heck et al., this method has been often used for paleoecological issues to estimate taxonomic richness from (sub)fossil records, where sample size is not well controlled. Examples include terrestrial plants (based on pollen e.g.), algae (based on diatoms), insects (from chironomids) or testate amoebae. However, there is increasing awareness of the application of rarefaction to fossil records. Among the assumptions made when estimating past richness using the number of taxa (rarefied or not) in sedimentary samples, the following factors have been critically addressed: (i) count sizes ; (ii) the effect of evenness, ; (iii) differential productivity and dispersal of taxa ; (iv) lack of taxonomic precision ; (v) the spatial scale for which the sub-fossilized remains are representative of the population actually producing them ; and (vi) the temporal resolution of samples. An additional restriction on estimating past taxonomic richness from paleoecological records is that only a fraction of species living in a particular area can become archived. Changes in floristic richness are commonly inferred from pollen records because of their efficient dispersal and their abundance in sediments. In contrast, despite the growing number of macroremain records that are suitable for reconstructing vegetation-stand dynamics, and biomass, taxonomic richness has been rarely estimated based on plant macroremains mainly because of mathematical difficulties associated with manipulating the low counts that are frequently recorded. However, tandem studies of pollen and macrofossils confirm the different spatial smoothing and taxonomic resolution of the two types of data, with macroremains providing records of vegetation composition and species ranges with higher spatial details and, for some taxa, higher taxonomic resolution –. Here we describe a new method to estimate and compare the taxonomic richness and evenness of vegetation based on pollen and plant-macrofossil records. Van der Knaap proposed the use of pollen influxes to assess palynological richness and to cope with differential evenness of pollen assemblages. In addition to van der Knaap, we introduced the use of interpolated influx that may be regarded as a way to deal with potential pitfalls associated with the species-time-area type relationships, and assumed that rarefaction would be more valuable when used to handle sampling-effort discrepancies within equally sized samples both in terms of volume and time. Finally, we specifically address some issues that are of concern for estimating richness from paleorecords in general, i.e., different temporal resolutions, sample volumes and low-count sizes. # Materials and Methods ## Species, Volume, Count Size, and Time Relationships Usually sediments are subsampled on cores collecting samples of equal volume (cm³) and thickness (cm on depth) at set intervals resulting in a series of snapshots each of which gives an integrated picture for different periods of time. Depending on the sediment-accumulation rate (yr.cm⁻¹), these snapshots represent anything between 1 and \>100 years. Furthermore, count sizes tend not to be standardized within a sequence and it sometimes happens that sample volume varies, e.g. for the analysis of plant macroremains. The potential bias for richness estimates related to the variations in sample volume, temporal sample resolution, and count sizes is, in principle, equivalent to “species-time-area relationships” in modern vegetation relevés. In fact, differences in taxonomic richness and evenness may occur between samples of the same sedimentary sequence because higher taxa numbers can be expected in samples of larger volumes, in samples that accumulated over a longer time span, and when count sizes are larger. Therefore, before estimating richness, counts were resampled to a constant sample resolution (yr.sample⁻¹) and sample volume (cm³). To do this, we first interpolated the initial sample depths at a constant time interval (*w*) by interpolating between dated top and base of each sample using the age∼depth model. Subsequently, we determined the contribution of initial counts and volumes to each interpolated sample; this step involved the calculation of a proportion matrix. Finally, the interpolated counts were transformed to concentrations (numbers.cm⁻³) that were in turn converted to influxes (numbers.cm⁻².yr⁻¹) by multiplying concentrations by sediment accumulation rate (cm.yr⁻¹) for each sample, resulting in a regularly spaced time series of pollen or macroremains. This procedure uses the pretreatment method of charcoal records for reconstruction of fires as proposed by Long et al. and was adapted for the pollen and macroremain records by modifying the interpolation procedure from the freely available CharAnalysis software {<https://sites.google.com/site/charanalysis/Higuera>, 2009 \#308}. ## Individual Based Rarefaction Method To estimate taxonomic richness \[*E*(*T_n*)\] based on rarefaction analysis, we used Heck et al. equation, which was first applied in the context of paleoecology by Birks & Line :where E(Tn) is the rarefied taxonomic richness, N is the influx sum of particles in a sample, Ni is the influx of particles for the ith taxon, and n the minimum influx along a sedimentary record. This method is particularly suitable for proxies characterized by large total numbers of particles and taxa per sample (e.g. pollen records) that allow high-influx sizes. In contrast, the total number of taxa and particles per sample of low- influx proxies (e.g. macroremains) are often equal to zero. In situations where n = 0, the rarefaction cannot be calculated. To avoid that within a resampled record n = 0, we used the procedure described in the previous paragraph (Species, volume, count size, and time relationships) with increasing resampling time intervals w and selected the minimal w for which all resampled samples displayed influx sums larger than zero. This procedure was not necessary for the pollen records where the minimum influx sums were always \>\>0, and the records could be resampled using a w equal to the median time resolution. Rarefaction is sensitive to sample size. Furthermore, calculating rarefaction on low influxes such as those usually observed for macroremain records (usually \<1 particle.cm⁻².yr⁻¹) would result in richness estimates difficult to interpret. To circumvent this problem, we used a simple conversion procedure of the interpolated taxa matrix to obtain a new converted matrix in which the initial minimum influx sum *n* was converted to *n’*. This procedure was replicated 500 times with increasing *n’* (*n’* = 1,2…500) and, rarefaction was calculated on each of those matrices. To analyze the long-term richness patterns, we rescaled the *E*(*T_n*) time series using a simple minimax transformation by subtracting the minimum *E*(*T_n*) value and dividing by the range of *E*(*T_n*) values. To assess the significance of *E*(*T_n*) changes through time, median *E*(*T_n*) values for adjacent 1000 years’ time windows were compared using a non-parametric Kruskal Wallis ANOVA. When a significant difference was revealed, post-hoc least-significant difference tests were realized to assess the significance between pairs of time periods. ## Evenness Biodiversity estimates are a combination of the taxa number and the species evenness. Therefore, we also estimated the evenness of fossil assemblages using Hurlbert’s probability of interspecific encounter (Δ1,) such as:were *p_i* is the proportion of the *i*^th taxon and *S* the total number of taxa in the sample. Finally, we considered important to distinguish between functional groups in the fossil assemblages, i.e. between tree, shrubs and herbs taxa. This allowed us to highlight the main temporal trends of plant community composition. The pollen data from Lac de Fully can be accessed from the European Pollen Database (<http://www.europeanpollendatabase.net/>), the macroremain data is available at <http://doi.pangaea.de/10.1594/PANGAEA.807921>. The statistical codes and functions were developed under Matlab language (tested using Matlab R2011b) and are freely available at <http://blarquez.com/page/Codes/>. ## The Test Dataset In order to test the new algorithm, we used the macroremain and pollen records from Lac de Fully (2135 m a.s.l), a small subalpine lake located on a south- facing plateau in the Valais (inner Swiss Alps). A full description of the site, sediment sampling, the age∼depth model, sample processing, and identification of plant macroremains (e.g. needles, leaves, seeds, bud scales) and pollen is presented in Finsinger and Tinner. The pollen and macroremain records from Lac de Fully cover the past ∼11,000 years. In total, 84 samples were analyzed for macroremains (\[min-max\] resolution: \[24–351\] years.sample⁻¹) and 62 samples were analyzed for pollen (\[50–379\] years.sample⁻¹). All ages are expressed in calibrated years before the present (hereafter “cal BP”; by convention the “present” year is AD 1950). In this study we chose to focus solely on the period between 11,000 and 3,000 cal BP, because only four macroremains samples were available for the past ∼3,000 years. # Results The minimal interpolation time windows *(w)*, that satisfied the condition that any resampled macroremain influx value must be greater than zero, was 91 years for the macroremain record. For the pollen record we interpolated the influxes using the median resolution w = 147 years. The *E(T_n)* values from the pollen record depict a larger dispersion compared to the macroremain record, which showed more conservative rarefaction reconstructions. Estimated richness for the pollen record range from c. 12 to 34 taxa during the 11,000–3,000 cal BP period, and was lower for the macroremain records which showed a maximum *E(T_n)* of c. 9 taxa. The pollen records displayed an evenness that ranged between 0.7 and 0.9 which is generally higher compared to macroremain Δ₁ that peaked at 0.8 at maximum. ## Pollen Richness and Evenness Pollen richness (Transformed *E*(*Tn*),) showed lowest values recorded between 11,000 and 10,000 cal BP (*p*\<0.05). After this, *E*(*Tn*) gradually increased up to 8,000–7,000 cal BP; this was accompanied by a gradual decrease in the percentage of shrubs in the pollen assemblages (mainly *Corylus*, data shown in Finsinger and Tinner). Then richness remained stable until the end of the record (p\>0.05). *E*(*Tn*) scores were maximal around 5,500 cal BP, when the percentages of tree pollen was high, mainly *Pinus cembra* and *Abies alba*. The evenness of pollen assemblages followed the same general trend as *E*(*Tn*) values. The main Δ₁ pattern is represented by a millennial monotonic increase in Δ₁ during the early Holocene and a maximum between 4,000 and 3,000 cal BP that is significantly different from the other periods (except the 7,000–6,000 cal BP period, *p*\<0.05). ## Plant Macroremain Richness and Evenness The transformed *E*(*Tn*) values for plant macroremains gradually increased from 9,000 to 7,200 cal BP in parallel with a decrease in the percentages of shrub and herb macroremains (mainly *Dryas octopetala* and *Juniperus communis* subsp. *nana*, data not shown) and an increase in tree taxa macroremains (mainly *Larix decidua* and *Pinus cembra*, data not shown), indicating either an increase in the altitude of the treeline or increasing woodland density at Lac de Fully. The onset of the Holocene (11,000 to 10,000 cal BP) showed high and gradually decreasing transformed *E*(*Tn*) values that are equivalent to the values recorded between 9,000–6,000 cal BP (*p*\>0.05). Finally, *E*(*Tn*) decreased between 7,200 and 3,000 cal BP, reaching values close to those from 10,000–9,000 cal BP at about 4,000–3,200 cal BP. The evenness of the macroremain record displayed the same temporal trend compared to richness; however we could report a greater stability in Δ₁ during the 9,000–6,000 period compared to *E*(*Tn*). Likewise the decrease in Δ₁ from 7,000 to 3,000 cal BP is less apparent but characterized by a higher variability of Δ₁ scores. # Discussion We developed an algorithm allowing the comparison of past trajectories of plant diversity, which is expressed in terms of richness and evenness, as inferred from pollen and macroremain records. We applied our original method to sedimentary data from a subalpine lake located near the Alpine treeline. Solving the problem of low-influxes, when assessing biodiversity, is not trivial. This process involved several steps including a pretreatment of the records and a replication of the rarefaction calculation using multiple deterministic parameters. These problems are important, particularly for plant macroremains, certainly for animal remains such as insects or land snails, and, testate amoebe in lake sediment in which they are few compared to peat in which they abound. Low-influx sum records pose difficulties for quantitative reconstructions of environmental variables and when investigating biotic responses to environmental changes, because the relative abundances of taxa cannot be accurately estimated. Clearly, low-influx sums also adversely affect the ability to reconstruct changes in taxonomic richness or evenness. The absence or the low concentration of macroremains of certain taxa in most archives used by paleoecologists (e.g. the central part of large lake) cannot be always attributed to their absence from the area surrounding the study site. Of course, if the absence of macroremains is purely due to taphonomic reasons (poor preservation, excessive distance from lake shore, etc.), then it would be advisable to exclude such samples from the record before estimating *E*(*Tn*). Macroremain records are largely dominated by woody taxa (trees and shrubs) because their plant tissues are better preserved in sediments than plant tissues of non-woody plants, which are mostly composed of cellulose, and because they are dispersed over longer distances. The attempts to calibrate the abundance of macroremains in terms of woody biomass demonstrate that such estimates may be affected by large uncertainty. Nevertheless, phases characterized by low concentration (or absence) of macroremains near the treeline can indicate treeline shifts in response to climate changes. In such cases conserving low- influx samples for rarefaction analysis is advisable since they can have an ecological meaning in such environments. Consequently, to solve the problem of low-influxes, our approach involving the calculation of influxes at regular intervals while maintaining the dependency through the original samples (based on count and volume interpolation) allows the use of rarefaction on the macroremain record. Our results indicate that, at the study site, floristic richness increased gradually between 10,000 and 7,000 cal BP. Pollen and macroremain records indicate a decrease in relative abundances of shrubs and an increase in trees, pointing to a treeline shift, likely in response to the early Holocene temperature increase. Shrub cover can constitute an important component of woody biomass in the subalpine belt. In such shrub-dominated ecosystems, floristic richness is lower than in adjacent conifer-forest stands because shrub cover reduces tree productivity and recruitment. The observed increasing floristic richness during the early Holocene at Lac de Fully may be related to the decrease in shrub cover that was associated with increasing tree cover. This gradual afforestation of the subalpine belt at Lac de Fully during the early Holocene, which contrasts with the more rapid afforestation at other sites in the Central Alps, can be interpreted as reflecting the effect of local climatic conditions. These may have delayed the expansion of closed stands of coniferous trees in the catchment of Lac de Fully until *c*. 8,200 cal. BP, when the climate shifted to more humid and less continental conditions. Diversity indices such as *E*(*T_n*) include both taxonomic richness and evenness. For both pollen and macroremain records evenness and richness follow similar trends ( *vs* 1c; 2b *vs* 2c), thus confirming the difficulty in distinguishing these two aspects of diversity. However, from our data, we cannot rule out that ecosystems have a higher richness when all pollen and macroremain types co-dominate. ## Conclusion The present study shows that low-influx paleoecological proxy records (here macroremain records) can be used to reconstruct local scale woody diversity. This suggests the need for further research, including the development of diversity indices and their application for plant communities as described herein for the rarefaction index. Such studies could provide a unique opportunity to analyze precisely the spatial and temporal dynamics of biodiversity across different ecological gradients, including productivity and elevation. Such reconstructions may also facilitate discussions about the long- term impacts of stresses (land-use, climate) or disturbances (e.g. fire, insect outbreaks, extreme climatic events) on the biodiversity of ecosystems and provide reference data for use in predicting relationships between environmental forcing, ecosystem functioning and biodiversity. We thank Thibaut Frejaville, Pim van der Knaap and the reviewers for stimulating suggestions, and SEES editing for proofreading. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: OB WF CC. Analyzed the data: OB WF CC. Contributed reagents/materials/analysis tools: OB. Wrote the paper: OB WF CC.

# Introduction Retaining functional activity and structural integrity of the enzyme is a desirable quality for a number of proteins finding application in biotechnology, food processing and other commercial industries, where high temperature processes are used. It is also important with regard to the insights that it might give on the overall stability of the proteins. Therefore, thermostability in proteins has been studied for a long time both experimentally as well as computationally. Experimental approaches primarily focus on sequence comparison between the mesophilic and thermophilic proteins followed by mutating the mesophilic proteins in order to gain thermostability. Computational studies involve structural comparison of the meso- and thermophilic proteins and correlating the differences to the stability of the proteins. Molecular dynamics simulations have also been employed to study the differences in the dynamics of these two types of proteins and correlating it to higher stability of the former. Thermostability is an emergent property of the protein that is imbibed through several structural factors that arise due to mutations of its mesophilic counter-part. Sequence and structure based comparison of the known thermophilic homologs of the mesophilic proteins has shed some light on the factors that might be contributing to the higher stability of thermophilic proteins. Although several factors can lead to stability of the proteins, some properties are found to be consistently correlated to the higher stability of proteins. These include, higher number of hydrogen bonds – and salt bridges, stabilization of secondary structures, introduction of disulphide bonds, higher number of Proline residues, higher polar surface area, shortening and stabilization of loops, and increased buried surface area after oligomerization. Recently, surrounding hydrophobicity has also been found to be an important property distinguishing the thermophilic from mesophilic proteins. However, there are certain confounding factors in such comparisons, and the differences observed may not be completely attributable to thermostability. Some of these differences might be present due to phylogenetic differences and thus be representative of just different ancestry rather than physico-chemical implications. In this regard, “directed evolution” has been proven to be successful in studying and designing enzymes with specific properties without being confounded by the phylogenetic differences inherent in the comparison of the natural homologs. This involves using a series of experimental techniques for mutation and recombination of the protein and selection on the basis of a certain property. Directed evolution has been used extensively to achieve substantial thermostability in proteins. The prominent examples being p53, p-nitrobenzyl esterase, xylanase, subtilisin E, β-glucosidase and *Bacillus subtilis* Lipase A. By studying the thermostable mutants produced by directed evolution one can study the structural features leading to thermostability in isolation as the protocol dictates only the selection of those mutations that contribute to thermostability. Usually, while studying the effect of mutations on protein functions, only the areas in the vicinity of mutation are analyzed in detail. However, studies have shown that small conformational changes occurring in one region of the protein can have an effect in a region quite far from the point of perturbation without affecting the functionality. Hence it has become pertinent to study the effects of mutations on the protein as a whole, rather than concentrating only on the regions of perturbation, because multiple small conformational changes may lead to an overall effect on the functionality of the protein. This becomes all the more evident when the overall conformational change occurring in the proteins, due to such distributed mutations, is small. Working on this premise, in the present study, we have chosen the wild-type and six *in-vitro* evolved thermostable mutants of *Bacillus subtilis* Lipase A protein, in which increasing thermostability has emerged due to distributed mutations (ranging from 2 to 12 positions) without causing any significant difference in their three-dimensional structures. We have analyzed the wild type (WT) *Bacillus subtilis* Lipase A structure, and the factors leading to thermostability in its six thermostable mutant structures, by modeling them using a combination of network-based approach (Protein Contact Networks or PCN) and molecular dynamics (MD) studies. Recently network based approach has been used extensively for studying the structure function relationship in proteins. Protein Contact Networks have been used to predict residues crucial for signaling within the protein, to study the allosteric communication in the proteins, to understand the folding kinetics, etc. However, in the present work we have attempted to utilize this approach to study small conformational changes occurring in the protein due to mutations, by analyzing the contact patterns and their positional information in all the six mutants and the WT, with respect to their structural stability and functionality at higher temperatures. Molecular Dynamics has been employed here to support the results drawn from the network analysis, as well as, to gain insights into the effect of mutations on the dynamics of the protein, and more specifically, the role that it might be playing in thermostability caused by the mutations. Our results indicate that this combined meso-scale network analysis and fine-scale dynamic simulation approach can be both useful and important in understanding the key factors influencing the structural correlates and allosteric information transfer in rendering thermostability in these proteins. # Results *Bacillus subtilis* Lipase A is a globular protein consisting of six β-strands in a parallel β-sheet surrounded by two α-helices on one side and three on the other. The catalytic site of the protein is composed of three residues S77, D133 and H156. However, other nearby residues also participate in the overall function of the protein. The crystal structures – of six mutants of Lipase A generated from the wild type (WT), having increased thermostablity, used in this study are listed in. The WT and the mutant structures were aligned and the cross structure Root Mean Square Deviation (RMSD) between them was determined. RMSD provides an average estimate of the overall structural variation between the molecules, and here the cross-structure RMSD ranges between 0.18–0.39 Å. This low RMSD signifies the minor changes that occur between the structures on acquiring 2 to 12 mutations. However, these nominal changes in the conformation and dynamics associated with them are responsible for the remarkable variation observed in the thermostability of the mutants. Since structural changes are quite small, inferences based on RMSD values alone are not informative in explaining the increased thermostability of the mutants. Hence we have utilized a network-based approach to quantify the small conformational changes occurring throughout the protein using simple assumptions. ## Network analysis of the WT and mutant protein structures Six network parameters, namely, Degree, Betweenness Centrality (BC), Closeness Centrality, Clustering Coefficient, Shortest Path and Modularity were calculated for the Protein Contact Networks (PCNs) constructed from all seven structures (see Methods). The global (average) network parameters show very little difference among the structures, and no correlation was found between the average network parameters and the temperature at which the mutant proteins were functional. This supports the diminutive changes in the overall structures as shown by the small cross-structure RMSD values in. To understand the role of small conformational changes occurring due to the mutations, the variations in local, residue-specific network parameters were studied in all seven PCNs of the WT and mutants of Lipase A. ### Degree On an average, 26% of the residues showed any change in degree in these PCNs (range: 23% (3D2A)–28% (1T2N and 3QMM)). shows the difference between the degree of each residue in the mutants and WT, and the maximum change observed (increase or decrease) was 2. The active site residues, N18 and M134, show increase in degree by 2 in four out of six mutants. Other non-active site residues showing increase of 2 are F41, K44, A81, T83, K88, N179, and N181. Of these, K44 and A81 are directly connected to the active site residues, N179 and N181 are C-terminal residues, and K88 lies close to a mutated residue N89Y. Thus these residues are responsible for the changes indirectly. As will be shown later, all these allosteric residues are involved in making new contacts in the mutants that can lead to small conformational changes, which can be correlated to the overall enhanced stability of the structures. ### Betweenness Centrality (BC) High BC signifies that the node resides on several shortest paths in a network, and hence is crucial for communication across the different modules of the network. Significantly high BC (z-score\>2.5) of all the residues in each PCN are listed in. The three hydrophobic residues (V100, L102, L159) in the core of the proteins consistently show high BC. The residue-wise differences in the BC between mutants and WT are shown in the. Ten residues from each mutant showing highest positive and negative differences are listed in. The contacts formed and lost in the mutants are crucial in the overall connectivity of the protein network, which can also be inferred from the centrality value of these residues. The role of contact patterns in the mutants and their influence on the BC is discussed in the next section. Closeness Centrality measure of the residues do not show appreciable difference between the WT and mutants. ### Clustering Coefficient (CC) CC measures the cliquishness of the neighborhood of a node. Residues with high clustering coefficient (z-score\>2.0) were determined for all the PCNs. The three hydrophilic non-active site residues, D34, N/Y89, and N120, that are solvent-exposed in all the structures, have consistently high CC in all PCNs. Residues I135 and G153 that are close to the catalytic triad residues D133 and H156, show increase in CC in all the mutants. Residues present near the active site - I12, G14, A/S15, N106, L108, S131 and V136 - also show increase in the clustering coefficient in different mutants. This signifies increase in the interconnectedness of the region near the active site, which can contribute to further stabilization of the region. A substantial decrease is observed in the clustering coefficients of the terminal residues due to addition of new neighbors resulting from the new contacts formed in mutants (keeping the older contacts among these neighbors the same). Analysis of residue-specific network parameters offers several important clues towards the residues that may have a role to play in governing structural stability under thermal stress in the mutant proteins. To elaborate these, in the next section, we analyze the contact patterns in the PCNs of WT and the mutant proteins. One important feature that is revealed through this analysis is that most changes in network parameters occur at the residues that are not part of the active site of the proteins, indicating the specific role of allosteric sites in these thermostable proteins whose overall conformations do not change due to the mutations. ## Analysis of Contacts in PCNs Since there is little overall conformational change among the structures , and the local network properties point towards important roles of specific residues at certain allosteric sites of the mutants, along with analysis of all mutants, we also performed an in depth analysis of changes in contact patterns in the most thermostable mutant (3QMM or 6B) in comparison to the WT. The PCN representation is particularly useful for highlighting the contacts among residues in the three-dimensional structure space. The ring graph representation of the two PCNs (WT and 6B shown) can easily demonstrate the changes in their contact patterns. The edges across the circle represent the long-range contacts in the protein tertiary structures. Consequently the contacts between β-strands in a β sheet are often visible as long edges in the centre of the circle (green circle). Helices contain short-range contacts (n to n+4) and hence are visible along the circle boundary. However inter-helical contacts in the tertiary structures are visible as long-range contacts in the centre of the ring graph. In a PCN, the occurrence of a new contact signifies decrease and loss of contact indicates an increase in the distance (below the threshold) between two residues. The contacts lost (Red) and made (Blue) in the mutant 6B have been indicated in the ring graphs of WT and 6B respectively. The figure shows that, compared to the WT, the long-range contacts made in the mutant are more than the long-range contacts lost. The ring graphs showing the contacts lost and made in the other five mutants are shown in. Analysis of the new contacts made and lost also revealed the fact that most changes in contacts in the mutants are at the allosteric sites and do not involve the active site residues. shows the number of new contacts made and lost, compared to the WT. The figure clearly indicates that the number of new contacts made is consistently higher than the number of contacts lost in all mutants (details given in). Among all the contacts made and lost mentioned in the and, there were a total of 36 unique contacts made and 27 unique contacts lost in all the mutants combined. Of these, 67% (24 out of 36) of the new contacts made are long-range contacts, and roughly half (52% or 14 out of 27) of the contacts lost are short range, in the mutated structures. Additionally, also shows that almost all contacts lost in the six mutant structures involve the loop region residues. This clearly suggests considerable reorganization in the loop regions of the thermostable mutant structures. An interesting point to note here (which is clear) is that the contacts are made and lost throughout the structure, and not necessarily at the sites of mutation or at the functional residues. This suggests that the conformational changes are distributed throughout the structure, and contribute to the resultant thermostability. This trend is also seen in PCNs constructed with higher and lower cut-offs (see Methods). In order to understand the role of these new contacts on the overall stability of the mutants, the location of these contacts is shown in the three dimensional structure. These new contacts in the PCNs are obtained using a purely geometric criterion, and not on the basis of the physico-chemical interactions. Below we categorize these contacts based on their likely effects on the structural stability and function of the proteins. ### (a) Termini stabilizing contacts show the specific contacts (in black) between residues (31 & 3, and 181 & 113/115) that help stabilize the N- and C-termini of the mutant protein 6B. All the mutants with L114P mutation (1T2N, 3D2A, 3D2B, 3D2C and 3QMM) show this terminal stabilizing contact. This also corroborates the structural studies that attribute L114P mutation to the stability of C-terminus of the Lipase A structure. Two new contacts observed between the residues N179 and K122, I123 in some mutant PCNs correspond to a salt bridge that has been described in a previous study between N181 and K122 as a result of the L114P mutation. ### (b) Loop stabilizing contacts About 53% of the total new contacts made (19 out of 36) involve one of the residues from a loop region, and 89% (24 out of 27) of the contacts lost are from the loop regions in the mutants. These flexible regions (loops and turns) of the proteins are vulnerable during the unfolding process. Both contacts made and lost in these regions lead to the reorganization of the loop regions that, along with the new contacts made as described in earlier and later sections, might be contributing to the overall stability of the proteins. For the mutant 6B PCN, 14 out of 20 new contacts made have at least one residue from the loop region, and 11 out of 12 contacts lost are from the loop regions. Similar trend is seen in all mutants. shows that a new contact (blue) between P114 and K88 (connecting the longest loop and the C-terminus of helix αC) occurs in all the five structures with L114P mutation. This, along with the termini stabilizing contacts between loop and the C-terminus residues, contributes to the overall stability of the protein. The contact (54 to 41) connecting the helix αB to a loop between β4 and αB ( orange contact), and the contact between residues 143 and 106 (connecting the loop between αE and β8 to the 3₁₀ helix αD) – also contribute to the reduction in the flexibility of the loop regions as is shown later in molecular dynamics studies (see section on RMSF). ### (c) Contacts leading to active site rigidity Rigidity of the region surrounding the active site in mutant proteins was suggested to be responsible for thermostability in a previous experimental and dynamic simulation study. Our analysis shows new contacts between G46-G11 and T45-G13 (connecting two loops) have residues 11 and 13 positioned close to two mutated residues A15S and F17S and the active site residues I12 and N18. Four mutated residues (A132 (D), M134 (E), M137 (P) and I157 (M) near the catalytic triad) and residues adjacent to them show four new contacts (137–133, 137–134, 150–131 and 155–134) in the mutant PCNs. Such anchoring of the flexible regions and presence of new contacts can lead to rigidity in the region surrounding the active site in the mutant protein. As mentioned earlier, the residues in this region also show an increase in their clustering coefficient, which indicates development of rigidity in the region. ### (d) Contacts connecting regular secondary structures shows two new contacts between helix αF and helix αA (166-25 and 161-22) present in all mutants but absent in the wild type. These have been shown in an earlier structural study on two mutants as parallel-displaced pi-stacking and gain of a hydrogen bond. Another new contact L173-D72 (connecting helix αF and β5) is present in all the mutants. Thus a simple meso-scale network representation can easily give a clear idea of the new contacts made for comparison between many proteins. These local effects of contacts made/lost are also visible in the network parameters at the residue-level. Increase in degree, clustering coefficient, and Betweenness Centrality are observed for most of the allosteric residues discussed above, which are implicated to contribute to reduction of flexibility and increase in rigidity of mutant structures with little change in overall conformation. Some of these have been discussed in. ## Betweenness Centrality and new Contacts High BC of a residue is crucial for communication across different modules of the PCN, as it resides on several shortest paths in a network. shows the structure of WT with the ten residues with highest increase in BC (z-score 2.74–3.21,) along with the new contacts made in 6B. Residue D72 (in central β5) has high Betweenness Centrality (z-score 2.74), and it also makes a new contact with residue L173 (in outer helix αF) in all the mutants. This contact increases the number of shortest paths passing through these nodes thus increasing their BC. Similar increase in BC is observed in both Y25 and Y166, and they make a new contact between them in all the mutants. Other residues showing increase in the BC are N18, H3 and K88, which also make new contacts in the mutants. This suggests that the contacts formed in the mutants are crucial in the overall connectivity and information transfer among residues in the mutant protein network. ## Community Structure Analysis of the PCNs Modularity is defined by groups of nodes within the network that are more connected among themselves compared to the other nodes. PCNs have been shown to possess modules, which correspond to the functional regions of the protein, and have been implicated in efficient transfer of information between different domains of the protein. Even though the WT and mutant proteins do not show much difference among their three dimensional structures, an analysis of the contacts points towards addition of new contacts and loss of some others due to specific mutations, that can influence information transfer. To assess if these changes, which may not induce overall structural changes, can lead to reorganization of the modules in the structures, we determined the community structures for all seven PCNs. This analysis gives five to six communities in all the structures, and the residues belonging to different communities are listed in. shows the structures of the WT and 6B and the corresponding communities with residues colour-coded according to the communities. The communities in the mutant PCN show extensive reorganization in membership of nodes compared to the WT. For example, Community 1 in WT (Residues 21–34 and 157–176, red nodes) is split in all the mutants. The probable cause for this split is the first mutation N166Y in the WT structure that separates the single community involving the two secondary structures (helices αA and αF) into two (red and orange residues) in. It is interesting to note that there are two new contacts that have been shown to occur between these two helices (αF and αA) in all mutants. This has been earlier implicated to increase the interaction between them, but our community analysis shows that these residues now belong to two different communities. This indicates that the new contacts may be helping predominantly in stabilizing the two helices, but have reduced communication between them. Interestingly, Community 3 (yellow) in WT, containing residues of the longest loop (114–123), merges with a larger community having the central β sheet (green residues) in the mutant. This is indicative of an increase in communication between the loop and the β sheet in the mutant due to changes in the contact patterns in nearby regions. This is reminiscent of the L114P mutation and the loop stabilizing contacts. One of the most important observations here is that the community (green) containing nodes from central beta sheet expands in the mutant. This inclusion of greater number of nodes in the central community shows that the edge density within the core structure has increased leading to strengthening of the core structure, thereby increasing the overall stability of the structure. The PCN analyses of the WT and mutant structures offer several important clues to the possible finer structural factors, involving residues from non-active site regions, which can lead to overall stability that aids in thermostability and retention of functionality in *Bacillus subtilis* Lipase A. ## Analysis of Molecular Dynamics Simulations In order to understand the role played by the entire molecule in the thermal stability and the factors responsible for it, we performed equilibrium MD simulations on the WT and two mutant structures (2D9 and 6B) having different numbers of mutations and exhibiting functionality at increasing temperature. Earlier biophysical and MD (20 ns) studies have focused only on the active site geometry and studied the rigidity and dynamics in the 6B mutant. In our study we have monitored all the new contacts (near and far from the active site) observed in the two mutants in molecular dynamics simulations in order to see the stabilizing role of these contacts in overall protein dynamics. First we observed the distance between the Cα atoms of the residues involved in the new contacts formed in the mutant 6B as compared to the WT, for the 100 ns trajectory. As shown in the distance between Cα atoms of residue 46 and 11 was lower for a large number of MD structures in mutant 6B than WT. The average value of this distance was 0.79 nm for WT and 0.47 nm for mutant 6B. A similar observation can also be made for contact between residues 137 and 33. The distances in this case were 1.02 nm for WT and 0.66 nm for 6B. The frequency distributions of structural snapshots of simulations for 100 ns trajectory, as a function of distances for all the new contacts showing considerable shift in the distance distribution in 6B mutant, are shown in. Twelve out of the twenty new contacts show a significant shift towards a lower distance between the residues (two sample KS test p-value\<0.001). The shift in the distance distribution towards lower distances suggests that these contacts lead to compactness in the protein in these regions. All the new contacts that show stability during MD simulation have either one or both the residues belonging to a loop. Since the majority of these contacts are long-range contacts, it suggests that the mutant molecule has become more compact/rigid compared to the WT. ### Root Mean Square Fluctuation (RMSF) Residue specific fluctuations in the protein are observed using RMSF. Flexible regions show higher values of RMSF as compared to the stable regions. Using 20 ns simulation of 6B mutant it was earlier concluded that the mutations cause rigidity in the structure, particularly in the active site region and this leads to the higher activity of the mutant at higher temperatures. Here we show that this observation also holds for longer simulation times (100 ns at 300 K), but the reduction in flexibility of the molecule is not restricted to just one mutant (shown for both 2D9 and 6B), and is distributed throughout the structure. It is interesting to note that the 2D9 mutant, which includes six of the twelve mutations present in 6B mutant, shows reduction in flexibility in the same regions as 6B, in the RMSF analysis. This indicates the effect of these particular regions of the proteins in conferring overall stability to the mutants. Comparison of WT RMSF profile with the two mutants highlights that the reduction in the flexibility is specific to few regions. gives the difference in RMSF values of 2D9 (Red bars) and 6B (Blue Bars) as compared to WT. There are five major regions where the RMSF shows reduction in case of the mutants. These regions comprise residue - 1) 10–18; 2) 43–49; 3) 77–81; 4) 152–159, and 5) 3 and 181. Four of these regions contain residues primarily in loops. The reduction in RMSF indicates the stabilization of these loops and consequently the increased rigidity in the active site. There is also a reduction of RMSF in the termini of the molecule (Region 5). There are seven new contacts that have formed in 6B that involve one of these regions. Thus the contacts made in these regions stabilize these regions and reduce the flexibility. Both rigidity of active site, loop regions and stabilization of terminal regions are in agreement with the results that were obtained by network analysis. This result was further corroborated by the principal component analysis of the WT and 6B MD trajectories at 300 K. The first ten PCs accounted for 71.7% and 57.0% of the total variance in the dynamics of WT and mutant 6B respectively. The values for the trace of the covariance matrix were 2.03 and 1.18 for WT and 6B respectively. Higher value in case of WT suggests higher flexibility associated with WT. This flexibility is contributed mainly by motions in the loops around the active site (77–80, 152–159) as well as away from the active site (45–49, 10–14) and also the termini (3,181). The motion described by the first PC in WT essentially represents the motion of these residues. However in the case of the mutant, the regions mentioned above show low flexibility. This may be due to the new contacts made in these regions (31-3, 45-13, 46-11, 81-48, 155-134, 181-113 and 181-115). Also the residues (I12, G14, A15, I151 and G153) belonging to these regions show an increase in their clustering coefficient suggesting increase in the cliquishness in network. The RMSF values were also compared to the residue wise RMSD obtained after aligning the WT and six mutant structures. The residue wise RMSD plot expectedly shows similarity to the RMSF plot obtained for the mutants. This suggests that the fluctuations observed in MD simulations of the mutants are similar to the deviations observed between X-ray crystal structures of different variants. The structural factors attributed to induce enhanced stability of thermostable proteins include hydrogen bonds, solvent accessible surface area (SASA), salt bridge and secondary structure content. Some of these important factors were studied in the MD simulations to compare the WT and 2D9 (3D2B) and 6B mutants (3QMM). ### Solvent Accessible Surface Area (SASA) Solvent accessible surface area represents compactness of a protein structure. Increase in the polar SASA and simultaneous decrease in the non-polar SASA has been implicated in the increased stability of the thermo stable proteins. We calculated the SASA values for the 100 ns trajectory of MD simulation for WT, 2D9 mutant and 6B mutant. The average total solvent accessible surface is significantly lower (p\<0.05) in case of mutated structure 6B (91.04 nm²) as compared to the WT structure (95.28 nm²). The fractional polar and non-polar SASA values, calculated from each frame of the 100 ns trajectory were significantly different (Wilcoxon ranksum test p\<0.05) between the WT and the mutant 6B. also shows the average fractional SASA values for the 2D9 mutant. The values for 2D9 were in intermediate range between WT and 6B. Thus, our simulation results show increase in the polar exposed surface area and decrease in non-polar exposed surface area, along with an overall reduction in the total SASA in the mutant. This indicates that the mutant structures become more compact in comparison to the wild type. However, it may be noted that the fractional polar and non polar surface area for the WT and 6B were found similar (∼0.4 and ∼0.6) for the crystallographic structure. This may be due to the fact that the crystal structure represents only one of the several conformations that the structures might be sampling in the MD simulations. ### Hydrogen Bonds The average number of hydrogen bonds per frame of the 100 ns MD simulations for WT, and thermostable proteins 2D9 and 6B were calculated. The average number of hydrogen bonds per frame is higher (125.6, Wilcoxon Ranksum Test, p\<0.05) in 6B mutant as compared to the WT (117.5– see 300 K). The average number of hydrogen bonds per frame in case of 2D9 mutant was found to be 123.9. Higher number of hydrogen bonds have been previously related to the increased thermostability of thermophilic proteins. shows that the average number of hydrogen bonds per frame increases with increasing thermostability in the 300 K simulations. ### Salt Bridges Electrostatic interactions, in the form of salt bridges, are known to play important role in the stability, folding and function of proteins. The role of salt bridges in thermostability has been thoroughly studied and has been found to be a major factor contributing to the enhanced stability of the thermophilic proteins. Salt bridges were calculated for the MD simulation trajectories of WT and 6B mutant at 300 K, as described in Methods. A total of twenty-one salt bridges were found in WT and twenty-nine in the mutant trajectory. Further, number of frames having the center of mass of the positively charged atoms in basic side chain and negatively charged atoms in the acidic side chains at a distance less than 5 Å was calculated for each trajectory. All the salt bridges along with this number have been listed in. As seen from this table, the number of salt bridges has increased in the mutant, which may contribute to the overall stability of the mutant. ## High Temperature Molecular Dynamics Simulations Since protein denaturation occurs typically in microsecond time scale, it is difficult to capture the unfolding of protein at normal temperatures using molecular dynamic processes. In order to study denaturation process in WT and mutant proteins within the feasible time limits, much higher temperatures are used. MD simulations at higher temperatures have been performed previously to study the thermostability in Xylanases. It has been shown earlier that higher temperature accelerates the unfolding process without changing the unfolding pathway, at least in the initial stages of unfolding. Working on this assumption, we have analyzed the unfolding of WT and the mutant Lipases 2D9 and 6B at higher temperatures. The optimum temperature of activity for the 6B mutant was reported as 65°C in comparison to WT protein, which is active at 35°C. The 6B mutant exhibited increased activity at all temperatures, as compared to WT or any other mutant. The major reason for this phenomenon was argued to be the rigidity of the active site using MD simulation and Fluorescence study. However, the factors leading to this rigidity, and contributions of the entire protein to this phenomenon, were not discussed. Here we attempt to shed some light on these questions by performing MD simulations of WT, 2D9 and 6B mutants at higher temperatures of 400 K and 500 K for 30 ns each. ### Root Mean Square Deviations (RMSD) Average RMSD for WT and 6B mutant for 300 K, 400 K and 500 K are shown in. At 300 K, the RMSD of the WT and 6B do not show much difference. However there is a clear trend in RMSD values at higher temperatures. RMSD values at 400 K and 500 K show a decrease with the increase in thermostability. shows RMSD traces for four high temperature MD simulations for the WT and 6B structures at 400 K and 500 K for 30 ns corroborating the results mentioned above. ### Secondary Structure Content Total number of residues occurring in regular secondary structure is an indicator of the stability of the protein. shows the number of residues in different secondary structures in each frame of the MD simulations at different temperatures. At all three temperatures simulated here, the average number of residues occurring in regular secondary structures are higher in mutant 6B compared to WT. ### Hydrogen Bonds The average number of hydrogen bonds in the mutant 6B is significantly higher (Wilcoxon ranksum test p\<0.01) than that in WT at both 400 K and 500 K ( 400 K and 500 K). As hydrogen bonds play an important role in the maintenance of the secondary as well as tertiary structure of the molecule, mutant proteins should show higher resilience towards denaturation at higher temperatures. Also at 400 K the average number of hydrogen bonds per frame increases with increasing thermostability as shown in for both 2D9 and 6B proteins. ### Solvent Accessible Surface Area Denaturation is marked by an increase in solvent exposed surface area as water solvates the unfolding molecule. shows that the average fractional polar SASA value at 400 K is higher in case of mutants 6B and 2D9. This shows that the extent of unfolding in the case of WT is higher at 400 K, as the non-polar regions start getting exposed to the solvent. At 500 K, WT and both the mutants show a higher fractional non-polar SASA. # Discussion The Lipase A protein and its carefully selected *in vitro* evolved thermostable and functional mutants – offer a convenient system to study the structural changes that enhance thermostability without inducing much changes in their overall three-dimensional conformation. We have used the Lipase A protein and all the six thermostable mutants to uncover the general features underlying evolution of increased thermostability in the absence of significant changes in the three-dimensional structures by - a) performing a comparative coarse-grained protein contact network-based study to extract the contact changes and their influence in network parameters, b) finding the structure stabilizing effects of the new contacts, c) analyzing the changes in the community structures, between the wild-type and mutant structures, and d) supplementing the network-based results with room temperature and high temperature molecular dynamics study of the proteins. Even though these different approaches give us important information on how the thermostability of the mutants is achieved due to mutations in the protein, we have aimed at arriving at a consolidated view by connecting these different approaches. The study on contact changes uncovered establishment of important stabilizing contacts due to mutations, which gives important clues to thermostability. The total number of contacts formed is higher than the number of contacts lost suggesting an increase in the compactness of the protein. Important contacts were found to be made near termini and active site residues, which stabilize these regions, and hence add to the thermostability of proteins. Both these observations were confirmed by the reduced RMSF values of the termini and the active site regions indicating that the formation of these contacts has actually reduced the flexibility of these regions. The results obtained by analysis of network parameters also corroborate these results. The residues near the active site show increase in the clustering coefficient, which further suggests the enhancement of rigidity caused by the formation of new contacts. The community structure analysis clearly shows reorganization of the membership of specific nodes in different communities occurring due to mutation, even when the overall structure shows negligible difference. This indicates changes in communication among the functional and allosteric residues in the whole protein. The establishment of more contacts can also yield a larger number of communities by fragmenting the existing ones, as seen in our analysis. All mutants show fragmentation of the first community of the WT into two different communities where a few residues have become part of a larger Community 4. As shown in the RMSF and RMSF difference plots, the residues (152–159) in community 1 and 4 in WT show high fluctuations, but in mutant 6B these residues now belong to a larger community (Community 4) and exhibit much reduced fluctuations. The expansion of the core region beta sheets into a larger, single community by incorporating a large part of the loop residues is a clear pointer to increased stability of the protein. The community structure analysis, in conjunction with dynamic simulations, strongly indicates features relevant to the establishment of structural stability in mutants. A similar result is obtained from the analysis of residue-wise Betweenness Centrality in mutants. Contact analysis also unveils certain important contacts (e.g P114 to K88) that play crucial role in the overall stability of the protein, but have not been noted in previous structural analysis In recent years, network-based approaches have been used extensively to understand the structure function relationship of proteins. However, our combined coarse-grained network and molecular-level dynamical modeling approach is an unusual way to analyze small conformational changes in the proteins through the loss or gain of contacts, and relating it to its stability and function. Our PCN analysis has shown that the gain of contacts can lead to the regional stability (termini and loop stabilizations), thereby increasing the overall stability of the protein. Such synergy between the ordered and disordered regions for functional versatility in proteins is increasingly being observed in different systems. Importantly, these small conformational changes are not only restricted to the sites of mutation, but throughout the protein, and thereby, contributing to the thermostability of the protein in a concerted manner. Thus, it becomes important to consider the small changes occurring throughout the proteins, in order to explain the structural and functional aspects of mutations. Surprisingly, certain new contacts and important structural factors leading to stability of the protein become apparent when dynamics is associated to the network analysis, in comparison to static crystal structures. This is noticeable in the analysis of the SASA where the average hydrophilic SASA is higher in case of thermostable mutants but was not evident from the static crystal structures. Also, other results obtained by the MD simulations, including number of hydrogen bonds and number of salt bridges, indicate towards the higher stability of the mutants. This suggests that the dynamics of the protein also plays a crucial role in defining a quality as important as thermostability. Thus the combined network and dynamics based analysis of protein structures gives novel insight into the enhanced stability of the thermostable mutants of the *Bacillus subtilis* Lipase A. One interesting direction that this approach can take is to construct the PCNs from the ensemble of MD simulations, and study the motion of those important residues identified from the various network based analysis of the WT and mutant structures. This mesoscopic-network approach and fine-grained molecular dynamics approach can be a convenient and useful scheme to elucidate the small but important conformational changes among the structures where altered allosteric communication pathways resulting from ligand/drug binding and mutations, underlie changes in functions without significant changes in the overall network structure. # Methods ## Dataset used Lipase A is an extracellular lipase that catalyses both hydrolysis and synthesis of long-chain triacylglycerols. It is a 19.34 kDa protein consisting of 181 amino acids and is much smaller than other lipases. It is a globular protein resembling α/β hydrolase fold. It consists of six β-strands in a parallel β-sheet surrounded by two α-helices on one side and three on the other. *Bacillus subtilis* Lipase A structures used in the analysis are listed in. All the protein structure coordinate files were taken from Protein Data Bank.([www.pdb.org](http://www.pdb.org)). The “directed evolution” method used to obtain the thermostable mutants offered a set of six mutants with increasing thermostability with each successive mutant retaining the earlier mutations in them. ## Protein Structure Alignment Multiple structure alignment of the proteins was performed using Protein structure comparison service Fold at European Bioinformatics Institute (<http://www.ebi.ac.uk/msd-srv/ssm>). ## Protein Contact Network Construction PCNs were constructed for all seven protein structures as described in,. Each Cα atom of the residues in the protein is considered as a node and the spatial proximity between them defines the link. In this coarse grained representation any two nodes are considered to be linked by an edge if the distance between them is \<0.7 nm. This cut-off has been widely used for constructing Cα based protein contact network and provides a reasonable estimate of the interacting residue pairs in the first interaction shell. However there are other cut-offs also that have been used when considering all atom network models. In this way a protein structure consisting of N amino acid residues can be represented as adjacency matrix (A) such thatwhere D_ij is the distance between Cα atom of i^th and j^th residue. The PCNs were also constructed with two other realistic thresholds - 0.65 nm and 0.80 nm, and the analysis repeated for the WT and mutant 6B. The absolute number of contacts made or lost obviously depends on the threshold used for constructing the PCNs, but the ratios of the two are very similar and statistically insignificant (Wilcoxon test with p-value 0.7 and 0.3 when 0.70 nm is compared to 0.65 nm and 0.80 nm, respectively). Additionally, in both cases, like the PCNs with 0.70 nm cut-off, most changes in contacts in the mutant are at the allosteric sites and do not involve the active site residues. ## Network Parameters The network parameters (Degree, Betweenness Centrality, Closeness Centrality, Clustering Coefficient, and Shortest Paths) were calculated using igraph package of R. The details of these network parameters are provided in. ## Community Calculation Communities in the PCNs of WT and the mutant structures were calculated using the “igraph” package. Fast greedy algorithm proposed by Clauset et. al. was used for calculating the best split of communities. ## Molecular Dynamics Simulations The X-ray structures of WT (PDB ID 1I6W) and 6B mutant (PDB ID 3QMM) were used as starting models for MD simulations. MD simulations were performed using GROMACS software package. Both proteins were simulated at 300 K, 400 K and 500 K for 100 ns, 30 ns and 30 ns respectively. The simulations were done by soaking starting structures in explicit solvent using TIP4P water model and OPLSAA force field. Periodic Boundary conditions were used with minimum distance between the solute and the box of 0.9 nm. The systems were neutralized by replacing appropriate number of water molecules by Cl- ions. We used only counter ions and not the explicit salt conditions as these are more important only for highly charged systems. The systems were subjected to energy minimization using steepest descent method for 50000 steps or till the maximum force on all atoms was less than 1000.0 kJ/mol/nm. Energy minimization was followed by equilibration first in an NVT ensemble for 200 ps and then in an NPT ensemble for 200 ps. After the systems were equilibrated, the production run was performed for 100 ns at 300 K and 30 ns at 400 K and 500 K. The equations of motion were integrated with a time step of 2fs and the coordinates were saved every 1000 time steps (2ps) resulting in total of 50001 frames for 100 ns simulations and 15001 frames for 30 ns simulations. Particle Mesh Ewald (PME) summation was used for calculating electrostatic interactions. All-bonds were constrained using LINCS algorithm. A constant temperature was maintained by weak coupling to modified Berendson thermostat and Parinello Rahman pressure coupling was used to maintain a constant pressure during simulations. 2D9 mutant (PDB id 3D2B) was also simulated for 100 ns at 300 K, and for 30 ns at 400 K and 500 K, following the same protocol and parameters as mentioned above for WT and 6B. ## Analysis of MD Simulation Trajectories All the analysis was done using the frames from the production run (50001 for 300 K and 15001 for 400 K and 500 K). RMSD, RMSF, Hydrogen Bonds and SASA were calculated using *g_rms*, *g_rmsf*, *g_hbond* and *g_sas* functions of GROMACS respectively. The secondary structure content in each frame of the trajectories was calculated using the *do_dssp* function of GROMACS which in turn uses DSSP to assign secondary structures. To extract the essential motions from the trajectory, principal component analysis was performed using the *g_covar* function of GROMACS. The eigen values and the eigen vectors obtained as a result of PCA were analyzed using the *g_anaeig* function of GROMACS. Salt bridges were calculated using the Salt Bridge Plugin in the VMD. The distance cutoff between the oxygen atom of the acidic side chain and the nitrogen atom of the basic side chain was selected as 4 Å. The distance between the centre of masses of the positive charges atoms in the basic side chain and the negatively charged atoms in the acidic side chains were also calculated for every fifth frame of the trajectory (total 10002 frames). All the programs for the construction of adjacency matrix from PDB files and contact analysis were written in MATLAB R2007b ([www.mathworks.com](http://www.mathworks.com)). Adjacency matrix calculated using MATLAB was used as input for network parameter calculation in igraph package of R. Two-sample Kolmogorov-Smirnov Test was performed in R and Wilcoxon ranksum test was performed in MATLAB. The protein structures were visualized and the figures were generated using PyMOL. Solvent accessible surface area (SASA) for the crystal structures of WT and the mutants was calculated using the POPS program. The SASA values calculated for the crystal structures showed no consistent trend with respect to compactness of the wild type and mutant structures. However the crystal structure represents only one of the several conformations that the structures might be sampling in actual solvated environment. Therefore, SASA values were calculated for the 100 ns trajectory of molecular dynamics simulation (50000 frames) for WT and 6B mutant. # Supporting Information The authors are grateful to the reviewers for many useful comments that have improved the analysis and presentation of the results. The authors acknowledge the use of IISER Mohali Computational Facility, discussions with Shashi Bhushan Pandit and C Suguna. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: AS SS. Performed the experiments: AS. Analyzed the data: AS SS. Wrote the paper: AS SS.

# Introduction The HIV and AIDS epidemic is a major pandemic that affects millions of people globally. The UNAIDS Global AIDS Update 2019 reported that 74.9 million people have become infected with HIV since the start of the epidemic, with 32.0 million deaths from AIDS-related illnesses. Nigeria has the second-largest HIV epidemic worldwide. In 2018, 130,000 new infections and 53,000 AIDS-related deaths were recorded. Nigeria alone accounts for more than half of the new infections and deaths from AIDS-related illnesses in the western and eastern Africa region in 2017. This high mortality is probably attributed to the large population size of Nigeria compared to other countries in the region. Considerable progress has been made in providing global access to antiretroviral therapy (ART), with 23.3 million people accessing therapy worldwide. ART has greatly reduced AIDS-related mortality and morbidity and increased the life expectancy of people living with HIV and AIDS (PLHIV); however, it has led to other consequences, including malnutrition. PLHIV are vulnerable to malnutrition due to intestinal damage, which causes impaired nutrient absorption and reduced food intake from vomiting and painful swallowing. Furthermore, malnutrition could result from food insecurity and the side effects of ART, such as appetite loss and abdominal pain. The adverse effects of HIV and malnutrition on the immune system are similar in that they both reduce CD4 and CD8 T-lymphocyte numbers, which eventually increase susceptibility to opportunistic infections. Opportunistic infections and malnutrition can affect intake, absorption, and metabolism of food, worsen disease progression and increase HIV-related mortality. PLHIV are encouraged to consume healthy diets rich in essential amino acids, unsaturated fats, and micronutrients at the recommended daily allowance (RDA) to achieve an adequate nutritional status vital for health and survival. Unfortunately, several studies reported a poor diet intake with inadequate nutrients among PLHIV in Sub-Saharan Africa, including Nigeria. A study conducted in Nigeria reported significant malnutrition in early HIV infection before ART initiation. Malnutrition is a public health challenge in Nigeria; available data showed that the country has the second-highest burden of stunted children worldwide. Two million children and 7% of women of childbearing age were also reported to suffer from severe acute malnutrition. *Moringa oleifera* Lam (syn. M. ptreygosperma Gaertn.) is a species of the monogeneric family Moringaceae. It has been documented to contain many nutrients and bioactive compounds in literature. The leaves are the part of the plant mostly used and with several nutrients often deficient in malnourished PLHIV. It is a rich source of both macro and micronutrients and natural antioxidants source. *Moringa oleifera* leaf powder is a novel, cheap, culturally acceptable, efficacious, and regionally produced plant and can reduce the malnutrition burden in Sub-Saharan Africa. Furthermore, in Nigeria, *Moringa oleifera* use is promoted based on the commendation of its nutraceutical benefits, and the Nigerian Federal Government Raw Materials Research and Development Council (RMRDC) has been actively encouraging farming and consumption of *Moringa oleifera*. Monera *et al*. reported the *in vitro* CYP3A4 inhibitory activity of *Moringa oleifera* leaf extracts, suggesting the potential for interaction with antiretroviral drugs. However, the *in vitro* data alone is insufficient to conclude the clinical significance of concomitant administration of *Moringa oleifera* with ART in PLHIV. Moreover, the interaction between tenofovir/lamivudine/efavirenz and *Moringa oleifera* leaf powder has not been reported. No adverse clinical effects have been reported in the literature despite its widespread use and concomitant use by PLHIV. Therefore, this double-blind, randomized study aimed to evaluate the effects of six months of *Moringa oleifera* leaf supplementation on the CD4 counts, viral load and anthropometric parameters of HIV-positive adults who were on ART in Kano State, Nigeria. # Methods ## Study location The study was conducted at the S. S Wali Virology Center at the Aminu Kano Teaching Hospital, Kano State (AKTH), Nigeria. AKTH is a tertiary health institution and referral center that operates a daily HIV clinic (5 days a week). It also serves as a center for clinical evaluation, laboratory tests, HIV counseling and testing (HCT), treatment, and care supported by the Federal Government and the Institute of Human Virology, Nigeria (IHVN) in partnership with its global partners, including the Centers for Disease Control and Prevention (CDC) and the Global Fund to Fight AIDS, Tuberculosis, and Malaria. The center attends to all patients with HIV infection diagnosed within the hospital or referred from outside the health facility. ## Type of study and participants The study was a double-blind, randomized control trial conducted between December 2017 and November 2018. Registered HIV-infected individuals receiving treatment and care at the S.S Wali Virology Center were invited to participate. Inclusion criteria for the study were: being HIV sero-positive, ≥ 18 years old, CD4 counts ≤ 500 cells/mm³, ART for at least three months (tenofovir + lamivudine + efavirenz combination), informed consent, and compliance with the study protocol. Exclusion criteria for the study were: known allergy or intolerance to *Moringa oleifera* or placebo (cornstarch powder), pregnancy, CD4 counts \> 500 cells/mm³, presence of active opportunistic infection, and intake of micronutrient or natural health product supplements within 30 days of screening. For ease of monitoring, patients who lived outside Kano State, where the study was conducted, were excluded. ## Sample size The sample size was calculated to ensure detection of medium effect size (Cohen’s d = 0.5) or 0.5 standard deviation in mean weight or CD4 by randomized control trial (RCT) arm with 90% power (1-β \[type 2 error probability\]) and 95% confidence (or 5% α error probability \[type 1\]), assuming a balanced 1:1 study design. A sample size of 172 patients was calculated, rounded up to 200 to give room for attrition. The sample size was calculated using G\*Power version 3.1.9.2. ## Randomization Block randomization was used to balance the groups throughout the enrollment period. PASS 12.0 software was used to develop the randomization list using Wei’s Urn algorithm by an independent statistician who held the randomization code. A random allocation sequence was generated to allocate and assign each patient to either the *Moringa Oleifera* group (MOG) or the control group (COG) when participants fulfilled the inclusion criteria and consented, with 100 patients in each group. All the research team members, including the principal investigator (PI) and the study participants, were blinded to the allocation of patients to the study groups. ## Preparation for study Fresh *Moringa oleifera* leaves were obtained from Prime Global Agricultural Industries Limited, Kano State, Nigeria. Fresh leaves were identified and authenticated by a botanist at the Department of Biological Science, Bayero University Kano (BUK), Nigeria. Confirmation of the taxonomic identity of the plant was achieved by comparison with voucher specimens kept at the Herbarium of the Department of Biological Sciences, BUK. The leaves were processed by HOMIP Spices and Foods Limited, Kano State, Nigeria. The procedure involved washing and drying the fresh leaves in a clean environment on a net mesh away from direct sun for days until it was completely dried. The dried leaves were cleaned, and the small branches were removed. The dried leaves were ground using a grinder and sieved using a 0.500 mm standard sieve (No. 35 mesh size) to obtain a fine powder. Fine Moringa leaf powder was stored in airtight containers. The placebo was obtained by coloring cornstarch powder with chlorophyll. It was manufactured and processed at Dala Foods Nigeria Limited, Kano State, Nigeria. Both the *Moringa oleifera* and the placebo were similar in presentation and were identically packaged to be indistinguishable. The interventions were packaged into small (15 g) sachets each. Thirty (30) individual sachets were further packaged in a bigger green-colored plastic bag to be used as supplements taken together with meals for one month. It was sealed, labeled, and stored in a dry place away from heat and humidity. Patients could simply put a sachet in the pocket or bag while going out for their daily activities. The supplements were taken together with meals. ## Intervention The interventions were provided in 15 g Moringa leaf powder sachets. Thirty sachets were given to the participants to represent one month prescription, and they were directed to divide each sachet into three and use it thrice daily (5 g), adding it into meals. They were asked to maintain their regular diet and not consume *Moringa oleifera* in any form from other sources during the study period. In Kano State, home visits of participants by research members to ensure adherence was not convenient for fear of stigmatization by family members. Thus, adherence was monitored by biweekly phone calls to the patients and interviewing them during their monthly visits to evaluate compliance. ## Data collection At the first visit, the research team interviewed the patients to obtain socio- demographic information, patient history, and other relevant information, including dietary information. A trained nurse at the virology clinic and a trained research assistant were responsible for all anthropometric measurements and data collection under the supervision of the PI. Weight was measured to the nearest 0.1 kg using a digital weighing scale (Tanita HD-372, Tanita Corporation, Tokyo, Japan), with participants wearing light clothing and without shoes. Height was measured to the nearest centimeter using a stadiometer (Seca 217, Seca Gmbh and co. KG., Hamburg, Germany). Body mass index (BMI) was calculated as the weight in kilograms divided by the square of height in meters. Anthropometric parameters were measured at baseline and each monthly visit. All laboratory evaluations were performed by a trained phlebotomist at the President’s Emergency Plan for AIDS Relief (PEPFAR) laboratory of the S. S Wali Virology Center at AKTH. The CD4 count was tested using a Partec flow cytometer (Partec, Munster, Germany). Five (5 ml) venous blood samples were aseptically collected from each study participant. Briefly, equal volumes (20 μL) of CD4 PE antibody and ethylene diamine tetraacetic acid blood were mixed and incubated for 15 min, and 800 μL of CD4 buffer was added before reading in the cell counter. The viral load was quantified by polymerase chain reaction (PCR). Ten (10 ml) of venous blood samples were aseptically collected from each study participant. The COBAS AmpliPrep/COBAS TaqMan HIV-1 Test version 2.0, manuals (Roche Diagnostics GmbH) was used as the standard operating procedure. The CD4 test was conducted at baseline and each subsequent monthly visit for each study participant, while the viral load test was conducted twice, at baseline and after the sixth month. ## Study outcomes The outcomes assessed were changes in immune status (CD4 cell count and viral load) and changes in anthropometric parameters (weight and body mass index \[BMI\]) and from baseline to the sixth month. ## Data analysis The data input was done in Microsoft excel and exported into SPSS and SAS statistical software for analysis. Findings from the analysis were reported in frequency tables, charts and descriptive analysis was done to estimate mean and standard deviation. Normality test for the data was conducted using Kolmogorov- Simonov and Shapiro-Wilk tests. Data which were not normally distributed were transformed through Box-Cox transformation. Independent t-test was used to determine the significance of mean difference in immunological and anthropometric parameters between the two groups at each stage of the experiment. To further confirm variability between the two groups, a repeated measure linear mixed effect model analysis was deployed to determine the difference in immunological and anthropometrics outcomes between the treatment groups overtime. An exploratory analysis was done to evaluate the influence of socio-demographic characteristics on the immunological and anthropometrics outcomes of the treatment groups over the study period. All statistical tests were carried out at 95% Confidence Interval. ## Ethical considerations This study was reviewed and approved by the ethics committee of Aminu Kano Teaching Hospital (AKTH) Kano, Nigeria (reference number NHREC/21/08/2008/AKTH/EC/2012), and the Biomedical Research Ethics Committee of the University of Kwazulu-Natal Durban, South Africa (reference number BFC294/16). The study was registered with the Pan African Clinical Trial Registry (identification number PACTR201811722056449). The study complied with the principles outlined in the Declaration of Helsinki. All participants provided oral or written informed consent before enrolling them in the study. The procedures of the study, together with the aims, were explained to the participants. Participants were also informed of their right to withdraw from the study at any time. # Results ## Participants flow shows the flow chart of participants in the study. Four hundred and ten patients were screened and assessed for eligibility. Two hundred and ten patients were excluded (204 did not meet the inclusion criteria for the study, and 6 refused to participate). Two hundred patients were randomized into two groups. One hundred patients were randomly selected and allocated to the group receiving MOG, and 100 patients were randomly assigned to the group receiving COG. In the MOG, 8 patients were lost to follow-up, and 3 discontinued the intervention. In the COG, 9 patients were lost to follow-up, and 3 discontinued the intervention. Overall, 177 patients (89 and 88 in the MOG and COG, respectively) completed the 6-month study. ## Nutritional contents of *Moringa oleifera* leaves powder The nutritional content of a 100 g *Moringa oleifera* leaf powder (Nigerian ecotype) was analyzed using a South African National Accreditation System (SANAS) accredited laboratory ASPIRATA Food and Beverage Laboratory. Each 100 g contained an average of 28 g protein, 3.9 g total fat content (total saturated, monounsaturated, and polyunsaturated fatty acids), and 22 g carbohydrate. It contained 1791.82 mg calcium; 4879.26 mg potassium; 24 mg sodium; 2.88 mg Zinc and 37.78 mg iron ( &). ## Characteristics of participants at study inception shows the socio-demographic characteristics of the study participants at baseline. Participants in the MOG’s socio-demographic, socioeconomic, nutritional status, and immunological characteristics were similar to those in the COG at baseline. Females were predominant in both groups \[MOG = 70 (78.7%); COG = 67 (76.1%)\]. The majority were between 30 to 39 years of age in both groups \[MOG = 37 (41.6%); COG = 36 (40.9%)\]. The majority were married \[MOG = 42 (47.2%); COG = 38 (43.2%)\]. Islam was the predominant religion of participants in both groups \[MOG = 64 (71.9%); COG = 66 (75%)\] with more than half of participants belonging to the Hausa/Fulani ethnicity \[MOG = 55 (61.8%); COG = 47 (53.4%)\]. A few of the participants were without any form of education in either group \[MOG = 15 (16.9%); COG = 16 (18.2%)\]. The majority of the participants in both groups earned a monthly income below the minimum wage of ₦30,000 (\$78.23) \[MOG = 67 (75.3%); COG = 66 (75%)\]. The baseline anthropometric and immunological characteristics of the study participants in both study groups showed a similar trend. The means of weight (kg) for both groups were \[MOG = 63.8 (± 14.8); COG = 61.9 (±12.5)\]. The mean BMIs for MOG and COG was 24.84 (± 4.76) and 23.75 (± 3.82), respectively. More than half of the patients had BMI within the normal range of 18.5–24.9 in both groups \[MOG = (51.7%); COG = (58%)\] while a significant number were overweight with BMI values of 25.0–29.9 \[MOG = (30.3%); COG = (31.8%)\] for both study groups. At baseline, the mean CD4 cell counts were statistically similar for both MOG and COG with values of 341.78 (± 106.06) and 352.34 (± 125.99) cells/μL, respectively. The majority of the patients in both groups had an undetected viral load at baseline (MOG = 76.4% and COG = 71.6%). A test of normality for the dependent variables was carried out using the Shapiro-Wilk and Kolmogorov-Smirnov tests. The data that were not normally distributed were transformed through Box-Cox transformation. The CD4 count was transformed through a lambda value of 0.5, weight by lambda value -0.1 and BMI by lambda value -0.2. ## Effect of nutritional supplement intervention on immunological and anthropometric parameters shows the results of independent samples test for the difference in immunological parameters anthropometric between the MOG and COG. The mean CD4 count between the two groups was not significantly different throughout the period of measurement except at the 6th month. From baseline to the 6th month, there was no significant (P\>0.05) difference in all the anthropometric parameters \[weight; BMI\] between the MOG and COG. In addition to the bivariate analysis test above, a linear mixed-effect model was used to examine the differences in anthropometric and immunological parameters between the MOG and COG. shows the linear mixed effect model results showing the differences in the CD4 counts, viral load, weight and BMI between the two groups over the study period. An unstructured correlation matrix was assumed for the model analysis. For CD4 counts, the treatment by time interaction shows a significant difference in CD4 counts by treatment group over time (p\<0.0001). A further estimate of fixed effects showed that the CD4 counts among MOG were 10.33 folds greater than COG over the study period. On the other hand, viral load (p = 0.9558) and the anthropometric parameters (BMI; p = 0.5145 and weight; p = 0.5556) between the two groups were not significantly different over time. shows the chart depicting CD4 cell count mean measurements by treatment group over the study period. Over the six months study period, there was significant increase in mean CD4 cell counts for MOG while the mean CD4 cell counts for the COG was relatively constant. An exploratory analysis of the influence of socio-demographic characteristics on the changes in immunological and anthropometric parameters by treatment groups over time was computed. The analysis of the impact of socio-demographic characteristics on the changes in CD4 counts by treatment group over time showed that ethnicity (p = 0.0491) and family size (p = 0.0483) had a significant influence. However, the changes in CD4 counts by treatment group remained significantly different over time (p = 0.0001). None of the socio-demographic characteristics explored significantly influenced the viral load, BMI, and weight overtime. The changes in these parameters between the treatment groups over time were not significant after controlling for socio-demographic characteristics. # Discussion This study examined the effect of the sixth month’s consumption of *Moringa oleifera* leaf powder supplement on the immunological profile (CD4 cell count and viral load) and anthropometric parameters (weight and BMI) of PLHIV that are on ART in Kano State, Nigeria. We studied 200 patients randomly divided into MOG (100 patients) and COG (100 patients). Our sample size was larger than that reported by Tshingani *et al*. (60 patients) in the Democratic Republic of Congo and Ogbuagu *et al*. (40 patients) conducted in Anambra State, Southeast Nigeria . Over the study period, a significant increase was observed in CD4 cell counts of the MOG participants than the COG using the linear mixed effect model. There was no significant difference in viral load between the two study groups. This improvement observed in the CD4 counts in the MOG was influenced by some socio- demographic characteristics of the study participants that include ethnicity and family size. However, the improvement was still detected regardless of their influence. This result suggests that *Moringa oleifera* leaf powder supplementation and ART effectively improved the CD4 cell counts of the study participants. Conversely, the *Moringa oleifera* leaf powder supplementation intervention was not effective in improving the weight and BMI of the patients when compared to the COG over the study period. Furthermore, none of the socio-demographic characteristics explored was observed to significantly influence any of the anthropometric parameters overtime. *Moringa oleifera* nutritional constituents and ART effect could be responsible for the increased CD4 cell counts in the MOG. *Moringa oleifera* leaves (Nigerian ecotype) analysis shows that they are rich sources of vitamins and micro-and macronutrients. Additionally, it contains minerals and trace elements reported to have multiple curative properties, improve the immune system, and act as strong antioxidants. Furthermore, the dried leaves of *Moringa oleifera* have been documented as a good source of polyphenol compounds, such as flavonoids. Flavonoid consumption has been reported to offer body protection against chronic diseases associated with oxidative stress. In addition, ART has proven effective in reducing morbidity and mortality related to HIV infection by reducing HIV viral load. This improved CD4 level is associated with a reduced number of opportunistic infections. Our results agree with a similar study from Eastern Nigeria by Ogbuagu *et al*.. A significant increase in CD4 counts was observed in PLHIV receiving ART and supplemented their local meals with *Moringa oleifera* leaf powder for two months. Local meals were prepared using palm oil. The limited information offered by the study article prevented an in-depth analysis of the results. However, the study demonstrated the potential of *Moringa oleifera* of improving the CD4 count of PLHIV within short period of supplementing it with regular meals prepared using local food items. Palm oil has been documented to contain palmitic acid which is a saturated fatty acid with health benefits. The lack of significant change in BMI observed in MOG in our study could be due to the fact that few participants (n = 5) were underweight with BMI\<18.5 kg/m² at study inception. This class of people would probably have benefitted more in terms of improvement in BMI from the *Moringa oleifera* leaves supplementation due to the vast amounts of nutrients constituted. Contrary to our results, Tshingani *et al*. did not report a significant difference in CD4 lymphocyte counts after six months of *Moringa oleifera* leaf powder supplementation between their study groups. This could be due to the small sample size. In addition, the presentation of their intervention in bags of 100 g could have reduced adherence, which resulted in a lack of significant increase in CD4 cell counts. This study’s outcomes can be attributed to factors related to the study design. This includes being a double-blinded randomized trial, a larger sample size, and the presentation of the intervention in individual sachets representing daily dose. In addition, the involvement of the virology clinic ‘support group’ members improved patient monitoring and adherence to the study protocol. Nevertheless, future studies using a more diverse population of PLHIV are recommended. The *Moringa oleifera* leaf intervention was not effective in decreasing the viral load of HIV-infected individuals accessing ART at the S.S Wali Virology Center. This could be attributed to some of the challenges encountered. The study protocol did not follow the standard operating procedure (SOP) of viral load monitoring conducted at the S. S Wali Virology Center. Viral load was monitored yearly for patients without any medical problems, whereas the study protocol was designed to have a viral load test conducted twice, at baseline and after six months of receiving the intervention. The study plan overcame this challenge. The use of viral load alone without CD4 cell counts to monitor treatment outcomes remains a challenge in resource-limited settings. Further challenges encountered during the study also include participants’ reluctance to keep monthly appointments to the clinic for the study. This is because at the S. S. Wali virology center, ART drugs were dispensed to last for two to three months for patients with stable medical conditions. This challenge was alleviated with the bi-weekly telephone calls performed to monitor the study participants and remind them of their hospital appointments. The stipend given for transport fare after each monthly hospital visit assisted the study participants in keeping their appointments. This is because of the high poverty level in resource-limited settings such as Kano State and the poor socioeconomic status of the study participants. No other incentives or gifts were provided to participants. Some limitations of this study must be noted. The distinguishable taste of *Moringa oleifera* could be a source of bias. Use of *Moringa oleifera* in capsules could forestall this limitation in future studies. In addition, the inclusion of patients who were only on one ART regimen (tenofovir + lamivudine + efavirenz drug regimen) limits the generalizability of our study findings. Lastly, the short duration of the study and compliance, which was monitored by the self-reporting of the study participants, are further limitations of the study. # Conclusion This study revealed an association between *Moringa oleifera* leaf nutritional supplementation consumption and increased CD4 cell counts among PLHIV on ART in a limited resource setting. Programs in low-resource settings, such as Nigeria, should consider nutritional supplementation as part of a comprehensive approach to ensure optimal treatment outcomes in PLHIV. # Supporting information The authors wish to acknowledge all the patients who participated in this study. We also thank the S. S. Wali Virology Center staff, particularly Nurse Murjanatu Abdulmumin who assisted in the study coordination. Support group members who are also PLHIV and are currently accessing treatment and care at AKTH supported the conduct of the study in ensuring adherence to the research protocol by the participants. We wish to acknowledge and convey our sincere appreciation to the management and staff of Dala Foods Nigeria Limited, Kano State, who assisted in producing the interventions. Lastly, we wish to acknowledge Marothi Peter Letsoalo of the Centre for the Aids Programme of Research in South Africa (CAPRISA) who assisted with the statistical data analysis. 10.1371/journal.pone.0261935.r001 Decision Letter 0 Poole Danielle Staff Editor 2021 Danielle Poole This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 18 Jan 2021 PONE-D-20-30776 A double-blind randomized control trial to examine the effect of Moringa Oleifera leaf powder supplementation on the anthropometric and immune status of adult HIV patients on antiretroviral therapy in a resource- limited setting PLOS ONE Dear Dr. Gambo, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. The manuscript has been evaluated by three reviewers, and their comments are available below. You will see the reviewers have commented on the importance of your work. However, the reviewers have also raised critical concerns and the manuscript will need significant revision before it can be considered for publication – you should anticipate that the reviewers will be re-invited to assess the revised manuscript, so please ensure that your revision is thorough. I have outlined some of the key concerns noted by the reviewers below, but you should respond all concerns mentioned by the reviewers in your response-to- reviewers document.  The key concerns noted by the reviewers relate to the study intervention and analysis. Specifically, the reviewers have suggested alternate approaches to the statistical analysis and presentation of results. Additionally, Reviewer 2 noted that the distinguishing taste of MOG may be considered a source of potential bias in this study. Finally, Reviewer 3 noted that inclusion was limited to participants on one ARV regimen, which limits the generalizability of the findings. These issues have limitations for the interpretation of the results and should be explored. Please submit your revised manuscript by Mar 02 2021 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <plosone@plos.org>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript: A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols> We look forward to receiving your revised manuscript. Kind regards, Danielle Poole Staff Editor PLOS ONE Journal Requirements: When submitting your revision, we need you to address these additional requirements. 1\. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at <https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main _body.pdf> and <https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_titl e_authors_affiliations.pdf> 2\. We suggest you thoroughly copyedit your manuscript for language usage, spelling, and grammar. If you do not know anyone who can help you do this, you may wish to consider employing a professional scientific editing service. Whilst you may use any professional scientific editing service of your choice, PLOS has partnered with both American Journal Experts (AJE) and Editage to provide discounted services to PLOS authors. Both organizations have experience helping authors meet PLOS guidelines and can provide language editing, translation, manuscript formatting, and figure formatting to ensure your manuscript meets our submission guidelines. To take advantage of our partnership with AJE, visit the AJE website ([http://learn.aje.com/plos/](about:blank)) for a 15% discount off AJE services. To take advantage of our partnership with Editage, visit the Editage website ([www.editage.com](http://www.editage.com)) and enter referral code PLOSEDIT for a 15% discount off Editage services.  If the PLOS editorial team finds any language issues in text that either AJE or Editage has edited, the service provider will re-edit the text for free. Upon resubmission, please provide the following: The name of the colleague or the details of the professional service that edited your manuscriptA copy of your manuscript showing your changes by either highlighting them or using track changes (uploaded as a \*supporting information\* file)A clean copy of the edited manuscript (uploaded as the new \*manuscript\* file) 3\. We note that you have indicated that data from this study are available upon request. PLOS only allows data to be available upon request if there are legal or ethical restrictions on sharing data publicly. For information on unacceptable data access restrictions, please see <http://journals.plos.org/plosone/s/data-availability#loc-unacceptable-data- access-restrictions>. In your revised cover letter, please address the following prompts: a\) If there are ethical or legal restrictions on sharing a de-identified data set, please explain them in detail (e.g., data contain potentially identifying or sensitive patient information) and who has imposed them (e.g., an ethics committee). Please also provide contact information for a data access committee, ethics committee, or other institutional body to which data requests may be sent. b\) If there are no restrictions, please upload the minimal anonymized data set necessary to replicate your study findings as either Supporting Information files or to a stable, public repository and provide us with the relevant URLs, DOIs, or accession numbers. Please see <http://www.bmj.com/content/340/bmj.c181.long> for guidelines on how to de- identify and prepare clinical data for publication. For a list of acceptable repositories, please see <http://journals.plos.org/plosone/s/data- availability#loc-recommended-repositories>. We will update your Data Availability statement on your behalf to reflect the information you provide. 4\. PLOS requires an ORCID iD for the corresponding author in Editorial Manager on papers submitted after December 6th, 2016. Please ensure that you have an ORCID iD and that it is validated in Editorial Manager. To do this, go to ‘Update my Information’ (in the upper left-hand corner of the main menu), and click on the Fetch/Validate link next to the ORCID field. This will take you to the ORCID site and allow you to create a new iD or authenticate a pre-existing iD in Editorial Manager. Please see the following video for instructions on linking an ORCID iD to your Editorial Manager account: <https://www.youtube.com/watch?v=_xcclfuvtxQ> 5\. Please include captions for your Supporting Information files at the end of your manuscript, and update any in-text citations to match accordingly. Please see our Supporting Information guidelines for more information: [http://journals.plos.org/plosone/s/supporting-information](about:blank). \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Partly Reviewer \#2: No Reviewer \#3: Partly \*\*\*\*\*\*\*\*\*\* 2\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: No Reviewer \#2: Yes Reviewer \#3: No \*\*\*\*\*\*\*\*\*\* 3\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: No Reviewer \#2: No Reviewer \#3: No \*\*\*\*\*\*\*\*\*\* 4\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: Yes Reviewer \#3: No \*\*\*\*\*\*\*\*\*\* 5\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: \*\*\* General comments: \*\*\* The manuscript presents interesting data from what appears to have been a well- run study. However, the statistical analysis is quite idiosyncratic and fails to control for multiple comparisons. It should be redone. In addition, the manuscript suffers from a variety of English syntactic and usage errors. These should be corrected. There are relatively few in the results section, they are more frequent in the discussion section, and they are numerous in the remaining sections. There are too many to list. \*\*\* Specific comments: \*\*\* 1\) The protocol is quite weak in its description of the statistical methods to be used for the data, as well as the hypotheses of interest and the specific tests or contrasts that were to be employed. This weakness appears to have carried over to the actual statistical analysis. The study design is fairly straightforward, so it is unclear why a better framework for evaluation could not be specified. Having said that, the most important test specified in the protocol is the t-test of the two treatments at the last time point, so this should be presented in any case. 2\) The current layout of the statistical analysis is quite idiosyncratic and simplistic. The following approach or equivalent should be used instead: Test baseline differences via the t-test as already performed. This is analogous to a two-factor analysis of variance (ANOVA) with one fixed effect --- that is, treatment. Assess the differences between groups in changes over time by using a linear mixed effect model framework. The fixed effects are treatment, time, and the treatment by time interaction. Include a correlation structure to account for the dependence structure caused by observing the same patient over time. Do not use a compound symmetry structure (equivalent to "classical" repeated measures) --- this is an unrealistic assumption here. Instead, either use an autoregressive correlation structure such as AR(1) (since time points are equally spaced) or use a completely unstructured correlation matrix. The test of the interaction of treatment by time provides the answer to the question of whether the response profile over time differs between treatments. If there is no interaction effect, that is not necessarily the end of the road, but it is arguably of primary interest. Next, perform post hoc comparisons using an adjustment for multiple comparisons such as the Tukey-Kramer adjustment. It is up to the authors, they can perform an analysis of either simple effects (comparison between treatments at each time point or comparison among time points within each treatment) or an analysis of all pairwise comparisons. Note that the test of the interaction effect is not technically required to be statistically significant for these tests to be performed, though this is the usual procedure. All linear models should be assessed for fit using graphical analysis of the residuals to check for symmetry, homoscedasticity, and independence. The Shapiro-Wilk or Kolmogorov-Smirnov test can be used, but is probably of less importance. If transformations are required, please report all transformations attempted. Even if fit is marginal on the original scale, if the results are qualitatively similar to a fit on a transformed scale then it may be best to report analysis on the original scale for interpretation. For completeness's sake, since the protocol specifies it, also include the t-test of the last endpoints. However, this is probably less powerful than the result that will be obtained from the linear mixed effects model approach. 3\) The evaluation of socio-demographic characteristics (Line 270 forward) should be redone. This should be done in the framework described above, as an exploratory analysis. These covariates can be entered as main effects in linear mixed effects model described above. This will yield answers to the question of whether the treatment effect remains if the effect of the covariate is statistically corrected, which is probably the question that the authors have in mind. 4\) In some cases, it is difficult to understand what exactly is being presented. For example, what does Table 5 present? What are the "differences" here? Each table and figure should have a caption that clearly describes its contents. 5\) Figure 2 is really terrible and should be replaced with a standard figure that shows the behavior of the expected marginal means over time for each group. 6\) Lines 317-318: This conclusion seems far too strong for this paper. Has it really been proved to this degree? 7\) Lines 338-344: The authors do not disclose whether similar claims were made by COG patients. This paragraph seems really close to misinformation as it is currently worded. If patients from both treatments reported these claims, then how can they be attributed to Moringa oleifera? All such claims in the discussion need to be backed up by data analysis in the results. Where is the presentation of these results? 8\) The methods section lacks critical details. As an example, on Line 175 the authors state that viral loads were measured "by the Laboratory scientist using Real-time PCR machine" (sic). Which machine? What was the protocol? Which laboratory? The same issue pertains to the CD4 counts. 9\) Lines 403-405 overstates the results. 10\) The conclusion and recommendation in Lines 417-121 do not seem to be fully supported by the manuscript. Reviewer \#2: 1. Moringa leaf powder packaged in sachets as used in the MOG potentially had a taste distinguishable from COG. This most likely violated the blinding process especially in a community where there was a “high awareness” level of Moringa’s usefulness as a nutraceutical. This should be considered as significant source of bias enough to dispute the conclusion of a causal relationship between the MOG intervention and the observed increase in CD4 count. A more firm conclusion could have been justified if the powder had been packaged in gelatin capsules which could mask the taste and also ensure more accurate and consistent dosing of the product. 2\. While literature references were made, the claims that Moringa leaf samples contained essential and non-essential amino acids, wide range of vitamins, minerals, carotenoids, polyphenols, phenolic acids, flavonoids, alkaloids, glucosinolates, isothiocyanates, tannins and saponins were unsubstantiated in this particular study. Perhaps a description of whether the manufacturer Prime Global Agricultural Industries Limited routinely analyzed the samples for the claimed ingredients would have been useful. 3\. Addition of the discussion of the potential of interactions between Moringa and antiretroviral therapy, particularly with efavirenz via alteration of liver metabolism would enrich the manuscript. 4\. It is more scientifically acceptable to use the terms such as “study participants” or “HIV infected individuals” instead of “HIV patients”. 5\. Sentences on line 174 and line 176 need reconstruction 6\. There may still be more work before a causal relationship can be claimed between the consumption of Moringa leaf powder and observation of increased CD4 count. Review recommends that the conclusion should limited to an association between the intervention and increased CD4 count, siting the observed shortcomings of the study design. Reviewer \#3: The manuscript reports findings from a randomised controlled double blinded trial assessing the effect of Moringa oleifera (MO) leaf powder supplementation on weight, BMI, MUAC, CD4 count and viral load. This is an important study given that MO is widely consumed and is an important herb among people living with HIV. Based on the data, the authors conclude that only the CD4 cell counts improve significantly (p=0.03) after 6 months concurrent MO leaf supplementation and ART. The manuscript however has some technical issues in some aspects of the trial design and statistical analysis. p11 Selection criteria L105 - Inclusion of patients on only one regimen limits generalisability of the effect on CD4 counts to patients on other regimen. Why was this necessary? L113 - Typically, randomisation follows screening and indicates a participant is enrolled into the study. However, under exclusion criteria authors state that patients taking micronutrient or natural health product supplements within 30 days of randomisation were excluded. Please clarify. p12 Sample size L122 - In the abstract the authors talk about significance of the mean increase of CD4 count from baseline at the 99% CI which is inconsistent with sample size justification based on a 5% margin of error, i.e. 95% CI. p12 - 13 Intervention L135 - How were the MO leaves positively identified? Was a botanist consulted? L156 - What seive size was used to prepare the intervention? 5g weight seems too heavy for 1 teaspoon of MO leaf powder. L158 - Specify that MO from other sources is what was disallowed. p14 L174, L177 - specify the type and volume of blood samples collected and a brief explanation or reference to the details of the laboratory tests. p15 L202 - It is the participants who provide consent to participant in the study, not the researchers. This presents a significant ethics issue for this study. p18-19 Why was the change in anthropometric and immmunological parameters within group important to analyse statistically? Given the objectives of the study, it is the difference between the intervention and control groups that is relevant. The results section is crowded with results that may not be important. In addition, the change from baseline at the end of each month may be more useful than the change between each month. Consider revising the results section and highlighting only the difference in change from baseline between the intervention and control group. One figure and one table, with legends, and focusing on the key results will further enhance the section. p24 L334 - There is no data on which the discussion on 'good adherence' is based. The strategy the authors describe on p13 L161 was to monitor adherence, rather than enhance or optimise it. No results are presented on the level of adherence. In fact a large portion of the discussion refers to perceptions from participants and other operational and implementation aspects that are core to the study and not presented in the results and are appearing for the first time as discussion. p21 L407 - From the power calculation, the study was adequately powered, perhaps a more diverse population of PLWH rather than a larger sample size could be recommended? The recommendations should be based on the effect on MO on CD4 and not extend to issues of food security and nutrition which have not been demonstrated by the current study. Several challenges and limitations are cited. To what extend were participants reluctant to keep clinic appointments? What was the effect on study outcomes and how was that accounted for in the analysis? The write up also has some grammatical errors and could benefit from revision by an English editor. Some comments that may be helpful Page 1 L1 - 'oleifera' is always written in lower case and itilised. Correct throughout the manuscript. L2 - Be consistent with the term used for study population. Ideally, use people first language such as 'people living with HIV (PLWH)' rather than HIV patients or HIV infected patients. Page 9 L46 - Change accountable to accounts L48 - revise grammar L71 - 'leaf' powder L72 - Sub-Saharan \*\*\*\*\*\*\*\*\*\* 6\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No Reviewer \#2: No Reviewer \#3: No \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <figures@plos.org>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0261935.r002 Author response to Decision Letter 0 4 May 2021 All comments have been responded to and attached as 'Answers to reviewers comments'. 10.1371/journal.pone.0261935.r003 Decision Letter 1 Monera-Penduka Tsitsi G. Guest Editor 2021 Tsitsi G. Monera-Penduka This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 24 May 2021 PONE-D-20-30776R1 A double-blind randomized control trial to examine the effect of Moringa Oleifera leaf powder supplementation on the anthropometric and immune status of adult HIV patients on antiretroviral therapy in a resource- limited setting PLOS ONE Dear Dr. Gambo, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria after your first revision. Therefore, we invite you to submit a further revised version of the manuscript that addresses the outstanding statistical issues raised during the current review process. Please submit your revised manuscript by Jul 08 2021 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <plosone@plos.org>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript: A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols>. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at <https://plos.org/protocols?utm_medium=editorial- email&utm_source=authorletters&utm_campaign=protocols>. We look forward to receiving your revised manuscript. Kind regards, Tsitsi Grace Monera-Penduka, PhD, MSc, MPhil, BPharm Hons Academic Editor PLOS ONE Journal Requirements: Please review your reference list to ensure that it is complete and correct. If you have cited papers that have been retracted, please include the rationale for doing so in the manuscript text, or remove these references and replace them with relevant current references. Any changes to the reference list should be mentioned in the rebuttal letter that accompanies your revised manuscript. If you need to cite a retracted article, indicate the article’s retracted status in the References list and also include a citation and full reference for the retraction notice. Additional Editor Comments (if provided): The authors have addressed the grammatical and other issue raise satisfactorily save the statistical issues. They should revisit additional comments from the current review and address them. \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation. Reviewer \#1: (No Response) \*\*\*\*\*\*\*\*\*\* 2\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Partly \*\*\*\*\*\*\*\*\*\* 3\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: I Don't Know \*\*\*\*\*\*\*\*\*\* 4\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes \*\*\*\*\*\*\*\*\*\* 5\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes \*\*\*\*\*\*\*\*\*\* 6\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: The authors appear to have at least partially implemented some changes to the statistical analysis. However, it is not possible to tell whether the analysis was performed correctly due to the lack of detail in the description of the statistical analysis and in the results shown. From what can be seen, it appears that the analysis was not specified correctly; or, if it was specified correctly, it was not reported correctly. Also, the authors should put some energy into evaluating whether the different types of analysis performed produce similar results. My suggestion to keep the planned analysis was because as a rule all pre-planned analyses should be presented unless completely inappropriate. However, that means that there will be potentially conflicting results, perhaps due to the effects of statistical correction, perhaps due to increased or decreased power, or perhaps due to different assumptions. In fact, this might result in relegating some part of the current work into a supplementary appendix on the grounds that it is less germane. The covariate analysis that is presented is also difficult to assess. The model structure (as in the other case) is not clearly specified. If transformations were used, they are not described. Note that "transformation" does not mean changing data format from wide to long format. Instead, it means such things as logarithmic transformation or other functional transformations, aggregation, or recoding. Simple manipulation of data shape is irrelevant to the manuscript because it is only a technical requirement of the data analysis software. As initially mentioned in the first review, all linear model analysis must be subjected to goodness of fit evaluation before interpretation. This is described in brief in the first review and does not appear to have been performed. My current recommendation is to obtain the services of a consulting statistician to work through the data analysis and review comments and to put everything onto a solid and consistent footing. \*\*\*\*\*\*\*\*\*\* 7\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <figures@plos.org>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0261935.r004 Author response to Decision Letter 1 3 Oct 2021 All comments have been addressed 10.1371/journal.pone.0261935.r005 Decision Letter 2 Monera-Penduka Tsitsi G. Guest Editor 2021 Tsitsi G. Monera-Penduka This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 15 Dec 2021 A double-blind, randomized controlled trial to examine the effect of Moringa oleifera leaf powder supplementation on the immune status and anthropometric parameters of adult HIV patients on antiretroviral therapy in a resource-limited setting. PONE-D-20-30776R2 Dear Dr. Gambo, We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements. Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication. An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at <http://www.editorialmanager.com/pone/>, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to- date. If you have any billing related questions, please contact our Author Billing department directly at <authorbilling@plos.org>. If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact <onepress@plos.org>. Kind regards, Tsitsi G. Monera-Penduka Guest Editor PLOS ONE Additional Editor Comments (optional): Reviewers' comments: 10.1371/journal.pone.0261935.r006 Acceptance letter Monera-Penduka Tsitsi G. Guest Editor 2021 Tsitsi G. Monera-Penduka This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 17 Dec 2021 PONE-D-20-30776R2 A double-blind, randomized controlled trial to examine the effect of *Moringa oleifera* leaf powder supplementation on the immune status and anthropometric parameters of adult HIV patients on antiretroviral therapy in a resource-limited setting Dear Dr. Gambo: I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department. If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact <onepress@plos.org>. If we can help with anything else, please email us at <plosone@plos.org>. Thank you for submitting your work to PLOS ONE and supporting open access. Kind regards, PLOS ONE Editorial Office Staff on behalf of Dr. Tsitsi G. Monera-Penduka Guest Editor PLOS ONE [^1]: The authors have declared that no competing interests exist.

# Introduction The number of individuals afflicted by severe mental illness (SMI) is increasing annually. Those with SMI, including schizophrenia, bipolar disorder (BD), and major depressive disorder (MDD), experience a life expectancy reduced by 10–20 years compared to the general population. The occurrence of cancers, including breast cancer (BC), exacerbates this shortened lifespan. Research shows that after adjusting for age, income, race, ethnicity, geographic location, and marital status, individuals with SMI face a twofold increased risk in all-cause mortality compared to those without mental disorders. Newly diagnosed female BC cases are more significant than the corresponding cases of lung carcinomas, depicting the leading position of BC in worldwide carcinoma incidence. Early diagnosis of BC can improve the prognosis. However, BC patients with SMI are more susceptible to being diagnosed with advanced BC and invasive tumor characteristics due to a lack of regular BC screenings. This highlights the importance of focusing on BC health management among individuals with severe mental disorders. Emotional abnormalities and psychological stress will affect the occurrence and metastasis of BC through various mechanisms. In addition, evidence reveals that patients with SMI have fewer BC screenings and less adherence to anticancer therapy, making BC treatment challenging in patients with SMI. Therefore, clarifying the linkage of SMI to the incidence of BC seems imperative, which can heighten awareness among patients and healthcare professionals, leading to early detection and treatment, thus enhancing BC survival rates. Clinical studies depict an elevated BC risk among the SMI population. However, they will be affected by clinical confounding factors, such as smoking, obesity, long-term use of anti-schizophrenic drugs, diabetes, physical activity, and alcohol consumption. Mendelian randomization (MR) has been widely utilized in causality inference. A two-sample MR is one of the essential strategies in which the single nucleotide polymorphism (SNP)-exposure and outcomes are acquirable based on two genome-wide association studies (GWAS) without overlapping samples, increasing the applicability of MR. Utilizing the summary data from two GWAS studies, evaluating the causal influence of exposure on the outcomes is possible. This is a huge advantage, allowing the performance of causal inferences between two traits even if their measurements are not from an identical sample set. It will enable us to take full advantage of statistics from various existing GWAS studies. The two-step MR, based on a two-sample MR strategy, investigates the possible roles of mediators between the exposure and the outcomes. In this study, we investigated the causal relationship between SMI and the incidence of BC using two-sample Mendelian randomization (MR), including Schizophrenia, major depressive disorder (MDD), and bipolar disorder (BD). Additionally, we employed a two-step MR approach to explore the mediating effects of smoking initiation, alcohol consumption, physical activity, type 2 diabetes, prolactin levels, and the body mass index (BMI) in the causal link between SMI and BC. The inverse-variance weighted (IVW) method was utilized as the primary algorithm to estimate the magnitude of causal effects. Understanding the association between SMI and BC occurrence could aid in enhancing BC screening and prevention among SMI patients, contributing to more effective health management strategies for individuals with SMI. # Materials and methods ## Study design This work is an Mendelian randomization (MR) research, with a two-sample MR and a two-step MR, whose implementation relied on three hypotheses: (1) There exist robust correlations of instrumental variables (IVs) with the exposure (association); (2) IVs are irrespective of the confounders (independence); (3) the effect of IVs on outcomes is achievable solely through exposure (exclusion restriction criteria). All the data comes from online databases. This study involves a reanalysis of publicly available GWAS data. All included studies have received approval from their respective academic ethics review committees, and written informed consent has been obtained from each participant. Therefore, no additional ethical approval is required. ## Genetic instrumental variables for exposures Summary data regarding the associations of schizophrenia BD and MDD, with single nucleotide polymorphisms (SNPs) were gathered from the Psychiatric Genomics Consortium (PGC) (<https://www.med.unc.edu/pgc/>), which is one of the largest, most innovative, and most productive consortia within the psychiatric domain. The PGC analyzes the genetic characteristics of schizophrenia, BD, MDD, and other mental diseases. A GWAS research is the source of summary data from schizophrenia-associated SNPs, which comprises up to 36989 cases and 113075 controls. A large GWAS meta-analysis provided summary data for BD-related SNPs, comprising 20,352 cases and 31,358 controls. Meanwhile, the most significant GWAS meta-analysis depicted MDD-related SNPs, including the three most extensive depression-related studies involving 246,363 cases and 561,190 controls (after elimination of overlapping samples), amounting to 807,553 subjects. We selected SNPs with p \< 5e-08 and removed linkage disequilibrium (LD) (r² \< 0.001, distance threshold = 10000 kb) to guarantee a strong association of SNPs with exposures. In addition, the F-statistics exceed 10 for the selected SNPs, indicating that the instrumental variables are associated powerfully with the exposure. The computational formula for F-statistic herein is Beta²/SE², where Beta refers to the genetic correlation with exposure and SE stands for standard deviation. We selected 147 SNPs as instrumental variables (S1 Table). ## Genetic instrumental variables for potential mediators The mediators-associated SNPs came from the GWAS summary data(<https://gwas.mrcieu.ac.uk/>), including smoking initiation, alcohol consumption, physical activity, type 2 diabetes, prolactin levels, and the body mass index (BMI) (S2 Table). The smoking initiation related-SNPs were obtained from the GSCAN, a worldwide consortium of genetic correlation meta-analyses, including up to 1.2 million individuals. The Within Family Consortium was the source of alcohol consumption-related-SNPs, consisting of researchers intending to comprehend the effects of the human genome on a broad spectrum of socioeconomic and medical traits based on the relevant individual datasets. The UK Biobank was the source of physical activity related-SNPs. The largest GWAS of physical activity was performed with three self-report-based measures and two wrist-worn accelerometry data-based measures. The type 2 diabetes and BMI- related-SNPs were obtained from the MRC Integrative Epidemiology Unit. A genome- wide meta-analysis was the source of prolactin levels related to SNPs, involving 21,758 participants. ## Genetic instrumental variables for BC The source of pooled data on BC-associated SNPs with 122,977 cases (69,501 ER+ cases plus 21,468 ER cases) and 105,974 controls was the Breast Cancer Association Consortium (BCAC). We performed subgroup analyses based on the status of estrogen expression. ## MR analyses A two-sample MR analysis exploring the exposures’ impact on BC employs three analytical approaches: the IVW (inverse-variance weighted; the foremost approach), the MR-Egger, and the weighted median methods. During the MR analysis, the IVW achieves meta-aggregation of the multiple loci’s effects (beta0). The application principle of IVW is that the SNPs are entirely irrespective of each other and are valid IVs. MR-Egger regression is a method for achieving pleiotropy identification and adjustment in the MR analysis, relaxing the requirement that there is no pleiotropy between genetic variants within IVW methods. In the case of MR-Egger regression, its intercept estimates become the mean of pleiotropic effect estimates within each genetic variant. With the weighted median approach, the medians are derived through weight- dependent sorting of all the independent values within the SNP effect. The weighted median can become a robust estimate if the MR core hypotheses are satisfied by a minimum of 50% genetic variation. In the first step of the two-step MR for the effect mediators, the causal effect of exposure to the mediator variable (beta1) was calculated using the two-sample MR method. The second step involves estimating the mediator’s causal influence on the outcomes (beta2) by the multivariate MR method. Beta0, beta1, and beta2 were all significant, indicating a causal relationship between exposure and outcome, and mediator variables could partly mediate this relationship. In general, we can take beta1\*beta2 as the mediating effect from exposure to the outcome and beta1\*beta2/beta0 as the proportion of the mediating effect within the causal relationship. Beta0 was not significant, but both beta1 and beta2 were significant, suggesting that this mediating variable entirely mediated the association from exposure to outcome. Beta0 is significant, but at least one of beta1 and beta2 is not significant, indicating that this mediating variable has no effect on the causal association from exposure to the outcome. ## Sensitivity analysis The causal effect estimates were biased due to the possible heterogeneity among IVs. Our study used the Cochran Q statistic of the MR-Egger and the IVW method to perform a heterogeneity test for detecting heterogeneity. P \> 0.05 indicated no heterogeneity in IV. Therefore, we applied a fixed-effects model in the MR analysis; otherwise, a random-effects model was utilized. Confounders are probable mediators of the exposure-outcome causality. We tested the horizontal pleiotropy of IVs by utilizing the MR-Egger’s intercept and its p-value. If this intercept approached zero (\<0.1) and the p-value exceeded 0.05, we excluded the effect of pleiotropy at the IV level. For better validation of the potentially abnormal SNPs and the horizontal pleiotropy, the Mendelian random pleiotropic residuals and outliers (MR-PRESSO) approach was incorporated. The eligible SNPs were subjected to sensitivity analyses through the aid of Leave-one-out. This method primarily calculates the MR results of the remaining IVs after excluding IVs one by one, regarded as reliable if unobvious alterations are noted in the overall error bars after the exclusion of every SNP. ## Statistical power We calculated the statistical power of MR through online statistics (<https://shiny.cnsgenomics.com/mRnd/>), finding a type I error of 0.05 and K (case composition ratio) values of 54%, 40%, and 17% for all BC, ER+ BC, and ER- BC, respectively. R²_xz is the variance proportion explainable for the SNP–exposure correlation, whose computational formula is: 2×EAF×(1-EAF)×beta². Here, EAF and Beta refer separately to the effect of allele frequencies and the predicted influence of genetics during exposure. The R²_xz for schizophrenia was 20.93%. The sample size for the BC pooled data was 228,951, having 122,977 cases and 105,974 controls. We were sufficiently competent (\>80%) when the expected odds ratio (OR) for schizophrenia was≥1.03. Power estimates of BC subtypes for schizophrenia IV are represented in S3 Table in. Data were analyzed in this study through the TwoSampleMR ver. 0.5.6 plus the R ver. 4.1.0. Differences were considered significant when P \< 0.05. # Results ## The two-sample MR estimates for schizophrenia, BD, and MDD with the risk of BC ### MR estimates for schizophrenia–BC susceptibility association Based on evidence, the genetically-predicted schizophrenia was associated with all BC (OR = 1.06, 95% CI 1.02–1.10, P = 0.001), ER-BC (OR = 1.06, 95% CI 1.01–1.11, P = 0.0123), and ER+ BC (OR = 1.06, 95% CI 1.03–1.10, P = 0.0009) risks. Nine palindromic SNPs were excluded (S4 Table). The F-statistics range for schizophrenia-related SNPs was 29.65–143.41, indicating a powerful IVs–exposure association. With the schizophrenia-related SNPs, 20.93% of the genetic variation was understandable. ### MR estimates for BD–BC risk association The causal association of BD with all BC risk was supported by little evidence (OR = 1.05, 95% CI 0.96–1.14, P = 0.2809). The association with ER-BC (OR = 1.00, 95% CI 0.86–1.15, P = 0.9578) and ER+ BC (OR 1.03, 95% CI 0.94–1.12, P = 0.4964) were similar. As displayed in S4 Table in, we eliminated five palindromic SNPs. Within an F-statistics range of 29.66–56.81, the BD-related SNPs were associated significantly with the exposure. It was observed that 4.92% of the overall genetic variation was explainable through the BD-related SNPs. ### MR estimates for MDD–BC risk association MR estimates showed that the genetically-predicted MDD was insignificantly associated with all BC (OR = 1.08, 95% CI 0.97–1.20, P = 0.1436). As revealed by the subgroup analysis, less evidence supported the causal correlations of MDD with the ER-BC (OR = 1.05, 95% CI 0.90–1.21, P = 0.5361) and the ER+ BC (OR = 1.03, 95% CI 0.91–1.17, P = 0.6453) risks. Additionally, nine palindromic SNPs were excluded (S4 Table). The F-statistics range for MDD-related SNP was 30–78.45, implying that the weak instrumental bias barely affects the SNP. It was observed that 1.78% of the overall genetic variation was explainable through the MDD-related SNPs. – Figs shows the scatter and forest plots of schizophrenia, BD, MDD, and BC. ### The two-step MR estimates for mediators-associated SNPs between schizophrenia and BC risk A two-step MR study explored the potential role of mediators through the effect sizes of schizophrenia to the mediators, in which schizophrenia and smoking had a significant correlation (beta1 = 0.0438, P = 0.0058). In a multivariate MR analysis of mediators to outcomes, smoking, alcohol consumption, and BMI were associated with the occurrence of BC. Among them, smoking revealed a significant correlation with all BC (beta2 = 0.1467, P = 0.0166). Significant connections of alcohol intake to all BC and ER-BC were noted. Besides its significant linkage to all BC, BMI was also found to be significantly linked to both ER-BC and ER+BC in the subgroup analysis. Beta0, beta1, and beta2 were all significant, indicating a causal relationship between exposure and outcome, and mediator variables could partly mediate this relationship. Therefore, in the causal relationship analysis of smoking in schizophrenia and BC risk, beta0, beta1, and beta2 were all significant. It depicted that smoking could mediate schizophrenia leading to an elevated all BC risk, revealing a final causal relationship of 11.29% mediated through smoking. ## Sensitivity analyses According to the heterogeneity test results, several significant differences were noted in the estimated causal associations of exposures with all BC risks (S5 Table). Thus, these models were applied with random-effects MR estimates when the Q p-value\<0.05; otherwise, fixed-effects MR estimates were used. Based on the horizontal pleiotropy assessment by MR-Egger intercept and the p-values, such pleiotropy did not affect the MR estimates for exposures and BC (S6 Table). The Leave-one-out sensitivity test revealed that the result alterations were unobvious no matter which schizophrenia, BD, and MDD-related SNP was eliminated, indicating the high robustness of the MR results (Figs –). No potential pleiotropy was detected in the MR estimation of exposures–BC risk casualty by the MR-PRESSO approach, despite discovering several outliers (S7 Table). For all BC and ER+ BC, after the SNPs removal in restrictive MR analysis, schizophrenia was found to be significantly associated with the risks of ER+ BC (OR 1.05, 95% CI 1.02–1.09, P = 0.0009) and all BC (OR 1.05, 95% CI 1.02–1.08, P = 0.0003). Outlier SNPs were not identified after applying the MR- PRESSO method in ER-BC. No potential pleiotropy was detected in the MR estimation during causal correlations of BD and MDD with BC risk by the MR- PRESSO approach, and it found little cause relationship. The scatter, forest, and Leave-one-out plots depending on MR-PRESSO applied after excluding outlier SNPs for exposures on BC are shown in – Figs. # Discussion As indicated by the present TSMR analysis covering over 122,977 BC patients and 105,974 healthy controls, BC susceptibility is influenced etiologically by schizophrenia. The schizophrenic population is more susceptible to BC, irrespective of the estrogen levels. As revealed by the sensitivity analyses, outcomes are influenced through IVs by exposure only, instead of pathways like confounding ones, suggestive of the absence of pleiotropy. The latest global burden of disease data indicates a significant increase in the raw prevalence, incidence, and burden of schizophrenia between 1990 and 2019.The latest meta-analysis depicted that the schizophrenic population is more susceptible to BC, but there is significant heterogeneity among the included studies, which may be due to confounding factors. Mendelian randomization (MR) research, unlike clinical studies, is less susceptible to clinical confounding factors. We assessed the schizophrenia–BC casualty based on a two-sample and two-step MR strategy, finding that BC was more prevalent among the schizophrenic population, consistent with previous MR studies. Some behaviors in patients with schizophrenia are risk factors for BC, including obesity, childlessness, and use of anti-schizophrenia drugs leading to elevated prolactin levels. We also performed a two-step MR analysis on confounding factors and found that smoking explained 11.29% of the causal relationship between schizophrenia and BC risk. Furthermore, accumulating evidence that the underlying mechanisms of schizophrenia and BC share a common genetic basis. Studies have shown that anti- schizophrenia drug use for more than 5 years is associated with BC risk but this association was not found in our study, which is limited by the selected database. BC patients with schizophrenia have a two-fold increase in the all- cause mortality hazard. Patients with schizophrenia have less BC screening than the general population, and poor treatment adherence, which may have contributed to higher mortality in BC patients with schizophrenia. It suggests that we should strengthen the screening and prevention of BC in patients with schizophrenia. Bipolar disorder refers to a range of psychiatric illnesses with both manic or hypomanic episodes and depressive episodes that are recurrent, chronic, and severe. Several researchers have now focused on its association with the incidence of carcinomas. The causality of BD and BC risk is controversial. In our MR research, BD failed to be linked to BC incidence, and our results differ from the published studies. Although more IVs for exposure-related SNPs were used in this article, the IVs used for exposure-related in this article included Asian populations. In contrast, the outcome-related IVs were from the European populations, leading to errors in the results. Major depressive disorder (MDD) is a common illness among women where the incidence is twice as high as in men. The 1-year prevalence of MDD is approximately 6%. Previous MR studies have shown that MDD increases the risk of developing BC, whereas our study did not observe a similar result. In our study, the linkage disequilibrium (LD) was removed according to kb = 10000, r2\<0.001 where there is less confounding and pleiotropy while the previous study set kb = 3000, r2\<0.1. Further, numerous studies have shown that MDD is more common after BC diagnosis and emphasize the importance of MDD in BC prognosis. Few works of literature have investigated the relationship between MDD and the BC occurrence, and clinical studies have not shown a positive association. Dysthymia with major depression is defined as having MDD during periods of dysthymia. It is speculated that persistent depression status and intensity of depression are additional risk factors in the development of BC, and the duration and intensity of depression are more critical. It is necessary for the stratified analysis of MDD and BC risk. Unfortunately, the data of MDD-related SNP we obtained cannot be stratified. Our study’s strength lies in identifying a causal relationship between schizophrenia and the incidence of BC using the MR method. Additionally, we discovered that smoking plays a certain mediating role in this relationship. The results indicated that smoking mediates schizophrenia to elevate the risk of BC, suggesting we should improve BC screening for mentally ill patients and manage unhealthy behaviors, especially smoking. Likewise, our study has some limitations. First, other confounding factors, those not available (eg age stratification of BC populations and duration and intensity of schizophrenia exposure) may also have influenced the results. Second, mediating exposure factors only explore the common influencing factors. Third, because this study relied on pooled data from GWAS of European ancestry, the generalization of this result should be validated in other ethnic groups. In future research, larger-scale clinical studies should be conducted to comprehensively assess the relationship between schizophrenia and the incidence risk of BC. Additionally, clinicians should delve deeper into exploring the underlying mechanisms linking schizophrenia and BC to facilitate the development of more effective intervention strategies. In terms of clinical application, BC risk should be integrated into the healthcare management plans of individuals with schizophrenia. This could include measures such as health education, smoking cessation support, and regular BC screening. Furthermore, psychiatrists and oncologists should actively engage in interdisciplinary collaboration to better manage the co-occurrence of these distinct diseases across different medical domains. # Conclusions In conclusion, our study utilizing the two-sample Mendelian randomization method underscores the significance of schizophrenia as a potential risk factor for BC. Smoking was identified as a mediating factor in this relationship. These findings highlight the need for enhanced health management strategies for individuals with schizophrenia. Further exploration of biological mechanisms and interdisciplinary collaboration will guide future interventions and public health policies in mitigating BC risk in this vulnerable population. # Supporting information [^1]: The authors have declared that no competing interests exist.

# Introduction Theoretical and empirical research has linked trends in species occupancy, abundance and population extinction with habitat alteration and community structure in a wide range of animal taxa. Specifically, over time, habitat fragmentation can result in increased abundance of rapidly dispersing species and the local extinction of poorer colonizers. The extinction debt hypothesis or theory (EDT) introduced the counter-intuitive idea that with continued permanent habitat destruction an ordered extinction starts with species that are the best competitors ending with the worst competitors but are the best colonizers. Tilman's model introduced a strong tradeoff between competitive and colonizing ability; thus, when habitat is fragmented and destroyed, the best competitors do not have the dispersive ability to colonize fragmented landscapes. Theory also suggests that under equilibrium conditions, species with good competitive abilities are also the species which are likely to dominate biomass and in some ways typify fish community assemblages. At the same time, the poorer competitors, but better dispersers or colonizers, would increase in relative abundance and spatial occupancy. For example, if a community experiences habitat fragmentation and loss, an extinction debt is realized that will result in the species with the greatest competitive ability and weakest colonizing ability to trend toward local extinction (extirpation) over multiple generations. With permanent habitat destruction, extirpation becomes inevitable once the amount of suitable habitat falls below the extinction threshold. The speed at which the debt is repaid (i.e. rate that the species abundance and occupancy disappears = amortization of the debt) is proportional to the distance from threshold. For populations close to but still below the threshold, the amortization period of the debt may be very long and not necessarily noticeable in the short term especially given the large natural variability of many animal populations. Additionally, for regions where suitable patches remain even with high connectivity, a debt may still be incurred, but it will be less noticeable in the shorter term and low abundance populations are more vulnerable to stochastic events. In situations where habitat can regenerate at time scales close to the the generation time of the species involved, then the extinction rate is variable and the extinction threshold may not be a one way barrier. When the habitat can regenerate faster than the generation time of the species, extinction is not inevitable and in fact unlikely, but full recovery of the population may take longer than expected based on a simple demographic analysis. In the marine environment habitat fragmentation and loss caused by bottom trawling and dredging, can reduce benthic biodiversity, community composition and biomass of megafauna and epifauna. Trawling impacts on benthic communities is complex and related to the degree and frequency of natural disturbance and resilience of the community. Sandy habitats may naturally experience a high frequency of disturbance and impacts from trawl fishing is comparatively low. In contrast, pebble and rocky habitats with infrequent natural disturbance are more sensitive to the effects of fishing and are more likely to experience a reduction of species diversity following disturbance. Georges Bank is a productive ecosystem in the northwestern Atlantic that has supported prolific fisheries for over two centuries. Since about 1980, the GB finfish community has gone through major changes related to fishing, large-scale species movement and climate change. GB has a wide range of habitat types including sands, granules-pebbles, cobbles and boulders. The effects of bottom trawling and dredging have been well documented showing declines in benthic productivity, species richness and increased homogenization. However, the expected effects of habitat homogenization on mobile fish are more complex compared to the physical disturbance of sedentary benthic communities. Temporal and spatial dynamics of mobile species depend on a range of habitats critical for successive life stages. Hanskii et al., pointed out that in the course of an extinction debt there will be a “transient” period with increased prevalence of rarer species. This results because of the time-lag between fragmentation and eventual extinction and is especially important for species close to their extinction threshold. Thus, evaluating the present conditions in an ecosystem may not reflect the actual extinction debt accrued or potential future population extinctions resulting from habitat destruction. When perturbations destroy habitat it becomes fragmented and less structured and eventually homogenized, and counter- intuitively, local species richness may actually increase during the transient period followed eventually by richness declines. As spatially structured habitat becomes increasingly fragmented, boundaries between heterogeneous habitat types are blurred and competitors with specialized niches are forced to occupy suboptimal habitat which reduces their fitness. The key feature that emerges is that a greater number of species will occupy a greater proportion of area leading to greater average richness per sampled area because of the habitat homogenization. Extinction debt theory was developed for terrestrial ecosystems that are comparatively easier to survey than marine systems. Frisk et al. 2011, studying the GB finfish and shellfish community, and the related concept of occupancy and abundance (O-A) relationships, found that the strengths and slopes of the relationship declined between 1963–2006 and was correlated to trawling effort. The changes in O-A relationships are consistent with habitat fragmentation as has been well documented for terrestrial systems. It is important to recognize that EDT is a theoretical model with a hard tradeoff between colonizing and competitive abilities. This hard tradeoff needs caution when applied in marine environments because of the a wide range of life histories and dispersal levels displayed from passive large-scale pelagic eggs, to nesters and live bearers and marked ontogenetic niche shifts that includes differences in the strength and types of habitat dependence. However, when the theoretical underpinnings of EDT are combined with empirical observations, the resulting analysis can serve as a tool to evaluate the state of the system and provide information critical for understanding trends in species richness. In this study we consider long term changes in species richness spatial distribution (occupancy) in the Georges Bank ecosystem in the context of potential habitat homogenization and transient signs of extinction debt that could be manifested in increased species richness and changing occupancy. We specifically examined OA trends for groups of species that correspond to Competitor-Colonizer groups to address the dynamics of competitive skill and dispersal in marine systems. The overall objective is to determine if long term trends in the finfish and shellfish community are consistent with a system experiencing extinction debt from habitat alteration and factors that confound or contribute to mechanistic understanding of the trends. # Materials and Methods ## Data We analyzed data from the autumn National Marine Fisheries Service’s (NMFS) bottom trawl survey from 1963–2008 for Georges Bank. This is a random stratified survey and consists of an average of 63 tows each year of 30 minutes each at 7 km/h. The gear used is a Yankee 36 bottom otter trawl with 12.7 mm codend liner. For every tow, all fish and invertebrates were separated into species counted and weighed. In 2009, new tows were added to the survey as well as changes to trawl mesh size, gear type and tow speed. These changes affected species relative catchability, diversity and abundance amongst other community measures; therefore, our analyses are based on data only up to and including 2008 rather than allowing a gear based artefact to influence results. Georges Bank is a relatively small area with a fairly high tow density therefore, we dealt with the results of individual station tows rather than using stratified means and the number of tows per stratum is relatively constant from year to year and most differences occurred in two consecutive years: 1978 and 1979. Furthermore, because strata are based on depth rather than relevant habitat features, information related to spatial structure and occupancy could be lost if stratified means were used in this analysis as depth does not necessarily equate to relevant habitat in the sense that habitat is being used here. ## Species filtering The NMFS database for Georges Bank has recorded the presence of 219 different categories of catch including “trash” or broad lumped taxonomic categories with many species. For the present analyses, it was necessary to apply appropriate filtering criteria to obtain subsets of the 219 categories that reduce the possibility that poor catchability of rarely caught species would lead to spurious conclusions. For analyses of abundance trends, we needed to confine the species to groups that were well caught by the survey and thus relative abundance and occupancy differences could be considered real. We therefore applied a filter following to the dataset where the only groups considered in the analysis were those that: \(1\) were caught in more than 40 of the survey’s 46 years; (2) where more than 1000 individuals were caught over the survey time series (in standardized tows); (3) identified to the species level; (4) were not shrimp as they are not well caught by the gear and are sometimes partially destroyed during sampling. This filter resulted in a dataset which included 39 species. For analyses of species richness, we need only to be confident in presence/absence thus the filter was relaxed to allow more rarely caught species into the analysis. Accordingly, only filtering criteria (1) and (3) above were applied. This filter resulted in a data which included 164 species. ## Competitor-Colonizer group classifications The reduced species dataset, consisting of 39 species meeting the filtering criteria above were classified as competitors (*CM*, n = 9), colonizers (*CL*, n = 9) or mixed (*MX*, n = 21). These qualitative designations were determined through examination of Fishbase.org as well as primary publications and DFO and NMFS species information sheets and. The main characteristics subjectively used for dividing species three different groups were: maximum age, fecundity, age at maturity, larval/juvenile phase dispersal. A species with a long life span and older age at maturity would be classified in the Competitors group as would species with relatively low lifetime fecundity and species having live birth. Conversely, Colonizer group species tended to have relatively high lifetime fecundity as well as an ability to disperse widely often at juvenile stages (e.g. a long pelagic phase found in many invertebrates for example). Colonizers were often of small body size. Mixed group species were those which displayed a mix of characteristics and it was not obvious how to classify a species. In some respects the mixed group is the null group which represents a lack of designation rather than a positive group assignment. There is some overlap with the groupings of, though the Competitor- Colonizer groups developed here do not correspond perfectly to their groupings of periodic, opportunist and equilibrium. The ecological interpretations here are based primarily on trends in the two ends of the Competitor-Colonizer classification: Competitors and Colonizers which have a more definitive classification and therefore the results have a certain degree of robustness to uncertainty in group classification. ## Analyses Occupancy was modelled using nested generalized additive models (GAM) models to account for abundance abundance and competitor colonize group abundance and competitor colonize group and year as 3 but with a random effect for species: $$O = s\left( {log(A)} \right)$$ $$O = s\left( {log(A)} \right) + s\left( {log(A)byCC} \right)$$ $$O = s\left( {log(A)} \right) + s\left( {log(A)byCC} \right) + s\left( {yearbyCC} \right)$$ $$O = s\left( {log(A)} \right) + s\left( {log(A)byCC} \right) + s\left( {yearbyCC} \right) + s\left( {species,random} \right)$$ Where O is occupancy which was calculated as the proportion of stations where a species was caught in any year. A is the mean abundance of that species in the survey for a year. CC is the competitor-colonizer group factor. *s* represents a cubic spline smooth. Because occupancy is a proportion, the GAM was fitted with quasi-binomial errors. 1732 datapoints were available for modelling the 39 species dataset. Model 1 is the most basic OA model with a single fitted smoother describing occupancy as a function of log abundance. Model 2 accounts for deviations from the Model 1 smooth within competitor- colonizer groups. Model 3, is the model 2 but accounts then for a temporal effect (year) on occupancy in groups. Model 4 is the full model (model 3) with a random effect for species. For Models 2, 3 and 4 we restricted smoothers to have similar degrees of freedom within groups and smoothing terms determined using generalized cross validation. Occupancy was logit transformed and models were fitted using the logit link function. The nested models were compared to adjacent order models with ANOVA (F-test) to determine the importance of terms for explaining occupancy. We examined the mean species richness at stations over time in the 164 species dataset. Changes in richness can be indicative of the effect on species of homogenizing habitat. All analyses were conducted in R and specifically using the library mgcv. # Results Occupancy vs abundance of the 39 species Georges Bank dataset for all years showed the standard log asymptotic relationship for all species combined as well as just for species in different colonization groups ( left column). The overall model (Model 1) and the within CC group model (Model 2) smoothing terms were significantly different than the overall group smooth and Model 2 was significantly better than Model 1 (ANOVA, p \< 0.0001). When the data were examined in relation to their position above and below the fitted GAM line, latter years tended to fall below the smoothed line while earlier years tended to be above. Therefore, a statistic called occupancy deviation (OD) was developed to show the proportion of data points larger than predicted (above the GAM line) and smaller than predicted (below the GAM line) fit for each year ( right column). OD represents the proportional deviation in any one year from the expected mean community value determined for all years and a trend in OD over time indicates that the OA relationship has also changed over time. OD showed a declining trend over time for all species and for Competitor and the Mixed groups. To the contrary, the OD for Colonizers increased. This indicates that for Competitors, their occupancy at the end of the time series was 10% below the mean fitted occupancy for competitor species while at the beginning of the series it was as much as 40% above the mean fitted occupancy. The GAM model analysis (Model 2) showed that the occupancy abundance relationship was significantly different within colonizers and competitors but not for mixed species compared to the overall model. The temporal trend in OD suggests that the OA relationship for the whole community as well as component groups changed between 1963 and 2008 and differently between Colonizers and Competitors. Model 3 formalizes a test of a temporal trend of OA within groups. Model 3 is Model 2 with an additive smooth term for year within the Competitor-Colonizer groups. Model 3 was significantly better than Model 2 (ANOVA, p \< 0.0001) and all the additive smoothing terms provided a significantly better fit. This result confirms the trend shown in OD where occupancy of Competitors significantly decreased over time even when that occupancy was standardized by abundance. It also indicates that the increase in occupancy in Colonizers over time was significant. Competitor abundance increased between the 1960s and 1970s while the occupancy abundance relationship remained fairly stable, but the relationship changed more so from the 1980s onward. Colonizers showed increasing occupancy and abundance from the 1960s until the 2000s and a change in occupancy abundance relationship during this time. Model 4, contains a random effect for species that explains 18% more of the variance in occupancy than Model 3 without a random effect. Inclusion of a random effect for species still showed significant differences in occupancy over time for all the Competitor-Colonizer groups, confirming the decline in occupancy over time of competitors and increase in occupancy of colonizers. The species random effect, however, does render the differences in abundance within species groups an insignificant explanation for occupancy differences. This suggests that changes in species assemblage over time may be an important explanation for occupancy changes on Georges Bank. Mean species richness per station increased from 1963 to 2008 by about 2.5 species or about 20% (linear model). The smoothed GAM fitted to the same data show a more interesting curve with similar lows of about 11.4 species in 1969 and 1984 and a maximum richness of 13.8 species in 2004. The increase in richness is primarily occurring after the mid-1980s. These results were based on the 164 species dataset which specified that a species had to be caught in 40 out of 46 years in the survey to be included. Therefore this analysis does not represent an increasing occurrence of incidental catches of unfamiliar species or a species richness creep problem (i.e. that surveys show increasing richness over time as implementation of new protocols may include rarer and incidentally caught species as interest in all components of the ecosystem grows in the research community). This represents a broadening of the assemblage such that species which may have been confined to a smaller subset of sampling stations in the past are now found at more stations and the greatest rate of change occurred from about 1984–2004. The random effect for species (Model 4) captures some of the effect of this change in assemblage which could be almost 20% of the variance in the occupancy abundance relationship. # Discussion Extinction debt theory (EDT) predicts that ecosystems experiencing permanent habitat destruction will accrue a debt on local species extinction (extirpation) where the best competitors will be the first species to disappear from the system. It is a debt in the sense that it does not occur in the generation first experiencing the destruction but is paid slowly over multiple generations through declining fitness. The best colonizers will be less susceptible to habitat destruction because their increased dispersal ability allows them to find suitable habitats in a fragmented landscape even when total suitable habitat is decreasing. There is thus a colonizer credit where the destructive force gives colonizer species a relative competitive advantage. In fact, this credit can lead to longer term increases in community diversity: a corollary of EDT that may not be well appreciated. In aquatic systems, one could expect EDT to be manifested in a decline in occupancy and abundance of competitor species at a faster rate than colonizer species. In marine systems this process is unlikely to lead to extirpation or extinction; instead, it more likely to be observed with long term changes in the occupancy of colonizers and competitors. Our analysis of the Georges Bank fish and invertebrate community shows a change in occupancy consistent with a moderated extinction debt where abundance standardized occupancy of competitor species decreased over time while it increased for colonizer species. Donovan and Flather showed that when habitats are fragmented and habitat quality is reduced there is a decreased occupancy of the fragmentation sensitive species (i.e., the equivalent of competitor species defined in this study). They speculated that the net result of local occupancy and a decrease in breeding success could lead to large scale population declines. Occupancy reduction of terrestrial animals in response to habitat destruction or fragmentation is common. It is not unreasonable that the ecological occupancy responses in relation to habitat fragmentation may operate also in marine systems similar to bird populations which show habitat dependence in forested landscapes. The habitat dependence, however, may be more difficult to uncover in marine systems because of the technical challenges in identifying and classifying marine habitats, difficulties measuring changes in habitat fragmentation and problems identifying causes of the change in habitat quality and fragmentation. Additionally, most marine animals have large ontogenetically changing ecological roles and dependencies which can be manifested in very different habitat preferences as individuals age. The fact that many marine species have highly dispersive larval stages and more sedentary adult stages means that extinction debt in a marine system would have terrestrial analogs as plants when young and birds when old. There are certainly added challenges to application of EDT in marine systems. Georges Bank has been highly targeted with benthic trawling gear since industrial fishing began in earnest in the 1950s. Benthic trawling gear is known to be destructive to habitats especially those that have long recovery times. Benthic trawling (scallop dredging and otter trawling) is the major human induced habitat modifier that has occurred on Georges Bank and if habitat deterioration and fragmentation has occurred on Georges Bank then it is reasonable to assume benthic trawling is an important driver of habitat change. Exacerbating any habitat destruction or loss; however, could be a change in suitability of the persistent Georges Bank fish and invertebrate community by warming temperatures and of course direct exploitation by commercial fishing activities. The decline in occupancy of competitor species on Georges Bank from the early 1960s until 2008 was undoubtedly affected by fishing activities which decreased their abundance. This alone would lead to a decline in occupancy of any species with a positive occupancy-abundance relationship. This decline could have also depleted some sub-populations changing their relative contribution to overall abundance and consequently how the population uses space as a whole. The strength of the trends in the occupancy-abundance relationship have recently been confirmed and recreated through simulations that assumed species statistical spatial distributions derived from natural communities and a simulated survey. However, these studies showed that abundance was not the only factor explaining the occupancy decline and the nature of occupancy abundance relationship itself changed over time. Even if abundances were the same in 2008 as in 1963, less space was used to support that abundance. Furthermore, refining this phenomenon showed that it was true for the whole community as well as within the competitor and mixed groups, but was the opposite for colonizer species. This decline in overall species occupancy, especially for competitor species, while colonizer species occupancy increased, is consistently explained by a decrease in habitat availability and/or suitability and is broadly captured in an ideal sense by EDT. George’s Bank has several closed areas to fishing established in 1994. These areas could have acted as refuges for competitor species in particular which may buffer a decline in competitor occupancy. The closed areas probably would have little effect on the transient effect of increased colonizer occupancy. The closed areas present an interesting possibility for future work in the sense that they create an experiment that could be subject to a relatively rigorous before-after press perturbation design. This experiment could possibly reveal impacts of the closed areas on the community from a perspective of EDT, i.e. the mix of community characteristics along the competitor-colonizer continuum. The contraction of abundance into a smaller area for competitor species means that these populations are more vulnerable to fishing activities than they were previously. Spatial contraction of a population is known to be an important factor to consider in the sustainable management of commercially exploited species especially if the fishery catch per unit effort is interpreted as an indicator of population abundance. The decoupling of population size with commercial catch per unit of effort that can occur with population spatial contraction can lead to over-exploitation of a harvested population. It must be noted though that much of this increased biomass since 1994 is likely to be in closed areas and therefore there may be less concern about overexploitation of the community since an important portion of that biomass remains unavailable to fishing. Sea surface temperatures (SST) have been anomalously warm on Georges Bank since 1970 compared to most of the previous 70 years and those anomalies were greater in more recent years suggesting a possible regime in SST. These increasing temperatures could have decreased the suitability of Georges Bank for colder water adapted species particularly demersal competitor species which may require greater environmental stability. This suggests that temperature increases could also contribute to observed declines in the Georges Bank fish and invertebrate community abundance but not so much occupancy, at least not as a direct impact; but occupancy changes could be mediated through foodweb alterations that new species may have on the assemblage. EDT is an approach first described in a modelled system based on mixes of species with perfect tradeoffs between competitive and colonizing abilities. EDT provides a useful basis for interpreting trends in occupancy-abundance relationships and species richness. However, the approach is limited in its current form for several reasons. Most importantly it is an equilibrium theory which assumes habitat destruction is permanent in a closed system without immigration or emigration. The cobble habitats, which are characteristic of Georges Bank fishing areas, have a regenerative capacity so their destruction is not usually permanent even if not immediately regenerated. Rescue effects from adjacent areas may also be possible, even if the areas adjacent to Georges Bank (Gulf of Maine, Scotian Shelf, Southern New England Banks) have experienced large declines in fish community abundance but these assemblage species still exist in these other areas. The second major issue with the tenets of EDT in a practical context is that most species have mixed strategies with both competitive and colonizing characteristics and cannot be so generally categorized. This is why our mixed group here contains so many species while start and end points of the continuum have fewer species. Application of EDT in a marine system led to a useful analysis of community occupancy over time and provided a broad causal mechanism for its change (habitat fragmentation and destruction). The habitat forcing mechanism may be exacerbated by climate change and could also affected by direct impacts of fishing. There are conservation implications for this work in that competitor type species presently may be more vulnerable to fishing activities than they were in the 1960s although protected areas still shelter significant important portions of this biomass. Our combined theoretical and empirical approach gives insights into possible causes of trends observed for the Georges Bank fish community when viewed from the perspective of competitors and colonizer functional groups. # Supporting Information This research was initiated as part of the work conducted by members of the International Council for the Exploration of the Sea (ICES) Working Group on Fish Ecology (WGFE). We would like to thank the members of WGFE for support, review and assistance on this research, particularly Brad Harris. We are grateful to NOAA/NMFS/NEFC Woods Hole and particularly Bill Kramer for providing us with fall survey data. Participation of M.F. in WGFE was supported through NSF grant (OCE-0738214.). The work of V.T. was partially funded by the EU project IMAGE (contract FP6–044227). We are grateful to two other unnamed colleagues who improved the analysis and manuscript. [^1]: The authors have declared that no competing interests exist. [^2]: **Conceptualization:** DED MGF VMT. **Data curation:** DED MGF VMT. **Formal analysis:** DED MGF VMT. **Funding acquisition:** DED MGF VMT. **Investigation:** DED MGF VMT. **Methodology:** DED MGF VMT. **Project administration:** DED MGF VMT. **Resources:** DED MGF VMT. **Software:** DED VMT. **Supervision:** DED MGF VMT. **Validation:** DED VMT. **Visualization:** DED MGF VMT. **Writing – original draft:** DED MGF VMT. **Writing – review & editing:** DED MGF VMT.

# Introduction Pancreatic cancer (PC) is one of the most aggressive and most lethal solid malignancies. The pancreatic tumor microenvironment favors the recruitment of immunosuppressive cells that dampen anti-tumor immune responses, allowing tumor cells to evade immune surveillance and leading to tumor progression. Understanding the mechanisms by which these anti-tumor immune responses, specifically those mediated by T cells, are regulated in PC is therefore critical for developing new, targeted treatment options. Effector CD4⁺ and CD8⁺ T cells play important roles in the host’s immune response to cancer. Early studies showed a conventional “helper” role for CD4⁺ T cells by primarily influencing immune responses by regulating CD8⁺ cytotoxic T lymphocytes (CTLs). The percentages and function of CD8⁺ T cells are significantly decreased in the peripheral blood of PC patients, compared to healthy controls. One contributing mechanism to this diminished anti-tumor response in PC patients is the induction and recruitment of suppressive cells by tumor-derived factors (TDF). In particular, immunosuppressive regulatory T cells (Tregs) are a subpopulation of CD4⁺ T cells that express the forkhead boxP3 (FoxP3) gene. Their main function is maintaining peripheral immune tolerance against self-antigens and foreign antigens by suppressing CD4⁺ and CD8⁺ T cell responses. The percentages of Tregs are elevated in PC in human patients as well as murine models of PC. Delineating the mechanisms by which this balance in T cells is lost is critical for the generation of effective anti-tumor immune responses in PC hosts. Alterations in transcription factors (TF) that play critical roles in the commitment and maintenance of lymphocyte development often promote malignant transformation. One such example is the Ikaros family of zinc finger TF that includes Ikaros, Aiolos, Helios, Eos and Pegasus proteins. Ikaros, Helios and Aiolos are restricted to the immune-cell lineages whereas Eos and Pegasus are found in lymphoid tissues. These TF regulate cell-fate decisions during hematopoiesis and are thus important players in the development of immune cells. In particular, Ikaros, the founding member is highly important for normal T cell development. Ikaros is regulated post-transcriptionally by alternative splicing, which produces functional and dominant-negative (DN) isoforms, which can inhibit its activity. Ikaros is also regulated by posttranslational modifications, which primarily include phosphorylation. Phosphorylation by protein kinase (Casein II) CK2 and dephosphorylation by protein phosphatase 1 (PP1) can negatively affect Ikaros’ stability, localization and function. Specifically, CK2 phosphorylation of Ikaros impairs its DNA binding ability, regulation of cell cycle progression, and its function in T cells. It also alters its subcellular localization and leads to its ubiquitin-mediated proteasomal degradation via phosphorylation in PEST sequence regions. On the contrary, dephosphorylation of Ikaros by PP1 maintains its stability and function. CK2 is a ubiquitously expressed and highly conserved serine/threonine kinase that regulates a number of critical cellular processes, including cell proliferation and apoptosis. CK2 is widely studied in blood and solid malignancies. Overexpression of its tetrameric subunits and deregulation of its activity have been linked to numerous cancers. Overexpression of CK2 in mice leads to T cell leukemia’s and lymphomas. However, limited studies have focused on CK2’s involvement in regulating immune responses. Apigenin (API) is a natural plant flavonoid and selective CK2 inhibitor that targets CK2-dependent signaling pathways. API has a number of reported biological effects including anti-proliferative, anti-oxidant, anti-inflammatory and anti-carcinogenic characteristics, which are thought to be an integral part of its anti-cancer attributes. Recently, there has been increased exploration of the use of API as a chemopreventive agent in a number of cancer models. More specifically, API has been shown to induce cell death and also enhance the anti- proliferative effects of chemotherapeutic agents in human PC cells, *in vitro*. We have previously shown that Ikaros undergoes proteasomal degradation, which may contribute to altered effector and regulatory T cell development in murine PC. Our studies suggest that a shift in the balance between CK2 and PP1, favoring CK2 activity may be responsible. Therefore, to further delineate CK2’s involvement in regulating Ikaros expression and thus T cell responses, we investigated the effects of API in our PC model. We found that API is able to stabilize Ikaros’ expression *in vitro* and *in vivo* while also restoring the balance between effector CD4⁺/CD8⁺ T cells and Tregs. This correlated with an increase in immune function as observed on splenocytes from API treated pancreatic tumor-bearing (TB-API) mice exemplified by increase in the *in vivo* production of INF-γ CD8⁺ T cells *in vivo* and by robust allogeneic CD8⁺ T cell responses, *in vitro*. This study highlights the importance of CK2 in regulating Ikaros expression and its possible influence on T cell immune responses in murine PC. # Materials and methods ## Cell line Panc02 murine pancreatic adenocarcinoma cell line was established by Corbett et al.. This cell line was maintained in complete RPMI 1640 medium supplemented with 10% Fetal Bovine Serum (FBS), (HyClone, Logan, UT), 2mM L-glutamine, 100μ/ml penicillin and 100μg/ml streptomycin (Gibco BRL, Rockville, MD) at 37°C in 5% CO₂. Cultured cells were tested and found to be negative for mycoplasma and viral contamination. ## Mice Female C57BL/6N mice (6–8 weeks) were purchased from Harlan Laboratories (Indianapolis). The Institutional Animal Care and Use Committee of the University of South Florida approved protocol T IS00000447 is in compliance with the Guide for the Care and Use of Laboratory Animals. All mice were maintained in a pathogen-free animal facility, fed and housed with other mice for 1 week before the start of *in vitro* or *in vivo* experiments. Mice were humanely euthanized using CO₂ and cervical dislocation according to the University of South Florida IACUC guidelines. ## CK2 inhibitor Apigenin (CK2 Inhibitor) (API) **(**4′,5,7-Trihydroxyflavone, 5,7-Dihydroxy-2-(4-hydroxyphenyl)-4-benzopyrone**)** was purchased from Fisher Scientific, USA and diluted in DMSO according to the manufacturer’s instructions. ## Proteasome inhibitor MG132 (proteasome inhibitor Cbz-LLL) carbobenzoxyl-L-leucyl-leucyl-L-leucine was purchased from Fisher Scientific, USA and diluted in DMSO according to manufacturer’s instructions. ## In vitro assay Control splenocytes from C57BL/6N mice were collected and co-cultured in the absence or presence of Panc02 cells, treated with and without API or MG132 for four hours at 10μM and 20μM, *in vitro*. Cells were harvested for protein lysates and western blot analysis. ## Mice Female C57BL/6N mice (6–8 weeks of age) were injected with 1.5 × 10⁵ Panc02 cells suspended in 100μl of PBS and administered subcutaneously (s.c.) in the lower, left abdomen. Mice from the control (CTRL) group were s.c. injected with sterile PBS only. Treatments of pancreatic tumor-bearing (TB) mice started immediately after the tumor onset (as evidenced by the appearance of palpable tumors). A group of TB mice received either 100μl of PBS (TB) or doses of 25mg/kg of API (TB-API) administered three times per week via intra-peritoneal injections (i.p.). Mice were monitored three times per week for weight, infection, abdominal swelling (due to ascites), impediment in locomotion, labored breathing, and any signs of discomfort (pain). At the experimental endpoint of TB and TB-API mice, from the survival or treatment studies that experienced, signs of suffering or pain, abdominal swelling due to ascites (the main cause of animal deaths), solitary tumor masses greater than 2 cm or necrotic tumors were humanely euthanized using CO₂ and cervical dislocation, according to the University of South Florida IACUC guidelines and approved protocol (T IS0000447). Tumors and spleens were harvested and weighed from all mice. In addition, spleens were processed for *in vivo* and *in vitro* biochemical experiments. ## In vitro CK2 kinase assay CK2 kinase activity was measured in splenocytes from CTRL, TB and TB-API mice using the CK2 assay kit (Millipore) according to the manufacturer’s instructions. CK2 activity was calculated by subtracting the mean counts per minute (CPM) of samples in the absence of substrate from the mean CPM of samples in the presence of substrate. ## Western blot analyses Protein lysates were prepared from splenocytes from CTRL, TB and TB-API mice. In addition, control splenocytes were co-incubated in the absence or presence of Panc02 cells treated with and without API and/or MG132. Cells were lysed with modified Radioimmunoprecipitation assay (RIPA) Buffer (Millipore) supplemented with Na₃OV₄ and protease inhibitor cocktail (Sigma- Aldrich). Protein concentrations were determined using the BCA Protein Assay Kit (Thermo Fisher Scientific). Approximately, 40 μg protein lysates were loaded and resolved using NuPAGE 4–12% Bis-Tris polyacrylamide gels (Invitrogen) and transferred to nitrocellulose membranes (Whatman). The membranes were blocked with 5% nonfat milk in PBS/0.1% Tween-20 and then probed with either anti-Ikaros (Cell Signaling), at a dilution of 1:1000, anti-CK2α (Santa Cruz Biotechnology) and anti-PP1 (Santa Cruz Biotechnology) at a dilution of 1:200. Primary antibodies were detected using their respective secondary IgG, HRP-conjugated antibodies (Jackson Immunoresearch), at a dilution of 1:10000. Secondary antibodies were identified using Super Signal West Pico and Femto Chemiluminescent Substrates (Thermo Fisher Scientific). As an internal control for equal protein loading, all blots were stripped and re-probed with anti-β- actin (Sigma-Aldrich) at a dilution of 1:20,000 or anti-GAPDH (Santa Cruz Biotechnology) at a dilution of 1:200. Membranes were either exposed to X-ray films (Phoenix) and developed using a Kodak M35-X OMAT Processor or imaged using a ChemiDoc XRS Imaging System (Bio-Rad). Band intensities were quantified using Quantity One 1-D Densitometry and Image Lab softwares (Bio-Rad). ## Flow cytometry Splenocytes from CTRL, TB and TB-API mice were lysed with red blood cell (RBC) lysis buffer (eBioscience) and counted for immunophenotyping. Cells were then suspended in 3% FBS-PBS and stained with fluorescent antibodies against murine T cell surface markers CD3 (FITC) (eBioscience), CD4 (PE-Cy7) (BD Pharmingen), CD8 (APC-H7) (BD Pharmingen) and CD25 (PE) (eBioscience). Subsequently, cells were intracellularly stained with anti-IFN-γ-PE (BD Pharmigen) after using a fixation-permeabilization kit from eBioscience according to the manufacturers protocol. Flow cytometry was performed using a BD LSRII (BD Biosciences Immunocytometry Systems) and data analyzed with FlowJo software (Tree Star Inc.). ## Allogeneic mixed lymphocyte reaction CTRL, TB and TB-API spleens and Balb/c spleens were processed into single cell suspensions, RBC lysed and counted. 4 × 10⁵/well Balb/c splenocytes (responders) were labeled with 1μM of Carboxyfluorescein diacetate succinimidyl ester (CFSE) and co-cultured with 8 × 10⁵/well irradiated (2000 rad) C57BL/6N splenocytes (stimulators) from CTRL, TB and TB-API mice. Culture wells were set-up in triplicate in a 96 well plate in a one-way allogeneic mixed- leukocyte reaction (MLR), and cultured for 4 days at 37°C. Proliferation responses of allogeneic CD8⁺ T cells from Balb/c mice assay were evaluated using flow cytometery. Cells were stained with murine anti-CD3 PerCP (BD Pharmigen), anti-CD8-APC-H7 (BD Pharmigen). The CFSE dilution profile of CFSE⁺CD3⁺CD8⁺ cells was acquired using BD LSRII flow cytometer and data analyzed with FlowJo software (Tree Star Inc.). ## Statistical analysis All *in vivo* and *in vitro* graph results described in this study are representative of the mean ± S.E.M. of at least three independent experiments analyzed with two-tailed Student’s *t* test and Kaplan–Meier survival curve using Prism 5 Software (GraphPad, San Diego, CA). Statistical significance and representative quantification of normalized densitometric ratios of western blot data was realized using the *J* software. Statistical differences were considered significant at p\< 0.05. # Results ## Apigenin prevents Ikaros downregulation, in vitro We previously published that MG132 is able to stabilize Ikaros expression *in vitro*, providing evidence that Ikaros undergoes ubiquitin proteasomal degradation. A balance between CK2 and PP1 regulates Ikaros stability and function. In particular, increased CK2 activity is thought to cause Ikaros degradation. Therefore, inhibiting CK2 should stabilize Ikaros expression and prevent its degradation, similar to MG132. We treated naïve splenocytes with the CK2 inhibitor, API, as well as MG132, both at 10μM and 20μM, to compare their effects on Ikaros expression. Both API and MG132 stabilized Ikaros expression ( Lanes 2 and 3 vs. Lane 1; Lanes 4 and 5 vs. Lane 1), respectively and displayed a significant synergistic effect, which shows accumulation of ubiquitination ladders ( Lane 6). The addition of murine Panc02 cells causes a reduction, although not significant, in Ikaros protein expression ( Lane 7) that is prevented by MG132 treatment. To determine whether API has the same activity as MG132 to prevent downregulation of Ikaros, we treated Panc02 cells and naïve splenocyte co-cultures with both 10μM and 20μM of API ( Lanes 10 and 11) and 10μM and 20μM of MG132 ( Lanes 8 and 9). Thus, our results show that API treatment also significantly prevented Panc02 reduction of Ikaros protein expression in splenocytes at the same concentration as MG132. However, adding both drugs did not result in a synergistic effect in this co-culture system ( Lane 12). Overall, these results suggest that API is able to stabilize Ikaros and prevent its downregulation in a pancreatic tumor microenvironment. Moreover, the similarities to MG132 and their additive effect, also further suggest that API may be preventing Ikaros’ proteasomal degradation, possibly via its inhibition of CK2. ## Apigenin increases survival and reduces tumor burden, in vivo API has been shown to have anti-tumor effects in a number of tumor models such as breast cancer and melanoma. To determine whether the effects of API on Ikaros *in vitro*, are also occurring *in vivo*, the impact of API treatment on survival and tumor burden was evaluated using TB mice. Treatment of TB mice with 25 mg/kg API caused an increase in their survival and a significant decrease in their tumor weight compared to vehicle-treated TB mice. Next, we evaluated whether API treatment may have any toxicity effects *in vivo* by weighing and observing all mice three times a week for the duration of the study. Results showed that there was no significant difference in the weights of API treated compared to untreated TB mice at the end of the study. A hallmark finding in TB mice is a pronounced splenomegaly, measured by a significant increase in spleen weights. We found that *in vivo* API treatment reversed this PC induced splenomegaly and caused a significant reduction in spleen weights in TB-API compared to TB mice. ## Apigenin partially stabilizes Ikaros expression, in vivo Since our *in vitro* data shows that API can stabilize Ikaros expression, especially in the presence of murine Panc02 cells, we evaluated the effect of API treatment on Ikaros protein expression in an *in vivo* pancreatic tumor microenvironment. Western blot analyses revealed that API partially restored Ikaros expression in TB-API mice compared to TB mice. More specifically, it appears that DN Ikaros isoforms, (described as less than 46 kDa), were increased in TB-API compared to TB mice. ## Apigenin inhibits CK2 activity and improves PP1 expression, in vivo We previously published that key regulators of Ikaros expression CK2 (increased activity) and PP1 (down-regulated expression) were altered in TB mice. Therefore we evaluated API’s effect on CK2 expression by western blot using an antibody specific to its catalytic alpha subunit. Splenocytes from TB-API mice showed a slight decrease in the molecular weight of CK2α compared to TB mice, similar to that seen in CTRL splenocytes. Furthermore, we observed a significant increase in CK2α expression in TB-API compared to TB mice **(**)**.** To further delineate the effect of API on CK2 in our pancreatic TB model, we evaluated CK2’s activity and found that API treatment caused a reduction in CK2 activity in TB-API, compared to TB mice. However, this inhibition was not significant. Next, we also evaluated PP1 expression (observed as doublets) in splenocytes found in CTRL mice, however the higher molecular weight isoform was absent in TB mice (**)**. In fact, the lower molecular weight PP1 isoform, which was present in splenocytes from CTRL, TB and TB-API mice was significantly increased in TB- API mice (**)**. These data strongly suggest that API is able to stabilize Ikaros expression *in vivo*, which may be mediated by its ability to inhibit CK2 activity and increase PP1 expression. ## Apigenin partially restores T cell homeostasis and immune responses, in vivo We previously published that T cell homeostasis was lost in TB mice. To determine whether API-mediated rescue of Ikaros expression had any effect in the previously observed shift in T cell numbers in TB mice, we measured effector T cells (CD4⁺, CD8⁺ and Treg) percentages in TB-API and TB mice. Flow cytometry results show that splenocytes from TB mice had significant reduction in CD4⁺ and CD8⁺ T cell percentages but an increase in Treg percentages compared to CTRL mice. However, flow cytometry results of splenocytes from TB-API mice had significantly increased CD4⁺ and CD8⁺ T cell percentages but reduced Treg percentages compared to TB mice. These results suggest that Ikaros expression may in fact influence T cell development in TB murine model of PC. Next, we asked whether API-mediated an increase in effector T cell percentages and whether the reduction in Treg percentages could impact anti-tumor immune responses in TB mice. Therefore, we performed one-way allogeneic mixed leukocyte reaction (MLR) to address this question. In the MLR assay, splenocytes from CTRL, TB and TB-API mice were used as stimulators and co-incubated with CFSE- labeled BALB/c splenocytes, which were used as responders. As expected, TB whole splenocytes were deficient in their ability to prime allogeneic CD8⁺ T cell immune responses compared to CTRL splenocytes. In contrast, TB-API whole splenocytes restored the ability to prime allogeneic responses compared to TB splenocytes. Based on this observation, we hypothesized that API treated TB mice may have a higher rate of activated CD8⁺ T cells that produce IFN-γ which is critical to their effector function in eliminating tumor cells. To address this question, we evaluated intracellular IFN-γ production of CD8⁺ T cells from splenocytes of CTRL, TB and TB-API using flow cytometry. Intracellular staining and flow cytometry analyses revealed that there were defects in CD8⁺ T cell IFN-γ production in TB mice compared to CTRL, which were significantly restored with API treatment. These findings suggest a correlation between Ikaros expression, T cell development and immune responses in a pancreatic tumor microenvironment and ultimately points to a possible involvement of CK2 as a key regulator. # Discussion Recently, Song et al reported that inhibiting CK2 restored Ikaros tumor suppressor activity in clinical samples and pre-clinical xenograft models of leukemia. Although widely studied in hematological malignancies, the role of Ikaros in solid cancers has not been fully investigated. We have previously identified the possible involvement of Ikaros in maintaining effector and regulatory T cell homeostasis in a pre-clinical PC model. Our previous published data suggested that loss of Ikaros was a result of its ubiquitin-mediated proteasomal degradation in response to increased CK2 activity versus PP1 in a PC microenvironment. In this current study, we make use of a selective CK2 inhibitor, apigenin (API), to further delineate the mechanism by which Ikaros is regulated and to provide functional evidence for its involvement in modulating T cell anti-tumor immune responses. *In vitro*, API stabilized Ikaros expression (IK1 and IK 2/3) in naïve splenocytes and prevented its downregulation in the presence of murine Panc02 cells, similar to MG132 treatment. *In vivo*, API treatment of TB mice improved survival, reduced tumor burden, reduced CK2 activity, increased PP1 expression and restored expression of some Ikaros isoforms. In addition, API treatment of TB mice increased CD4/CD8⁺ effector T cell numbers while decreasing Treg numbers compared to TB mice. Also, API treatment of TB mice restored the splenocytes’ ability to prime allogeneic CD8⁺ T cell responses in MLR. More importantly, API treatment of TB mice showed an increase in anti-tumor immune responses correlated with the increased production of intracellular IFN-γ from CD8⁺ T cells, *in vivo*. Our study sheds insight into Ikaros’ regulation of T cell immunity in PC and demonstrates evidence for a possible mechanism by which it is regulated. Regardless of the mechanism, the results of this study suggests that pharmacological CK2 inhibition restores Ikaros expression and can influence T cell immune responses in a murine PC model and other solid tumor models. Phosphorylation of Ikaros by CK2 induces Ikaros degradation while dephosphorylation by PP1 maintains its stability. *In vitro*, we found that API appeared to mimic the effects of MG132 by stabilizing Ikaros expression, causing the accumulation of its ubiquitination ladders. These data suggest that API may be similarly preventing ubiquitin-proteasomal degradation of Ikaros via its ability to inhibit CK2 activity. The combined effects of MG132 and API further provide evidence for this mechanism. As a result, our current working hypothesis is that API may be inhibiting the upstream effector of the pathway, CK2 and its ability to hyper-phosphorylate Ikaros further leading to its ubiquitination and degradation. On the contrary, we propose that MG132 works downstream of this pathway by inhibiting the proteasome. Ultimately, both inhibitors would lead to a more stable Ikaros expression. In addition, regarding our model Ikaros function is important for T cell homeostasis. Alternatively, API has also been reported to regulate proteasomal degradation. More specifically, API has been shown to potentially inhibit the chymotrypsin- like activity of the proteasome, similar to MG132. It is therefore possible that API may stabilize Ikaros expression by inhibiting both CK2 and/or proteasomal activity, which needs to be further, investigated. Furthermore, clinically available proteasomal inhibitors exhibit some toxic effects, highlighting the need for safer alternatives such as natural, non-toxic compounds like API. *In vivo*, API treatment improved survival and significantly reduced tumor weights of TB-API compared to TB mice. These findings suggest that API may have anti-tumor properties in murine PC. Although the frequency and dosage of API administered in our TB mice reduced CK2’s activity to half; this value however was not significant (p = 0.053). Therefore, in-depth pharmacokinetics and dose- dependent studies need to be done to determine a more effective dosage of API treatment for targeting CK2’s expression and/or activity in TB mice. Western blot analyses of CK2α showed an increase in its expression in splenocytes from TB-API compared to TB mice. However, CK2 expression in splenocytes of TB-API mice was accompanied by a reduction in MW, similar to that observed in splenocytes from CTRL mice. This suggests that API treatment may be inhibiting a posttranslational modification event of CK2. Phosphorylation of CK2 by kinases increases its activity. Therefore, this observation opens the intriguing possibility that API may reduce the activity of CK2 rather than the level of CK2. It will be crucial to evaluate other kinases such as ERK, and CDK1-cyclinB1, which may be responsible for modulating CK2’s activity in TB mice. Furthermore, western blot analyses from splenocytes showed a higher MW of PP1 isoform (demonstrated PP1 expression as doublets) found in CTRL and TB-API mice but absent in TB mice. PP1 protein expression was down-regulated but not significant (p = 0.0752) in TB mice compared to CTRL mice. API was able to significantly increase PP1 expression in TB mice with the restoration of the upper MW PP1 isoform (appearance of doublets) that was originally observed in CTRL mice. This data suggest that PC negatively impacts the phosphatase activity, which may be specific to only one of the several isoforms of PP1. Therefore it will be equally crucial to determine which PP1 isoform(s) activity is lost as a result of transcriptional or posttranslational modification events in PC microenvironments. In addition, it would be paramount to determine which PP1 isoform(s) responds to API treatment in TB (PC) mice. API treatment appeared to increase the expression of Ikaros isoforms *in vivo*. Ikaros expression is critical for T cell immune balance. We previously published that full-length Ikaros isoforms (IK-1 and IK-2/3) in enriched CD3⁺ T cells were degraded in TB models. Typically, the overexpression of DN isoforms is known to inhibit the activity of Ikaros and it is associated with T cell malignancies. However, in this study we observed less degradation of Ikaros isoforms from splenocytes of TB-API compared to TB mice. Therefore, the identification of tumor-specific full-length and/or DN Ikaros isoforms and their functional impact on immune regulation and PC progression are warranted and are currently being investigated. We observed that API caused a decrease of CK2 activity in splenocytes from TB-API mice but there was no significant difference when compared to TB mice. However, it is possible that an increased dosage or more frequent treatments with API, may lead to a significant decrease in CK2 activity, which could lead to the up-regulation and the stability of more full- length Ikaros isoforms and consequently an increase in anti-tumor immune responses. API treatment significantly increased CD4⁺ and CD8⁺ T cells but decreased Tregs percentages in TB mice. Our results showed functional evidence that API modulates immune responses in our TB mice since API treatment increased IFN-γ production of CD8⁺ T cells. This is an indication of restored CD8⁺ T cells’ activity and cytotoxic function. In addition, our data show that API also significantly increased the ability of antigen- presenting cells (APCs) to prime allogeneic CD8⁺ T cell immune responses. In a one-way MLR, allogeneic BALB/c CD8⁺ T cell responses were stimulated by APCs, from CTRL, TB and TB-API mice. Dendritic cells (DC) are the most potent APCs. Their function is often evaluated by their ability to induce proliferation of allogeneic T cells in MLR assays. Therefore, the ability of TB-API splenocytes to effectively stimulate allogeneic CD8⁺ T cell proliferation may be a result of API’s effects on DC function, which has previously been reported. API’s effects on DC function may be a result of its reduction of Treg percentages, which can inhibit DC function and T cell immune responses. However, these Treg percentages were not fully restored to those of CTRL mice, which may also explain why allogeneic CD8⁺ T cell proliferation was not fully restored to CTRL levels. In addition, we have previously published that other immunosuppressive cells such as myeloid derived suppressor cells (MDSC) are expanded in TB mice. MDSC are immature macrophages, dendritic cells and granulocytes. API reduction in MDSC percentages may be a result of maturation or differentiation of these immature cells, thus producing mature DC, macrophages and other APCs, which could also account for the increased allogeneic immune responses. Our unpublished findings suggest that API reduces MDSC percentages, which may also account for the increased proliferation of allogeneic T cells from TB-API splenocytes in MLR assay (*unpublished* Ghansah et al., 2017). Overall, our results with API provide evidence that Ikaros may be specifically involved in regulating T cell immune responses in TB (PC) model. In conclusion, this study highlights the importance of CK2 in regulating Ikaros expression and T cell immune responses in a solid pancreatic tumor microenvironment. Our results suggest that the natural flavonoid, API, may be therapeutically beneficial in stabilizing Ikaros expression via regulation of CK2 activity, thus restoring T cell homeostasis and enhancing anti-tumor immune responses in pancreatic cancer. We would like to thank Drs. Shengyan Xiang and Xiaohong Zhang for their guidance regarding proteasomal degradation experiments. [^1]: The authors have declared that no competing interests exist. [^2]: **Conceptualization:** TG NN. **Data curation:** CI IR GJ ON. **Formal analysis:** NN KS CI IR AM GJ ON TG. **Funding acquisition:** TG. **Investigation:** NN KS CI IR AM GJ ON TG. **Methodology:** TG NN. **Project administration:** TG. **Resources:** KS. **Software:** KS. **Supervision:** TG. **Validation:** TG. **Visualization:** NN TG. **Writing – original draft:** NN. **Writing – review & editing:** NN TG.

# Introduction Coronaviruses (CoVs) are found in a wide variety of animals in which they can cause respiratory, enteric, hepatic, and neurological disease of varying severity. In humans, CoVs cause a large fraction of common colds, as well as more rare, but serious disease such as the outbreak of Severe Acute Respiratory Syndrome (SARS, caused by SARS-CoV) in 2003, and the recent Middle East Respiratory Syndrome (MERS, caused by MERS-CoV). Both SARS and MERS are zoonotic diseases as they emerged following spill-over from animal reservoirs. CoVs belong to the family *Coronaviridae* and are divided into four genera or groups: *Alpha*-, *Beta*-, *Gamma*- and *Deltacoronavirus*. SARS and MERS are *Betacoronaviruses*; current phylogenetic evidence suggest that bats are the genetic source of *alpha*- and *betacoronaviruses* which include human and most other mammalian specific viruses. In contrast, *gamma*- and *deltacoronaviruses* are dominated by viruses found in birds. The most prominent representatives of *gammacoronavirus* are infectious bronchitis virus (IBV) and Turkey CoV, which cause substantial economic losses to the poultry industry. Avian coronaviruses, most of which are distinct from IBV and Turkey CoV, circulate in wild birds, but the taxonomic status of “avian coronavirus” is contentious. Regardless, avian gammacoronavirus has been detected across an array of avian orders including *Anseriformes*, *Ciconiiformes*, *Charadriiformes*, *Columbiformes*, *Galliformes*, *Passeriformes*, *Pelecaniformes*, and *Psittaciformes*. Members of the *deltacoronvirus* group are also composed of viruses detected in wild birds, which suggests that wild birds play an important role in the epidemiology of *gamma*- and *deltacoronaviruses*. Despite this assertion, limited studies have been performed to assess host species breadth, seasonal prevalence, geographic distribution, or genetic structuring. In this study we aimed to sample a range of waterbirds from Scandinavia for the presence of CoV. For this, we utilized samples collected from wild birds at Ottenby Bird Observatory (56°12′N, 16°24′E), which is located on Öland, an island in the Baltic Sea off the south-east coast of Sweden. It is an important stop over site for migrating birds, including *Anseriiformes* and *Charadriiformes* migrating between breeding areas in Scandinavia, the Baltic Countries, Finland, Russia and more southern wintering grounds. Furthermore, CoV have been detected and characterized from Mallards (*Anas platyrhynchos*) at this study site previously, allowing us to compare viruses circulating in both 2007 and 2011. Finally, we assessed interspecies transmission and geographic subdivision of avian CoV in wild birds. # Materials and Methods ## Sample collection All trapping and sampling was carried out in accordance with regulations provided by the Swedish Board of Agriculture. The capture and sampling protocols were approved by the Swedish Animal Research Ethics Board ‘‘Linköpings djurförsöksetiska nämnd”, reference number 8–06. All trapping and sampling was conduced by trained ornithologists, and all birds were released following sample collection. All waterfowl species in this study were captured using a large baited trap. Shorebirds and gulls were captured using standard trapping techniques, including the use of walk-in shorebird traps and nets. Fecal and/or cloacal samples were collected using a sterile tipped applicator and were placed in Virus Transport Media (Hanks’ balanced salt solution containing 0.5% lactalbumin, 10% glycerol, 200 U/ml penicillin, 200 mg/ml streptomycin, 100 U/ml polymyxin B sulfate, 250 mg/ml gentamycin and 50 U/ml nystatin; Sigma); and stored at -80C within 4–6 hours of collection. ## Screening and characterization of CoV Viral RNA was extracted using the MagAttract Viral RNA M48 extraction kit and the M48 Robot (Qiagen, Hilden, Germany). A previously published pan-Cov Real- Time Reverse Transcriptase PCR (QRT-PCR) using the iScript Kit (BioRad, Hercules, USA) was utilized with IBV as a positive control, with a cutoff threshold of 39. Sequences of the RNA-dependent RNA polymerase (RdRp) were generated using methods described in or, and sequences generated in this study have been deposited in GenBank under accession numbers KT882615-28. Resulting sequences were aligned using the MAFFT algorithm implemented in Geneious R7 (Biomatters, New Zealand). The nucleotide substitution model was determined in MEGA 5.2 and maximum likelihood trees were constructed using PhyML with aLRT branch support in SeaView. Trees were projected using FigTree v1.4. # Results ## High prevalence of coronavirus in *Anseriformes* A total of 764 birds from 11 species of *Anserifomes* (ducks, geese, swans, n = 691) and 11 species of *Charadriiformes* (gulls, terns, shorebirds, n = 143) were sampled in this study. Overall, the prevalence of CoV was high, at 18.7%, but species, groups and orders were unevenly represented. Diving ducks had the highest prevalence (39%), driven by high prevalence in Greater Scaup (*Aythya marila*; 51.5%), however the sample size was small (n = 37). Dabbling ducks of the genus *Anas* also had high prevalence; Mallard, in particular, had a prevalence of 19.2%. While *Anseriformes* had the highest prevalence, in the order *Charadriiformes*, CoV was detected in a Black-headed Gull (*Chroicocephalus ridibundus*), but absent in tern and wader species sampled. Eleven sequenced Mallard CoV and three sequenced Scaup CoV were *gammacoronaviruses*. ## Putative maintenance, despite no global spatial, temporal, host species patterns A global phylogeny of all sequenced avian *gammacoronaviruses* (not IBV-like) RdRp show no spatial, temporal or host species differentiation between clades. That is, all clades contain sequences from multiple geographic locations, different years of sampling, and different host species (and Figs). Indeed, viruses sequenced in this study were placed in a clade containing sequences from Hong Kong, China, Beringia, the United States of America and Sweden. However, despite limited global patterns of clade differentiation, 8/10 sequences from Mallards in 2007 were most closely related (within 98–99% identity) to CoV sequences from Mallards at Ottenby in 2011, suggesting local circulation of these RdRp. Mallard CoV sequence 69998 had the highest pairwise identity to viruses from Hong Kong, but is nested in the broad and highly similar clade containing the largest number of sequences from Ottenby in 2011. A single Mallard CoV sequence (69740) was in a differentiated clade, most similar to sequences from Hong Kong and more distantly China. Sequences from Scaup CoV were not similar to Mallard CoV; Scaup 67699 was similar to sequences from Beringia and China (within 98% identity;). Two Scaup CoV sequences (67693 and 67703) were highly similar to only one sequence in Genbank, which was collected from a Tufted Duck (*Aythya fuligula*) in Hong Kong (JN788847). All other sequences were less than 92% similar, indicating a differentiated lineage that is poorly sampled. # Discussion In this study we aimed to further contribute to the natural history of avian coronaviruses by screening an array of waterbirds for these viruses. We found a prevalence of 18.7% CoV, which is higher than the 0–15% reported previously in wild bird studies. This may be due to the high proportion of dabbling ducks, particularly Mallard, in our study, and the temporal and spatial features of the dabbling duck sampling. More specifically, sampling of dabbling ducks occurred at a migratory stopover site, and thus high prevalence could be driven by migrants which not only import viruses through infected individuals, but also replenish the pool of susceptible individuals across the migratory period, resulting in high prevalence at these locations. Directly comparing prevalence estimates to previous studies, however, is challenging due to the array of different methods utilized, ranging from conventional PCR to multiplex real-time PCR, in addition to species distributions and seasonal variations. Regardless, this study corroborates previous studies suggesting the importance of dabbling ducks (*Anas* sp) as hosts of avian coronaviruses, but also highlighting the importance of diving ducks. The high diving duck prevalence in this study was driven by high prevalence in Greater Scaup, which could be due to congregation in very large flocks facilitating transmission, but could also be due to other factors such as a local variation in disease prevalence or reflect outbreak dynamics wherein the virus rapidly expands across the susceptible population. Overall, diving ducks have only been sampled in this study and by Chu *et al*. (2011), so it is unclear whether this pattern of high prevalence is present globally. We found a low prevalence in waders, which is in contrast to and, which found a very high prevalence (\>20%) in waders, but corroborates findings in. This may be due to differences in congregation and feeding patterns of waders between the locations, but has to be interpreted with caution, as numbers of waders sampled in our study, and others, were small. Finally, while assessment of gulls is limited, this study corroborates findings with Muradrasoli et al (2010) indicating that this group requires further scrutiny. In order to better assess CoV dynamics, resampling the same site across time is imperative, and this is the first study to do so. Coronaviruses from Mallards were previously assessed using samples collected at Ottenby in 2011 and we detected high similarity between viruses from 2007 and those from 2011. Indeed, 8/10 CoV were most closely related to sequences from 2011, which strongly suggests maintenance and/or location transmission of these RdRp lineages at this location. This is unlikely to be a coincidence as Swedish Mallard CoV make up less than 1/3 of those available sequences, and all current RdRp clades are comprised of sequences from Asia. The waterfowl host appears important in these samples, however assessing host species biases is challenging due to the overwhelming number of sequences from *Anas* species. However, coronaviruses from diving ducks were rather different from existing diversity as demonstrated by the similarity between two of three Scaup CoV sequences from this study and Tufted Duck CoV sequence from Hong Kong. The limited number of sequences in this clade could be due to sampling shortcomings, or that currently developed primers do not amplify this clade effectively. Despite few studies, small samples sizes and differences in prevalence, what is clear, is that in the Northern Hemisphere waterfowl species, especially dabbling and diving ducks are important in the epidemiology of avian CoVs. It is interesting to note that these patterns are very similar to those found in low pathogenic influenza A viruses: high prevalence in waterfowl and gulls in the Northern Hemisphere, and little host species and temporal structuring within waterfowl derived viruses in the conserved polymerase genes (such as PB2, PB1). Further, detection in fecal or cloacal samples suggest these viruses may also replicate in the gastrointestinal tract, which is the true for Turkey Coronavirus. As demonstrated by years of influenza sampling, to better understand the eitiology, epidemiology and ecology of these viruses more systematic surveillance needs to be undertaken and more viruses need to be sequenced, particularly full genomes. Given the importance of coronaviruses as human pathogens, as exemplified by SARS and MERS-CoV and the potential for wild birds as reservoirs, spreaders and mixing vessels, further studies of coronaviruses in wild birds are warranted. # Supporting Information We wish to thank and the duck trappers at Ottenby Bird Observatory and Jonas Waldenström for collecting and providing samples used in this study, Jonas Blomberg for kindly providing lab facilities, and Erik Salaneck for his contribution. This is article 290 from Ottenby Bird Observatory. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: MW SM AN JDJ. Performed the experiments: MW SM AN JDJ. Analyzed the data: MW SM AN JDJ. Contributed reagents/materials/analysis tools: SM JDJ. Wrote the paper: MW SM AN JDJ.

# Introduction Recurrent seasonality in disease incidence is common in infectious diseases. Generally, incidence of bacterial enteric pathogens is greater during warmer months in both human and animal populations, but respiratory and viral enteric pathogens incidence peaks during colder months. Seasonality is caused by various mechanisms. For example, environmental factors can influence pathogen abundance, survival and virulence or influence host susceptibility, and changes in host behavior and aggregation in different seasons can alter contact patterns, leading to different transmission dynamics. Investigating mechanisms underlying disease seasonality is challenging because different mechanisms often interact in multiple causal pathways. In infectious disease transmission models, seasonality is often included phenomenologically (e.g. sinusoidal function) to evaluate its implications on disease dynamics, but not the causes leading to seasonality fluctuations. Hence, mechanistic, system-based methods are necessary to identify and quantify the contribution of different transmission drivers and causal pathways involved in seasonality of disease incidence. Despite the need for such approaches to address environmental drivers of disease, the application of system-based methodologies such as agent-based modeling (ABM) has been limited. Shiga-toxigenic *Escherichia coli* (STEC) are human zoonotic enteric pathogens that can cause severe illnesses including hemorrhagic colitis and hemolytic uremic syndrome. Healthy cattle and their environment are the main reservoirs for STEC. For humans, the most common source of exposure and subsequent infection is food and water contaminated by cattle feces containing STEC. Understanding the biological and environmental factors leading to the persistence and transmission of STEC among cattle populations is necessary to design and improve control strategies. One of the most consistent patterns observed in STEC epidemiology is the strong seasonal pattern of fecal STEC shedding and increased prevalence during the warmer months in all cattle production systems, including grazing systems. Multiple potential mechanisms have been proposed, but their relevance in contributing to the seasonal variations in STEC prevalence remain inconclusive (Alam and Zurek, 2004. Proposed mechanisms include enhanced transmission mediated by water and flies at higher temperature, and increased susceptibility linked to hormonal fluctuations associated with changes in day length. Overall, evidence is limited for these proposed mechanisms. One possible mechanism that has received less attention is behavior changes in cattle in response to temperature fluctuations. There are several pathways of STEC transmission in cattle, including both direct host-to-host transmission and indirect environment-to-host transmission. Mutual grooming, aerosols, and contact with excretions (feces, urine) on the bodies of other individuals are important sources of direct transmission. Consumption of contaminated water and food are potential sources of indirect transmission. As temperature increases, cattle spend less time grazing and more time resting under the shade as a group to seek relief from heat stress. In addition to escaping heat from solar radiation, feeding is reduced to decrease the heat associated with feed fermentation, with thresholds for reductions in feeding activity reported between 25 and 30°C. In addition, cattle tend to consume more water as temperature increases. For example, temperature increasing from 10°C to 32°C leads to 2.5 times more water consumption by calves. This tends to occur via drinking larger volumes of water during each drinking, but not necessarily through more frequent drinking (MS, personal observation). These behavior changes are highly interrelated and can shape cattle exposure to pathogens and subsequent disease transmission. Computational models are important tools to study complicated and dynamic systems. Agent-based models (ABM) provides a flexible framework to investigate relationships between observed patterns and hypothesized mechanisms underlying these patterns in complex ecological and epidemiological systems. Agent-based models are particularly useful in linking transmission drivers and pathways with epidemiological patterns. Furthermore, spatially explicit ABMs allow for the explicit consideration of spatial and temporal heterogeneity in host behavior as well as pathogen distribution in the environment, thus unifying the epidemiological triad: host, pathogen, and environment. Therefore, to investigate mechanisms underlying seasonality in STEC incidence in grazing cattle, we developed a spatially explicit ABM for the transmission of STEC that incorporates varying mechanisms linking temperature and STEC transmission. In this study, we aimed to evaluate how changes in animal behavior in response to temperature may influence the transmission patterns and the prevalence of the pathogen in the population of cattle in a pasture. In particular, we used the model to investigate two mechanisms by which temperature is hypothesized to influence transmission of STEC, including 1) by affecting the relative amount of time spent engaged in different activities (i.e., grazing vs. resting), and 2) by influencing the volume of water consumed by cattle. # Materials and methods ## Model overview We constructed a stochastic, spatially explicit agent-based model (ABM) to simulate transmission of STEC among grazing cattle. The model was written and executed in NetLogo 5.3.1, an open-source agent-based modeling software. The model scope was a group of grazing beef cattle in an intensively managed pasture. A detailed model description in accordance with standard ODD (Overview, Design concepts, Details) protocol for individual- and agent-based models is provided in. We provide a brief overview below. The purpose of this model was to quantify how temperature fluctuation caused changes in STEC incidence among grazing cattle by influencing diurnal behavior patterns. The model represented a 20-acre typical pasture consisting of patches (3.6 m² (i.e., 1.9 x 1.9 m)), including 19 acres of a 80%/20% mixture of edible grass and inedible weeds, a 1-acre large pond with a constant depth of 0.5 m, and 5 trees that each provided a 4–patch radius (R = 7.6 m) of shade. See supplemental information for a graphical representation of the model environment. The model simulated a closed cattle population (N = 25) as it engaged in different distinct activities throughout a model day (grazing, resting, drinking, sleeping). How cattle participated in these activities was influenced by the social state of an individual (dominant or subordinate), air temperature, and in the case of grazing, grass presence and length. In the first case, a single dominant individual influenced the movements of subordinate individuals during drinking and resting behaviors. In the second case, air temperature was included as an input variable (supplied via an external data file), and explicitly influenced aspects of cattle behaviors. In particular, the amount of time spent grazing versus resting was reduced, and the volume of drinking was increased with increased temperature. A temperature threshold also determined resting behavior, with temperatures above the threshold resulting in resting in groups under trees, while temperatures below the threshold resulted in cattle resting in the open field. Finally, grass grew at a constant rate over the course of the simulation to a maximum height, and was reduced in length by grazing to a minimum height, at which point it could not be grazed until it regrew. All major cattle activities recurred on an hourly schedule that repeated each model day. Sub-models governed stochastic cattle movements and transmission dynamics occurring during these activities on a 10-minute time-step. Simple rules governed daily animal activities and movements to generate realistic patterns of animal aggregation and fecal-pat distribution in the model environment. Some rules were derived from direct field observation of grazing cattle at the East Tennessee Research and Education Center—Blount Unit in August 2013 while others were based on existing literature on the topic. The schedule of the model, including the sequence of actions and their corresponding sub- models, is shown in and described in greater detail in. Superimposed on cattle activity was a Susceptible-Exposed-Infected-Recovered (SEIR)-type transmission model that simulated the transmission dynamics of STEC between cattle, and between cattle and the environment. Cattle could take one of 4 epidemiological states, including susceptible, colonized and in a latent period (without shedding), colonized and shedding, and partially susceptible (after becoming colonized once). Colonized cattle shed STEC in their feces, and susceptible cattle could become colonized through the daily accumulation of colony forming units (CFU’s) of STEC via direct contact with colonized individuals, through eating contaminated graze, or through drinking contaminated water. To gain STEC through direct contact, susceptible cattle needed to come within a proximity threshold (*ddt****)*** of a colonized, shedding individual, at which point a quantity of colony forming units (CFU) of STEC was randomly drawn from a Poisson-lognormal distribution. Shedding individuals were assumed to shed a constant amount of CFU’s per fecal pat over the entire infectious period. To gain STEC through drinking, cattle had to drink from a water patch contaminated with infectious feces. The CFU’s up-taken during a drinking session was proportional to the concentration of STEC in the water patch (total CFU’s deposited/volume of patch, assuming homogenous distribution in patch), and the volume of water consumed. The volume of water consumed was based on a non- linear, temperature-dependent function derived from data presented by Parish and Rhinehart (2008): $$Liters = 33.51213 - 0.74978*avg\ daily\ temp + 0.05806{*avg\ daily\ temp}^{2}.$$ In this function, average daily temperature (calculated as the average of daily maximum and minimum) determined the total expected liters consumed per day, which was distributed over the total minutes cattle were expected to drink, adjusted for travel time to the lake. To gain STEC though grazing, cattle had to graze from a patch contaminated with infectious feces. The CFU’s up-taken during a grazing event was proportional to both the amount of CFU’s in the patch, and the amount of grass eaten: $$STECexposed = Patch\ CFUs*\left( {grass\ units\ consumed} \middle| {pregraze\ units\ available} \right)*p_{GrazeInfect.}$$ Populations of STEC shed in feces into water and onto grass were modeled as CFU’s per patch. STEC dynamics in the environment can vary by both environmental factors like temperature and environmental substrate, and these differences may contribute to environmental transmission dynamics. STEC are known to decay in the environment as a factor of increasing temperature. Therefore, we modeled the CFU per patch to decay according to temperature, but at a substrate- dependent rate. In particular, we assumed that STEC decayed faster in water in an agricultural setting than in fecal-pats due to greater competition from other microbial organisms in water in agricultural settings. Decay was modeled in both substrates followed a common Q₁₀ function: $$k_{T} = k_{r}{Q_{10}}^{\frac{T - Tr}{10}}$$ where k_T was the bacterial decay rate at given temperature *T* (°C), *k*_*r* was the bacterial decay rate at the reference temperature (*Tr*), and Q₁₀ was the temperature coefficient that gave the rate of change for each temperature increase of 10°C. Although there is evidence that suggests that a short period of growth can occur in STEC in various substrates following deposition, the accumulated CFUs may be relative insensitive to this initial growth because of the prolonged decay, and thus was ignored here. The probability of colonization (*p*_*co***l**) was based on daily accumulated CFU’s and was calculated using a re-arranged form of the Hill-1 dose-response equation presented in: $$p_{col} = \frac{1}{1 + {K/{CFU}}}$$ in which *K* = median infectious dose of the population, and *CFU* was total CFU’s accumulated over the day. Whether or not colonization occurred was determined by drawing a random value from a uniform distribution between 0–1, and assessing whether the value fell below (colonized), or above (not-colonized) p_col. The source of colonization (direct, water, graze) was assigned based on the category contributing the majority of CFU’s. In addition, the infectious individual responsible for direct transmission or that excreted the cow pat resulting in an indirect colonization was noted during the infectious period of the initially infected individual in order to calculate the basic reproduction number (R₀), which is the average secondary infections produced by a single infectious individual in an otherwise susceptible population. ## Sensitivity analyses and calibration To characterize the parameter space of the model and to assess the relative influence of different parameters on dsease dynamics, a two-stage sensitivity analysis was conducted. Both used the count of incident cases at the end of each simulation as an output. First, a local sensitivity analysis was carried out for a subset of parameters (18; 12 deterministic, 6 probabilistic using either mean values from literature sources (when available) or assumed values as starting values. In this method, the effect of perturbations to each parameter was assessed individually by holding all parameters at their starting values except for the test parameter, which was set to be either greater than or less than its mean or assumed value, with the range tested varying by parameter. In these analyses, the effect of variable shedding rates between cattle was incorporated by randomly sampling a normal distribution with mean equal to the mean *C* value and variable standard deviation. Based on data presented in, the effect of variable shedding within cattle was accomplished by assuming that cattle shed the most upon initially becoming infected (i.e., the selected *C* value) and applying an exponential decay function with a variable daily decay rate. All model parameterizations were run under 4 constant temperatures, including cool (20°C) and warm (30°C) temperatures, at *T*_*thr* (24°C), and marginally above *T*_*thr* (25°C) to differentiate influences on model output due to absolute temperature rather than differences in temperature threshold-dependent Rest behavior. At each selected temperature, each parameter set (37; mean conditions + 36 sets with 1 permuted parameter) was run 100 times for a total of 3700 simulations per temperature. The numbers of incident cases over the simulation period were evaluated for parameter sensitivity, with increases in total cases by at least 100% or decreases by at least 50% indicating a potentially sensitive parameter. From this local analysis, 7 parameters identified as “sensitive” (shown later in results section) were included in a global Latin Hypercube Sampling (LHS)-based sensitivity analysis. Parameters were randomly sampled in each 0.1% of their parameter space, resulting in 1000 unique parameter sets. Model runs were completed for each parameter set under the same set of 4 constant temperatures as previous (20°C, 24°C, 25°C, and 30°C). Then, partial rank correlation coefficients (PRCC) and corresponding 95% confidence intervals (via 100 bootstrapped samples) were computed for each parameter in each temperature set using the package sensitivity in *R*. To calibrate the model, we used a method similar to the “best fit” method suggested by Railsback and Grimm (2012), in which model outputs were calibrated against aggregate estimations of winter (95% CI: 1.50–9.49%) and summer (95% C: 7.98–16.25%) STEC prevalence reported for pastured beef cattle in the United States. This proceeded by selecting the subset of LHS model runs (at 20°C) with a prevalence over the 60 day run that fell within the range (1–2 new colonizations) of reported winter prevalence. Because the model population was small (25), and all model runs resulted in at least 1 infection (the starting individual), prevalence at the end of the run was assessed as: the number of incident cases / 24. The median of these parameter values was then used as the parameter set of a series of simulations run at 20°C, but with incremental changes (0.01–0.04 in increments of 0.005, 100 simulations apiece) to the value of *P*_{*GrazeInfect*}. This parameter was shown by PRCC to have a relatively strong individual influence on incidence at 20°C, to have relatively low influence at 30°C, and has considerable uncertainty, making it a desirable candidate for calibration. From these simulations, a value of *P*_{*GrazeInfect*} was selected that resulted in the average prevalence (of the *P*_{*grazeinfect*} level) at the end of the model run falling in the calibration range, and near to the mean value presented in Ekong et al. (2015). Next, model simulations were run at 30°C using the resulting parameter set, but in which the value of the distance threshold at which direct transmission was possible (*ddt*) was incrementally adjusted (0.2–0.5, in increments of 0.05, 100 simulations apiece). Similar to *P*_{*GrazeInfect*}, *ddt* was chosen as a variable to manually calibrate because PRCC analysis showed that it was relatively influential at 30°C, relatively non-influential at 20°C, and to have high uncertainty. As before, a value of *ddt* was selected from model runs that resulted in the average prevalence (of the *ddt* level) at the end of the model run falling in the calibration range, and similar to the mean value presented in Ekong et al. (2015) for summer STEC prevalence in the United States. Lastly, 1000 simulations were run using the final calibrated set at both 20°C and 30°C to verify that simulation outputs fell into the expected ranges. Lastly, to gauge the influence of several simplifying assumptions made in the calibrated model, we conducted a second LHC-type sensitivity analysis in which aspects of the physical environmental structure and cattle population were varied. These included the number of trees, the radius of the shade patch, the size, arrangement, and location of the water source, the proportion of weeds to grass, and the number of cattle per simulation. See for a description of this analysis and detailed results. All parameter values used in model simulation, as well as ranges tested during sensitivity analyses used for calibrated are shown in. All analyses were conducted using program *R* with various packages. Simulations were prepared using the RNetLogo package, and run on either a desktop computer or the North Carolina State University High Performance Computing (NCSU-HPC) cluster. The model can be found in and its associated temperature files and an example R script to run it can be found in. ## Factorial analysis We evaluated our hypotheses regarding how temperature influences STEC dynamics using a fully factorial design that compared simulation results from scenarios assuming three different seasonal temperature conditions, including spring (beginning cool with a warmer end), summer (warm throughout), and fall (beginning warm with a cooler end) conditions. Within each temperature condition, assumptions of temperature-dependent and temperature-independent cattle behaviors, including daily resting behavior conditions (3 states) and drinking behaviors conditions (2 states) were systematically varied for a total of 6 factorial combinations. Resting behavior states included 1) always exhibiting ≤ T_thr Rest/Graze behavior (“Rest Cool condition”), 2) always \> T_thr behavior (“Rest Warm condition”), and 3) exhibiting temperature dependent behavior (“Rest Dep condition”). Drinking behavior states included 1) temperature-dependent (“Temp Dep condition”) or 2) constant (“Temp Indep condition”), assuming drinking rates at 20°C. For each factorial combination, historic temperature in each of 10 years (2002–2011) were used to run 100 simulations for two spring months (April and May), two summer months (June and July), and two fall months (October and November) of that year, with a total of 18000 simulations, each of 60 days apiece. During this 10-year period, summer temperatures were relatively stable, with an overall mean of 25.9 ± 1.2°C, and mean highs and lows during summer were 31.3 ± 1.5°C and 20.4 ± 0.9°C, respectively. Spring temperatures increased over the course of the 60-day model run, with a mean temperature of 17.9± 0.9°C, an average high of 23.9± 1.01°C, and an average low of 11.8 ± 0.9°C. Fall temperatures decreased over the course of the 60 day model run, with a mean temperature of 12.8± 1.1°C, an average high of 18.8± 1.2°C and an average low of 6.7± 1.3°C. Model outputs from each simulation included the count of incident cases over the duration of the simulation (and prevalence over the period) and on a daily basis the relative proportion of incident cases originating from each transmission pathway of colonization (i.e. direct, indirect water, indirect graze), and *R*_*0*. Because colonization pathways are generally unknown in real- world systems, non-source specific incident cases over the duration of the simulation were the primary output of interest for analysis. However, the explicit accounting of colonization pathways in the model enabled us to relate general patterns of incident cases with the pathways driving those patterns. Initial exploration showed a high proportion of simulations resulted in zero new colonizations, depending on the model conditions. To understand drivers of both epidemics occurring and their extent, we modeled incident cases in two stages using an approach similar to that of a zero-altered generalized linear model (ZAGLM) (otherwise known as a hurdle model) with the glmmADMB package in *R*. In the first of this two-part approach, the probability of 0 new colonizations was modeled with a binomial distribution. In the second part, incident cases \> 0 were modeled with a zero-truncated negative binomial distribution. This approach acknowledged that zero new colonizations could emerge as a result of both underlying system stochasticity and model conditions (with particular probability), but unlike observed systems in which false zeros occurred and in which zero-inflated approach would be appropriate, all colonizations were captured by the model. For both the binomial and count models, the β and ϒ vectors were the same, with the linear predictor structure: $$\eta\left( Y_{i} \right) = \mu + R + D + \left( {R*T} \right) + \left( {D*T} \right) + ᴇ$$ In this equation, the link function *η(Yi)* is either the logit link ($\frac{1}{1 + e^{- \pi}}$) of the binomial distribution (giving the probability of 0 colonizations in year *i*), or the log link (ln *π*) of the negative binomial distribution (giving the actual count of colonized individuals in year *i)*; μ was the mean of the binomial or zero-truncated negative binomial distribution; R, D, and T were the fixed effects of rest behavior, drinking behavior and temperature, respectively; R\*T and D\*T were interactive terms between drinking and rest terms and temperature, respectively, and *E* represented independently distributed error. A comparison to a similar mixed model structure that included year as a random intercept term found the two model structures did not significantly differ in their fit (according to a Likelihood Ratio Test, data not shown), and that the non-mixed models had a lower AIC value. Therefore, the non-mixed model was selected for analysis here. Other model outputs, including incident cases from direct and indirect pathways, and *R*_*0* were summarized in relation to the incidence model. Lastly, an explicit comparison was made of model outputs across seasons using the temperature-dependent parameterizations of both resting and drinking behavior. # Results ## Sensitivity analysis The LSA procedure identified 7 particularly sensitive variables for inclusion in the LHS, including the distance threshold at which direct transmission was possible (*ddt*), the probability of ingesting STEC via grazing a contaminated patch (*p*_{*GrazeInfect*}), the recovery time following colonization (**γ**), the mean of the Poisson-lognormal distribution sampled to determine the quantity of STEC transferred during direct transmission events (*pln*_*mean*), the concentration of STEC in a new fecal-pat (*C*), the concentration of *STEC* expected to infect 50% of exposed individuals (*K*), and the factor (*SI*_*mult*) by which *K* was multiplied to simulate partial immunity for previously infected individuals to secondary colonizations. Boot-strapped estimates of PRCC’s at each temperature set showed very low bias (\<0.01) for each variable (indicating high stability of estimates), and the 95% CI’s of all variables except *SI*_*mult* did not include 0, indicating statistically significant correlations. PRCC estimates demonstrated that the sensitivity of incident cases to changes in parameters generally depended on the temperature. At the cooler temperature sets (20°C, 24°C), *C* and *p*_{*GrazeInfect*} were strongly positively correlated with counts of incident cases while only weakly correlated at warmer temperatures (25°C, 30°C). Inversely, *ddt* and *pln*_*mean* were strongly positively correlated with counts of incident cases at warmer temperatures while only weakly correlated at cooler temperatures. Lastly, sensitivity to some parameters was independent of temperature, including the strongly negatively correlated *K*, and the weakly correlated parameters *SI*_*mult* and **γ**. Differences in PRCC between simulations at 20° and 24° C and between simulations at 25°C and 30°C simulations were minimal, indicating that sensitivity to parameters was more strongly influenced by model behavior determined by the 24°C temperature-threshold than continuous changes in temperature. The secondary LHS sensitivity analysis found grass-to-weed ratio to have a significant, but weak positive correlations (PRCC ≤ \|0.3\|) with incident case counts at both 20° and 30°C. Meanwhile, the total number of cattle in the simulation was moderately positively correlated with incident case count (PRCC = 0.44) at 30°C, and more weakly correlated (PRCC = 0.34) at 20°C. In addition, although the size of the lake did not have significant influence on incident cases, there were consistently more incident cases if the lake was positioned at the corners of the rectangular pasture than if the lake was positioned at the sides or the center. ## Factorial simulations The binomial model found that the probability of zero new transmissions relative to baseline conditions (spring temperatures, Rest Cool, Drink Indep) was driven by significant interactions between temperature and Rest behavior condition. Together, the interactions show that the probability of zero transmission is reduced (relative to spring temperatures) when the Rest Warm or Rest Dep conditions occurred with summer temperatures. In contrast, increasing temperature alone increased the probability of zero new colonizations, likely due to higher bacterial degradation rates at higher temperatures. Drinking behavior had no clear impact on the probability of zero new transmissions. Results of the incident case model (using non-zero counts) were similar to the binomial model, with significant interactions between rest behavior condition and temperature defining the number of non-zero incident cases over the model run. In general, interactions with rest behavior condition had the largest impact, with the Rest Warm and Rest Dep conditions resulting in higher average incident cases than the Rest Cool condition. Within the Rest Cool and Rest Warm conditions, the count of incident cases were similar across seasons, although higher bacterial degradation rates resulted in the lowest incidences with summer temperatures under both Rest Cool and Rest Warm conditions. With the Rest Dep condition, a large positive interaction with temperature resulted in the highest average count of incident cases during the summer. This was due to warmer temperatures leading to Rest Warm behavior (i.e., resting under trees) occurring on most days in summer simulations. Average counts of incident cases were also higher with spring temperatures than fall temperatures under the Rest Dep condition, likely due to warmer overall temperatures during the spring than the fall (particularly during the latter half of it), and therefore more days with Rest Warm behavior than conditional Rest Cool behavior. In contrast to the binomial model, a significant interaction also occurred in which the count of incident cases as higher with summer temperatures under a temperature-dependent drinking condition. ### Differential transmission pathways Considering contributions to counts of incident cases from different transmission pathways showed that the distribution of colonizations between pathways depended largely on rest condition and seasonal temperature. When considering the proportion of total transmission occurring within a simulation, transmission under Rest Warm conditions largely occurred through direct transmission, and the grazing pathway accounted for the majority of new colonizations for seasons under the Rest Cool condition. In contrast, the majority pathway under the Rest Dep condition depended on season, with the majority of new colonizations during the spring and fall temperatures transmitted through grazing, and the majority of transmission during the summer occurring through direct contact. Overall, transmission via water was generally minimal, and drink behavior condition had little discernable impact on the distribution of colonizations between the transmission pathways. For all seasonal temperatures, secondary transmission did not appreciably contribute to the count of incident cases, accounting for an average low of 0.93% (Spring) to an average high of 1.1% (Fall) of incident cases overall. ### Basic reproduction number The *R*_*0* varied significantly with temperature and Rest/Graze conditions, largely mirroring the pattern found for incident cases. Under Rest Cool conditions the distribution of R₀ was similar cross seasons. Of note here was that the average *R*_*0* under Rest Cool conditions across seasons (0.43) is \<\< 1, reflecting the high proportion of zero new transmissions under this condition. In contrast, while average *R*_*0* with summer temperatures was similar under both Rest Warm and Red Dep conditions (0.82 ± 0.1 SE), average R₀ values were higher with cooler seasonal temperatures under Rest Warm conditions (Fall: 0.86 ± 0.1 SE; Spring: 0.95 ± 0.1 SE), and lower with cooler seasonal temperature under Rest Dep conditions (Fall: 0.55 ± 0.08 SE; Spring: 0.58 ± 0.08 SE). This reflected the influence of bacterial degradation in the environment on transmission, with higher rates of decay occurring during the warmer months. ### Temperature-dependent conditions Of the total set of factorial model runs, those operating under model conditions of Rest Dep and Drink Dependent reflected comparisons of fully temperature- dependent parameterizations of the model during different seasons; that is, models operating under conditions most representative of real-world conditions. When considered in terms of prevalence (considered here as the count of incident cases / 24), the 95% CI for average prevalence using summer temperatures (0.09–0.15) fell within the empirical 95% CI of summer STEC used as a validation range (7.98–16.25). Meanwhile, the 95% CI’s for average prevalence during the spring (0.06–0.1) and fall (0.055–0.089) largely overlapped with empirical 95% CI of winter STEC prevalence used as validation range (0.015–0.0949). That the means of the intervals of the spring (0.084) and fall (0.072) simulations were higher than the validation range mean (0.048) is likely a result of temperatures \> *T*_*th*r during the end of the spring and beginning of the fall that drove colonizations higher than in the calibration sets, all run at a constant 20° C. # Discussion Higher STEC prevalence in environmental substrates, feces, and beef carcasses, as well as an increase in shedding of STEC by cattle during warmer seasons of the year has been commonly reported. For example, Barkocy-Gallagher et al. (2003) reported mean STEC prevalence of 12.9%, 6.8%, and 3.9% during Summer, Fall, and Spring seasons in samples taken beef carcasses in a beef processing facility. Van Donkersgoed et al. (1999) reported higher prevalence of STEC in fecal samples from cattle at slaughter during the summer months than cooler months, and a large-scale review (used for calibration purposes here) reported mean values of winter and summer prevalence in US pasture-range beef cattle of 4.84% and 11.82%, respectively. Various mechanisms have been postulated to explain this phenomenon, and two hypotheses were comparatively evaluated here, including increased drinking with higher temperature (and thus more water-based transmission) and more frequent aggregation under shade trees as temperatures increases, promoting more direct transmission. The results of the factorial analysis of model simulation outputs found that temperature-induced changes in rest behavior most strongly drove overall patterns of new colonizations. Counts of incident cases were significantly higher when either 1) cattle always rested under shade trees versus resting in place and grazing an extra hour, or 2) when rest-behavior was temperature dependent and temperatures were more frequently above the temperature threshold (*T*_*thr**)* causing cattle to rest under trees more often. The average temperature under summer conditions was above *T*_*thr* (25.6°C) while the average temperatures during spring (17.9°C) and fall (12.8°C) were below it, resulting in the highest average count of incident cases with summer temperatures. Further, direct transmission was the dominant transmission pathway during these higher incident case situations, where it accounted for \>74% of new colonizations for all Rest Warm factorial combinations, and for \> 79% of new colonizations for Rest Dep conditions during the summer. Meanwhile, graze-based transmission was the primary pathway in all other situations. In addition, the probability of no new colonizations occurring was significantly negatively impacted by the Rest Dep/Rest Warm conditions, particularly under summer conditions, meaning that temperature-driven spatial aggregation both significantly increased the probability of an epidemic occurring at all, as well as determining its extent. In contrast, higher drinking rates had a more limited effect on STEC incidence in which an interaction with summer temperatures resulted in slightly higher counts of incident cases through a drinking-pathway during this season. However, the proportion of incident cases due to water was low small relative to other pathways, accounting for a maximum of 5% of colonizations in any factorial combination (Fall, Rest Dep Conditions). This contrasts with previous research showing that drinking water is a plausible transmission pathway for STEC. The water resource in the model system was a single 1-acre lake of uniform 0.5 m depth instead of water troughs, as in previous work. Thus, the dilution of concentration and dispersion of fecal-pats due to volume within water patches likely contributed towards the reduced contribution of water-based transmission. However, partially compensating for this was the assumption that STEC concentration within a particular water patch was homogeneous throughout the water column, and was directly proportional to the concentration in the deposited fecal-pat. In real systems, the concentration of STEC cells would be partitioned between aquatic and sediment phases due to adsorption, and therefore the availability of STEC for ingestion may be limited. On the other hand, the relatively high daily decay rate at 20°C (0.388) compared to the low rate (0.042) used for decay in manure may have underestimated the persistence of STEC in water. On the balance, however, STEC in water was likely more available for uptake than in a real aquatic system, increasing the likelihood of transmission. Thus, of the two hypotheses, temperature-driven spatial aggregation that promotes a greater frequency of direct contact between individuals provides the more plausible mechanism to explain seasonality in STEC prevalence in grazing systems, as least when water resources are similarly structured. Increased direct transmission through temperature-mediated spatial aggregation is a plausible explanation for seasonal patterns in STEC transmission for several reasons. First, increasing animal density is well understood to be positively associated with the transmission of infectious disease, and there are previous reported instances in which higher STEC prevalence in cattle may have been due to increased aggregation in the absence of warmer temperatures. In particular, cattle may have a higher risk of shedding STEC when housed than pastured, even during cooler months. Secondly, climate varies widely in space, and the temperatures (collected near Knoxville, TN), schedule and spatial structure of the model assumed here are not representative of conditions in many other locations. Therefore, it is not unexpected that there are reported instances in which the pattern of STEC prevalence did not vary strongly with season, or was not clearly associated with increasing temperature. In one of these cases, however, the prevalence of STEC in feed lot cattle was found to increase with time after cattle arrived in the yard. Thus, changes in spatial aggregation patterns with temperature, rather than changes in temperature alone, may be a reasonable underlying mechanism to explain seasonal STEC prevalence where it occurs. Less clear, however, is whether direct transmission would be the dominant pathway, as suggested by the model. Direct transmission is generally thought to occur via a fecal-oral route, either from social interactions (e.g. grooming activities) which result in transfer of STEC via direct ingestion, or from incidental contacts due to proximity that can result in the transfer of feces between hides. The transmission of STEC via aerosols between cattle in close proximity has also been suggested. In the model, these forms of contact are not differentiated, with a quantity of CFU’s per contact drawn from a distribution whenever a contact occurred due to the breach of the distance threshold. The Poisson-lognormal distribution sampled to simulate the transference of STEC during a contact is integer-valued and over- dispersed, meaning that large numbers of STEC are rarely transferred when direct contacts occur. Because they are directed and may last several minutes, social interactions such as allo-grooming may have the potential to transfer enteric pathogens more efficiently than incidental contacts. However, social interactions occur non-randomly and often occur hierarchically, with less dominant individuals being groomed by more dominant ones. The Poisson-lognormal distribution used approximates the condition that most contacts between cattle are incidental (transferring smaller quantities of STEC), while some are social (transferring larger quantities). Because the nature and context of cattle contacts were not explicitly modeled here, there is uncertainty in understanding how aggregation, beyond the simple proximity rules used in this model, may influence direct transmission. Additional work explicitly incorporating more complex social structure into direct contact behavior could be helpful in reducing this uncertainty. Indirect transmission through the ingestion of contaminated grass emerged as the most important pathway during spring and fall under the Rest Dep condition, and under the Rest Cool condition. That the graze-based pathway was more important than the water pathway may have partially been because even though STEC decayed with rising temperature in both substrates, STEC decayed faster decay in agricultural water than in fecal-pats due to greater competition from microbial organisms in the former. It should be noted that this may not be the case in more pristine water like lakes in non-agricultural settings. When considering model simulations under temperature-dependent conditions, the grazing pathway in all seasons tended to develop at a much slower, approximately linear rate compared to the approximately logistic growth of the direct pathway in the summer. Thus, the grazing pathway may contribute towards maintaining enteric pathogens within a population in an endemic state, particularly at cool temperatures which promote a slower decay of STEC populations in the environment. The greater proportion of graze-based transmission occurring during fall simulations than spring simulations under the Rest Dep condition was a result of fall temperatures that were cooler on average than spring, resulting in more days of additional grazing. In contrast, higher counts of incident cases during the spring than the fall was the result of more frequent warm weather, resulting in more aggregation and direct-transmission. Overall, these results suggest that higher exposure during grazing can result in a greater proportion of graze-based colonizations even though the chances of ingesting STEC via grazing are low, and that contributions from indirect pathways may maintain low- levels of colonization during cooler weather. ## Model limitations and considerations While factorial simulations from our model suggest that higher prevalence of STEC in cattle during the warmer months may be due to more aggregation that drives direct transmission, the global sensitivity analyses suggested that the parameter space exists in which environmental transmission may drive counts of incident cases to be similar to or higher during cooler months. The PRCC’s calculated as part of the global sensitivity analysis at each static temperature (20, 24, 25, 30°C) showed clear temperature-driven patterns in sensitivity reflecting differences in the importance of transmission pathways between cool (20 and 24°C) and warm (25 and 30°C) model conditions due to rest behavior below and above the *T*_thr. When temperatures were \< *T*_thr, the parameters with the strongest correlations were those influencing transmission via a graze-based pathway (the grass infection factor (*p*_{*grassinfect*}) and the starting concentration of STEC in a fecal- pat (*C*)). When temperatures were \> *T*_thr, the strongest positive correlations were those involved in direct transmission, including the direct distance threshold (*ddt*) and the mean of the Poisson-lognormal distribution used to determine the CFU’s of STEC transferred between cattle during direct transmission events **(***p*_lnmean). So, factors increasing the probability of transmission from a grazing-route could potentially result in higher counts of incident cases during cooler weather than warmer weather. Of the two variables, increasing concentrations of STEC in fecal-pats (*C*) is more likely to occur in natural grazing systems, as cattle are known to avoid eating grass contaminated with feces. Although the quantity of STEC resulting from the calibration process (10.36 g/CFU), was well within the range of values reported (most probable number) of STEC gram^-1 of cow manure in a study by Fegan et al. (2004), it was far below the maximum value reported in that study (4.3 x 10²). It also below the 10³ CFU gram^-1 suggested as the threshold for an individual to be classified as a “super- shedder”. Indeed, an examination of our LHC simulation outputs in which values of *C* \> = 10.36 CFU’s gram^-1 showed mean incident cases to be 20 ± 0.62 SE and 18.66 ± 0.45 for simulations with temperatures of 20°C and 30°C, respectively. In this “high *C*” subset, grazing-based transmission made up 83% of new colonizations at 20°C and 61% of new colonizations at 30°C, respectively. Many animals colonized with enteric pathogens shed heterogeneously over the course of their infectious periods, and evidence suggests that super-shedders may largely be animals sampled near the high-shedding points of their infectious periods. Although variable shedding between or within individuals was not found to have a large effect in on incident case count in our local sensitivity analysis, we did not directly include the presence of “super-shedding” individuals as a parameter or as a hypothesis to be explored. This effect may be further heightened if potential growth of STEC after deposition is considered. Therefore, the current model may not adequately capture the role of *C* in STEC transmission dynamics. More frequent graze-based transmission may also occur due to the structure of the model environment and cattle density. Although mostly weak correlations were found between counts of incident cases with environmental structures of the model, positioning the lake into the corners of the property versus the sides or center resulted in more graze-based cases. This appears to occur because when the water source is in a corner of the rectangular model area, cattle tended to concentrate at one end of the property while grazing and were exposed more often to contaminated graze. Although an artifact of model structure here, increasing distance between water and grazing forage has been shown to reduce the use of available forage in grazing cattle, and to increase the unequal distribution of manure in pasture systems. Although the model predicts that there are potential pathways for high graze- based transmission, there is currently limited evidence of food-based infections/colonizations. There is also limited evidence that super-shedders contribute highly to increased risk from environmental pathways. In addition, since cattle are known to avoid contaminated graze (accounted for implicitly here by making the CFU of STEC taken up by grazing very low compared to the amount in the plot), it is possible that the model overestimates the potential of graze-based colonization. If so, it would help explain why the 95% CI of prevalence values under temperature dependent conditions during Spring and Fall temperatures fell on the higher end of the calibration range. However, the influence of water location versus graze availability on graze-based exposure worth may be worth exploring in future work, particularly in situations where graze quality widely varies. Lastly, the model assumed that cattle behaved according to simple rules, and that all individuals were of indeterminate adult age. Although sensitivity analyses suggested that increasing cattle density could increase direct transmission at high temperatures due to denser clustering around shade-trees during rest, this assumes that all cattle would always rest under the same tree, and that inter-cattle distances between individuals would not be maintained as cattle density increases. However, maintenance of a minimum personal space is an important aspect of cattle social behavior, and cattle may maintain larger distances as herd size increases to reduce aggression. So, it is likely that cattle that do not fit under the shade of a tree due to lack of space would find another tree, and that some minimum inter-cattle distance may be maintained during resting. Thus, the influence of increasing density on direct transmission may be overestimated. In contrast, the more moderate effect of increasing cattle density on graze-based transmission may be more mechanistically plausible (i.e., more cattle produce more manure), but because actual stocking rates are determined by pasture yields, and transmission was found to decrease as the grass to weed ratio decreased, this relationship may not be practically relevant to lower producing pastures. Finally, the model was calibrated using a meta- analysis of adult beef cattle prevalence data that listed relatively low average values for summer (11.83) and winter (4.84) prevalence in the United States. However, the prevalence of cattle may be significantly influenced by individual factors such as age, with the highest expected STEC prevalence during the first year of life, and parity status influencing STEC colonization thereafter. Thus, the current model structure and calibration may not adequately capture the temporal dynamics of shedding patterns for juveniles, or different age and parity classes of adult female cattle. Despite its limitations, the current model structure is quite flexible, and additional ecological, behavioral or biological aspects of agents or the environment can be readily incorporated in order investigate additional hypotheses, or to more closely model particular conditions. Additionally, distinguishing between direct and indirect transmission pathways using empirical data is difficult, particularly if the time-scales of epidemiological dynamics and pathogen dynamics in the environment are convergent. In simulation model- based approaches like the one used here, uncertainty associated with transmission sources within simulations can be eliminated or greatly reduced, making it a useful tool for inferring the role of different pathways in epidemiological dynamics. # Conclusions Model simulations suggest that seasonal patterns of higher STEC prevalence during warmer months in some grazing systems may be driven by temperature- mediated aggregation that promotes direct-transmission of STEC between individuals. In the model, this hypothesis is contingent on the presence of shade-providing structures, such as trees, under which cattle aggregate for temperature relief, a centrally located water source, and on the assumptions that cattle follow a rigid social structure in which individuals in the herd follow a dominant individual to resting locations. Therefore, determining ways to reduce rates of close contact between cattle under shade or while being housed could be beneficial to reducing rates of STEC transmission. # Supporting information The authors would like to thank Brandon Beavers from the East Tennessee Research and Education Center—Blount Unit for his valuable insight on cattle behavior and management. [^1]: The authors have declared that no competing interests exist.

# Introduction The human *SLC22A5* gene, which encodes a 63 kDa organic cation/carnitine transporter 2 (OCTN2), is located in the cytokine cluster region on chromosome 5q31. OCTN2 is expressed in various tissues, including kidney, skeletal muscle, heart, colon, brain, liver, etc. Functional defects of OCTN2 are associated with various diseases including primary carnitine deficiency, Crohn's disease, and asthma. OCTN2 not only transports carnitine, but also recognizes clinically important therapeutics such as mildronate, verapamil, pyrilamine, oxaliplatin, imatinib and cephaloridine. OCTN2 is associated with oxaliplatin accumulation and cytotoxicity in OCTN2-HEK293 transfected cells. The two alleles of *OCTN2* (rs2631367 and rs2631372) may be important predictors in gastrointestinal stromal tumor patients receiving imatinib therapy. These reports indicate that the functional defects and/or aberrant expression of OCTN2 may affect the disposition and subsequent therapeutic efficacy of its substrates. Several reports suggest the involvement of peroxisome proliferator-activated receptor alpha (PPARA) and gamma (PPARG) in the transcriptional regulation of OCTN2 in various tissues. However, down-regulation of OCTN2 has been reported in tumors with high expression of PPARA and PPARG. A recent study found that the decreased levels of OCTN2 in several epithelial cancer cell lines could be restored by the demethylating reagent 5-aza-cytidine. These findings imply that other machineries cooperate with the transcription factor network to modulate the expression of OCTN2, such as DNA methylation. DNA methylation is an important epigenetic mechanism that modulates gene expression. The CpG dinucleotide near transcriptional start sites is abundant in gene promoters, and is referred to as CpG islands. The methylation of CpG islands is associated with repressed gene transcription and abnormal DNA methylation can lead to aberrant gene expression. Unlike gene mutation, DNA methylation can be reversibly altered by demethylating agents such as decitabine (5-aza-2′-deoxycytidine, DCA) and 5-aza-citidine. These agents are incorporated into the DNA and inactivate DNA cytosine C5-methyltransferases. Thus, we hypothesized that the differential methylation status of *OCTN2* may be correlated with the aberrant expression of OCTN2 in cancer cells. In this study, we investigated whether the methylation of CpG islands acts as a possible mechanism responsible for the down-regulation of OCTN2 in cancer cell lines. By using methylation-specific PCR (MSP), bisulfite genomic sequencing, and *in vitro* methylation assays, we have provided evidence that promoter DNA methylation is an essential mechanism suppressing OCTN2 expression in cancer cell lines. Application of a demethylating reagent, which modulated the methylation status of the *OCTN2* promoter, increased the expression of OCTN2 and made cancer cells more sensitive to oxaliplatin. # Materials and Methods ## Chemicals and Reagents Decitabine, sodium bisulfate, hydroquinone and oxaliplatin were purchased from Sigma-Aldrich (St. Louis, MO). TRIzol reagent and Lipofectamine 2000 were obtained from Invitrogen (Carlsbad, CA). ## Cell Culture and Treatment with DCA The hepatoma cell line HepG2, colon cancer cell line LS174T, glioma cell line U251, bile duct cancer cell line QBC-939 and African green monkey kidney cell line COS-7 were obtained from American Type Culture Collection (Manassas, VA). Cell lines were cultured in DMEM supplemented with 10% fetal bovine serum (FBS) at 37°C in a humidified incubator containing 5% CO₂, except the QBC-939 cell line, which was cultured in RPMI 1640. Cells were treated with DCA at a final concentration of 0.5 μM or 1 μM and renewed every 24 h for one week. ## RNA purification, cDNA Synthesis and Quantitative Real-time PCR Total cellular RNA was isolated from the treated cells using TRIzol reagent. cDNA was synthesized from 1 μg of total RNA using the first cDNA synthesis kit (Takara, Dalian). Quantitative RT-PCR was performed using SYBR Green (Bio-rad) and the primers are listed in. RT-PCR was performed over 40 cycles at 95°C for 30 s, at 60°C for 30 s, and 72°C for 30 s followed by a final extension at 72°C for 2 min. The relative mRNA expression was normalized to the housekeeping gene *β-actin* and the analysis was carried out using the 2^−ΔΔCt method. ## Western Blotting Total cellular proteins were extracted and the concentrations were quantified by the BCA protein assay (Thermo Scientific, Rockford, IL). Proteins were subjected to SDS-polyacrylamide gel electrophoresis on 10% polyacrylamide gels and transferred to PVDF membrane. The membrane was probed with either an anti-rabbit OCTN2 antibody (1∶400, Abcam Labs, Cambridge, MA) or β-actin antibody (1∶1000, Abcam Labs). Subsequently, membranes were incubated with DyLight 800-labeled rabbit polyclonal antibody (1∶7500, KPL Inc, Gaithersburg, MD), protected from the light. The stained membranes were scanned by Odyssey Infrared Imaging System (LI-COR Biosciences). The ratios of OCTN2 against β-actin were calculated. ## Cell Viability Assay Cancer cell lines were treated with 1 μM DCA or 0.1% DMSO for one week. Subsequently, cells were seeded into 96-well plates at a density of 2×10⁴/well, and exposed to oxaliplatin at various concentrations (0, 0.3, 1, 3, 10, 33, 100 or 330 μM) for 48 h. Cell viability was measured by the MTS assay (Promega, Madison, WI) and IC₅₀ values were calculated using SPSS 13.0 version. ## Methylation-Specific PCR (MSP) and Bisulfite-Sequencing PCR (BSP) To predict regions rich in CpG dinucleotides close to the *OCTN2* transcriptional start site, we used Methyl-Primer software (<http://www.urogene.org/methprimer/>) according to the following criteria: the length of CpG island \>100 bp, observed/expected CpG ratio \>0.6 and percentage of G plus C\>50%. We searched the *OCTN2* genomic sequence including 1.5 kb upstream and 2 kb downstream of transcriptional start site (TSS). According to the predicted CpG islands, MSP and BSP primers were designed and are listed in. Genomic DNA of cancer cell lines was isolated and DNA bisulfite modification was performed as described. Bisulfite modified DNA was recovered by Wizard DNA Clean Up System (Promega) and treated with 5.5 μl of NaOH then precipitated with ethanol. Precipitated DNA was re-suspended in water, and then amplified with MSP or BSP primers. The PCR condition was as follows: 94°C for 5 minutes, followed by 40 cycles of 94°C for 30 s, annealing temperature for 30 s, 72°C for 60 s, and finally 72°C for 5 min. The amplified DNA fragments were run on 2% agarose gels and stained with ethidium bromide. BSP products were purified by a DNA purification kit (Takara, Dalian) and inserted into pGEM-T easy vector (Promega). Five clones for every cell line were randomly selected and sequenced. The methylated sequence of *OCTN2* was checked for alignment using MethBLAST. ## Plasmid Construction and *in vitro* Methylation Assay According to the predicted CpG islands in the genomic sequence, we divided the promoter into three regions. Three regions containing different CpG islands were amplified (Primers listed) and inserted into the pGL4.17 vector (Promega). These recombinants, pGL4.17-Region1, Region2 and Region3 were verified by automated sequencing. The respective plasmid was linearized by *Xho* I and subjected to *in vitro* the methylation assay as described in our previous work. The methylation products were then digested by *Kpn* I and inserted into the pGL4.17 vector. These new recombinants containing the methylated promoter regions were named pGL4.17-mRegion 1, pGL4.17-mRegion 2 and pGL4.17-mRegion 3, respectively. ## Transient Transfection and Luciferase Report Assay COS-7 cells were seeded in 24-well plates at a density of 10⁵/well in 500 μl DMEM and 10% FBS. After 8 h, cells were transfected with constructed luciferase vectors (0.2 μg/well) using Lipofectamine 2000. The pGL4.17 vector lacking a promoter served as the negative control. Internal control vector pRL- TK (0.02 μg/well) was used to normalize luciferase activity. After 8 h, each well of transfected cells was washed twice then 100 μl of reporter lysis buffer was added to lyse cells. Each group had triplicate wells and was repeated more than three times. Dual luciferase activity (Relative light units, RLU) was determined by a luminometer. ## Statistical Analysis All data were expressed as mean ± standard deviation. Statistical analyses were performed with a one-way ANOVA for significance followed by a post-hoc test using SPSS 13.0 software. Data comparisons with a *p* value of less than 0.05 (*p*\<0.05) were considered significantly different. # Results ## OCTN2 mRNA and Protein Levels in Different Cancer Cell Lines Total cellular RNA and protein were extracted from cancer cell lines and OCTN2 expression was measured by Quantitative RT-PCR and Western blotting, respectively. The highest expression of OCTN2 was in QBC-939, followed by U251, LS174T and HepG2 cells. The expressions of OCTN2 in QBC-939 and U251 cells were significantly higher than that in U251, LS174T cells (*p*\<0.05). ## Effects of DCA on the Expression of OCTN2 in Cancer Cell Lines To examine whether the discrepant expressions of OCTN2 in different cancer cell lines was due to promoter methylation, we treated cancer cell lines with DCA and determined the restored expression of OCTN2. In , DCA remarkably up-regulated the relative OCTN2 mRNA and protein expression in HepG2 cells with 0.5 μM (1.38-fold, 1.61-fold) and 1 μM treatment (1.52-fold, 2.07-fold) (*p*\<0.05), compared to the control group. Similarly, treatment of DCA lead to an increase of OCTN2 levels in LS174T cells. However, DCA did not change the expression of OCTN2 in QBC-939 and U251 cells, which already had high OCTN2 expression levels. ## OCTN2 Genomic Sequence Harbors Three CpG Islands We searched the *OCTN2* genomic sequence for the location of CpG dinucleotides using Methy-primer software. CpG dinucleotides were found extensively located near the TSS. According to the location intensity of CpG dinucleotides, we observed three putative CpG islands located in the *OCTN2* genomic sequence. CpG island-2 is located close to UTR and exon 1, CpG island-3 is located in intron 1, and CpG island-1 is located upstream of the TSS. ## Methylation Status of CpG Islands within OCTN2 by MSP To further elucidate the methylation status of CpG islands within *OCTN2* in different cancer cell lines, we analyzed three CpG islands by MSP which distinguishes methylated DNA from un-methylated DNA. In, CpG island-1 (CpG1) was completely methylated in HepG2 cells, partially methylated in LS174T and U251 cells, and un-methylated in QBC-939 cells. Partially methylated products of CpG island-2 (CpG2) were found in LS174T and QBC-939 cells, but these were un- methylated in HepG2 and U251 cells. In almost all cell lines, CpG island-3 had more methylated products than unmethylated. In, after treatment with DCA, un- methylated products of CpG1 were increased in HepG2, LS174T and U251 cells. This finding indicates that DCA can cause de-methylation within CpG1. A similar result was observed within CpG2. DCA treatment increased un-methylated products of CpG3 in HepG2 cells, but did not cause significant changes in other cells. ## Effect of Methylated and Un-methylated Promoter Regions on Transcription Activity of OCTN2 using an *in vitro* Luciferase Reporter Assay To determine which methylated CpG islands play essential roles in down- regulation of OCTN2, we constructed luciferase reporter plasmids bearing either un-methylated or methylated regions. Constructed plasmids were transfected into COS-7 cells and used to assess transcriptional activity. In comparison with pGL4.17 (Control), pGL4.17-Region1 displayed markedly increased luciferase activity of 102.5-fold (*p*\<0.01), while luciferase activity from pGL4.17-Region2 and Region3 were only increased 11.4-fold and 3.5-fold (*p*\>0.05). This result indicates that Region1 spanning −354 to +85 bp plays an essential role in promoter activity. Interestingly, only the luciferase activity of un-methylated Region-1 was higher than its methylated vector (29.7-fold, *p*\<0.01). Un-methylated Regions-2 and -3 did not show significant differences in luciferase activity compared to their corresponding methylated regions. It is possible that the methylation statuses of Region-2 and -3 containing CpG island-2 and -3, respectively, are irrelevant to *OCTN2* expression. Thus, these findings demonstrate that increased methylation can inhibit promoter activity of Region-1. ## Determination of DNA Methylation Profiles of OCTN2 by BSP For assessing the methylation profile of CpG sites in Region-1, bisulfite genomic sequencing in various cancer cell lines was carried out. In, the genomic sequence of OCTN2 promoter was displayed and CpG sites were marked. One methylated sequence of *OCTN2* spanning −325 to −92 bp in HepG2 cells was checked for alignment using MethBLAST and is presented in. Other methylated sequences from LS174T, QBC-939 and U251 cells are presented in Fig S1, S2 and S3. In, the methylation profiles showed that the rates of methylated CpG sites (mCpG/total CpG number) were 72.2% and 60.0% in HepG2 and LS174T cells, respectively, which is much higher than methylation rates measured in QBC-939 (20.0%) and U251 (11.1%) cells. Moreover, after 1 μM DCA treatment in HepG2, the rates of methylated CpG sites decreased to 13.3%. Since the hypermethylated CpG sites within OCTN2 are rescued by DCA treatment in HepG2 and LS174T cells, these findings provide direct evidence that individual methylated CpG sites in the promoter may precisely regulate the expression of OCTN2. ## Effects of DCA on Enhancing Sensitivity of Cancer Cells to Oxaliplatin To further understand if methylation status of the *OCTN2* gene affects the sensitivity of cancer cells to oxaliplatin, cancer cells were pre-treated with DCA followed by exposure to oxaliplatin. Cell viability was assessed and the IC₅₀ values were calculated. In. A, when cells were exposed to oxaliplatin at 3, 10, 33 and 100 μM, cell viability of DCA-treated HepG2 and LS174T was decreased compared to DMSO-treated cells (*p*\<0.05), but did not change in U251 and QBC-939 cells. In, the IC₅₀ values for DCA-treated HepG2 and LS174T cells were decreased to 51.6% and 44.8% (*p*\<0.05), respectively, compared to DMSO-treated cells, but were unchanged in QBC-939 and U251 cells. Therefore, it is possible that DCA sensitizes HepG2 and LS174T cells to oxaliplatin through increasing OCTN2 expression and enhancing the uptake of oxaliplatin. # Discussion In this study, we found that hypermethylation within the *OCTN2* promoter is responsible for the down-regulation of *OCTN2* in HepG2 and LS174T cells. We precisely detected the degree of methylation of individual methylated CpG sites in the promoter region by MSP and BSP, which was inversely correlated with the expression of OCTN2 in different cancer cells. Moreover, DCA pre-treatment of HepG2 and LS174T cells increased oxaliplatin-induced cell death via OCTN2 demethylation. The essential role of DNA methylation has attracted much interest as a possible mechanism underlying the aberrant expression of oncogenes, tumor suppressor genes and DNA repair genes in carcinoma cells. Recent findings show a role for DNA methylation in regulating the expression of transporters, such as SLC22A1, A2, A3 (OCT1, OCT2, OCT3), SLC22A18, SLC5A8, Ntcp, Oatp1b2, Bsep, ABCG2, ABCG5/G8. To understand the mechanism by which OCTN2 expression is repressed in HepG2 and LS174T cells, we treated these cell lines with the demethylating reagent DCA and observed an increase in OCTN2 mRNA and protein levels in HepG2 and LS174T cells, while no change occurred in QBC-939 and U251 cells. A recent study also found that the low levels of OCTN2 were increased by DCA in human papillomavirus16 E6- and E7- expressing keratinocytes. DCA inhibits DNA methyltransferase enzymes (DNMTs) and further corrupts the integrity of the methylation imprint, thereby leading to the disruption of epigenetic signatures and regulation of gene expression. Thus, these findings suggest that the down regulation of OCTN2 in HepG2 and LS174T cells may be due to aberrant promoter methylation. Analyzing the three putative CpG islands within *OCTN2*, we observed that CpG islands displayed different methylation statuses in different cancer cells, and that hypermethylation could be reduced by DCA. Therefore, we sought to identify which CpG island plays a pivotal role in the regulation of OCTN2 expression. In the luciferase reporter assay, only the promoter Region-1 spanning −354 to +85 bp showed increased promoter activity. Moreover, methylation of Region-1 that mimics the methylation patterns in cells, completely inhibited the promoter activity. This result provides the evidence that Region-1 is a possible determinant of OCTN2 expression, because hypermethylation of Region-1 inhibited transcriptional activity of OCTN2 in LS174T and HepG2 cells. Bisulfite genomic sequencing precisely detected the individual methylated CpG sites within promoter Region-1, which was significantly hypermethylated in HepG2 and LS174T compared to QBC-939 and U251 cells. The degree of DNA methylation was inversely correlated with the expression of OCTN2 in these cancer cells. These findings may reasonably explain the lower levels of OCTN2 in HepG2 and LS174T cells, compared to the expression in QBC-939 and U251 cells. Moreover, DCA reversed the methylation of CpG sites in Region-1 in HepG2 cells consistent with the increase in OCTN2 expression. Methylated CpG dinucleotides within promoter Region-1 inhibit the expression of *OCTN2* probably through two mechanisms. First, many transcription factor bindings may be blocked by methylated CpG dinucleotides within the *OCTN2* promoter. We analyzed promoter Region-1 of *OCTN2* by TFSEARCH software (<http://www.cbrc.jp/research/db/TFSEARCH.html>), and found that the consensus binding sites of transcription factor specificity protein 1 (Sp1) and myeloid zinc finger 1 (MZF-1) overlapped some key CpG sites (Fig S4). Further work is in progress to investigate whether the methylation of specific CpG sites blocks these transcription factors. A second possible mechanism is the aberrant regulation of DNA methyltransferase enzymes and methylation-dependent recruitment of nucleoprotein factors such as the methy-CpG binding protein MeCP1 and MeCP2. We will investigate how the hypermethylated CpG sites of OCTN2 recruit these methylation-related proteins and repress the expression of OCTN2. Given the essential role of OCTN2 in cancer cell uptake and thus treatment efficacy of oxaliplatin, methylation of *OCTN2* might be used as a target for enhancing therapeutic efficacy. After treatment with 1 μM DCA, HepG2 and LS174T cells became more sensitive to oxaliplatin, but not QBC-939 and U251 cells. Moreover, in DCA treated cells, the increased expression of OCTN2 conferred greater oxaliplatin uptake and more cytotoxicity than non-DCA treated cells. Although DCA in high concentrations causes global genomic instability and cytotoxicity, DCA in low concentrations acts mainly to inhibit DNMT rather than to induce cell death. Thus, DCA was concomitantly administered with oxaliplatin, and the resulting restoration of OCTN2 expression enhanced the uptake and efficacy of oxaliplatin. # Conclusion In conclusion, we revealed the epigenetic mechanism of the down-regulation of *OCTN2* in different cancer cells. The promoter region spanning −354 to +85 was a determinant of OCTN2 expression. The degree of methylated CpG sites within the region was inversely correlated with the levels of OCTN2 in cancer cells. Application of the demethylating agent, DCA, reversed the hypermethylation status of the *OCTN2* promoter and increased OCTN2 expression, thereby leading to enhanced uptake of oxaliplatin-causing cancer cells to be more sensitive to oxaliplatin. Given the essential role of OCTN2 in cancer cell uptake and thus treatment efficacy of several chemotherapeutics, the pretreatment of a demethylating reagent is a possible strategy for optimization of pharmacotherapy against cancers. # Supporting Information We are grateful to Dr. Jie Liu and Dr. Helen Renaud, who provided language help. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: QQ HHZ. Performed the experiments: QQ JQ YWZ XYL MZ. Analyzed the data: QQ LXW. Contributed reagents/materials/analysis tools: QQ LJF. Wrote the paper: QQ HHZ.

# Introduction Monocytes/macrophages (Mo/MΦ) represent one of the important targets during dengue infection and are important in viral dissemination. It has been shown that dengue virus (DENV) is capable of inducing oxidative stress in humans – suggesting that the interaction of DENV with Mo /MΦ could play a role in the pathogenesis of dengue. It has also been reported that the immune alterations can influence oxidative metabolism and vice versa. In this regard, monocytes from neonates and elderly individuals have been shown to have immunosuppressive status against infections, suggesting a possible altered oxidative response. Nitric oxide (NO) plays an important role in inflammation and in the regulation of immune responses. This nitrogen reactive species are greatly produced by Mo/MΦ during inflammatory processes. Since, an altered immune response has been documented in neonatal and elderly monocytes, we hypothesized that neonatal and elderly monocytes probably have an altered oxidative response to dengue infection. Therefore, the aim of this study was to analyze the oxidant (nitric oxide) and antioxidant (catalase, superoxide dismutase and reduced glutathione) responses of monocyte from neonates, young adults and elderly subjects during an *in vitro* dengue virus infection. During this study both elderly and neonatal monocytes had lower oxidant/antioxidant responses to dengue virus infection. These findings are probably important in the pathogenesis of dengue disease in individuals from those age groups. # Materials and Methods ## Preparation of virus stock and virus titration DENV prototype laboratory strains; DENV-1 (Hawaii), DENV-2 (New Guinea C), DENV-3 (H-87) and DENV-4 (H-241) were propagated in mosquito C6/36HT cells that were cultured in Eagle's MEM medium containing 10% FBS prior to viral monocyte infection. The virus culture medium was harvested after 5 days of incubation and after removal of cell debris by centrifugation, the supernatant containing virus was aliquoted and stored at −70°C until used. Virus was titrated by plaque formation assay on Vero cells. Cells were planted at 1×10⁶ cells/well in 24-well plates and subsequently, serial dilutions of virus were added and incubated at 37°C for 7 days. Afterwards, the plaques were visualized by staining with 1% crystal violet solution. Virus concentration is given as plaque-forming units (PFU)/ml. Virus stock was free of endotoxin as determined by limulus amebocyte lysate assay (Charles River, MA, USA). ## Monocyte cultures Monocytes were isolated from heparinized peripheral blood obtained from male human healthy neonates (umbilical cord), young adults (35–45 years old) and elderly (65–70 years old) subjects (N = 10 each group). All individuals were tested for circulating NS1 protein and anti- DENV antibodies (Dengue NS1 Ag + Ab combo, Standard Diagnostic, Inc. Bioline, Korea). Subjects positive to DENV NS1 protein or anti-DENV antibodies were excluded from this study. Mononuclear cells were obtained by density centrifugation over 1.077 Histopaque (Sigma Chemical Co, St. Louis, MO). Individuals or parents were informed about the study procedures and their written consents were obtained before enrollment in the investigation. In this context, individuals or their relatives were informed of the scope of the study, samples to take and obtained results. Written consents were approved by the Bioethical Committee of Medical School (Universidad del Zulia, Maracaibo, Venezuela). Total mononuclear leukocytes recovered from the interface were washed and suspended in RPMI 1640, 10% fetal bovine serum and penicillin/streptomycin. Afterwards, 300 μl/well of a cellular suspension (4×10⁶ cells/ml) were layered on 24-well plastic tissue culture plates (Nunc, Roskilde, Denmark) and incubated for 3 hours at 37°C and 5% CO₂. Non adherents cells were washed out with warm medium and adhered cells (approx. 3×10⁵ cells) were used for experiments. The monolayers of adherent cells were reacted with an FITC conjugated anti- human CD14, monoclonal antibody (Sigma Chemical Co., St. Louis, MO, USA) to determine monocyte percentage, using a microscopy with epifluorescence system (Zeiss, Germany). ## Infection of monocyte cultures Monocytes from each subject were infected with the different DENV types at a multiplicity of infection of 1 (MOI: 1) and incubated for 1 and 3 days at 37°C and 5% CO₂. Controls represent monocytes cultured with supplemented medium without virus. In addition, monocyte cultures were incubated with LPS (50 ng/ml) (Sigma-Aldrich Company, St. Louis MO, USA) for the same period of time. Doses of MOI: 1 and LPS (50 ng/ml) were chosen, since they were capable of inducing high production of proinflammatory cytokines under the same culture conditions (unpublished data). ## Determination of nitrite/nitrate production Total nitrite concentration in monocyte homogenates was used as an indicator of nitric oxide (NO) synthesis. Nitrates in samples were reduced to nitrites by incubating with nitrate reductase. Nitrite concentration was measured using a commercial kit following the manufacturer's indications (Nitric Oxide Quantichrom, Bioassay Systems, Hayward, USA). Optical density was measured at 550 nm and results expressed as μM/mg of cellular protein. ## Determination of thiobarbituric reactive substances (TBARS) Monocyte malondialdehyde (MDA) content was assessed by the thiobarbituric acid assay (NWLSS, Malondialdehyde Assay. Vancouver, WA). Absorbance was measured at 532 nm. As external standard the MDA bisdimethyl acetal (Sigma – Aldrich, St. Louis, MO, USA) was used and results were expressed as nmol per mg of cellular protein. ## Determination of enzyme activities and reduced glutathione (GSH) Treated monocyte cultures and controls were homogenized and catalase activity was determined using a commercial kit (Cayman Chemical Company, Michigan, USA). Results are expressed as nmol/min/mg of cellular protein. Superoxide dismutase (SOD) activity was determined using a commercial kit (Cayman Chemical Company, Michigan, USA) and results were expressed as as U /mg of cellular protein. Content of GSH was also determined in monocyte homogenates using a commercial kit following the manufacturer's indications (Cayman Chemical Company, Michigan, USA). Results were expressed as μM/mg of cellular protein. Total protein content was measured in the monocyte homogenates by the method of Bradford (Bio Rad, USA). ## Determination of apoptosis Controls, LPS and infected monocyte cultures were fixed with 1% paraformaldehyde in PBS for 10 min at room temperature and permeabilized with acetic acid: ethanol. The percentages of apoptotic cells were assessed by TUNEL reaction using the *In situ* Apop Tag kit (Chemicon International, USA & Canada) according to the manufacturer's instructions. ## Statistical analysis Data were expressed as mean ± standard deviation. Differences between groups were determined by ANOVA followed by Bonferroni posttest. Significance was assumed to be at two tailed p\<0.05. # Results In this study, monocytes (purity \>95%) were co-cultured with different DENV types or with LPS. In order to determine the response capacity of monocytes according to the age of the donor, monocytes from neonates, young adults and elderly individuals were tested for the oxidant/antioxidant response during infection by dengue virus. In general, lower values in NO and MDA productions, catalase and SOD activities and GSH content were found in neonatal-monocytes, following by elderly-monocytes and the highest values were observed in adult- monocytes; except increased activity of SOD observed in elderly monocytes. These findings were observed at days 1 and 3 post-infection and the pattern of monocyte response induced by dengue virus was similar to that observed in LPS- treated cultures (data no shown). Since oxidative stress is related to apoptosis, TUNEL assay was used to determine the degree of apoptosis in the different monocyte cultures. Apoptosis was increased in LPS and dengue virus infected-monocyte cultures, regardless of the monocyte source with higher values observed at day 3. To analyze the potential of different DENV types on the oxidative stress induction, monocytes cultures were infected with laboratory strains of DENV-1 to 4. Increased production of NO, MDA, catalase and SOD activities and GSH content were induced by the different viral serotypes and LPS in monocytes from neonatal, adult and elderly individuals. (as described in -). DENV-2 induced the highest production of NO accompanied with high production of MDA. Anti-oxidant response was also observed in all DENV types. The highest values of catalase activity were observed for DENV-1 and DENV-4 and SOD activity in DENV-1. The highest content of GSH was observed in monocytes infected by DENV-4. The grade of oxidant and anti-oxidant responses was influenced by monocyte source. # Discussion NO and reactive oxygen species have modulating effects on inflammation and in the regulation of immune responses. During inflammatory reactions, Mo/MΦ are capable of producing increased amount of oxidants including NO. Sustained inflammation resulting in inflammatory sequelae in neonates, is well known, suggesting altered monocyte function. In this study, we assessed the potential contribution of monocytes from neonates, elderly and young adult subjects for the production of NO and enzymatic and nonenzimatic antioxidants after infection with DENV. The neonatal monocyte response was lower than those observed in elderly and young adult monocytes, suggesting impaired response to dengue virus. There is little information about oxidative response by neonatal monocytes. However, it has been shown a relationship between immune and oxidative mechanisms, suggesting that altered immune response could be associated to altered oxidative response. Neonates are born with quantitative and qualitative defects in both adaptive and innate immune responses. In this regard, lymphocyte subset percentages in cord blood from neonates and cytokine responses to bacterial antigens were observed to be diminished when compared to peripheral blood from adults. Plasma cytokine concentrations and cytokine production by neonatal monocytes after lipopolysaccharide stimulation *in vitro*, have been found decreased compared to adult plasma and monocytes. In addition, blocking cytokine condition has been reported in neonates. Plasma interleukin-1 receptor antagonist was significantly higher in neonates than in plasma from adults. The reduced response of neonatal monocytes to DENV could also be related to the increment of distinct inhibitory receptors on neonatal peripheral blood immune cells that could play a role in regulation of the neonatal immune system. However, previous report has shown that the cord blood mononuclear phagocyte has a respiratory burst quantitatively comparable to that of the adult cell. The oxidant and antioxidant responses of elderly monocytes were also lower than those observed in monocytes from young adults. Ageing may contribute to the oxidative metabolism dysregulation that affects the elderly. In this regard, controversial information has been reported. Oxidative stress is commonly observed in the elderly and could be involved in age-related diseases. The aging brain undergoes a process of enhanced peroxidative stress, as shown by reports of altered membrane lipids, oxidized proteins, and damaged DNA that could explain age-related neurodegenerative diseases. Experimentally, it has been reported decreased antioxidant enzyme levels with age in murine brain. However, other studies have shown decreased oxidative stress in the elderly. Regarding this, increased activities of antioxidant enzymes (Catalase, SOD) in healthy human neutrophils have been reported to be age –dependent. Decreased plasma membrane fluidity of lymphoid cells and monocytes in advanced age, may be one contributor factor to decreased production of NO observed in elderly monocytes, since, it has been reported an association between decreased membrane fluidity in red blood cells of hypertensive patients and low plasma NO-metabolite levels . In addition, monocytes from elderly subjects had a decreased accessory function for PHA-stimulated T cells compare to those obtained from young subjects. Cytokine production and expression of costimulatory T cell proteins (CD80) on monocytes from older adults were lower than those on cells from young individuals. These monocyte alterations could be involved in the course of dengue disease. In this regard, monocytes from neonatal and elderly subjects had decreased production of cytokines after dengue virus interaction compared to young adults, suggesting impairment in the production of cytokines (unpublished data). Since, both immune and oxidative mechanisms could be related, immune alterations could influence oxidative metabolism and vice versa. Thus, the immunosuppressive status in neonatal and elderly monocytes could be involved in decreased oxidant and antioxidant responses after DENV infection. As an interesting finding, DENV-2 induced higher stimulatory effect on NO production compared to other viral types. This observation could be reflected in patients, since, symptoms, signs, and laboratory findings appear to be different for patients infected with DENV-2. NO affects virtually every step of the development of inflammation. Low concentrations of NO produced by constitutive nitric oxide synthases (NOS), inhibit adhesion molecule expression, cytokine and chemokine synthesis and leukocyte adhesion and transmigration. Large amounts of NO, generated primarily by the inducible NOS (iNOS) can be toxic and pro-inflammatory. In our study, the amount of NO produced by the different DENV types could be considered toxic, since, increased monocyte content of MDA (lipid peroxidation) was found. High values of NO and MDA were accompanied by increased activities of catalase and SOD and high content of GSH, suggesting an antioxidant response. In this report only NO was studied as an oxidant molecule; however, the induction of oxygen reactive species cannot be discarded, since enzymes such as catalase and SOD were found increased in this study. Of interest, SOD could modulate the oxidant effect of NO, since the interaction of superoxide anion with NO produces peroxynitrites, a high reactive radical. SOD induces the dismutation of superoxide anion diminishing the production of peroxynitrites. The outcome of viral infections depends on viral and host factors. Host cells are thought to respond to viral infection by initiation of apoptotic cell death. There is mounting evidence that dengue virus can trigger the host cell to undergo apoptosis in a cell-dependent manner. During dengue virus infection, cell death is also modulated by the virulence of the infecting strains. In this study, the increased content of NO and MDA in monocytes during infection with all DENV types was accompanied by apoptosis, suggesting that NO was an apoptosis inducer during dengue infection. Dengue viruses generally induce apoptosis in mammalian cells in part, due to oxidative stress and the NO inducer apoptosis role has been reported. The induction of monocyte apoptosis by DENV has previously been shown. NO can inhibit dengue virus replication by inducing apoptosis, but, other viral inhibitory effects of NO have been reported. The viral inhibitory effects of NO have been reported in infections by members of different viral families including dengue virus, retrovirus and vesicular stomatitis virus. The capacity of all DENV types to induce oxidant/antioxidant effect in monocytes could be relevant during human dengue infection. In this regard, oxidative stress has been reported in dengue associated to severity of disease, thrombocytopenia and increased activity of glutathione peroxidase. In addition, increased plasma content of NO has also been reported, suggesting a role of NO in the oxidative stress during dengue infection. During this study, both neonatal and elderly monocytes had lower oxidant and antioxidant responses to dengue virus infection than young adult monocytes, suggesting a reduced oxidative response in both ends of life. However, the balance between oxidant/antioxidant effects resulted in lipid peroxidation and apoptosis regardless of monocyte source. Further investigation is required to determine the DENV-induced oxidative/antioxidative responses in monocytes from different sources after macrophage differentiation. # Supporting Information [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: NV JM GA. Performed the experiments: AL RM. Analyzed the data: NV JM MAM. Contributed reagents/materials/analysis tools: MAM. Wrote the paper: JM. [^3]: Current address: Center for Biologics Evaluation and Research, Food and Drug Administration, Bethesda, Maryland, United States of America

# Introduction Selective immunoglobulin A (IgA) deficiency is the most common primary immunodeficiency in Caucasians with a frequency of 1/600 in the general population. The current definition, established by the Pan-American Group for Immunodeficiency and the European Society for immunodeficiencies, defines the disorder as serum IgA levels below or equal to 0.07 g/L with normal serum levels of IgM and IgG in individuals of 4 years of age or older. Two-thirds individuals with IgA deficiency are clinically asymptomatic, but the defect may be associated with recurrent respiratory and gastrointestinal tract infections and selected autoimmune disorders. Celiac disease (CD) is a chronic immune-mediated disease which affects genetically susceptible individuals exposed to dietary gluten. It is characterized by intraepithelial lymphocytosis, crypt hyperplasia and villous atrophy on a gluten-containing diet and mucosal recovery on a gluten free diet (GFD). The reported prevalence of CD varies widely, ranging from 0.3% to 2.4% in Europe. However, the frequency of CD might be even higher as many individuals remain undiagnosed, probably due to the absence or atypical nature of symptoms. There is a strong genetic predisposition in CD, where the major histocompatibility complex (MHC) region contributes about 40% of disease susceptibility, especially the human leukocyte antigen (HLA)-DQ2/DQ8 alleles, are present in more than 95% of the patients. However, the presence of HLA- DQ2/DQ8 is a necessary, but not sufficient, prerequisite for development of CD. In studies on small cohorts, IgA deficient patients have been shown to have a 10 to 20-fold increased risk of developing CD. IgA deficiency has been reported to be strongly associated with the B8-DR3-DQ2 haplotype, which is the strongest recognized risk haplotype for CD as well. CD is also associated with another IgA deficiency associated haplotype, DR7-DQ2, albeit to a much lesser degree. Taken together, this might explain the overlap between CD and IgA deficiency, indicative of a common genetic background. The histological findings with small bowel mucosal villous atrophy and crypt hyperplasia have long been the gold standard for diagnosing CD. However, serological screening tests are valuable tools in selecting patients to undergo a diagnostic small bowel biopsy. Initially, IgA anti-native gliadin antibodies (AGA) were widely used but have in recent years been replaced by anti-endomysial antibodies (EMA) which were found to be a higher specific diagnostic marker. In 1997, tissue transglutaminase (tTG) was identified to be the auto-antigen targeted by EMA in patients with CD. As the diagnostic accuracy of IgA antibodies against recombinant human tTG is high and markedly increased levels are highly predictive of CD, specific anti-tTG antibody tests have been widely used for CD screening. Recently, it has even been suggested that a small bowel biopsy is not necessary for the diagnosis of pediatric CD in symptomatic children with high levels of IgA anti-tTG. Deamidated gliadin peptides (DGP) have been developed as the second generation anti-gliadin tests, including assessment of specific IgG antibodies. This is a considerable improvement as tests based on detection of IgA antibodies alone yield false negative results in IgA deficient patients. Several reports have confirmed the increased diagnostic specificity of using DGP instead of native gliadin although the source of antigen and the performance may differ among the available commercial DGP assays. However, Olen et al recently argued that anti- tTG is superior to anti-DGP for diagnosing CD since a combination of anti-tTG and anti-DGP did not provide a higher diagnostic accuracy than anti-tTG alone even in young children. Several screening studies have been performed to evaluate CD serological markers in relation to small bowel biopsy findings among IgA deficient individuals with suspected and/or confirmed CD. However, the number of patients has been limited and mainly restricted to children. In the present study, we recruited the hitherto, by far, largest cohort of adult IgA deficient individuals with suspected CD in order to estimate the prevalence of IgG antibodies against both tTG and DGP, and to investigate the correlation between different serological markers and the HLA types of the recruited individuals. # Materials and Methods ## Ethics statement The study was approved by Regionalaetikprövningsnämnden i Stockholm (Regional ethical review board in Stockholm) and Forkningsetikkommitte Syd, Karolinska Institutet (Research ethical committee South, Karolinska Institutet) which specifically encompassed the study presented in this manuscript (ethical permit numbers: 353/03; 356/03; 548/03; 04-261/4; 2008/1316-32; 2010/696-31/2; 2011/69-31/3). All adults have given their written informed consent to participate in the study. ## Study group Originally blood samples from 488,156 individuals were referred to seven Swedish immunology centers between 1998 and 2012 for serological CD screening including measurements of total serum IgA levels. Altogether 1,414 adults were identified to be IgA deficient and 741 of them who could be traced were asked to submit new blood samples for follow-up testing. Of these, 356 individuals accepted to participate in this study. In addition, 47 IgA deficient blood donors from the blood transfusion center at the Karolinska University Hospital Huddinge, Sweden were included as controls. In total, 74 out of 356 patients and 6 out of 47 blood donors (n = 80) had undergone a small bowel biopsy and 58 (72.5%) were diagnosed with CD (54 patients and 4 blood donors). ## Celiac disease serological tests Serum IgA levels of all IgA deficient individuals with suspected CD were determined by routine turbidimetric or nephelometric methods. Serum IgG anti-tTG and anti-DGP antibodies were analyzed in the newly collected serum samples at the clinical immunology unit at Linköping University Hospital, Sweden. Commercially available immunoassays for CD diagnostic testing were used according to the manufacturer instructions: (1) IgG antibodies against tTG (EliA™ Celikey®IgG, Phadia, Freiburg, Germany) were analyzed on ImmunoCAP 250 and the recommended cut-off \<7 EliA U/mL was used. (2) Two different ELISAs were performed for detection of IgG antibodies against DGP including Gliadin IgG II (Quanta Lite™, INOVA Diagnostics, San Diego, CA, USA) and GAF-3X (Euroimmun, Lübeck, Germany). The recommended cut-offs 20 AU/mL (INOVA) and 25 RU/mL (Euroimmun) were used. ## HLA typing HLA typing was performed at the clinical immunology laboratory, Karolinska University Hospital Huddinge, Sweden. DNA was extracted by standard procedures from whole blood samples of 355 IgA deficient individuals with suspected CD and 46 IgA deficient blood donors (2 samples were not available). All samples were subjected to whole genome amplification according to the manufacturer instructions (GenomePhi V2 DNA amplification kit, GE Healthcare, Buckinghamshire, UK) and subsequently HLA typed at the HLA-B, DR and DQ loci using polymerase chain reaction-single strand polymorphism (PCR-SSP). The kits used included HLA-B low resolution, DQ-DR SSP Combi Trays and DQB1\*03 high resolution kits (Olerup SSP AB, Stockholm, Sweden). ## Questionnaire A questionnaire regarding GFD was mailed to 359 IgA deficient individuals (44 individuals from one clinic were excluded due to ethical constraints). ## Statistics As our subgroups were of unequal sizes and the levels of different serological markers in different subgroups showed unequal variances, we used Welch's *t* test, which is an adaptation of Student's *t* test. For comparison of numbers of individuals in different groups, Chi-squared test was used. Statistical significance was accepted as *p*\<0.05. # Results ## Serological marker levels IgA deficiency (serum IgA≤0.07 g/L) was confirmed in all resampled individuals with suspected CD and blood donors. Out of the IgA deficient individuals with suspected or confirmed CD (n = 302 and 54 respectively), 67 (18.8%) were positive for IgG anti-tTG antibodies, 56 (15.7%) for IgG anti-DGP with the INOVA assay and 76 (21.3%) with the Euroimmun assay. In total, 38 of these individuals (10.7%) were positive for all three serological markers and 108 individuals were positive for at least one marker. In the IgA deficient blood donor group (n = 47), 4 (8.5%) were positive for IgG anti-tTG, 4 (8.5%) and 8 (17.0%) were positive for IgG anti-DGP using the INOVA assay and the Euroimmun assay respectively, whereas only 2 individuals (4.8%) were positive in all three assays. In total, significantly more samples were positive for IgG anti-DGP with the Euroimmun assay (n = 84) than with the INOVA assay (n = 60) (*p*\<0.001, Chi-squared *t* test). Interestingly, there was no individual with positive IgG anti-DGP (INOVA) alone. In our cohort, 39 of the 71 IgG anti-tTG positive individuals had very high antibody levels (\>10 times above cut-off) even including 16 patients reporting to be on a GFD. Out of the 58 individuals (54 patients and 4 blood donors) with biopsy confirmed CD, 34 (63%) were positive for IgG anti-tTG, 28 (52%) for IgG anti-tTG with the INOVA and 35 (65%) with the Euroimmun assay at the time of resampling. ## Genetic correlation Our cohort was divided into different subgroups based on HLA types, including those homozygous for DQ2 (DQ2,2), heterozygous for DQ2 (DQ2,X), those carrying DQ8 but not DQ2 (DQ8,X which includes one individual homozygous for DQ8) and non-DQ2/DQ8 alleles (DQX,X). The elevated antibody levels in different HLA subgroups were compared for each assay. Individuals homozygous and heterozygous for DQ2 represented the majority of samples with elevated antibody levels in all three assays. There was no significant difference of the antibody levels in homozygous/heterozygous DQ2 subgroups regarding IgG anti-tTG and anti-DGP with the INOVA assay (*p* = 0.151 and 0.054 respectively). The levels of IgG anti-DGP with the Euroimmun assay were significantly higher in the homozygous DQ2 subgroup as compared to the heterozygous DQ2 subgroup (*p* = 0.032, Welch's *t* test). Individuals carrying DQ8 or non-DQ2/DQ8 alleles showed markedly lower antibody levels than those carrying DQ2 with exception for the Euroimmun assay where there was no significant difference between the antibody levels in the non-DQ2/DQ8 (DQX,X) compared to the DQ2 subgroups (*p* = 0.114 and 0.807 respectively, Welch's *t* test). DNA samples from 69/71 individuals (66 patients and 3 blood donors) with positive IgG anti-tTG (with or without positive anti-DGP) were available for HLA typing and 68 (66 patients and 2 blood donors) were DQ2/DQ8 positive. Out of the 37 individuals that were only positive for anti-DGP, 5 carried neither DQ2 nor DQ8 and 4 of the latter were only positive for anti-DGP using the Euroimmun assay. The association between positive antibody assays and the common IgA deficiency risk haplotypes (DR3-DQ2, DR7-DQ2 and DR1-DQ5) was also analyzed regarding to having GFD or not. Most of the enrolled individuals carried at least one of the three major IgA deficiency risk haplotypes, mostly the DR3-DQ2 haplotype. ## Small bowel biopsy test and questionnaire results Since our 58 biopsy confirmed CD patients (54 from the patient group and 4 from the blood donor group) showed varying antibody levels, questionnaires were sent to the individuals for information on dietary restriction (GFD). Completed questionnaires were received from 261 out of 359 patients (6 individuals had died after re-sampling) showing that 68 reported to be on a GFD whereas 193 reported not. shows the distribution of IgG antibody levels among patients with/without GFD. Surprisingly, there was no significant difference between the two groups (*p* = 0.408, 0.938 and 0.476 respectively in three assays, Welch's *t* test). Twenty of the biopsy confirmed CD patients were positive in all IgG antibody assays, seven of whom reported to adhere to a GFD and seven not (one patient had died and 5 were not sent questionnaires due to ethical constraints). There were 16 confirmed CD patients negative for all markers and only four of them reported being on a GFD (4 were still on a gluten containing diet, 4 did not reply to the questionnaire and 4 were not sent questionnaires due to ethical constraints) as shown in. Fifty-four out of 58 biopsy verified CD patients carried HLA-DQ2/DQ8 (one DNA sample was not available). Interestingly, 45 of the 54 confirmed CD cases from the patient group carried at least one copy of the B8-DR3-DQ2 haplotype, which is a common IgA deficiency risk haplotype and contains the CD risk allele DQ2. Moreover, 16 of 45 patients carried two copies of the B8-DR3-DQ2 haplotype. Two of the 4 confirmed CD patients from the blood donor group carried one copy of the B8-DR3-DQ2 haplotype and one individual carried the other CD risk allele (DQ8). # Discussion IgA deficiency is the most common primary immunodeficiency in Caucasians. IgA deficient individuals demonstrate a higher incidence of silent forms of CD and a 10 to 20-fold increased risk of developing overt CD, thus highlighting the need to evaluate CD diagnostic antibodies of the IgG class rather than the IgA class. However, to date only a limited number of cases have been investigated for CD related IgG antibodies, mainly restricted to children. Therefore, we investigated the prevalence of IgG antibodies against both tTG and DGP, the latter including two commercial kits (INOVA and Euroimmun), in a large cohort of IgA deficient adults, detected by routine serological screening for CD. In IgA competent individuals, high IgA anti-tTG levels (\>10 times above cut- off) are strongly associated with CD as reported by Vermeersch *et al* as they found likelihood ratios of \>100 for CD among individuals with high levels of IgA anti-tTG. In children, a small bowel biopsy is no longer considered to be required for diagnosing CD provided that the individuals also carry CD risk HLA alleles and have symptoms suggestive of CD. However, whether this is true regarding high levels of IgG anti-tTG in individuals with IgA deficiency is still unknown. One striking finding in our cohort was that more than half of the IgG anti-tTG positive individuals had very high levels (\>10 times above cut- off), even including 16 patients reporting to be on a GFD. Already in 2003, Korponay-Szabo and coworkers noted that the decrease in IgG anti-tTG antibody levels was slow in CD patients with IgA deficiency and most of them were still positive after more than two or three years on a GFD, in contrast to IgA competent CD patients where levels decreased markedly or returned to normal for IgG anti-tTG/DGP antibodies after one year on a GFD. Recently, Chow et al also reported that only half of their CD patients with IgA deficiency had normalized their serology after having a GFD for a mean of 7.25 years. We observed that there was no difference in IgG antibody levels in patients with/without GFD. High levels of IgG anti-tTG and/or anti-DGP in patients who did not adhere to a GFD (n = 184) may reflect undiagnosed CD patients. However, 36 of 67 enrolled IgA deficient patients with suspected CD on a GFD were still positive for at least one marker, 27 of whom were positive for IgG anti-tTG antibodies and 16 tested positive in all three assays. These data are supported by a previous study showing persistent small bowel villous atrophy, defined as Marsh class 3, in 43% of CD patients when they were re-biopsied 0,5–5 years after diagnosis. The explanations for this could be that, despite being diagnosed with CD, these patients were not motivated for a strict GFD, presumably due to a combination of mild gastrointestinal symptoms and that adhering to GFD is expensive and difficult in a social context; second, consistent with previous publications, levels of IgG antibodies appear to decrease slowly in IgA deficient patients with CD despite being on a GFD for a long period. In our 27 patients who were positive for IgG anti-tTG antibodies and adhering to a GFD, 26 carried at least one copy of the B8-DR3-DQ2 haplotype and one did not carry the complete haplotype but still carried DR3-DQ2. This might be part of the immune-regulatory defect seen in IgA deficient individuals, as individuals carrying the B8-DR3-DQ2 haplotype have alterations in their immune responses, regardless of their IgA levels. Thus, it is still not clear if the slower decrease in IgG anti-tTG antibodies is directly due to IgA deficiency itself or its associated HLA haplotype. HLA types were obtained in 401 IgA deficient individuals (two samples were not available). Individuals carrying DQ2 were markedly overrepresented among the individuals who were positive for IgG antibodies to CD related antigens and also showed considerably higher IgG antibody levels as compared to other subgroups, which is in concordance with another study on IgA competent adult CD patients positive for IgA anti-EMA, suggesting that the genetic association between DQ2 and CD might be the same in IgA deficient and IgA sufficient adults. With exception for the Euroimmun assay, individuals carrying DQ8 not DQ2 showed very low IgG antibody levels, supporting the notion that the DQ2 allele itself may play a main role for the induction of these antibody specificities. In 2009, Prause et al claimed that the high accuracy of IgG anti-DGP (Euroimmun) might offer a chance to detect CD in IgA deficient patients without further diagnostic efforts as three IgA deficient children with negative IgA anti-EMA and anti-tTG, had high concentrations of IgG anti-tTG and anti-DGP (Euroimmun). Other studies have also shown that IgG anti-DGP antibodies using the Euroimmun assay were of value for diagnosing CD in IgA deficient children though the authors suggested that further studies were necessary to confirm the diagnostic accuracy of the assay in IgA deficiency. As shown in, significantly more samples were positive for IgG anti-DGP with the Euroimmun assay than with the INOVA assay and, six out of 20 individuals who tested only positive for IgG anti-DGP with the Euroimmun assay were DQ2/DQ8 negative (1 out of 7 DQ2/DQ8 negative patients was positive in both IgG anti-DGP assays). This implies that the diagnostic accuracy of IgG anti-DGP in the Euroimmun assay is questionable in IgA deficient adults. A 3-fold higher frequency of IgA deficiency was found in our cohort (approximately 1 in 200, including 1180 IgA deficient children) as compared to expected in the general population (1/600) and it is likely to reflect the higher prevalence of gastrointestinal problems, including CD, in IgA deficient individuals. In addition, a recent study showing that individuals with IgA deficiency were at higher relative mortality risks in the first 10–15 years after diagnosis, potentially associated with the higher mortality noted in patients with CD, underlines the need for a reliable diagnostic markers of the IgG class. To date, IgG anti-tTG antibody seems to be the most reliable marker and the importance of isolated presence of IgG anti-DGP is still unknown in IgA deficient cases. In our study, almost all cases with IgG anti-tTG positive levels also carried the CD risk alleles, that was not the case for individuals with an isolated presence of IgG anti-DGP, again indicating a rather lower diagnostic specificity. Approximately 500,000 individuals were initially screened for CD related antibodies and serum IgA levels in Sweden during 1998–2012, indicating that more than 5% of the Swedish population was suspected of having CD during this period, which is markedly higher than the published prevalence of CD in other European populations. Furthermore, this shows that serological CD screening is liberally performed and that awareness is high among physicians that IgA anti-tTG is a reliable marker for CD. Our results suggest that many individuals with IgA deficiency, who may have CD or latent CD could have escaped further investigations with IgG antibody screening and/or small bowel biopsy test and may therefore not be properly diagnosed. Thus, it seems not to be equally recognized well among clinicians, not specialized in gastroenterology, that the IgG anti-tTG antibody is a strong diagnostic CD marker in IgA deficient individuals, conferring a major risk to underdiagnosed CD in such cases. This assumption is supported by the fact that we have small bowel biopsy results in only 80 of 403 enrolled IgA deficient individuals. It is the laboratory's responsibility to screen for IgA deficiency in all samples referred for CD screening and to inform the referring clinicians that a negative result for CD markers of the IgA class is only valid provided that the individual does not have IgA deficiency. Löwbeer and Wallinder recently reported that testing total IgA levels in the screening of CD was unnecessary by using the EliA Celikey assay since their results showed that samples with IgA deficiency presented very low values in the IgA anti-tTG test. This is also consistent with our own experience when analyzing serum samples from 238 IgA deficiency individuals for IgA anti-tTG antibodies using the EliA Celikey assay since all tested sera gave very low responses (unpublished data). This procedure may facilitate the CD screening laboratory routine by reducing the number of samples needed to be analyzed for total serum IgA levels. Due to the ethical constraints, we were not be able to follow up all recruited individuals and information about small bowel biopsy results were only available in few IgA deficient subjects, limiting the solidity of our conclusion. However, our findings suggest that, although within the limitations of a retrospective questionnaire on dietary restrictions (261/359), the diagnostic values differ between different commercial IgG anti-DGP assays and many CD patients continue to eat gluten containing food. Furthermore, 4 of 6 our IgA deficient blood donors who had undergone small bowel biopsy were found to have CD, underlining the increased prevalence of silent CD in IgA deficient individuals. Thus, awareness of CD among IgA deficient individuals still needs to be increased in order to promote strict GFD, even in clinically silent cases, thereby preventing long term serious clinical consequences. We are grateful to all IgA deficient individuals who participated in this study, and technician Anne Hagman at the Department of Clinical Immunology and Transfusion Medicine, Linköping University, Sweden for performing the serological tests. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: NW CD LH. Performed the experiments: NW CD. Analyzed the data: NW CD AL. Contributed reagents/materials/analysis tools: LT KE BA JR LM-N JL. Wrote the paper: NW CD LH.

# Introduction Investigation of the association between socio-economic status (SES) and mental disorders has been a research focus since the 1930s. Results have confirmed a relationship between SES and various mental disorders (e.g., depression) in Western societies, leading to hypotheses about “social stress” or “social causation”. Depression commonly has a dynamic relationship with SES. There is a negative association at an individual level when surveys of SES have primarily used income as a variable; the lower the income, the greater the risk of depression. The mechanism may be explained absolutely by one’s ability to secure basic needs (e.g., in purchasing goods and services), including health, or relatively by implying social position and emotional satisfaction in terms of SES producing happiness. It may also imply the modality of higher income (compared with other modes of SES) having a substantial impact on psychological problems such as depression, mainly at low SES levels. However, SES factors such as education and employment have frequently been found associated with depression, as have other SES factors, as well as sociodemographic factors. Those *socio-economic-demographic* (SED) factors could be independently or dependently—directly or indirectly—associated with depression. There may also be negative societal perception or adverse socio- economic impacts if a person consults with a mental health professional or is hospitalized for a mental disorder in a society that stigmatizes such disorders. Additionally, those with low SES in developing countries may not be able afford medical treatment fees. Depression is the seventh most common cause of disability in Indonesia, with a high prevalence of 6.1%. To our knowledge, while some studies have investigated SED factors related to depression (commonly these are community-based studies), there are no reviews of those factors with regard to depression among hospital outpatients (hospital-based study) in Indonesia. Since the implementation of universal health insurance in Indonesia (named *Badan Penyelenggara Jaminan Sosial*), access to medical facilities has improved for people of low socioeconomic status. However, there may be risks of lost opportunity cost from undergoing medical treatment (e.g., because of time spent for hospitalization or because of prejudice toward patients with depression). It is therefore important to explore characteristics of patients at hospitals. # Materials and methods ## Study design and participants This study was conducted in the city of Makassar in South Sulawesi, Indonesia. The prevalence of depression in South Sulawesi Province is reported as 7.8% (95% confidence interval \[CI\]: 7.3%–8.4%), higher than the average nationwide prevalence (6.1%). This study uses a case–control study design, with the case group comprising 160 patients diagnosed with clinical depression at the subject hospital. The control group was recruited by visiting communities within a \~30-minute walk of the Hasanuddin University Hospital and sampled until the number of participants matched the number in the case group; these communities were located in residential areas near the hospital. The total sample was 320 (160 cases, 160 controls), with no significant difference in age (*p* = 0.089) or sex (*p* = 0.247) compositions between the two groups. ## Instrument and procedures The medical procedures and subsequent interviews were conducted at general hospitals in Makassar, involving clinical depression outpatients who regularly visit hospitals’ psychiatric departments (regular patients) or first-time/new patients. Psychiatrists diagnosed depression with reference to the third edition of the Indonesian mental disorder guidelines for classification and diagnosis (*Pedoman Penggolongan dan Diagnosa Gangguan Jiwa edisi 3*). Patients who had been diagnosed with depression (case) were then asked to undergo an interview. Psychiatrists took about 10–15 minutes to conduct face-to-face interview to each participant. The interview questionnaire collected information about age, sex, ethnicity, individual cash income in the preceding year, educational background (highest educational level completed), occupation, average daily hours of work, number of cohabitating family members, and others. The same questionnaire was given to the control group. All research was performed after obtaining written informed consent from each participant. This study was approved by the Ministry of Education and Culture (the former Ministry of Research, Technology and Higher Education), Ethics Committee of Medical Research, Indonesia (approval letter number: 01/H4.8.4.5.31/PP36-KOMETIK/2017 to Faculty of Medicine, Hasanuddin University and Hasanuddin University Hospital) and by the Kyoto University Graduate School and Faculty of Medicine, Ethics Committee (G1099 to Graduate School of Asian and African Area Studies, Kyoto University). All participants could refuse to answer any questionnaire item and could withdraw from the research at any time. All participants in both the case and control groups were provided with snacks and beverages, but no remuneration. ## Data analysis Data analyses comprised (1) multiple imputation methods for handling missing values, (2) univariate and bivariate analyses of the variables, and (3) multivariable model analyses. Missing data and refusals to answer are unavoidable; however, analyses without replacing missing values (i.e., excluding participants with missing values) yield misleading conclusions. Several statistical approaches have been proposed to counteract this. shows the proportions of missing values for the SED variables. The greatest amount of missing values (39.06%) was observed for individual cash income. Initially, we performed logical replacement by replacing missing cash income with 0 if the participant, for occupation, answered “unemployed” or answered “housewife” and worked \<1 hour per day. This slightly reduced the proportion to 30.94%. We then conducted multiple imputation analysis as follows. We regarded all missing values as neither caused by complete randomness (missing completely at random) nor by systematic bias (missing not at random), though they can be incorporated using partially observed variables (missing at random \[MAR\]). The multiple imputation analyses for MAR were based on a Bayesian approach; the overall association among all SED variables was simulated from the observed values through multiple imputation processes by using a fully conditional specification method. For example, we replaced missing cash income values for each occupation with random values that were generated with the same probabilistic distribution of the observed cash income for the respective occupation at each imputation. We replaced the missing values for occupation with random values in accordance with the same probabilistic distribution of the observed occupation. We used fraction of missing information (FMI) and relative efficiency (RE) score to indicate the output of five imputations was appropriate. Scores closest to 0 for FMI and 1 for RE indicated the best output. In this study, we mainly analyzed the average of five imputations as a full set of data, while for the purpose of transparency, we also showed the original data (with missing values) and the result of each imputation. For classifying ethnicities, the major ethnic groups in South Sulawesi Province (Bugis, Makassar, and Toraja) were first defined, and mixed ethnicity included any combination of the three primary groups. Other ethnic groups comprised an additional, “others,” group. We converted all ordinal and continuous data (e.g., education, income) to nominal data. Age and hours worked per day were classified for each quartile. We used a chi-square (χ²) test for detecting raw associations between the case or control and each variable; continuity correction was applied in the case of 2×2 cross-comparison tables. We also measured the strength of association symmetrically using Phi (φ) and Cramer’s (ν) test, and directionally using Goodman and Kruskal’s lambda (λ) or Goodman and Kruskal’s tau (τ) test as an alternative for proportional reduction in error (PRE) detection. In further investigation, we used the Mantel–Haenszel method to measure the estimated odds ratio (OR) toward the magnetic SED variable. We compared pooled crude OR with original and complete case analysis (CCA) for greater precision. Finally, we performed multiple logistic regression analysis including all variables in one model to confirm a variable’s effect after adjusting for all other variables. All data analysis was conducted using IBM SPSS for Windows, Version 26.0 (IBM Corp., Armonk, NY, USA). # Results First, we investigated the sociodemographic characteristics of the case group (patients with depression) compared with those of the control group and. No significant between-groups difference was found in age composition or sex. The ethnic composition differed between the groups especially in that the case group had fewer people of mixed ethnicity and more “others.” Among the socio-economic characteristics, only the income level differed between the groups, showing more with “high” income group and fewer with “middle” income for the case group. The strength of symmetric association for ethnicity or income for depression was \>0.20. Directionally, income contributed to PRE for depression at 18.1%. shows an association between income and patients with depression. Pooled data (OR = 1.723, \[95%CI = 1.076–2.758\]) and original and CCA data showed a significant association in high probability of depression for the high-income group. shows the association between high income and the case group, and other SED variables. The greater proportions of high-income patients with depression (case) were \>52 years old, female, of Makassar ethnicity, had middle-level education, were private sector workers, and were cohabitating with four or more family. shows the results of the multiple logistic regression analyses. Among all SED variables, high income (OR = 2.088 \[95%CI = 1.178–3.700\]), middle-level education (1.688 \[1.042–2.734\]), and cohabitating with four or more family members (1.632 \[1.025–2.597\]) were significant predictors of depression. # Discussion The results suggest that income has a substantial impact on depression. High- income status was related to a greater risk of depression then was lower or middle income, even after adjusting with other SEDs. This is despite many studies having suggested that high income does not correlate with a high risk for depression, such as in it facilitating the ability to manage and cope with financial stress and to foster individual and social well-being. This study had limitations in the frequency of missing values for cash income. Income has typically been a limitation to competing questionnaires, but exclusion either of participants with the missing values in the data or the income variable may cause substantial bias; accordingly, multiple imputation methods for missing values have been recommended. The results in the present study were consistent among the original data and each imputation; therefore, we can reasonably deem that the results from the pooled data of the five imputations were reliable. There are various reasons participants were reluctant and even embarrassed to inform of their income. We sought to avert that, though we did not entirely succeed. Some studies have shown similar experiences and it was validated by Mossakowski. Another limitation was the difficulty sampling control participants to match the case participants. We initially attempted to recruit control participants with identical neighborhood characteristics to the case sample. However, most individuals declined to participate because they lived too far away from the hospital. As an alternative approach, we recruited participants by visiting communities close to the hospital. Thus, the control participants may have had different neighborhood characteristics to the case participants. Although there were no significant differences in age or sex between the groups, the difference in the ethnicity may owe to sampling bias. Ethnicity only differed for mixed ethnicity or “others,” but not among the major ethnic groups (Bugis, Makassar, and Toraja). Our analyses regarded ethnicity as a confounding factor in the final model, rather than the explanatory variable. Despite the study’s potential limitations, we deemed the final result showing effects of income, education, and the number of cohabitating family members on patients with depression to be reliable. Since implementing a national social security system (*Sistem Jaminan Sosial Nasional* for health care in 2014 in Indonesia, all citizens acquired *de facto* no-cost primary care consultation at hospitals. However, accessing hospitals involves other costs, such as transportation. There is also an opportunity cost that citizens lose by visiting a hospital; e.g., day laborers will lose 1 day of income for every day they visit a hospital and ordinary employees with lower earned income hesitated to take leave for mental disorders. All the above factors might have led to the indication that patients with depression were more likely to have higher income. These findings raise a critical issue about how lower income people experience and express depression and common care-seeking pathways. Although few studies have examined this issue, symptoms of depression are often not recognized as a treatable medical condition among individuals with lower socioeconomic status and/or lower levels of education in Indonesia (Byron Good, personal communication). Presumably, some individuals in this group experience depression symptoms, but their condition may not be recognized as a treatable medical disease. In addition, depression may be expressed in local cultural and religious terms. Hence, people in such conditions may prefer to receive treatment from traditional or spiritual healers rather than going to the hospital. In some cases, restraints are still used for treating severe depression in this group. The bivariate analyses and multiple regression analyses also revealed that patients with depression most commonly had achieved a middle level of education and/or cohabitated with a larger number of family members. This may indicate that education could increase expectations toward achieving a certain income threshold in the community (a perception of economic position) posing a more substantial goal for patients with depression to achieve, thus creating a complex situation, as ten Kate et al. noted in the context of cultural entitlement, while also requiring more family members to bear the expenses and care for these patients. These can be regarded as exclusive characteristics regarding depression among hospital outpatients. Another possible interpretation involves social causation, in that high income itself is related to higher risk of depression. It should be noted that our income analysis used personal income, which reflects personal social and economic status. Unexpectedly, the current findings revealed that high personal income did not protect individuals from depression. Instead, high personal income may have caused depression independently, in accord with a phenomenon known as the Easterlin Paradox. Previous studies have reported that individuals may have a high income but still experience financial pressure and even hold high levels of debt because of loss-aversion, which causes them to experience hardship and lead to depression. Individuals’ inability to reach a certain income threshold within the social groups in which they live may fail to meet exceptionally high standards, which may be a particular problem when compared with what exists in other groups. This is also called “relative deprivation”. Hence, the implications of relative income related to the “fear of losing status” may be correlated with the occurrence of depression. The phenomenon described above occurs more commonly in Western societies. We argue this may be the effect of an “evolutionary mismatch” or “evolutionary cleft stick” phenomenon via modernity like that which takes place in Western societies (e.g., competition) and is already being mirrored in Indonesian society (e.g., Makassar society). In Indonesia, therefore, there is an association between income and depression, in which high relative income may impact the high risk of depression because of the effects of socio-cultural transformation. In the discussion above, we critically assessed possible explanations for the correlation between high income and depression, either as an exclusive determinant at the hospital due to the lack of access of lower income people to the hospital, or through relative social causation related to socio-cultural transition. Thus, this study proposes that the government of Indonesia reevaluate the effectiveness of hospital-based service in terms of (more) coverage of service and access for low- and middle-income households (e.g., cost, health insurance, procedures). Future research should examine hospital- related paradigms in society (i.e., lower or middle society) in depth and investigate why socio-cultural transformation causes some high income individuals to be at greater risk of depression. # Conclusions Cash income appears to be a determinant of depression in outpatients in hospitals in Indonesia. This study offers health policy implications toward enabling better hospital access and service for people with depression in low- or middle-income households, and recommendations for considering high income (and SED) as a determinant for high risk of depression occurrence, due to socio- culture transition. # Supporting information We deeply thank the medical team in the psychiatry department of Hasanuddin University Hospital for case (patients) collection, as well as the field contributors in control sampling. We also thank the Directorate General of Higher Education, Ministry of Education and Culture, Republic of Indonesia, which extended the opportunity, especially to Andi Agus Mumang as a doctoral student at Hasanuddin University, to engage in the Short-time International Exchange Program at the Graduate School of Asian and African Area Studies, Kyoto University through the sandwich-like program for doctoral students (named PKPI- PMDSU). We also thank Edanz Group ([www.edanzediting.com/ac](http://www.edanzediting.com/ac)) for editing a draft of this manuscript. 10.1371/journal.pone.0244108.r001 Decision Letter 0 Palazón-Bru Antonio Academic Editor 2020 Antonio Palazón-Bru This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 22 Sep 2020 PONE-D-20-17192 Socio-economic-demographic determinants of depression in Indonesia: A hospital- based study PLOS ONE Dear Dr. Mumang, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. Please submit your revised manuscript by Nov 06 2020 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <plosone@plos.org>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript: A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols> We look forward to receiving your revised manuscript. Kind regards, Antonio Palazón-Bru, PhD Academic Editor PLOS ONE Journal Requirements: When submitting your revision, we need you to address these additional requirements. 1\. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at <https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main _body.pdf> and <https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_titl e_authors_affiliations.pdf> 2\. Please include additional information regarding the survey or questionnaire used in the study and ensure that you have provided sufficient details that others could replicate the analyses. For instance, if you developed a questionnaire as part of this study and it is not under a copyright more restrictive than CC-BY, please include a copy, in both the original language and English, as Supporting Information. \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Yes \*\*\*\*\*\*\*\*\*\* 2\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes \*\*\*\*\*\*\*\*\*\* 3\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes \*\*\*\*\*\*\*\*\*\* 4\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes \*\*\*\*\*\*\*\*\*\* 5\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: This study compares a sample of persons being treated in hospital- based psychiatric outpatient clinics in Makassar, Indonesia, with a matched randomly selected population of persons living within a 30 minute walk of the university hospital. The study is important for two reasons. First, there is so little research on depression in Indonesia (the 4th most populous nation in the world), compared with studies of depression in most societies in the world, and compared with studies of psychotic illness in Indonesia. It is thus very welcome to find a study examining basic characteristics of persons being treated for major depressive disorder in a psychiatric outpatient setting Makassar. Second, the study is important because the most crucial finding of the study -- that persons being treated for depression in this sample has higher income than in the control population -- is unexpected and counterintuitive. It thus raises important questions for further study, suggesting in particular the need for further population-based studies in Indonesia, as well as studies of care- seeking patterns and access to care for persons with depressive illnesses. There is a great deal of focus on methodology, in particular methods for dealing with missing values (particularly for individual income), in this study. Overall, I leave it for others to discuss the statistical procedures. However, I do wonder why the study gathered information about individual income rather than household income. Because wealth and social status accrue to households, not individuals, particularly in Indonesia, there may be conflation of income and gender (although number of 'housewives' is quite low). One other methodological issue -- in selecting the comparison population. It seems that data were not collected about where the patient population resides in Makassar, and whether the neighborhood "within a 30 minute walk" is indeed a good comparison. We know that major cities in Indonesia are stratified by neighborhoods. It would be useful for the authors to add a note to indicate whether the neighborhoods around the university hospital have particular characteristics and whether most of the patients come from this same area. I appreciate the hypotheses the authors considered in explaining why this treated population has higher income than the comparison population. These are important hypotheses that should indeed generate future studies. And the suggestion that the public health services need to focus more attention on identifying and treating depression among lower income groups is an important one -- and suggests further studies of how best to do this. This then raises one critical issue that should be added to the discussion session. In what ways do persons from lower SES groups experience and express major depression, and what are the common pathways for care-seeking among these groups? There is remarkably little research on this topic in Indonesia, compared with many parts of the world. However, as an anthropologist who has worked on issues of mental illness in Indonesia for years, I would say that symptoms of depression are often not recognized as a treatable medical condition among many persons in lower socioeconomic and lower educated groups. To the extent they are viewed as illness, they are often treated in the popular and religious healing sectors rather than conceived as matters for treatment through medical settings. The primary health care system has recently been mandated to provide care for persons with severe mental illness in Indonesia, but there is no mandate to recognize and treat depression. Rates of recorded depression in medical records in the public primary health clinics are extremely low. My point is simply that in the discussion session, where the authors consider critical hypotheses for explaining an important unexpected finding, this set of issues should be recognized and put on the agenda for future research. Overall, I find this an important paper that with minor revisions should be published. \*\*\*\*\*\*\*\*\*\* 6\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: **Yes: **Prof. Byron Good, Dept of Global Health and Social Medicine, Harvard Medical School \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <figures@plos.org>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0244108.r002 Author response to Decision Letter 0 24 Oct 2020 Response to Editor’s comments: 1\. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at <https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main _body.pdf> and <https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_titl e_authors_affiliations.pdf> Response: We carefully checked our manuscript and made corresponding changes to meet the journal’s style requirements. 2\. Please include additional information regarding the survey or questionnaire used in the study and ensure that you have provided sufficient details that others could replicate the analyses. For instance, if you developed a questionnaire as part of this study and it is not under a copyright more restrictive than CC-BY, please include a copy, in both the original language and English, as Supporting Information Response: We have included our interview form in both languages as supporting information, labelled as S5 in the Appendix. Response to Reviewer’s comments: Part 1. This study compares a sample of persons being treated in hospital-based psychiatric outpatient clinics in Makassar, Indonesia, with a matched randomly selected population of persons living within a 30 minute walk of the university hospital. The study is important for two reasons. First, there is so little research on depression in Indonesia (the 4th most populous nation in the world), compared with studies of depression in most societies in the world, and compared with studies of psychotic illness in Indonesia. It is thus very welcome to find a study examining basic characteristics of persons being treated for major depressive disorder in a psychiatric outpatient setting Makassar. Second, the study is important because the most crucial finding of the study -- that persons being treated for depression in this sample has higher income than in the control population -- is unexpected and counterintuitive. It thus raises important questions for further study, suggesting in particular the need for further population-based studies in Indonesia, as well as studies of care- seeking patterns and access to care for persons with depressive illnesses (paragraph 1). I appreciate the hypotheses the authors considered in explaining why this treated population has higher income than the comparison population. These are important hypotheses that should indeed generate future studies. And the suggestion that the public health services need to focus more attention on identifying and treating depression among lower income groups is an important one -- and suggests further studies of how best to do this (paragraph 3). Response: We thank the reviewer for their positive comments regarding our research. Part 2. There is a great deal of focus on methodology, in particular methods for dealing with missing values (particularly for individual income), in this study. Overall, I leave it for others to discuss the statistical procedures. However, I do wonder why the study gathered information about individual income rather than household income. Because wealth and social status accrue to households, not individuals, particularly in Indonesia, there may be conflation of income and gender (although number of 'housewives' is quite low). Response: Household income is appropriate for measuring an individual’s household wealth and social status in society, but does not necessarily reflect their personal social status. In addition, household income is not always known by all household members (e.g., a patient with depression patient may not be aware of the income of other household members). Personal income is known to influence the specific opinions and decisions of individuals (Kuhn, 2019)\*. According to Ross and Huber (1985)\*\* : “Personal earnings, by adding to family income, decrease economic hardship and depression. But personal earnings, independently of their purely financial aspects, may also affect depression. Family (or households) income, from whatever source – a spouse's earnings, social security, or public assistance – is important to psychological well-being because it allows one to pay the bills, and feed, clothe, and care for the health of one’s family. But the money one earns oneself may affect well-being in other ways (p. 315)”. In addition, the introduction section of our manuscript describes a similar issue: “The mechanism (of income cause depression) may be explained absolutely by one’s ability to secure basic needs (e.g., in purchasing goods and services), including health, or relatively by implying social position and emotional satisfaction in terms of SES producing happiness (p. 3; line 59-62)”. Thus, we felt that it was better to gather information about personal income than household income to understand better how people think or decide (absolute vs. relative), and how their income affects their mental health. However, we thank the reviewer for raising this point, and we apologize that we did not describe this issue clearly in the previous version of the manuscript. We have addressed this issue in more depth in the revised discussion section (p. 16-17; line 271-278). \*Kuhn, U. (2019). Measurement of income in surveys. FORS Guide No. 02, Version 1.0. Lausanne: Swiss Centre of Expertise in the Social Sciences FORS. doi:10.24449/FG-2019- 00002 \*\*Ross CE, Huber J. Hardship and depression. J Health Soc Behav. 1985;26(4):312–27. Availabe at: <https://www.jstor.org/stable/2136655> Part 3. One other methodological issue -- in selecting the comparison population. It seems that data were not collected about where the patient population resides in Makassar, and whether the neighborhood "within a 30 minute walk" is indeed a good comparison. We know that major cities in Indonesia are stratified by neighborhoods. It would be useful for the authors to add a note to indicate whether the neighborhoods around the university hospital have particular characteristics and whether most of the patients come from this same area. Response: In the discussion section, we described the difficulty of matching case and control participants as a limitation. However, in the previous version of the manuscript, we did not explain why we chose participants from communities “within a 30-minute walk” for comparison with the case participants, which represents a potential limitation of the current study. We initially attempted to recruit control participants in a way that closely matched the neighborhood characteristics of the case group; however, most individuals refused to participate because of difficulties related to distance. As an alternative method, we searched communities that were located close to the hospital, which we assumed would have similar characteristics to the case group. Nevertheless, we agree that this issue should be clearly discussed as a limitation in the manuscript. Thus, we have added an explanation to address this issue in the revised discussion section (p. 14-15; line 232-236). Part 4. This then raises one critical issue that should be added to the discussion session. In what ways do persons from lower SES groups experience and express major depression, and what are the common pathways for care-seeking among these groups? There is remarkably little research on this topic in Indonesia, compared with many parts of the world. However, as an anthropologist who has worked on issues of mental illness in Indonesia for years, I would say that symptoms of depression are often not recognized as a treatable medical condition among many persons in lower socioeconomic and lower educated groups. To the extent they are viewed as illness, they are often treated in the popular and religious healing sectors rather than conceived as matters for treatment through medical settings. The primary health care system has recently been mandated to provide care for persons with severe mental illness in Indonesia, but there is no mandate to recognize and treat depression. Rates of recorded depression in medical records in the public primary health clinics are extremely low. My point is simply that in the discussion session, where the authors consider critical hypotheses for explaining an important unexpected finding, this set of issues should be recognized and put on the agenda for future research. Overall, I find this an important paper that with minor revisions should be published. Response: We thank the reviewer for raising this critical issue about how people with lower incomes experience and express depression, and their common care- seeking pathways. We agree with the reviewer’s suggestion regarding the possible explanations of this issue, and have elaborated on this point in the discussion section (p. 15-16; line 251-260). Furthermore, we realize that it is important to consider a critical hypothesis for explaining this important and unexpected finding. Therefore, we modified our discussion of this point at the end of the discussion section. With Professor Good’s permission, we would like to cite the reviewer’s comment itself as personal communication with Professor Byron Good, because we feel that this insight from an anthropological perspective is valuable. 10.1371/journal.pone.0244108.r003 Decision Letter 1 Palazón-Bru Antonio Academic Editor 2020 Antonio Palazón-Bru This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 3 Dec 2020 Socio-economic-demographic determinants of depression in Indonesia: A hospital- based study PONE-D-20-17192R1 Dear Dr. Mumang, We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements. Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication. An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at <http://www.editorialmanager.com/pone/>, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to- date. If you have any billing related questions, please contact our Author Billing department directly at <authorbilling@plos.org>. If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact <onepress@plos.org>. Kind regards, Antonio Palazón-Bru, PhD Academic Editor PLOS ONE Additional Editor Comments (optional): Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation. Reviewer \#1: All comments have been addressed \*\*\*\*\*\*\*\*\*\* 2\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Yes \*\*\*\*\*\*\*\*\*\* 3\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: I Don't Know \*\*\*\*\*\*\*\*\*\* 4\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes \*\*\*\*\*\*\*\*\*\* 5\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes \*\*\*\*\*\*\*\*\*\* 6\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: The authors have responded appropriately to my questions. Overall, I continue to believe this is an important study and should be published. One very small item: On pp 2-3, lines 54-56, the authors say that depression is almost non-existent in hunter gatherer or 'traditionally collectivist societies'. I suggest that either the authors add a reference to this, or simply omit this sentence. It is not important to the study here, and is a matter of some controversy. In particular, I would say that the term 'traditionally collectivist societies' would require definition and referencing to understand what is being claimed. This is not important -- not important to the study, and not a critical matter. However, I would just say the matter is controversial and probably better not to be stated as though it is a known fact. I am happy to have my name referenced as 'personal communication' concerning the issue of depression not being viewed as a treatable medical condition among many Indonesians, which in turn has a strong effect on care-seeking patterns (and in particular whether they would seek care for such a condition in a hospital outpatient clinic). On p 16, lines 257-258, I would not use the phrase 'cultural context of animism' (whatever that means). Many persons who are devout Muslims seek help from a variety of healers, including Islamic healers, for symptoms which might be diagnosed as depression (though being a devout Muslim and 'animist' may be quite compatible). I would simply say 'may be expressed in local cultural and religious terms'. Final note: there are times that the authors write as though they have found prevalence rates of depression to be higher among persons with higher income (e.g. in lines 283-289) in Makassar, Indonesia. I believe the findings indicate a link between rates of depression and higher income higher among persons who seek care in a hospital outpatient clinic. What this method cannot tell us is whether this is true of the population in general, or is the result of care- seeking, in particular differential patterns of care-seeking among those with higher and lower incomes. Running throughout the paper is a general suggestion that this study suggests that higher income may cause higher rates of depression in Makassar, without the qualification that the findings may be accounted for by either care-seeking patterns or differential risk for depression among individuals with higher income in the population. If I am correct in my interpretation of the findings, and if the authors agree, they may want to examine the language in the paper that at times seems to suggest this is true of the population at large rather than of a particular clinical sample. None of this lessens my enthusiasm for publication of this paper. \*\*\*\*\*\*\*\*\*\* 7\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No 10.1371/journal.pone.0244108.r004 Acceptance letter Palazón-Bru Antonio Academic Editor 2020 Antonio Palazón-Bru This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 7 Dec 2020 PONE-D-20-17192R1 Socio-economic-demographic determinants of depression in Indonesia: A hospital- based study Dear Dr. Mumang: I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department. If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact <onepress@plos.org>. If we can help with anything else, please email us at <plosone@plos.org>. Thank you for submitting your work to PLOS ONE and supporting open access. Kind regards, PLOS ONE Editorial Office Staff on behalf of Dr. Antonio Palazón-Bru Academic Editor PLOS ONE [^1]: The authors have declared that no competing interests exist.

# Introduction Rabies is a viral infection that infects all mammals. Estimations on the global burden of rabies indicate a health impact of 59,000 human deaths per year, a loss of 3.7 million DALYs per year and about 8.6 billion USD of economic losses mainly due to premature deaths (productivity losses) and post-exposure treatment (PET) costs. Most of the human death cases occur in Asian and African countries. Of these countries, Ethiopia is one of the worst affected, with domestic dogs being the major sources of the infection to humans. Dog management is, however, poor and anti-rabies dog vaccination is generally limited to a small number of dogs found in urban settings. The large dog population size in combination with poor dog management contributes to a high endemicity of canine rabies in Ethiopia. Basic information on the health and economic burden, e.g., on the amount of lives lost due to rabies and on the costs of treatment amongst those at risk, is crucial in the development of sustainable control programs. In most rabies endemic countries, reliable reports of incidence data on rabies and rabies exposure are lacking. Official reports generally underestimate the true number of human rabies cases and hence the true burden. For instance, in sub-Sahara African countries like Tanzania, the incidence of human rabies predicted on the basis of active surveillance data on bite incidences was up to 100 times larger than the officially recorded number of deaths. In Ethiopia, the national annual estimates from official reports indicate 12 exposure cases per 100,000 population and 1.6 rabies deaths per 100,000 population. However, the actual numbers are expected to be higher as many cases are not reported. The reliability of the reported incidence data is expected to differ among the regions in Ethiopia due to geographical as well as cultural differences. For instance, in rural Ethiopia, individuals who are exposed to rabies often prefer to see traditional healers for the diagnosis and treatment of the disease because of cultural background, lack of knowledge or limited accessibility to medical treatment. These widespread traditional practices of handling rabies cases might interfere with medical treatment seeking practice, resulting in an underreporting of the actual number of rabies cases and its related health burden. To fill the disparities between officially recorded and likely occurring rabies cases, researchers have applied approaches like extensive animal bite case searching, and predictive modeling. Rabies in exposed humans is preventable when an effective post-exposure treatment (PET) is applied immediately after exposure which includes post- exposure prophylaxis (PEP), wound washing, antibiotic and tetanus antitoxin administration. With respect to the applied PEP vaccine WHO recommends to use inactivated rabies vaccines produced in cell culture or embryonated eggs (CCEEV). In Ethiopia, the PEP vaccine used to treat most of the rabies exposures is produced from nervous tissue composed of rabies virus-infected sheep or goat brain inactivated with phenol and also called Nervous Tissue Vaccine (NTV). This type of vaccine is less immunogenic than the CCEEV and known for its fatality rate and disability rate due to the occurrence of severe and sometimes fatal allergic encephalomyelitic reactions. The health burden of rabies in Ethiopia, therefore, not only includes years of life lost due to the premature death following upon infection but also years of life lost due to the premature death following vaccination by NTV and years of life lived with a disability as a consequence of neurological complications, following NTV administration. At present, although rabies control associated working groups have been established across the regions of Ethiopia, there is no national strategic plan to clearly define targets for rabies control. The relative burden of rabies compared to other neglected tropical diseases is also not known. In countries like Ethiopia, where financial resources are limited, there is a need for quantitative data on the costs of certain diseases to support intervention prioritization. Moreover, quantification of the health burden enhances the understanding of its long term effects and of the comparative advantages of different levels of treatment and prevention. Therefore, in this study, we aim to estimate the status of rabies in Ethiopia by assessing its health burden and associated PET costs using data obtained by an extensive bite case search. # Materials and methods ## Study design A retrospective study was conducted to assess the burden of rabies by collecting data on the incidence of human rabies exposure over the period of one year (September 2013 to August 2014) through an extensive animal bite case search. Animal bite victims were traced using data collected from the recorded cases at health centers as well as by information obtained from questioning the local community to trace unregistered bite cases. After the tracing, victims (or their family) were contacted and questioned about their use of PET, the health impact of the bite, the incurred costs and on the behavioral manifestation of the biting animal. Based on these data, the public health burden of rabies as well as the costs incurred because of the applied PET were estimated. ## Description of the study area This study was conducted in three purposively preselected districts of Ethiopia, i.e. Bishoftu, Lemuna-bilbilo and Yabelo to account for the geographical variation within the country. Bishoftu is an urban district located at 45 km South East of Addis Ababa, the capital, with a relatively good infrastructure and health centers available within a reasonably short distance. Lemuna-bilbilo is a rural district located in the central highlands of Ethiopia, where most people live on a mixed crop and livestock farming system. Yabelo is a rural district in the lowlands of southern Ethiopia, where the majority of the people are pastoralists keeping livestock for their living. Both rural districts have poor road and health facility coverage. Bishoftu, Lemuna-bilbilo and Yabelo contain, respectively, 140000, 187000 and 101000 inhabitants, located in, respectively, 9, 27 and 25 villages/administrative divisions. Each district has one or more public health centers delivering medical services including PET and every village within a district has local health extension workers. ## Data collection A list of animal bite victims registered by the districts’ health centers for the period of one year between September 2013 and August 2014 was obtained. Animal bite incidences were claimed in 55 out of the total of 61 villages in the three districts. In September 2014, local health extension workers of all villages, including those from the 6 villages without any registered suspected animal bite cases, were trained in techniques of detailed questionnaire interviews as described by Hampson et al.. After this training, the extension workers contacted and interviewed all registered dog bite victims using a structured questionnaire. In cases where the victim was deceased or a child, an adult family member was interviewed. The health extension workers also performed an extensive search for non-registered animal bite victims by asking the contacted victims whether they knew other people that were bitten but did not visit a health center. In Ethiopia, most neighborhoods are socially close to each other and are therefore well informed about events happening to households in their village. Something serious as suspected rabid animal bite cases would, therefore, be known within the community. The indicated non-registered bite victims were subsequently contacted and interviewed in the same manner. The questionnaire in English as well as translated to the local language (Oromifa) is attached as supplementary information. ## Questionnaire survey The bite victims were interviewed using a structured questionnaire on: 1) demographic features including age and gender, 2) body part bitten, 3) characteristics of the dog that had bitten them to reflect its behavioural manifestation and ownership (either own, neighbours’ or unknown), 4) whether the victim visited a health centre and received PET, 5) whether the person died (within a month after exposure) or survived and 6) the expenditures incurred related to treatment of the bite. The behavioral manifestation and clinical signs of the biting dogs as described by the bite victims were subsequently used to categorize the dog as potentially rabid or non-rabid based on the six- criteria method of Tepsumethanon et al.. The same procedure was followed in interviewing the non-registered bite victims or their families. To check whether the bite actually occurred during the study period of September 2013 and August 2014, cross checking questions were asked by referring the incidence to specific verifiable (family) events. ## Statement of ethical clearance and informed consent The study protocol and consent procedure were approved by the scientific and ethical review committee of the Ethiopian Public Health Institute (Reference No. EPHI-6-13-824) (Ethical clearance approval letter is attached as Supplementary information). An oral consent procedure was applied as illiteracy was expected among part of the respondent. Potential respondents (bite victims or their family members) were informed on the purposes of the research, the procedure followed during the interview and on the subsequent use of their responses. It was emphasized that their participation was completely voluntary and that they could stop with the interview at any time. All respondents approached agreed verbally to participate in the conducted interviews. Their consent was recorded (marked) on the questionnaire paper, which was used to register the answers of a respondent. Permission to conduct interviews was also obtained from the district village health posts and administrative officials in all the study areas. ## Estimating the burden of rabies The health and economic burden of rabies was estimated based on the cost classification of Jo with the PET costs reflecting the direct costs, consisting of healthcare and non-healthcare costs expressed in monetary terms, and indirect costs in terms of DALYs as shown in. ## Estimating post-exposure treatment costs The costs of rabies post-exposure treatment were classified into two direct costs categories: 1) healthcare costs reflected by the expenditures for rabies vaccine, antibiotics, tetanus immunizations and disinfection (but not for rabies immunoglobulin as this was not used in the study areas) and 2) non-healthcare costs as the expenses for transport, food, and accommodation, while seeking PET. Costs were estimated given the interview responses of the bite victims on the number of PEP doses received, the number of visits to the health centres to receive PEP (in case the patient or their parents was not sure or forgot the number of doses we asked the health centres) expenditures during the medical treatment related to transportation, accommodation, food and communication, and the time spent by themselves and their caregivers (if applicable) to account for productivity losses while seeking treatment. For patients older than 15 years, the productivity losses while seeking treatment were valued in monetary terms using the GDP per capita daily income. When patients were accompanied during their treatment by an adult caregiver (older than 15 years), related productivity losses were also captured. Considering 568 USD as the Ethiopian average income per capita, we valued the daily average income at 1.6 USD per day. All the costs collected were in Ethiopian birr (ETB) and later converted to US dollars (USD) using the average annual exchange rate for the study period published by the National Bank of Ethiopia (1 USD = 19.22 ETB). For the estimation of PET costs, we considered PET costs per dog bite irrespective of the rabid status of biting dog and PET costs per sufficient treatment. The average PET costs per dog bite were defined as the sum of health care and non-health care costs including the opportunity cost of time spent while seeking for treatment divided by the total number of bite victims, including those who did not seek treatment. A complete PEP dose of NTV consists of 17 doses of vaccine administered consecutively for the first 14 days, with the remaining 3 doses at intervals of 10 days i.e. at day 24, 34 and 44. A treatment consisting of at least 14 of the 17 PEP doses was considered to be "sufficient", while any treatment consisting of less than 14 doses was regarded to be "insufficient", based on the minimum required number of doses to produce the neutralizing antibody level. Consequently, the average costs per sufficient treatment only accounted for the costs of patients who successfully received the minimum recommended doses (at least 14 out of the 17 doses) of PEP. ## Estimating the health burden of the disease We adopted the disability-adjusted life years (DALYs) estimation method developed by the World Health Organization to calculate the rabies health burden. In accordance, DALY is the sum of years of life lost (YLL) due to rabies and adverse effects of NTV and of years lived with disability (YLD) due to adverse effects of NTV. <img src="info:doi/10.1371/journal.pone.0192313.e001" id="pone.0192313.e001g" /> D A L Y = Y L L \+ Y L D = ∑ i = 1 18 ( ( N d e a t h i \* e i ) \+ ( N d i s i \* t \* D W ) ) YLL is calculated as the number of human deaths (*Ndeath*) within age category (*i*) multiplied by the life expectancy (*e*) of the concerned age category (*i*), cumulated over all age categories. Eighteen age groups with a 5 years age interval from 0 to 85 years and above were considered (*i* = 1…18) according to the age classification of WHO life expectancy table. The expected years of life lost at death were derived from the standard life table as given by WHO using the projected frontier life expectancy of 2050. Similarly, YLD is calculated by multiplying the total number of disability cases (*Ndis*) of the concerned age group (*i)*, with the duration of the disability (*t* in years) resulting from the NTV vaccine and its corresponding disability weight (*DW*), cumulated over the 18 age categories. Computation was performed by the WHO programmed spreadsheet template for DALYs estimations. As the most recent global health estimates do not account for age weighting or time discounting, the baseline DALY estimation was done without these social weighting functions. The YLL due to the disease were estimated from the number of deaths captured from the field survey. The disability (YLD) due to the disease itself was considered insignificant as the disease is very acute, leading to death once it develops. The field survey primarily indicated the number of death due to rabies disease as specifically asked for. It was, however, not possible to derive a direct DALY estimation due to the adverse effect of NTV because the time lag between NTV administration and occurrence of adverse effects could have been longer than the duration of the field survey and because of the difficulty to attribute health problems specifically to the NTV application. As a consequence, the health burden due to NTV was indirectly derived from the number of persons receiving PEP corrected for the number of deaths due to rabies and estimated likelihoods of adverse effects based on the study of Knobel et al.. YLL was derived from the number of patients who got at least one dose of NTV PEP, multiplied by the expected rate of neurological complications (0.4 out of 100 patients receiving NTV) and case fatality rate (17 out of 100 cases of neurological complications). The YLD due to NTV was calculated by multiplying the number of patients who got at least one dose of NTV PEP by the expected rate of neurological complications (0.4 out of 100 patients receiving NTV) and corresponding disability weight (DW = 0.613) for a most likely duration of disability of 8 days (t = 0.022 years). ## Extrapolation districts estimates to national level The health and economic burden of rabies was extrapolated to national scale by accounting for the proportion of the human population living in urban areas and rural areas. Of the total Ethiopian population of 95.7 million only 19% lives in urban areas like Bishoftu. Information on the distribution of people among rural high and low land is lacking; hence the incidence estimates of both rural districts, Lemina-bilbilo and Yabelo, were used to obtain a general estimate for the rural area where 81% of the people are living. Under the assumption of a random occurence, stochastic estimates of the true incidence rates were estimated using the Poisson confidence interval. Given these confidence intervals, the sensitivity of the estimations on average health loss and treatment costs was evaluated. ## Statistical analysis Descriptive statistics were used to summarize proportions of rabid suspected dog bite cases per age category, health center visits and sufficient doses of PEP receiving rates. Chi-square test and analysis of variance (ANOVA) were used to compare statistical difference between categorical variables and between group means. Tukey post hoc multiple testing for significance was used for additional exploration of the differences among means in cases of three or more group for which means were significantly different from each other. The statistical analysis was done in R statistical software (R 3.3.2). # Results ## Rabies exposure A total of 655 bite victims was traced and interviewed with 210 cases in Bishoftu (urban), 262 cases in Lemuna-bilbilo (rural highland) and 183 cases in Yabelo (rural lowland) districts. From this total, 23 cases originated from bites by animals other than dogs, i.e. cat (2), cattle (7) donkey (7), horse (1), wild animal/rodents (3), and human (3). These cases were excluded from further analysis, as animal other than dogs contribute less than 4% of the total bites. Moreover, no proven means of diagnosing rabies in these animals without laboratory confirmation were available as for dogs. Therefore, we focused on the remaining 96.5% cases originating from dog bites. From the total number of dog bites, 208 (95.7%), 250 (61.6%) and 174 (56.3%) were reported to the health centers from, respectively, Bishoftu, Lemuna- bilbilo, and Yabelo. The remaining cases did not seek medical attention and were identified through contact trace search. Based on the six- evaluation criterion of a biting dog, 189, 189 and 87 cases in the respective districts were found to be potentially caused by a rabid dog, indicating that on average 73.6% of the biting dogs were expected to be potentially rabid. schematically presents the number of dog bite victims who visited a health center (450) and those that did not visit health center (182) as identified through the ‘‘snow balling” exercise during the contact tracing. All (632) bite victims were interviewed using a structured questionnaire. Out of these 632, according to our six evaluation criteria, 465 (73.6%) were bitten by a potentially suspected rabid dog of which 360 victims visited a health center and 105 did not visit a health center and consequently did not receive medical care. Victims who visited health centers received either a sufficient number of PEP doses or an insufficient number of PEP doses or no PEP at all. Similarly, victims who did not visit a health center either went to traditional healers or nowhere. All the 12 deaths resulted from bites from dogs suspected to be potentially rabid. The table below the schematic diagram presents the types of costs considered under each action taken (Direct health cost related to modern PET (post-exposure treatment) and non-PET (traditional/spiritual healers), Direct non-health care costs, and indirect costs in terms of DALY loss (Years of Life Lost (YLL) due to rabies and or adverse effect of the vaccine) and Years Lived with Disability (YLD) due to adverse effect of the vaccine). When considering only rabid suspected dog bite cases, our data suggest that per 100,000 population, annually 135, 101 and 86 were exposed to potentially rabid dog cases in Bishoftu, Lemuna-bilbilo and Yabelo, respectively. Using chi-square test, we found a significant difference (p\<0.01) in the incidence rate between the urban and rural districts. However, there was no significant difference between the highland (Lemuna-bilbilo) and lowland (Yabelo) rural districts (p\>0.05). The overall rabid suspected dog bite exposure ratio of male to female was 0.56:0.44. There were no statistical differences between the gender exposure rates in the urban area (Bishoftu), but significantly higher (p\<0.01) exposure rates were observed in males than in females in both rural districts (p\<0.01). The urban district (Bishoftu) has the highest rate of bite victims visiting a health center and receiving sufficient PEP compared to the rural districts (Lemuna-bilbilo and Yabelo). The proportion of rabies suspected dog bitten victims visiting health center was 97% in Bishoftu, 64% in Lemuna-bilbilo and 66% Yabelo. The proportion of suspected rabid dog bite victims receiving sufficient PEP was 78% in Bishoftu, 50% in Lemuna-bilbilo and 29% Yabelo. Conversely, in the same district order, 2.6%, 35.5% and 31% of the rabies suspected dog bite victims visited traditional/spiritual healers. Rabies-related human death cases compared to the number of total dog bite cases were higher in rural areas than in the urban area. Out of the 12 deaths counted during the survey, one was from Bishoftu, eight were from Lemuna-bilbilo and three were from Yabelo rural districts. Age wise, the exposure to rabies suspected dogs in Bishoftu and Lemuna-bilbilo was highest in the 15–29 years age category and in Yabelo in the 5–14 years age category. Very young children and elderly people, i.e. 0–4 years of age and older than 60 years of age, were the least exposed age categories in all the districts. There was no significant difference among age groups across districts regarding the proportion of bite victims that received sufficient treatment (p\>0.05). Overall, the age category of 15–29 was the group with the highest proportion of bite victims visiting a health center (85%) and receiving sufficient PEP (67%). In terms of visiting a health center and receiving sufficient PEP, the variations between age categories within district were significant in the rural districts (p\<0.05) but not in the urban area. An overview of bite cases and deaths by bite site and age of patient is given in. Out of the 465 rabies suspected dog bite cases, 6% occurred in the head/neck region, 19% on arms/hands, 70% on the legs and 6% on the trunk. Overall, proportions of RSDB cases were significantly different across body parts bitten with the highest proportion of bites on the legs and lowest proportion on the trunk. Based on age category, these proportions demonstrated a similar pattern except that for the age category beyond 15–29 years with the lowest proportions of bites on the head/neck part of the body. Of the 12 recorded deaths, 75% occurred in victims being bitten on the legs. The two death cases following from a bite to the head/neck and the arms/hands received a sufficient number of PEP doses. Three of the deaths cases following from a dog bite on the trunk or the leg received an insufficient number of PEP doses, while the rest of the death cases did not receive any vaccine. ## Rabies burden ### Post-exposure treatment (PET) costs The average PET costs per bite case were estimated at 21.3, 19.1 and 22.6 USD in Bishoftu, Lemuna-bilbilo and Yabelo, respectively. These estimates included health care and non-health care costs of all 632 dog bite cases which account for all type of treatment received (NTV, traditional or none) and did not significantly differ across the three districts (p = 0.12). The average total PET costs for those victims who received sufficient PEP doses of NTV were 23.4, 31.5 and 40.1 USD in the same district order. For these average costs per sufficient treatment, significant statistical differences were observed among the three districts. indicates the distribution of health care costs and non-health care costs per dog bite case and per sufficient treatment. In addition, indicates detailed description on the cost of treatment along with types of treatment. Generally, the non-health care costs account for the largest part in total PET costs. Healthcare costs per bite case were generally higher in the urban area than in the rural areas due to a higher application level of supplementary treatments such as tetanus antitoxins in addition to the basic wound care. The non- healthcare costs per sufficient treatment were higher in the rural districts than in the urban area as the distance traveled to the nearest health center was longer than in the urban area. The largest portion of non-health care costs in the rural districts was attributed to the productivity losses due to the time spent on searching treatment. The number of days spent by victims in search of treatment ranged from 0 to 30 days per case with averages of 7.0, 9.0 and 8.4 days per bite case, and 8.5, 16.5 and 16.7 days per sufficient treatment in Bishoftu, Lemuna-bilbilo and Yabelo, respectively. Out of the total dog bite victims, about 35% was at least once accompanied by an adult family member (older than fifteen years of age) to the health center. The opportunity costs of time spent seeking for treatment per bite case ranged from zero to 54 USD with averages of 7.6, 10.0, and 8.7 USD for the victims and 1.6, 2.4, and 3.0 USD for the caregivers in Bishoftu, Lemuna-bilbilo and Yabelo, respectively. However, per sufficient treatment, the average opportunity costs of time seeking for treatment were equal to 9.2, 18.4, 18.1 USD for the victims and to 0.9, 3.5 and 4.1 USD for the caregivers in the same district order. The number of days required to get a single PEP dose in the urban area was considered half a day, while in the rural areas it was about one day. The average number of days spent and the related opportunity costs while seeking treatment per case were comparable across districts due to a lower proportion of victims that was seeking for medical treatment in rural areas. ### Health burden Based on the total of 465 rabies suspected dog bites, the numbers of deaths due to rabies disease out of 100,000 population in Bishoftu, Lemuna-bilbilo and Yabelo districts were 0.7, 4.2 and 2.9, respectively. The health burden due to rabies was estimated to be 48.7, 297.5 and 239.8 DALYs per 100,000 population per year for Bishoftu, Lemuna-bilbilo and Yabelo districts, respectively. The numbers of human deaths and DALYs were significantly higher in rural areas compared to the urban area (p\<0.05). Additionally, the estimated health burden due to the NTV application was 6.1, 2.9, and 2.9 DALYs per 100,000 population per year in Bishoftu, Lemuna-bilbilo, Yabelo, respectively. The largest contribution to this health burden resulted from YLL (i.e. 99%). The overall health burden (aggregated over the three districts) due to rabies and NTV was estimated to be 202.5 DALYs per 100,000 population per year. ### National estimate Given the surveyed death and exposure rates, the estimation on the true district incidence of rabies per year varied according to the Poission 95% CI as indicated in. Extrapolation of these estimations to the national level by weighing for the proportions of population living in urban (19%) and rural areas (81%), resulted in an estimated death rate of 3.0 cases per 100,000 at an estimated espossure rate of 101 cases per 100,000. With an Ethiopian population of 95.7 million these rates coincided with 2,871 (Poisson 95% CIs 574–8326) human rabies deaths and 96,657 (Poisson 95% CIs 78,474–117,711) rabies-exposed persons per year. When accounting for the uncertainty in incidence rates and considering average treatment costs of 21 USD per exposure case, the annual post exposure treatment costs at national level equalled an amount of 2 mln USD (1.6 mln– 2.5 mln). The corresponding national estimate on health loss due to the disease was estimated at 190,016 DALY per year (164,412–218,292). The health impact due to adverse effects of the NTV was estimated at 3,357 (765–9,091) DALY per year which made the total national health burden equal to 193,748 (168,049–222,406) DALY per year. # Discussion This study aimed to assess the burden of rabies in Ethiopia in terms of DALYs and PET costs while accounting for variation across different agro-ecologies. We recorded an average annual human mortality rate due to rabies between 0.7 (in Bishoftu) and 4.2 (in Lemuna-bilbilo) cases per 100,000 population with exposure rates ranging from 19–137 per 100,000 population with the highest number in Lemuna-bilbilo. Our evaluated numbers of exposures and deaths were higher than previous national estimates consisting of 12 exposure cases per 100,000 population and 1.6 rabies deaths per 100,000 population, respectively. These estimates were calculated from confirmed exposure cases that were reported to health centers. Our estimates could have overestimated the actual number of bite cases by rabid dog as suspected cases were determined by the six-criterion method and not by laboratory confirmation techniques. The six criteria method has, however, been reported to be rather accurate (94.6%) in the clinical diagnosis of rabies. Moreover, this retrospective study occurred up to a year after patients or family members had been bitten and were asked to provide details about the bite and the dog. Hence, responses could have been subjected to recall bias which would imply an under or over estimate of the burden. Recall error increases with the length of the recall period. Rabies is, however, one of the most feared diseases with almost 100% fatality rate. For this reason it was expected that the bite victims were rather accurate in remembering rabies exposures within the selected time frame of a calendar year. Our suspected rabid dog exposure estimate was similar to the estimate made by Teklu et al., in North-western Ethiopia but higher than the estimate of 1–5 exposure rates per 100,000 population as reported by Yibrah and Damtie. Underreporting in Ethiopia primarily occurs due to the deep-rooted traditional practice of treating rabies by healers, which as such interferes with assessing the real magnitude of the problem. In line with this, we found that about half of the bite victims we contacted in rural areas did not report to health centers but visited traditional or spiritual healers. In the urban area, this proportion was less than 4%. All exposure victims who seek treatment at least once but died later were not reported officially to the health centers or municipalities. This lack of follow-up by the health centers to monitor whether cases have recovered or died and been reported accordingly also contributes to the under reporting of rabies cases. The evaluated differences in treatment seeking behaviour among the urban and rural districts could, beside the differences in traditional practises, also be related to other socio-economic factors including difference in level of knowledge, accessibility to health centres or differences in the availability of financial means to call for further investigation (forthcoming paper Beyene et al.) in an attempt to improve the rate of sufficient PET in the rural areas. The health burden of rabies ranged from 49–298 DALYs in urban and rural areas with an average of 3 human deaths per 100,000 population (Poisson 95% CIs 0.6–8.8), which is comparable to findings in countries in the region like Tanzania where 4.9 rabies caused mortalities led to 154 DALYs per 100,000 population. Our study as well as the study in Tanzania, did not consider the temporary disability related to the time lived with severe open wounds which was given a disability weight of 0.005 (0.002–0.013) DALYs per patient. The health burden due to the adverse effects of NTV is equivalent to 1% of the total health burden due to rabies disease. This estimate is comparable with a recent study on the global burden of rabies by Hampson et al., reporting the neurological adverse effect constituting 0.8% of total rabies-related DALYs contributed by the 10 countries that reported to use NTV at the time of the study. The health impact per 100,000 population and post-exposure treatment costs per sufficient treatment were estimated to be higher in rural districts. Consequently, the health and financial burden of rabies for rural victims were higher than the urban area, although the human rabies exposure rate to potential rabid dogs was higher in the urban district. People in the age category between 15 and 29 years followed by children in the age between 5 and 14 years represented the largest groups among the rabid suspected dog bite cases. Children of 0–14 years of age had the highest probability of being bitten on the head/neck compared to other age groups. In most communities, children play with dogs. This predispose a higher risk for children to be bitten on the head/neck region. Exposure on upper parts of the body like head/neck is also associated with higher risk of developing the disease. As acknowledged by WHO, the high life expectancy allocated to children and the elevated chance of exposure result in a higher burden of rabies in these age groups. Comparable proportions across age groups were reported by other studies for Northern Ethiopia and countries in the East African region. In the evaluated rural districts, males were shown to be at higher risk of exposure to rabies than females. This could be due to the socio-cultural influence of allocating most of the outdoor activities to males while females mostly work indoor. Similar findings were reported in other locations of Ethiopia In this study, the average treatment costs per dog bite case irrespective of treatment status were more or less comparable among the districts. However, the proportion of health costs and non-health costs varied among the districts due to the fact that a higher proportion of the victims in the urban area sought medical treatment and needed to travel a shorter distance towards the nearest health center than victims in the rural areas. Health costs would generally increase if rabies immunoglobulin would also be administered for those who received PET. This is rarely applied in Ethiopian situation but recommended to all severe bite cases described by a single or multiple transdermal bites, scratches or contamination of mucous membrane with saliva (i.e. licks). The non- health care costs which include opportunity costs of productivity/time losses while seeking for treatment represented the most relevant costs for treatment. The non-health care costs varied across districts and were significantly higher in rural districts that in the urban districts. On average, the PET costs of rabies was equal to approximately 4% of the GDP per capita. Considering only out of pocket spending, i.e. expenditures without considering the opportunity cost of time spent looking for treatment, the economic burden of PET would reduce with 50% to 2% of the GDP per capita, which is still a severe financial burden. Application of the WHO recommended cell culture PEP by a 5-dose vaccine regimen instead of the currently applied 17 doses NTV, would result in considerably higher health care costs (average of 62.5 USD per five doses WHO-PEP) versus 2.3 USD per 17 doses NTV-PEP). The non-healthcare costs will, however, be reduced due to the reduction in number of doses and therefore the required number of visits to the health centers (5 visits versus 17 visits). These costs would have been reduced further if the Essen 4 dose IM regimen was considered. An estimation of the non-health costs under the WHO regimen can be obtained from the average non-health care costs spent per PEP dose (i.e. treatment day) under the NTV regime. Accordingly, the average estimated PET costs of the WHO 5 dose regimen the average PET costs is estimated to be 71.8, 71.9 and 74.7 USD per sufficient treatment in Bishoftu, Lemuna-bilbilo, and Yabelo, respectively. This is approximately 3.5 times higher than the current average PET costs under NTV. This increase in costs is primary related to the difference in price of the imported cell culture based PEP and the locally produced NTV. Comparable costs were also reported in Tanzania for 5 doses of WHO recommended cell culture based PEP. Currently, the Ethiopian government is working towards establishing and producing cell culture based PEP locally that could lower the health care costs. While the costs of cell culture vaccines are high, we only estimated a small saving of DALYs as a result of avoided adverse NTV effects; 6.1, 2.9 and 2.9 DALYs per 100,000 population per year in, respectively, Bishoftu, Lemuna- bilbilo, and Yabelo. Even though the health burden of the NTV compared to the total rabies burden is minimal, it could be eliminated by catalysing the current initiative of the Ethiopian government to replace the NTV to modern cell culture vaccine at the lowest cost possible. Although the extrapolation to the national scale provides an indication of the health and economic burden in the country, the obtained estimate should be interpreted with caution, given the relative limited number of districts studied and the implicit assumptions made by weighting the district incidence rates to obtain estimates for the overall urban and rural areas. Our national estimate on the health burden is, however, in line with the estimation on the Ethiopian burden obtained from the most recent global rabies burden study (175,945 DALYs). In Ethiopia, priority has been given to diseases like Malaria, HIV/AIDS, and TB that attract funding for achieving Millennium Development Goals. In addition, these diseases are rated as top priority in the current Ethiopian Growth and Transformation Plan (GTP). Although rabies is not comparable with these diseases which cause very high health and economic burden, our findings demonstrated that rabies causes a health and economic hardship and with a high likelihood of fatality for exposed rural inhabitants calling for national and global attention. Similarly, a recent study in Ethiopia prioritized rabies as the most important zoonotic disease in Ethiopia. However, in practice rabies control is often overlooked by both public health and veterinary authorities. Considering the goal set by WHO to eliminate dog-mediated human rabies by 2030 and meeting the Millennium Development Goals list of combating disease burden and reducing extreme poverty, including rabies control into the Ethiopian Growth and Transformation Plan would be very crucial. Studies in sub-Saharan Africa countries, where rabies is a public health concern, have shown that investments made to vaccinate a sufficient proportion of dogs would likely minimize the health burden and expenditures related to PET. Strategies to reduce human rabies in Ethiopia should, therefore, include mass dog vaccination. On the other hand, the implementation of a sufficient mass dog vaccination requires extensive knowledge of the dog population, the expected cost-effectiveness of vaccination measures and above all a strong integration of the public health and veterinary authorities for a progressive evaluation of one-health effectiveness in real time. In conclusion, this study evaluated the health and economic burden of rabies in three districts representing geographical diversity assuming possible variation in human exposure rates. The health burden varied among the districts from 49–298 DALYs per 100,000 population with an average of 3 human deaths per 100,000 population (Poisson 95% CIs 0.6–8.8), Extrapolation of the results to the national level demonstrated that rabies, on average, causes 2 million USD treatment costs per year. These results indicate a relevant health and economic burden for Ethiopia. Human exposure rates to potential rabid dogs were higher in urban than in rural districts but the health impacts and post-exposure treatment costs were higher in rural districts. Consequently, the health and financial burden of rabies for rural victims are substantially higher than for urban victims. Under the current NTV regimen, non-health care costs are the major contributor of PET costs. Our estimations of the burden of rabies on the Ethiopian society could provide important information for evidence based decision making process with insights into the potential benefits of implementing cost-effective and coordinated intervention activities. # Supporting information We would like to thank the dog bite victims and their families who participated in the field survey, the health extension workers who collected the data and the district health centers. [^1]: The authors have declared that no competing interests exist.

# 1. Introduction Vietnam is also somewhat unique in the range of state institutions that have been explicitly established to advance the position of women in terms of their political representation. Women’s involvement in decision-making and the paid workforce is higher in Vietnam than in comparable countries. Thus, it is commonly claimed that Vietnam has already achieved gender equality. However, some commentators question the gaps between traditions or laws on the one hand and practices on the other. Vietnam is also a rapidly growing economy with an emerging cohort of ‘gender specialists’ and a growing literature on gender inequality. are concerned about the persistence of gender inequalities in access to resources such as education, services, credit and land. However, the authors also note that most studies have been descriptive, and there is still a lack of sophisticated quantitative and qualitative analyses. Thus, there is still a need for research, development activities, and curriculum materials addressing gender issues in Vietnam as it grows in economic and political importance. A limited number of academic studies has focused on intra-gender differences (i.e., within-a-single gender), in which the inequality among female-headed households (FHHs) has not been systematically investigated, particularly in the Vietnamese context. Among FHHs, widows are an important sub-group commonly viewed as amongst the poorest of the poor. There are some reliable reasons to expect that widows might be especially susceptible to poverty. The death of a household’s breadwinner will likely lead to a significant loss of income, which a surviving spouse might find difficult to replace. Furthermore, the households most exposed to the risk of widowhood are, on average, those already most likely to have unfavourable economic characteristics, given the strong correlation between socio-economic status and life expectancy. However, conflicting empirical evidence exists on the link between female household headship and poverty. In any case, it may also be possible that the risks of poverty for widows are different from other FHHs. Widows’ route into household headship differs from that of, for example, women who are heads due to separation, divorce, or remaining single. Additionally, not all widows become household heads, with many women joining the household of other family members (such as a married son) when their partner dies. Based on household income or expenditure measures, it is possible that the poverty risks for widows in these latter circumstances might be lower than for other widows and FHHs. The task of enquiring into which widows are exposed to poverty and why is thus far from straightforward. It is important to avoid a ‘blanket generalization’ that a link between widowhood and poverty exists and that this link applies to all widows in all contexts. The extant literature also highlights the importance of examining the diversity in widows’ risk of poverty, taking account of possible differences across groups of widows defined by whether they are a household head or not, the composition of their household in terms of, for example, the number and age of dependent children, the age of the woman herself, rural versus urban location, and more. It is also important to investigate poverty risks using a broad informational base that allows for the possibility that data on household income might poorly capture a widow’s well-being (and that of her children). This will be the case if a woman is living in a relatively affluent household (with high per-capita income) but, due to a lack of bargaining power within the household, she has little ability to control how resources are used. It can also be the case that a woman in a less affluent household will achieve a high level of well-being if, by being the household head, she can achieve more independence and control over her life’s outcomes. The Vietnamese culture has been heavily influenced by Confucian ideology. Widows have been discouraged from remarriage to be faithful to their deceased partner. More importantly, they are responsible for caring for the husband’s family. Nowadays, family size in Vietnam tends to reduce, and the quantity of multi- generational families also declined toward migration and modernization. Using the VHLSS 2018 data, we find that 12.1 per cent of our research sample are widow-headed households (WHHs). Among WHHs, 365 households live alone (equivalent to 32 per cent), and 257 WHHs live with one other adult (equivalent to 23 per cent), in which nearly half of these adults are dependent members who are household members but do not have a job, do not contribute to household income, and do not do housework. On the other hand, 23 per cent of WHHs are living with one child who is under 16 years. These statistics reveal that many WHHs live alone or reside with many dependent members, including children and adults who cannot make a living. Besides, in 2018, more than two third of the WHHs were living in rural areas. Particularly, 18.4 per cent of WHHs are classified as poor households. The figures indicate that key sources of poverty are associated with WHHs in Vietnam. Because WHHs in Vietnam have somewhat unique characteristics, studies within this context can help highlight the diversity in such households and thus help counter the tendency to generalize their characteristics and outcomes. WHH is a relatively small group. While widowhood is exogenous with the household head’s decisions, it depresses the widow’s wellbeing. To this extent, WHHs in Vietnam may suffer a significant risk of poverty. However, the probability of WHHs falling into poverty has not been investigated in the current literature, particularly for emerging countries such as Vietnam. Studying widows’ experience of poverty in the Vietnamese context provides novel insights into whether ostensibly pro-women policies and institutions translate into favourable economic outcomes for key groups. This motivation warrants this study to examine the poverty risks for widows in the Vietnamese context. Following this introduction, the remainder of this paper is structured as follows. Section 2 discusses and synthesizes relevant empirical studies highlighting the research gaps in the existing literature. Section 3 discusses data and research methodology. Empirical results are presented and discussed in section 4, followed by concluding remarks and policy implications in section 5 of the paper. # 2. Literature review ## 2.1. Sources of poverty There are four typical sources causing poverty, including regional characteristics, community characteristics, household characteristics, and individual characteristics. The regional sources of poverty are related to country or regional characteristics, which are involved with macroeconomic factors and geographical and natural conditions. Ethnic minorities and regions which are isolated and are inhospitable climatic conditions tend to be easily vulnerable to poverty. Besides, poor infrastructure and lack of public service availability are also correlated with a high probability of poverty. Inversely, urban regions with high population density, the Mekong Delta and coastal regions with developed infrastructure have advantages to overcome poverty. Regarding household characteristics, household size, employment status of household members, dependency ratio, and the head’s gender are considered significant determinants of household poverty. A larger household size implies a higher efficiency in household expenditure toward the economy of scale. On the other hand, household income relatively depends on the quantity of economically active members in the household. As such, adding these economically active members to the household tends to increase household net income as per capita income is higher while per capita expenditure reduces. As a result, a higher proportion of households can avoid poverty. In contrast, dependent members who cannot work do share household expenditures. However, they do not earn income for the households. Ethically, dependent members are family members and shall be treated fairly like others. As such, the household size and the dependency ratio contribute to household wealth differently. Besides, a recent study from confirms the gender gap in household wealth in Vietnam. These authors also confirm the heterogeneous wealth across different marital statuses. Particularly, widowhood is a special marital status which is not controlled by the couple, while other family statuses, including single, married, separated or divorced, are. WHHs are in an exceptional situation where the head’s partner, who played a significant role in the household, passed away. Losing a member reduces income sources and the efficiency of numerous families resulting in many difficulties in financial balance in WHHs. However, did not particularly focus on the difficulties of WHHs. ## 2.2. The overviews of the economic circumstances of widows Early studies have shown that widows face a high risk of poverty because of the loss of family wealth after their husband’s death. The death of a spouse can be a devastating and life-altering experience, and widows often face numerous challenges after their loss. These challenges include financial difficulties. Besides, widowhood leads to higher poverty rates for all women, regardless of pension eligibility or annuity choice for retirement. This is because when a husband dies, much family wealth is lost, including private pension income. Also, the loss of wealth is not always compensated by other resources such as life insurance, and poor widows often had little housing wealth when they were married. In their study regarding the economic situations of middle-aged and older widows, found that widows in these age ranges face fewer economic resources after their spouse’s death than their married counterparts. Recent research by also confirms the negative impact of widowhood on household income. They conclude that the magnitude of the effect depends on individual and household characteristics as well as social and contextual factors. A female’s loss of a spouse can also mean a loss of social support and the need to assume responsibilities previously handled by their spouse. Thus, the labour participation of women is significantly affected by widowhood. studies the impact of widowhood on Indian women’s labour force participation using panel data that tracks women before and after the event of widowhood. The results confirm a strong age pattern, where women widowed before the age of 52 saw an increase in the number of days worked, while women widowed after the age of 52 saw a reduction. In addition, widowed women are more likely to be employed in the non-agricultural sector. They have higher incomes than married women if they become the household head after their spouse’s death. The economic gap between widows and married individuals may be due to the lower levels of financial literacy and experience in financial planning among women in general. Therefore, couples need to prepare psychologically and have a clear retirement plan to avoid the risk of poverty after widowhood. Many studies found that the most significant risk of falling into poverty occurs in the initial stage of widowhood, highlighting the importance of having sufficient resources before retirement to decrease the chance of facing poverty in retired years and after the loss of a spouse. Vietnamese studies that specifically address the economic circumstances of widows are rare, and typically, when widows are mentioned, it is only as part of a broader analysis of the situation of women. For example, includes widow households in his study of Vietnamese FHHs, and mentions them in her study of the impact of lone motherhood on children’s education outcomes. also include widows in their broad study of older persons in Myanmar, Vietnam, and Thailand, finding evidence that widows are more likely to live solo, which makes them more vulnerable to stress, disabilities, and financial pressure.’s study adds evidence to the disadvantaged economic position of widows by showing that widowed lone mothers have a relatively high prevalence of disability, a relatively low level of educational attainment, and a relatively high level of economic activity. However, they mostly work in low-skilled sectors.’s study also shows that children of widowed lone mothers have the lowest level of school performance, as measured by school enrolment and completion rates, pointing to concerning impacts of widowhood on the outcomes of both the widows themselves and their children. Widows were also an important group included in’s qualitative investigation of gender and land entitlements in northern Vietnam. This study provided more pointers to the possible sources of poverty among widows. It highlighted the barriers faced by widows in exercising their inheritance rights over land and other assets. Scott also found that widows faced barriers in accessing the information on land allocation. Thus, they commonly experienced difficulties acquiring land for production and as collateral to secure bank loans. Recently, approached widowed households as a moderating factor while investigating a gender wealth gap and a wealth return on education in the Vietnamese context, respectively. In these recent studies, widowhood significantly and negatively impacts household wealth accumulation compared to other households with different marital statuses. Particularly, at the upper asset threshold (from 440.25 million VND in 2018), found that wealth accumulation favours single households over widowed households. However, the authors found relatively weak differences in wealth accumulation between the two cohorts. On the other hand, found that WHHs consistently stay in the vulnerable group across all net worth quantiles. To our knowledge, the only prior Vietnamese study with a primary focus on widows is, in which the authors use the 2009 census data. This study identified that widows who were relatively old, had a disability, or attained a relatively low education level were more likely to live independently. In addition, ethnicity was also identified as an important correlate of living on one’s own in widowhood, with women in minority groups more likely to be in this situation than other Vietnamese widows. Our study adds to analysis by focusing squarely on the factors linked to widows’ vulnerability to poverty and expanding the analysis to widows living in FHHs and other household situations. First, we compare the likelihood of poverty for widows and other Vietnamese women. We also assess whether widows who head up their households face different risks of poverty from those who live in other households. Finally, we examine the effect on the risk of poverty of a range of social, demographic and locational characteristics of widow households in Vietnam. Our literature review indicates that selected social and economic aspects, such as the wealth and welfare of widows, have been investigated in developed countries. In addition, efforts have been made to propose several solutions to mitigate poverty after widowhood. However, widowed households suffering from poverty have largely been neglected in the existing literature, especially in Confucian countries such as Vietnam. This research gap warrants the analysis below to be conducted. Barriers due to traditional Confucian norms prevent women, especially widows, from asset ownership leading to vulnerability. We employ the probit regression for various samples, including the entire sample, a sub-sample of widowed households where at least one widow is living, and a sub-sample of female-headed households to examine the probability of suffering from poverty of the widowed households. Besides, other proxies such as land value and income are also used as the alternative dependent variable to investigate the effects of widowhood on a household’s wealth. We also extend our analysis by estimating the marginal effects of widowhood on several forms of insufficiency, including suffering from food, foodstuff, electricity, water, housing, and clothing. Our analytical approach is discussed in detail in the next section. # 3. Data and research methodology ## 3.1 Data This study utilizes data from the 2018 Vietnam Household Living Standard Survey (VHLSS2018) to examine the economic outcomes of widows in Vietnam and to assess the factors that influence their risk of poverty. The VHLSS has been conducted by the General Statistics Office of Vietnam since 2002 with technical assistance from the United Nations Development Programme and the World Bank. It plays a key role in Vietnam’s attempts to monitor the living standards of Vietnamese households, and it is widely used to inform policy decision-making. Data from the VHLSS has been used to evaluate developmental programs associated with the Comprehensive Poverty Reduction and Growth Strategy, the Millennium Development Goals, and Vietnam’s socioeconomic development goals. ## 3.2 Measuring poverty There are two typical kinds of poverty in the literature: (i) absolute poverty and (ii) relative or multi-dimensional poverty. Absolute poverty is usually measured by income. In this criteria, the poverty line is a benchmark to identify "a poor household". Besides the poverty line, the bottom 10 or 20 per cent of income is usually perceived as the ’a poor group’. However, assigning a proportion of households whose income is lower than the average or median income within the region is sensitive to personal preference. On the other hand, income-based measurement is typically underestimated. Regarding the second kind of poverty measurement, argue that poverty is essentially the lack of the means to live. The fundamental of poverty relies on a serious problem of what is needed to live "a decent life" and, more critically, what it is to be human. From this perspective, multi-dimensional poverty is more appropriate because it captures many social-economic aspects (including incomes, social needs, and many others) to identify "a poor household". Therefore, measuring poverty by a constant poverty line or a proportion of income is inadequate. In 2015, the Vietnamese government formally released official documents to identify a poor household by multi-dimensional measurements which are officially announced throughout the Decision number 59/2015/QD-TTg dated 19/11/2015 by the Prime Minister about the multi-dimensional poverty standards. There are two major pillars to assign whether a household is poor: (i) the income-based measurement and (ii) norms on deprivation of access to basic social services. *First*, regarding the income criteria, the Vietnamese government assigns a poor household based on monthly income. *Second*, regarding the norms on deprivation of access to basic social services, there are five categories (including (i) health; (ii) education; (iii) housing; (iv) clean water and sanitation; and (v) information), which are measured by ten indices (including (i) access to medical services; (ii) health insurance; (iii) education level of adults; (iv) school attendance of children; (v) housing quality; (vi) average housing area per capita; (vii) residential water sources; (viii) hygienic latrines and toilets; (ix) telecom services; and (x) assets to serve information access). The government also assigns different benchmarks of poverty in rural and urban areas. In the rural area, a poor household is identified if its monthly income is lower than 700 thousand VND or if its monthly income is from 700 thousand VND to 1 million VND, and the household fails in at least three indices of the second criterion as discussed above. In urban areas, the two milestones of monthly income are 900 thousand VND and 900 thousand VND to 1.3 million VND, respectively. The conditions of norms on deprivation of access to essential social services in rural and urban regions are identical. The VHLSS 2018 has a question to identify a poor household: “Have the local authorities classified your household as ‘poor’ in the commune/ward in 2018?”. As such, poor households have been classified by the local authorities. The local authorities have to strictly follow the official guidance from the prime minister’s decision to classify a poor household. Based on the above ground, the multi-dimensional poverty measurement is appropriate to identify poverty. Besides, the process of identifying a poor household is strictly managed by the government authorities in the Vietnamese context. As such, we consider it appropriate to use the current identification of poor households in the VHLSS for our analysis. ## 3.3 The analytical framework In the first part of our analysis, we examine whether Vietnamese households with a widow are more vulnerable to poverty than households that do not include a widow. Our key outcome measure is based on the responses to a VHLSS question: “Have the local authorities classified your household as ‘poor’ in the commune/ward in 2018?” In this sense, the outcome variable is binary, at which the value of “0” is proxied for non-poor households and the value of “1” denotes the poor household. In further stages of our analysis, we extend the set of outcome variables to include household income and land ownership. We assess the household income and land holdings differences between the widow and non-widow households. These additions allow us to account for the possibility that there may be widow households that are not subject to poverty *per se* but still experience a relatively high level of economic hardship due to low income/wealth. It is important to identify these risks if they exist, especially when the poverty threshold is so low. Household income is measured as the total labour income of all household members, plus other income from non-labour sources, over the previous 12 months. Labour income comprises salaries, wages, cash, benefits, allowances and in-kind payments from all jobs. Other sources of household income include agricultural activities, non-agricultural activities, gifts and presents. Regarding land ownership, the VHLSS has a range of measures, including the size and value of each land plot owned by the household. In this study, we use the total value of all land plots owned within the household. In each stage of our analysis, we incorporate controls to account for the likely influence of a range of individual and household characteristics on widows’ chances of poverty. We use VHLSS data on household size, the number of absent members, the household dependency ratio, and information on the household head’s characteristics, such as their level of education, age, injury status, and labour force participation status. These factors have been discussed as determinants of a household’s consumption and welfare by. Following, we measure the household size by its equivalent scale adjustment, considering the number of individuals in the household, their age and likely household economies of scale. We measure the equivalent scale adjustment recommended by Deaton and Zaidi (2002): $$EA = \left( {A + \alpha C} \right)^{\theta}$$ where EA denotes the equivalent-adjusted household size; A represents the number of adults, and C is the number of children under the age of 15 years; *α*, which ranges from 0 to 1, represents the weighted expenditure requirement of a child relative to an adult, and *θ*, ranging from 0 to 1, measures the household economies of scale. argue that raising a child can be less costly in developing countries. The economies of scale can be larger because they spend a larger proportion of income on food (while maintaining a fixed expenditure). Thus, these authors suggest that in developing countries, *α* shall be low, in the range of 0.3–0.5, and *θ* shall be high, from 0.9 to 1. In addition, since 2002, Vietnam’s average income has increased substantially, and raising a child has also been more costly. For this reason, we use a value of *θ* = 0.9, which is less than 1, and *α* takes a value of 0.5 as the highest weight suggested by. We measure absent household members as members who have not stayed in the household for at least six months. The dependency ratio measures the proportion of all household members that do not participate in the labour market, do not do housework and do not receive pensions, unemployment allowances, one-off severance pays, or an allowance for loss of working capacity the past twelve months. Finally, the injury variable identifies the total times household members have had such a severe injury that they have needed care or had to stop working, studying, or participating in other normal activities in the previous 12 months. We also include locational variables to help control for spatial heterogeneity. Of particular relevance in the current study is the relatively high rate of poverty in the Midlands and the Northern Mountainous Area, as well as in the Northern and central coastal and Highlands regions. The high prevalence of poverty in the Northern Mountainous Areas is partly due to its geography, which prevents the region from diversifying its economic activities, accessing major markets and improving its infrastructure. In addition, these northern regions are also the home of several ethnic minorities, among the poorest of the poor in Vietnam. The regional breakdown of our data is eight-fold, comprising the Red River Delta, the Northern and Southern regions, the Highlands, and the Mekong Delta. An urban/rural categorization of locations identifies cities in Vietnam according to their official ranking, which reflects their function and role, the structure and level of their social-economic development, and their size. The highest-ranked cities (type I) have central roles in the social-economic development of their region and the nation. These cities also have the largest populations (more than 1,000,000 inhabitants), the highest population density (more than 2,000 inhabitants per square kilometre) and the highest proportion of non-agricultural labour (more than 65%). The cities of lower ranking perform less substantial political and economic roles and have lower populations and population densities. ## 3.4 Research methodology In the above section, poverty has been described as a dummy variable at which “0” stands for a poor household, and “1” represents a non-poor household. Then, to analyze the vulnerability of widow households to poverty, probit regression and logit regression are the two appropriate methods for dealing with the binary dependent variable. On the one hand, logit regression can perform well regardless of whether the dependent variable is multinomial or binary. On the other hand, the probit regression mainly focuses on the binary dependent variable. In our study, both models are appropriate to analyze the probability of suffering poverty in widow households. As such, the probit regression is used to conduct our empirical analysis as the basis case, whereas logit regression is used as the robustness analysis. The empirical results of this robustness analysis are presented in the Appendix. Derived from a latent variable model, the regression model takes the form: $$y^{*} = \alpha Widowhh + \varepsilon$$ Where *y*\* is a latent variable, the household is classified as poor when *y*\*\>0. Widowhh is a categorical variable identifying whether the household includes a widow, and *ε* is the error term (assumed to have a standard normal distribution where its cumulative density function is *f*(*x*)). The probability of a household being classified as poor is defined as: $$P\left( {y^{*} > 0} \middle| {Widowhh} \right) = P\left( {\varepsilon > - \alpha Widowhh} \middle| {Widowhh} \right) = f(\alpha Widowhh)$$ Because the heterogeneity in household and individual characteristics can potentially explain differences in poverty status, we extend the probit model to include measures of these characteristics as control variables. <img src="info:doi/10.1371/journal.pone.0285595.e004" id="pone.0285595.e004g" /> P ( y \* \> 0 \| W i d o w h h , X ′ ) = f ( W i d o w h h , X ′ ) As noted, the control variables in *X*′ consist of: measures of household size, the number of absent household members, the household dependency ratio, the head’s age (including a squared term to account for possible non-linearity), the value of the land owned within the household, and locational and educational variables. Educational attainment is measured by years of education, based on the highest grade of education the person completed. Respondents who left school at Grade 12 are assigned 12 years of education, whilst those with a College, Bachelor, Master, or PhD degree are assigned 14, 15, 17, or 18 years of education, respectively. Urban location is measured with dummies for large and medium cities, with the rural location being the reference category. Also included are eight regional variables, with the Red River Delta being the reference category. Because we are also interested in whether WHHs are more vulnerable to poverty than other widow households, we run a similar probit regression model on the VHLSS data, including variables to identify households headed by a widow (WHH) and, secondly, other widow households. The other elements of the model remain unchanged. To address whether widow-headed households are more susceptible to poverty than other female-headed households (FHHs), we apply the regression model to the sub-set of FHHs. Our analysis of the correlation between widowhood and household income and land ownership deploys Ordinary Least Square (OLS) regression models, with both household income (*hhincome*) and the value of land owned by the household (*land*) logged to enable us to focus on relative differences where zero values are replaced with a number infinitesimally close to zero (e.i. 0.000000001). The models use the same key explanatory variable (*Widowhh*) as the probit models. The set of control variables is also similar. However, in the model of household income, land value is one of the control variables, whilst, in the land value model, household income is retained as a control variable. In the final step, we examine the factors that affect widows’ relative chances of being exposed to different forms of poverty. Different probit regression models target food, housing, energy and other forms of insufficiency. The control variables identify household and individual characteristics, including the head’s education, age, injury status, whether he/she is participating in the labour force, and household size, location, and dependency ratio. Summary statistics on the variables included in the regression models are provided in. These clearly show a relatively high incidence of poverty amongst WHHs. In all samples, Columns 2 and 3, 10.6 per cent of the households are considered poor. Within the sub-sample of the WHHs, 18.4 per cent of them is considered poor household, as presented in the last column. On the other end, for the widow households where the head is not a widow, results presented in columns 6 and 7 indicate that 16.1 per cent of the households are considered poor. In contrast, for the sub-sample of non-widow households, only 9.4 per cent are considered poor, as presented in columns 4 and 5. As could be expected, the heads of WHHs are relatively old. The average age of household heads in this group is 64.7 years, while in the broader group of widow households, the average age of heads is 58.5 years. In non-widow households, the average head age is 49.1 years. The head of widow-headed households also has a relatively low level of educational attainment. The average years of education for widows who are household heads are only 4.7, while heads in the broader group of widow households have, on average, 5.9 years of education. On average, the heads of non-widow households have had 7.8 years of education. Widows who are household heads also have a relatively low average level of labour income. As a result, these households tend to be more reliant on the labour income of other household members. The average annual labour income of the heads of widowed-headed households is 6.5 million VND, and the average earnings of other household members in these households is 29.3 million. In comparison, in non-widow households, the average annual earnings of household heads are 21.4 VND million, and other members’ earnings only reach 29.5 VND million. Pointing to the possible role of assets in determining the risk of poverty, non- labour income in WHHs is also relatively low (83.8 VND million per annum, compared to 120 million VND in non-widow households). The data in show that WHHs appear to be most susceptible to food and foodstuff poverty, with this rate reaching 2.6% and 7.1% compared to the average rate in other households of 2% and 5.6%, respectively. The incidence of other forms of poverty (clothing, electricity, housing) in WHHs is generally similar to that experienced in the other groups. # 4. Empirical results and discussions We commence our presentation of the results with the estimates of the relationships between the risk of *household poverty* and various economic and demographic factors across widows and other households. reports the marginal effects and robust standard errors of variables in four regression models of the probability of household poverty. The key variables of interest identify whether a widow is present in the household (Widowhh) and whether the household head is a widow (WHH). Column 1 presents the model results that examine whether the poverty risk for widow households is different from all other Vietnamese households. Column 2 has the results from the model that isolates the relative risk of poverty in two groups of widow households–those headed by a widow and those headed by someone else. Column 3 focuses on the possible differences in poverty risks between WHHs and other widow households. Column 4 presents results from a model designed to examine whether WHHs have different poverty risks to other FHHs. The results in Column 1 of show that households where a widow is present, have, at mean values, a risk of poverty that is 2.5 percentage points higher than other Vietnamese households. The figures in Column 2 reveal that this risk differential applies especially to WHHs, who have a risk of poverty 2.8 percentage points higher, at mean values, than non-widow households. The risk of poverty in households that include a widow who is not the household head is also higher, but the gap is not statistically significant. The results in Column 3 show that, in the presence of controls for a range of economic and other characteristics, the risk of poverty in WHHs and other widow households is similar. The results in Column 4 reveal, further, that the poverty risks of WHHs are similar to those of other FHHs. The results in the remainder of help to identify some of the risk factors for poverty in widow households and other household types. Larger household size appears to be a protective factor, reducing the odds of poverty, especially in widow households. For example, across all Vietnamese households, an increase in household size by one reduces the chances that the household will be poor by 1.3 percentage points. However, in the sub-sample of widow households, a similar change in household size reduces the chance of poverty by 3.1 percentage points. This pattern of results implies that widows’ chances of poverty are relatively strongly affected by whether other adults are in their household. Moreover, it speaks to a relatively high economic dependence of widows on the contributions of other household members. Notably, whilst a higher dependency ratio tends to be positively related to a household’s chances of poverty (Columns 1 and 2), this factor is less important in widow households (Column 3). Across the sample of Vietnamese households as a whole, one point rise in the dependency ratio is associated with a substantial 5.2 percentage point increase in the risk of poverty. However, in the sample of widow households (Column 3), this factor is not a statistically significant source of variation in the risk of poverty. This result is intriguing. In the broad sample, it could be the case that households are more susceptible to poverty if/when an adult member within the household is not economically active. The results differ from, in which the dependency ratio is associated with the wealthier household, especially at the top of wealth distribution. In widow households, the dependency ratio–of differences in members’ economic participation—is less likely to be at play, and, thus, the dependency ratio is less consequential to poverty risks. From the VHLSS 2018 data, we find that in a part of WHHs–who live with at least one dependent member in the household, 82.8 per cent of these households have one dependent adult and up to three children. These figures indicate that a majority of dependent members in WHHs are children. As raising a child is less costly than a dependent adult in the Vietnamese context, dependent members with many children might have minimal effect on the risk of poverty in WHHs. The results in also show a strong influence of education on poverty risk, and again this factor appears to be especially consequential in widow households. In the sample as a whole, a one-year increment in the grade of education completed by the household head reduces the odds of household poverty by 1.6 percentage points at mean values; in the sub-sample of widow households, the drop is 2.1 percentage points. Finally, age is a further protective factor against poverty, with each additional year of age-associated, at mean values, with a slight (but statistically significant) reduction in the likelihood of poverty (0.9 points in widow households and 0.6 points in all households). These results highlight the relatively high risks of poverty for older widows and those with lower levels of education. The results on the locational variables show, first, that urban location is generally associated with a lower risk of poverty. However, this relationship is not statistically significant in the sub-sample of widow households. There are also substantial regional differences evident in the results, with households in the Mountainous and Northern Areas, the Coastal Central Region and the Central Highlands having a relatively high risk of poverty and relatively low-income levels. Outcomes for widows follow a similar pattern, although the poverty risk for widows in the Central Highlands is not significantly different from that in the Red River Delta. The results for widows in the Mountainous and Northern Areas (“North of the Middle” area) are of particular concern. Their widow households face a risk of poverty that is 10.0 percentage points higher than that experienced by their counterparts in the Red River Delta. Other households in this region are also, on average, economically deprived, but the relative disadvantage of widows appears to be particularly large. Empirical results from the probit regression (in) are now compared with those from the logit regression (in the Appendix). We note that the empirical results under both estimation techniques largely remain the same. The vulnerability of widow households is consistently confirmed using these two regressions. Widowed households are more likely to fall into poverty, especially WHHs than others. This result aligns with findings from. Other demographic characteristics also reveal the similarity between empirical results from the probit and logit regressions. Household size, years of education, and value of owned land reduce the probability of falling into poverty, especially in widow households. A geographical location such as mountainous or isolated areas and regions of ethnic minority (including Western North and North of the middle regions) is associated with a higher probability of poverty. These results are identical to. presents complementary results–showing the estimated coefficients on variables included in an OLS regression model of household income and household land value, respectively. The data in Columns 1 and 5 show, among other things, that widow households have lower household income (by 10.0% at mean values) than households where a widow is not present. However, their land holdings are not significantly different from non-widow households. The results in Columns 2 and 6 reveal that WHHs have similar incomes and land holdings to non-widow households, whilst other widow households tend to have lower incomes and hold more land. Thus, as also apparent in the results in Columns 3 and 7, WHHs tend to have more income but less land than other widow households. The results in Columns 4 and 8 suggest that WHHs have similar income and land asset levels as other FHHs. The pattern in the results across Tables and indicates that at mean values and in the presence of controls for a range of demographic and other factors, being a WHH increases the risk of poverty. However, it has only a marginal effect on income. One possible explanation is that a relatively high proportion of WHHs are ’pushed over’ the poverty threshold by their lower income. At mean values, other widow households are associated with lower incomes than non-widow households but (as the results in show) somewhat similar risks of poverty. This could show that a smaller proportion of these households’ incomes are close to the poverty threshold. The results on the land variables also suggest that WHHs are characterized by relatively low levels of land ownership, whilst the opposite is true of other widow households. Thus, the pattern of coefficients across the variables that measure household type in the income and poverty models could also reflect the influence that lands assets have on the chances of households being categorized as poor, as the results in suggest the case. The results in for the various control variables follow a more consistent pattern. As was the case in the results of the poverty model, household size is positively related to income–and this relationship is relatively strong in widow households. Education has a measured positive impact on income, although the size of this effect is relatively weak in widow households. Age also has a positive effect on income in widows and other households. However, it is not positively related to land holdings in widow households (although it has substantial positive impacts in other households). Urban location is a further positive factor for household income, but, again, the strength of this influence is relatively small in widow households. The dependency ratio is negatively related to income in widow and non-widow households. This result is somewhat at odds with the pattern of results in (where the dependency ratio was not a statistically significant source of variation in poverty risk in widow households). Nevertheless, it adds to the questions about the underlying relationships. In the broad sample, the data in Tables and show that higher dependency is associated with lower income and higher poverty risk. In the sub-sample of widow households, this factor–which we can link to differences in women’s economic participation–tends to lower household income but not to the extent that it adds to the risk of being classified as poor. The results of the locational variables confirm that the Middle of the North region is relatively economically disadvantaged. Widow households in this region experience relatively low incomes, as do non-widow households. Land values are also lower, but not to the extent that they are in other northern areas. For example, the average land holding of widow households in the Middle of the North is 97 per cent lower than in the Red River region, but in the Western North, they are 347% lower. The final results of this study–presented in Tables and –are on the risk of various types of resource insufficiency. Generally, these do not reveal significant differences between the widow and other households. However, where widows are exposed to poverty, the risks are highest for WHHs in the areas of water and housing insufficiency. WHHs have a chance of water insufficiency that is one percentage point higher than other widows at mean values; and a chance of housing insufficiency that is 1.5 percentage points higher. The results in these tables also confirm the vulnerability to poverty of households where the head has a low level of education, and they further illustrate the regional disparities in poverty risks. Widow households in the Eastern North appear especially vulnerable to food and foodstuff insufficiency. # 5. Concluding remarks This study has contributed an analysis of the economic circumstances of Vietnamese widows, adding to a very small extant literature on the incidence of poverty amongst widows and the factors affecting this risk. The results from our analysis of the VHLSS 2018 data confirm that widows are relatively vulnerable to poverty. This risk is relatively large for widows that are household heads and, in this way WHHs are similar to other FHHs in Vietnam. Higher levels of education, older age, urban location, and a larger household size also appear to provide some Vietnamese widows with protection against the risk of poverty. However, the disparity level in widows’ poverty risks is high, with an especially strong regional variation. The high poverty rate amongst widows in the Middle North regions is particularly concerning. These findings indicate that, despite strong efforts at developing state institutions for gender equality, additional efforts are needed to address the disadvantaged economic position of key groups of women. This study has also contributed some fresh insights into the economic situation of female-headed households in Vietnam. Whilst some prior studies have shown that FHHs in Vietnam are better off than male-headed households, we find that widow-headed households are relatively disadvantaged, experiencing a relatively high risk of poverty. Thus, it would seem that when the route into female headship is through widowhood, at least, the risk of poverty for Vietnamese women is real. The study has focused only on statistical patterns in existing data sets. It will be important that future work involves the collection of new quantitative data and that complementary qualitative investigations take place. On the quantitative side, new measures are needed to assess the distribution of resources within the households in that widows reside. Our results suggest that households that include a widow who is not the head are not as economically disadvantaged as WHHs; however, it is not clear whether the widows in these households are protected from poverty. Our results also show, for example, that widows who live in the Northern Mountainous areas experience high poverty risks. However, we cannot be sure about the institutional or other factors that contribute to these risks. There is a need, therefore, to keep adding to the informational base on widows’ economic circumstances, perhaps using the current study results as a guide to the key areas of policy concern. # Supporting information [^1]: The authors have declared that no competing interests exist.

# Introduction Studies in humans have shown that cardiac contractile function declines with age, even in the absence of overt cardiovascular disease. Although contractile function is relatively well preserved at rest, the ability to increase contractile force under conditions of higher demand, such as exercise, declines with age. Studies in intact hearts and cardiac tissues from aged rats also show that the ability to augment force in response to positive inotropic stimuli is compromised in the aging heart. This age-related decrease in cardiac contractile function is due, at least in part, to a decrease in the ability of individual ventricular myocytes to contract. Most previous studies of the impact of age on cardiac contractile function in animal models have used hearts, tissues and myocytes from male animals. However, there is evidence that the effect of age on cardiac contractile function differs between the sexes. Studies have shown that contractile force, fractional shortening and left ventricular function deteriorate with age in male rats and non-human primates, but are unaffected by age in female animals. Previous work from our laboratory and others has shown that the ability of individual ventricular myocytes to contract declines with age in male but not female rats and mice. This arises as a consequence of a reduction in the magnitude of the Ca²⁺ transient required to initiate contraction. These findings suggest that sex differences in cardiac contractility in the aging heart may be linked to effects of sex steroid hormones on myocardial Ca²⁺ handling. However, little is known about the influence of sex steroid hormones such as estrogen on cardiac Ca²⁺ homeostasis in the setting of aging. Cardiac myocytes possess estrogen receptors and evidence suggests that chronic exposure to estrogen modifies intracellular Ca²⁺ homeostasis. Studies have shown that Ca²⁺ transients and contractions are smaller and slower in ventricular myocytes from young adult female rats when compared to age-matched males. However, bilateral ovariectomy (OVX) of young adult females increases the speed and magnitude of Ca²⁺ transients and contractions compared to sham-operated controls but cf.. We have shown that this is not due to an increase in Ca²⁺ current, but arises from increased sarcoplasmic reticulum (SR) Ca²⁺ release as a consequence of increased SR stores and larger Ca²⁺ sparks. These findings suggest that removal of ovarian estrogen in young adult females enhances SR Ca²⁺ release and leads to Ca²⁺ transients and contractions that are similar to those seen in myocytes from young adult males. It is possible that long term estrogen deprivation, starting early in life, may lead to age-associated Ca²⁺ dysregulation and contractile dysfunction as seen in myocytes from aged males. However, whether long-term OVX alters Ca²⁺ homeostasis and causes deterioration in cardiac contractile function in the aging female heart has not been investigated. The overall aim of this study was to determine whether long-term OVX modifies myocardial Ca²⁺ homeostasis and disrupts contractile function in the aging mouse heart. Studies used very old (e.g. \~24 month old) female C57BL6/J mice that received either a bilateral OVX or sham surgery at an early age (e.g. one month of age). Ventricular myocytes were loaded with Ca²⁺-sensitive fluorophores to investigate specific Ca²⁺ handling mechanisms. Contractions, Ca²⁺ transients, Ca²⁺ currents, sarcoplasmic reticulum (SR) Ca²⁺ content and Ca²⁺ sparks were compared in myocytes from aged sham and OVX mice. *In vivo* cardiac function was evaluated with echocardiography and myofilament Ca²⁺ sensitivity was assessed by measurement of actomyosin MgATPase activity. Ca²⁺ handling proteins were assessed by Western blot analysis. Our results showed that long-term OVX reduced myofilament Ca²⁺ sensitivity, promoted cardiomyocyte Ca²⁺ dysregulation and increased spontaneous SR Ca²⁺ release in the aging female heart. # Materials and Methods For full details of Methods, please refer to online. ## Ethics Statement Protocols were approved by the Dalhousie University Committee on Laboratory Animals (No. 12-022) and followed Canadian Council on Animal Care Guide to the Care and Use of Experimental Animals (CCAC, Ottawa, ON: Vol 1, 2nd edition, 1993: Vol. 2, 1984). Sodium pentobarbital anesthesia was used and all efforts were made to reduce suffering. ## Myocyte studies Myocytes were isolated from 20-24 mos female C57BL/6 mice that had either sham surgery or OVX at 1 mos. Ventricular myocytes were isolated by enzymatic digestion as described in detail previously. OVX was confirmed by uterine atrophy. All experiments were conducted at 37°C. Contractions (unloaded cell shortening), transmembrane currents and Ca²⁺ transients (fura-2 AM) were measured simultaneously as described in our previous studies. In voltage clamp experiments, membrane potentials and transmembrane currents were recorded with an Axoclamp 2B amplifier (switch clamp, 5-8 kHz). Transient outward K⁺ current was inhibited by 4-aminopyridine (4 mM) and steps were made from -40 mV to inactivate Na⁺ current. SR Ca²⁺ load was measured by rapid application of 10 mM caffeine in 0 Na⁺/0 Ca²⁺ buffer to inhibit Ca²⁺ extrusion via Na⁺-Ca²⁺ exchange. Action potentials were measured in separate experiments, in cells not loaded with fura-2. In field stimulation experiments, cells were paced with platinum electrodes. Ca²⁺ sparks were measured in fluo-4 AM-loaded myocytes as described previously and analyzed with SparkMaster. ## Echocardiography Two-dimensional guided M-mode echocardiography was performed in anesthetized mice (2% isoflurane). ECG electrodes were placed subcutaneously and mice were assessed with a high-resolution linear transducer connected to a Vivid 7 imaging system. ## Myofilament studies Myofilaments were isolated from the ventricles and frozen until use as described previously. Actomyosin MgATPase activity was determined with techniques that have been previously described. Myofilaments (25 mg) were incubated in activating solutions containing varying levels of free Ca²⁺ as described earlier. ## Western blots Ventricles were homogenized (60 s) in buffer (5 mM EDTA, 20 mM HEPES, 2% LDS, 10% glycerol, 1mM AEBSF, 800 nM apoprotin, 20 µM leupeptin, 10 µM pepstatin, 50 µM bestatin, 15 µM E-64). Tissues were sonicated 3 times and flash frozen in liquid nitrogen after each sonication step. Samples were then heated (5 mins, 70°C), cooled on ice (5 mins), centrifuged (12,000 rpm, 8 mins, 4°C) and the supernatant was stored (-80°C). Protein concentration was determined with a DC Protein Assay (BioRad). Antibodies used were rabbit anti- Ca_v1.2 polyclonal antibody (Alomone, ACC-003-AG; 1:2000), mouse anti-NCX monoclonal antibody (SWANT, R3F1; 1:1000), mouse anti-SERCA2 monoclonal antibody (Affinity Bioreagents, MA3-919; 1:2000) and rabbit anti-Na/KATPase polyclonal antibody (Abcam; 1:2000). Secondary antibodies were goat anti-mouse HRP-conjugated polyclonal antibody (Abcam, 1:30,000) and goat anti-rabbit HRP-conjugated polyclonal antibody (Abcam, 1:30,000). Protein was resolved on an 8% polyacrylamide gel, transferred to nitrocellulose membrane and blocked. Loading controls were anti-Na/K ATPase antibody (Ca_v1.2 and NCX) or amido black (SERCA2). Signals were visualized with Immuno-Star™ WesternC™ Kit (BioRad) and quantified with Quantity One software. ## Statistics Statistical analyses were performed with SigmaStat (version 3.1, Systat Software, Inc.). Differences between means were evaluated by a t-test or two-way repeated-measures analysis of variance. Graphs were constructed with Sigmaplot (version 8.0, Systat Software, Inc). Data are presented as means ± SEM and differences were considered significant if p\<0.05. # Results ## Physical characteristics of sham and OVX mice Selected physical characteristics of the sham and OVX mice used in this study were compared as shown in. Sham and OVX mice were the same age (24.5 ± 0.5 vs. 23.2 ± 0.5 months for sham and OVX mice, respectively; n=22 mice per group) and had similar body weights. Ventricle weight and ventricle-to-body weight ratios were also similar in the two groups , while uterine dry weights were significantly reduced in OVX mice. These findings show that while sham and OVX aged females were similar in weight and had similar heart sizes, OVX caused marked uterine atrophy. To determine whether OVX affected cardiomyocyte size, myocyte capacitance and myocyte volume were compared in sham and OVX animals. Cell capacitance was 24% lower in OVX myocytes compared to sham cells. Cell volume, calculated from cell capacitance as described previously, also was reduced in myocytes from OVX mice when compared to sham cells. These findings demonstrate that cardiomyocyte volume and membrane area were reduced by long-term OVX in the aging female heart. ## OVX modifies Ca²⁺ homeostasis and contractions in field-stimulated cardiomyocytes To determine whether long-term OVX influenced Ca²⁺ homeostasis and contractile function, contractions and Ca²⁺ transients were simultaneously recorded in sham and OVX myocytes that were field-stimulated at 2 Hz. shows examples of Ca²⁺ transients and contractions recorded from sham and OVX myocytes paced at 2 Hz. Mean peak contractions, normalized to cell resting length, did not differ between sham and OVX cells. However, time-to-peak contraction and time-to-50% relaxation were reduced by OVX. Intracellular Ca²⁺ concentrations also were compared in sham and OVX myocytes. Mean diastolic Ca²⁺ levels were similar in both groups, but Ca²⁺ transients were 41% larger in OVX cells than in sham controls. In addition, the average rate of rise of the Ca²⁺ transient was increased by 43% in OVX myocytes. The rate of decay also was increased by 71% in cells from OVX mice. To determine whether these differences in contractions and Ca²⁺ transients were present when cells were challenged with more physiological pacing rates, we conducted similar studies in cells paced at 8 Hz. As in cells paced at 2 Hz, contractions were similar in magnitude but had a faster time course in OVX cells compared to sham controls. In addition, Ca²⁺ transients were larger and had a more rapid time course in OVX cells compared to sham controls, as we observed when cells were paced at 2 Hz. The only difference between results obtained at these two pacing frequencies is that the resting levels of diastolic calcium were significantly higher in OVX cells compared to sham controls at 8 Hz, whereas this was only a trend in the 2 Hz data. Together, these observations show that Ca²⁺ transients were larger and faster in myocytes from aged OVX females when compared to sham controls. However, despite the marked increase in peak Ca²⁺ transients, OVX had no effect on peak contraction in cardiomyocytes. The mechanistic basis for this was explored. ## In vivo cardiac contractile function To determine whether *in vivo* contractile function was affected by long-term OVX, two-dimensional guided M-Mode echocardiography was performed. illustrates representative M-mode recordings from sham and OVX mice. Mean results showed all structural parameters measured in systole (e.g. LVIDs, LVPWs and IVSs) were similar in sham and OVX hearts. Although most structural parameters measured in diastole (e.g. LVIDd and LVPWd) were similar in the two groups, IVSd was smaller in OVX mice when compared to sham controls. Thus, there were few differences in structure between sham and OVX hearts. However, mean values for ejection fraction, left ventricular fractional shortening and heart rate also were not affected by long term OVX. These measurements showed that *in vivo* ventricular function was comparable in aged sham and OVX mice. ## Myofilament Ca²⁺ sensitivity is reduced by long-term OVX As OVX had no effect on cardiomyocyte contraction even though Ca²⁺ transients were enhanced, it is possible that OVX reduced myofilament Ca²⁺ sensitivity. To evaluate myofilament Ca²⁺ sensitivity, phase-loop plots of individual shortening-\[Ca²⁺\] relationships were compared in sham and OVX cells; the descending portion of the loop provides a dynamic index of myofilament Ca²⁺ sensitivity. Representative plots show that OVX shifted the descending portion of the loop to the right compared to sham controls. This shift was quantified by comparing the Ca²⁺ concentration at 50% relaxation, as in previous studies. Ca²⁺ levels at 50% relaxation were increased by 46% in OVX myocytes compared to sham controls. These data are consistent with a decrease in myofilament Ca²⁺ sensitivity in the aged OVX group. Next, myofilament Ca²⁺ sensitivity was assessed directly with a Carter assay to measure actomyosin MgATPase activity. shows that the actomyosin MgATPase activity-Ca²⁺ curve was shifted to the right by OVX, when compared to sham controls. Maximal actomyosin MgATPase activity at saturating levels of free calcium (\~10 mM) was not different between the two groups. However, the average concentration of Ca²⁺ required to produce half the maximum response (EC₅₀ values) increased by 26% in OVX when compared to sham controls. Cooperativity, measured by the Hill coefficient, did not differ significantly between the two groups. These results indicate that long-term OVX reduced myofilament Ca²⁺ sensitivity in the aging female heart. ## EC-coupling mechanisms are disrupted in cardiomyocytes from OVX mice To establish the cellular basis for the increase in peak Ca²⁺ transients in OVX myocytes, specific EC-coupling mechanisms were evaluated. First, action potential configurations were compared in myocytes from sham and OVX mice during steady state pacing at 2 Hz. shows representative action potentials recorded from sham and OVX myocytes. Mean data indicate that the resting membrane potential (RMP) was similar in sham and OVX cells. Action potential durations at 50 and 90% repolarization (APD₅₀, APD₉₀) also were similar in the two groups. These data demonstrate that changes in cardiomyocyte action potential configuration were not responsible for the increase in Ca²⁺ release in OVX myocytes. As SR Ca²⁺ release is proportional to the magnitude of the Ca²⁺ current, an increase in Ca²⁺ current could explain the larger Ca²⁺ transients in myocytes from aged OVX mice. To test this idea, voltage clamp experiments were conducted. shows representative Ca²⁺ transients and Ca²⁺ currents recorded from sham and OVX myocytes during a test step from -40 to 0 mV. Mean data show that OVX caused a marked increase in peak Ca²⁺ transients and Ca²⁺ currents. This effect was dramatic, as OVX caused a 48% increase in Ca²⁺ current and a 91% increase in Ca²⁺ transients at the peak of the IV curve. However, because OVX increased both Ca²⁺ current and Ca²⁺ transients, the gain of EC-coupling (Ca²⁺ transient amplitude/Ca²⁺ current) was not affected by OVX. By contrast, diastolic Ca²⁺ levels recorded in sham and OVX myocytes were similar at all test potentials examined. To further evaluate the overall change in Ca²⁺ influx in sham and OVX myocytes, both the time constant for inactivation and the integral of the Ca²⁺ current (step to 0 mV) were compared in the two groups. The time constants were similar in sham and OVX myocytes, but the integral of the Ca²⁺ current was larger in OVX cells when compared to sham-operated controls. These observations indicate that larger Ca²⁺ currents in myocytes from aged OVX mice triggered a larger release of SR Ca²⁺ when compared to age-matched sham controls. ## Impact of long-term OVX on Ca²⁺ handling proteins To determine whether a change in the expression of Ca²⁺ handling proteins contributes to disruptions in Ca²⁺ homoeostasis in the aging OVX heart, we compared the expression of Ca_v1.2, NCX and SERCA2 in sham and OVX hearts. As shown in, Ca_v1.2 protein expression was actually reduced in the OVX group when compared to sham-operated controls. This indicates that an increase in the expression of Ca_v1.2 does not account for the increased Ca²⁺ current in cardiomyocytes from OVX mice. We also compared the expression of NCX and SERCA2 in ventricles from sham and OVX mice. Results showed that the expression of NCX and SERCA2 was not affected by long term OVX in aging mice and suggest that increased expression of these proteins does not contribute to the enhanced rate of relaxation seen in cardiomyocytes from aged OVX animals. ## Long term OVX increases alters unitary SR Ca²⁺ release To determine whether an increase in the size of unitary SR Ca²⁺ release events contributed to the increase in peak Ca²⁺ transients in cardiomyocytes from aged OVX mice, spontaneous Ca²⁺ sparks were compared in the two groups. Representative Ca²⁺ sparks recorded from sham and OVX myocytes are shown in. Mean data demonstrate that Ca²⁺ spark frequency was 76% higher in OVX cardiomyocytes when compared to sham controls. Furthermore, Ca²⁺ sparks recorded in cells from aged OVX mice were 43% larger than those from sham controls. Long-term OVX caused a 22% increase in spark width, measured as the full width half maximum (FWHM), when compared to sham myocytes. Long-term OVX also affected the time course of Ca²⁺ sparks. While spark time-to-peak was reduced in myocytes from OVX mice, the time constant of spark decay (tau) was prolonged. However, the full duration at half maximum (FDHM) did not differ between the two groups. Taken together, these data demonstrate that long-term OVX increases the amplitude and width of the unitary Ca²⁺ release events that underlie the Ca²⁺ transient. OVX also caused a marked increase in the frequency of Ca²⁺ sparks. The mechanism underlying this increase in spark frequency following long-term OVX was explored. ## Long-term OVX augments SR Ca²⁺ loading and promotes spontaneous SR Ca²⁺ release To determine whether the increase in spark frequency in cardiomyocytes from aged OVX mice was caused by an increase in SR Ca²⁺ content, intracellular Ca²⁺ stores were compared in myocytes from sham and OVX animals. shows representative caffeine-induced Ca²⁺ transients recorded from sham and OVX myocytes. The profile of these responses is comparable to caffeine- induced transients recorded under similar conditions in our previous studies. Mean data showed that caffeine-induced Ca²⁺ transients were 90% larger in OVX myocytes compared to sham controls. Long-term OVX increased both Ca²⁺ transients and caffeine-induced transients by approximately 90%, so OVX had no effect on fractional SR Ca²⁺ release (Ca²⁺ transient/caffeine-induced transient;). Furthermore, diastolic Ca²⁺ levels measured at -80 mV in these experiments were similar in myocytes from aged sham and OVX mice. These data demonstrate that long-term OVX promoted SR Ca²⁺ loading, which can explain the increased Ca²⁺ spark frequency recorded in myocytes from OVX mice. To determine whether elevated SR Ca²⁺ content in OVX myocytes enhanced spontaneous SR Ca²⁺ release, spontaneous Ca²⁺ transients were compared in myocytes from sham and OVX mice. shows a representative spontaneous Ca²⁺ transient following a stimulated transient (S) in an OVX myocyte. Long-term OVX did not significantly increase the incidence of spontaneous Ca²⁺ release when compared to sham controls. However, OVX significantly increased the magnitude of spontaneous Ca²⁺ transients by 93% when compared to sham controls. These data confirm that long-term OVX increased SR Ca²⁺ loading and promoted spontaneous SR Ca²⁺ release in the aging heart. # Discussion Our study provides the first evidence that cardiomyocyte Ca²⁺ homeostasis and contractile function are altered by long-term OVX in aging female mice. Myocytes from aging OVX mice had larger and faster Ca²⁺ transients when compared to sham operated controls. Interestingly, this dramatic increase in SR Ca²⁺ release did not enhance contractile function either in ventricular myocytes or *in vivo*. Thus, long-term OVX disrupted the relationship between intracellular Ca²⁺ and cardiac contraction in the aging heart. We examined the underlying mechanisms involved and showed that long-term OVX reduced myofilament Ca²⁺ sensitivity in the aging heart. Results also demonstrated that increased Ca²⁺ current density along with larger, wider Ca²⁺ sparks amplified Ca²⁺ transients in myocytes from OVX mice. The increased Ca²⁺ current in OVX myocytes was not due to an increase in Ca²⁺ channel protein expression. Furthermore, the rapid decay of the Ca²⁺ transient in OVX cells was not due to increased expression of NCX or SERCA2. However, elevated intracellular Ca²⁺ in OVX cells led to higher SR Ca²⁺ loads, increased spark frequency and spontaneous SR Ca²⁺ release. Thus, myocytes subjected to long-term OVX exhibit reduced myofilament Ca²⁺ responsiveness, Ca²⁺ dysregulation and spontaneous SR Ca²⁺ release. Previous studies have shown that SR Ca²⁺ release and contractions are larger in cardiomyocytes from young adult males when compared to age-matched females (reviewed by). Interestingly, short-term removal of ovarian estrogen increases Ca²⁺ transients in myocytes from young adult females but cf., which suggests that short-term OVX leads to changes in SR Ca²⁺ release consistent with conversion to a male phenotype. As peak contractions and Ca²⁺ transients decline with age in cardiomyocytes from males but not females, it is possible that long term OVX actually suppresses SR Ca²⁺ release in the aging female heart. However, the present study reports the novel finding that long-term OVX did not suppress SR Ca²⁺ release but rather enhanced it, even in very old female mice. Thus, long-term deficiency of ovarian-derived estrogen, including the lack of normal pubertal estrogen, does not produce changes in SR Ca²⁺ release consistent with conversion of a female to a male phenotype. The impact of age on contraction in older males might be due to lower testosterone levels that can inhibit SR Ca²⁺ release and suppress contractions, as seen in myocytes from younger males after gonadectomy. We and others have previously shown that SR Ca²⁺ release is augmented in cardiomyocytes from young adult mice, 3-26 weeks after OVX. In young OVX mice, we found that increased SR Ca²⁺ release was not due to an increase in Ca²⁺ current. Instead, this was the result of higher gain due to elevated SR Ca²⁺ content and an increase in the amplitude of Ca²⁺ sparks. summarizes the major effects of OVX on key Ca²⁺ handling parameters in young adult OVX mice as reported in our previous study. The present study extends these findings to demonstrate that enhanced SR Ca²⁺ release persists in the aging heart, even almost two years after OVX. A key observation in this study is that effect of OVX on specific mechanisms involved in cardiomyocyte homoeostasis differed markedly in the aging heart when compared to young adult hearts, as shown in that compares our present study and our earlier work. While the impact of OVX on SR Ca²⁺ content was similar regardless of age, Ca²⁺ transients increased in parallel with Ca²⁺ current in aged OVX myocytes, with no effect on the gain of SR Ca²⁺ release. As the magnitude of SR Ca²⁺ release is proportional to the size of the Ca²⁺ current, our results indicate that, unlike young OVX mice, an increase in Ca²⁺ current is a key mechanism responsible for augmenting SR Ca²⁺ release following long-term OVX in the aging heart. We also demonstrated that the increase in peak Ca²⁺ current was not due to an increase in the expression of Ca_v1.2, which has been reported in hearts from younger animals subjected to short term OVX. Indeed, we showed that Ca_v1.2 expression was actually reduced following long-term OVX in the aging heart. The increase in the activity and expression of protein kinase A (PKA) that occurs following OVX may be implicated in the increase in peak Ca²⁺ current, as discussed in more detail below. The results of the present study also demonstrate that that long-term OVX had much more dramatic effects on Ca²⁺ spark properties than short-term OVX, as shown in. While OVX in young adult mice increased spark amplitude by 11%, long-term OVX increased spark amplitude by 43% and spark width by 22%, resulting in a marked increase in the overall size of these release events. Long-term OVX also modified the spark time course, by abbreviating time-to-peak and prolonging spark decay, although these changes did not affect overall spark duration. However, the shorter time-to-peak may underlie the increased rate of rise of the Ca²⁺ transient and shorter time-to-peak contraction observed in OVX myocytes in this study. Together, these observations suggest that a marked increase in the size of individual Ca²⁺ release events contributes to enhanced SR Ca²⁺ release following long-term ovarian estrogen deprivation in the aging heart. As OVX increases SR Ca²⁺ release, it would be expected to increase the magnitude of cardiac contraction, although this was not explored in our earlier study. A novel and important finding in the present study was our observation that the relationship between intracellular Ca²⁺ and contraction was dramatically altered by long-term OVX in the aging mouse heart. Even though peak Ca²⁺ transients were increased by 91% following long-term OVX, this did not augment cardiomyocyte contraction, even when cells were paced at rapid stimulation frequencies. Furthermore, this marked increase in SR Ca²⁺ release at the myocyte level had no effect on cardiac contractile function *in vivo*, as measured by echocardiography. We found that the underlying mechanism was a decrease in myofilament Ca²⁺ sensitivity in the aging heart. Reduced myofilament Ca²⁺ sensitivity would normally be expected to reduce cardiac contractile function. However, contractile function was preserved by a marked increase in the amount of Ca²⁺ available to the myofilaments in the aged, estrogen-deprived heart. Our observation that contractile function is maintained by elevated Ca²⁺ in the aging, estrogen-deprived female heart may help explain why older women are predisposed to heart failure with preserved ejection fraction rather than systolic heart failure, which occurs characteristically in older men. Elevated SR Ca²⁺ content in the aging OVX heart could arise from increased Ca²⁺ influx, as reported here, together with increased SERCA2a levels. The rapid decay rates of Ca²⁺ transients and contractions observed here and previously are compatible with increased SR Ca²⁺ uptake and/or extrusion in OVX myocytes. However, as shown in the present study and previously, the abundance of SERCA2 is unaffected by OVX and some studies have reported it is actually reduced by OVX. Furthermore, as demonstrated in our study and previously, the expression of NCX is either unaffected by OVX or is reduced. Together these observations indicate that the rapid decay of the Ca²⁺ transients observed following long-term OVX is not due to an increase in the expression of SERCA2 or NCX. However, it is possible that the observed decline in myofilament Ca²⁺ sensitivity explains the rapid decay of Ca²⁺ transients in OVX myocytes. A decrease in myofilament Ca²⁺ sensitivity could make intracellular Ca²⁺ more rapidly available to SERCA in the OVX heart. When SR Ca²⁺ load is high, the SR releases Ca²⁺ in the form of spontaneous Ca²⁺ sparks to limit SR Ca²⁺ content. It is likely that the increase in Ca²⁺ spark frequency we observed in myocytes from aged OVX mice occurred in response to the marked increase in SR Ca²⁺ load. Previous studies have shown that advanced age increases SR Ca²⁺ content in myocytes from female rodents. The results of the present study suggest that increased SR Ca²⁺ load in aged female myocytes is mediated by declining estrogen levels in aged female rodents. It is well established that high levels of SR Ca²⁺ lead to Ca²⁺ overload and induce spontaneous Ca²⁺ release from the SR. Indeed, we observed spontaneous SR Ca²⁺ release in myocytes from both sham and OVX animals, although the magnitude of this effect was significantly larger in the OVX group. Ca²⁺ overload and spontaneous SR Ca²⁺ release can disrupt myocardial function and lead to abnormal electrical and contractile activity. Our results demonstrate that ovarian estrogen deprivation leads to profound Ca²⁺ dysregulation and the initiation of spontaneous SR Ca²⁺ release. It is possible that disruptions in myocardial Ca²⁺ homeostasis induced by long-term estrogen deprivation may increase susceptibility to cardiovascular diseases such as arrhythmias in the aging female heart. Estrogen levels have been shown to decline with age in female rodent models and this is exacerbated in the setting of OVX. This suggests that the profound Ca²⁺ dysregulation observed in the aging OVX heart is linked to low estrogen levels. Still, the pathway by which estrogen may modify Ca²⁺ handling has not yet been identified. We found no evidence for increased expression of the key Ca²⁺ handling proteins Ca_v1.2, NCX or SERCA2, which suggests that post-translational modifications of these and possibly other proteins may be important. One central pathway in the regulation of Ca²⁺ handling in cardiomyocytes is the cAMP/ PKA pathway. Previous studies have shown an increase in both basal and β-agonist stimulated-PKA activity in hearts from rats 6 weeks after bilateral OVX as well as an increase in PKA expression. Furthermore, estrogen replacement has been shown to restore PKA activity and expression levels to control values. Increased PKA activity is known to phosphorylate various downstream targets and could modify Ca²⁺ handling in the OVX heart in a manner consistent with that seen in the present study. For example, L-type Ca²⁺ channel phosphorylation will increase peak Ca²⁺ current and thereby increase Ca²⁺-induced Ca²⁺ release from the SR. Furthermore, phosphorylation of phospholamban will increase SERCA2a activity and thus enhance the rate of SR Ca²⁺ uptake. Phosphorylation of troponin I at N-terminal serines promotes faster relaxation by facilitating dissociation of Ca²⁺ from the myofilaments and reducing myofilament Ca²⁺ sensitivity. Phosphorylation of the ryanodine receptor, although controversial, could help explain the increase in Ca²⁺ spark amplitudes reported in the present study. Increased PKA activity may be particularly important *in vivo*, as there is evidence that OVX enhances depolarization-induced norepinephrine release and elevates sympathetic tone in the heart. Stimulation of the cAMP/PKA pathway also could account for the increase in spontaneous Ca²⁺ release in myocytes from aged OVX mice and could promote arrhythmias in the aging female heart. Further exploration of the role of this cAMP/PKA pathway in modifying Ca²⁺ handling in the OVX heart could be illuminating. There is recent evidence that the production of reactive oxygen species (ROS) is increased in the aged OVX heart and this may explain some of our findings. For example, increased ROS activity has been shown to increase ryanodine receptor activity, which could contribute to the increase in Ca²⁺ sparks reported in our study. Increased ROS activity also has been reported to increase peak Ca²⁺ current in some models and could contribute to the enhanced Ca²⁺ current we observed in myocytes from aged OVX mice. Furthermore, as oxidative stress has been shown to reduce myofilament Ca²⁺ sensitivity in the heart, it is possible that an increase in production of ROS leads to the decrease in myofilament Ca²⁺ sensitivity observed in our study. On the other hand, previous studies have reported that short-term (10 weeks) OVX increases myofilament Ca²⁺ sensitivity, so prolonged ovarian estrogen withdrawal may be required to desensitize myofilaments in the aging female heart. ROS also reduces SERCA2 activity, which is not compatible with the faster time courses of contraction and Ca²⁺ transients observed in our study. Additional experiments to investigate the role of ROS in Ca²⁺ dysregulation in the estrogen-deprived heart could be informative. Previous studies have provided evidence that OVX can modify the structure of the heart. While heart weight-to-body weight ratios are similar, echocardiography has revealed increased IVSd thickness, increased wall thickness and reduced LVID in young adult mice after 10 weeks of OVX. In agreement with previous studies in mice and rats, we found that ventricle weight-to-body weight ratios were similar in sham and OVX mice. By contrast, long-term OVX had no effect on LVPW thickness or LVID measured in systole or diastole and actually reduced IVSd thickness. These observations suggest the structural changes in the whole heart observed early after OVX may be transient. The effects of OVX cardiomyocyte structure may also depend on the duration of ovarian steroid withdrawal. Shorter periods of steroid withdrawal (e.g. \<26 weeks) have no effect on cardiomyocyte capacitance. However, we found that long term OVX reduced cell capacitance and cell volume, thereby reducing cardiomyocyte membrane area. Whether this is due to long term remodeling of membranes such as the t-tubules or caveolae remains to be determined. There are limitations to the experimental approaches used in this study. Our studies did not investigate OVX mice treated with estrogen replacement, although this would be an interesting area for additional investigation in future studies. In addition, we removed the ovaries early in life to determine whether long-term estrogen withdrawal would result in an aging phenotype characteristic of that seen in aging males with respect to Ca²⁺ handling. In consequence, the period of estrogen withdrawal was prolonged and the effects on myocardial Ca²⁺ homeostasis were dramatic. This also produced a model that was characterized by the lack of exposure of the heart to normal pubertal systemic estrogen modeling. This model of early ovarian steroid withdrawal has been used previously in other studies, but it contrasts with the more commonly used approach where the effects of OVX are investigated in adult animals where tissues have already been exposed to estrogen. Additional experiments with other time frames for estrogen deprivation could be explored in the future. Although the ovaries are the primary source of estrogens, other tissues such as adipose tissue, vascular tissue and bone, express the enzyme aromatase that can convert testosterone to 17ß-estradiol. These non-gonadal sources of estrogen could be important in OVX animals. Furthermore, aromatase is expressed in neonatal tissues and cardiomyocytes and in the adult rodent heart. It is possible that androgens can be locally converted to estrogens in the myocardium. Whether exposure of cardiomyocytes to sex steroid hormones can be regulated at the local tissue level is an important area for further investigation. In summary, our study showed that long-term deprivation of ovarian estrogen disrupted myocyte Ca²⁺ homeostasis and contractile function in the aging female heart. Although Ca²⁺ transients were larger in OVX myocytes, *in vitro* and *in vivo* fractional shortening were similar in sham and OVX mice. The underlying mechanism involved a decrease in myofilament Ca²⁺ sensitivity in the aging OVX heart. The increase in peak Ca²⁺ transients in OVX myocytes was mediated by an increase in both Ca²⁺ current and the size of unitary Ca²⁺ release events. Higher intracellular Ca²⁺ led to an increase in SR Ca²⁺ load, an increase in spark frequency and spontaneous SR Ca²⁺ release. These results demonstrate that long-term ovarian estrogen deprivation reduces myofilament Ca²⁺ sensitivity, promotes Ca²⁺ dysregulation, and increases spontaneous Ca²⁺ release in the aging female heart. # Supporting Information The authors express their appreciation for excellent technical assistance provided by Dr. Jiequan Zhu, Peter Nicholl, and Rick Livingston. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: EF WGP GR RAR EMDW RPC SEH. Performed the experiments: EF WGP RPC GR SEH. Analyzed the data: EF WGP GR RAR EMDW RPC SEH. Contributed reagents/materials/analysis tools: SEH RAR EMDW WGP. Wrote the manuscript: EF WGP GR RAR EMDW RPC SEH.

# Introduction Numerous studies provide evidence that the human immunodeficiency virus 1 (HIV-1) infection is sex biased. In women, lower viral loads and higher CD4 T cell counts are measured in all stages of the disease. At the same viral load, women are at a 1.6-fold higher risk of progression to AIDS compared to men. Various factors contribute to these differences. A major role is attributed to sex differences in the immunity against HIV-1. HIV-1 infection profoundly interferes with the immune system and elicits persistent immune activation and inflammation, which are associated with progression to AIDS. A critical driver of this pathology is the interferon response of plasmacytoid dendritic cells (pDC) to HIV-1: sensing of viral components elicits an interferon alpha response by pDC, which is stronger in women compared to men. This translates to an increased immune activation and inflammation in women with downstream effects on components of the adaptive immune system such as B or T cells. B cells and more importantly T cells play a pivotal role in the pathogenesis of HIV-1 infection. Also, they are subject to immunological sex differences. A better understanding of sex differences in the pathogenesis of HIV-1 infection warrants the investigation of sex differences in B and T cells in HIV-1 infection. Therefore, this study aimed at identifying sex differences in B and T cell biology and investigating their relationship to markers of disease progression. Gene expression profiles obtained from the NCBI Genome Expression Omnibus (NCBI/GEO) and literature mining were integrated to identify candidate genes for analysis by reverse transcription quantitative polymerase chain reaction (RT- qPCR) and flowcytometry. We identified a sex bias in *dipeptidyl peptidase 4* (*DPP4*) expression in T cells, which correlated with markers of disease progression. These data indicate a new line of research on the sex bias in HIV-1 pathogenesis. # Materials and methods ## Screening of microarray data for sex differences B and T cells constitute a significant quantity of cells circulating in the peripheral blood. We assumed that a sex bias in their gene expression might be reflected in an overall sex bias in the peripheral blood. Therefore, three large-scale and publicly available gene expression profiling microarray datasets of the peripheral blood were investigated for sex differences in gene expression. Datasets were selected based on the following criteria: The datasets had to be derived from single-color gene expression profiling by microarray. Experimental design had to be available. Information on the corresponding data analysis pipeline had to be accessible and the data had to be preprocessed, normalized and quality checked. The samples had to be acquired by PaxGene tubes from whole blood. Information on sex, age and health status had to be available for each sample. The included samples had to be obtained from healthy individuals aged between 17 and 43 years of age. The datasets had to include at least ten samples of each, female and male donors. Three single-color microarray gene-profiling datasets matching these criteria were selected and downloaded from NCBI/GEO. The respective datasets were provided to NCBI/GEO by Joseph E. Powell from Brisbane (Australia, GSE53195), Thomas L. Ortel from Durham (USA, GSE19151) and Guillemette Duchateau-Nguyen from Basel (Switzerland, GSE16028). In total, 245 gene expression profiles of healthy individuals (HI) aged between 17 and 43 years were analyzed for sex differences in gene expression employing the GEO2R algorithm (NCBI/GEO). Of these samples, 53% were obtained from women. The respective analyses were carried out using R/Bioconductor software. The GEO2R script was used to investigate the microarray datasets obtained from NCBI/GEO. Sex biased genes were filtered for consistent up or downregulation in the three microarray datasets and for statistically significant differences in gene expression. To reduce the number of false positives, only genes were selected, which were highly significantly sex biased (p ≤ 0.01) in at least two of the three datasets and significantly sex biased (p ≤ 0.05) in all datasets. ## Analysis of gene ontology categories Gene ontology (GO) category analysis was used to investigate the biological context of sex biased genes observed in the microarray data. GO analysis was performed using the Molecular Signatures Database published and maintained by the Broad Institute. ## Literature mining of sex biased genes An algorithm written in R was used to count occurrences of sex biased genes in distinct sets of abstracts about HIV, B cells and T cells (R code). The abstracts were obtained from PubMed. To identify gene occurrences, each set of abstracts was searched for the occurrence of gene names, symbols, and aliases of the respective genes. The required symbols, names and aliases of genes were obtained from the Human Genome Organization’s (HUGO) Gene Nomenclature Committee (HGNC) webpage. Test datasets were used to optimize the algorithm’s performance and to test its specificity and sensitivity. A specificity of 99.71% could be achieved at a sensitivity of 97.34%. PubMed was searched for publications about HIV, B and T cells. The respective abstracts were searched for occurrences of sex biased genes as observed in the microarray. Genes occurring in abstracts about HIV and B or T cells were selected for subsequent analysis by RT-qPCR and flowcytometry. ## Patient samples for RT-qPCR and flowcytometry Altogether, 33 healthy individuals (HI) and 68 persons living with HIV-1 (PLHIV) were included at the University Hospitals in Munich and Hamburg. All were included after informed consent was obtained. All participants were of full age (≥ 18 years) and the groups were age matched. PLHIV were treatment-naïve and started combined antiretroviral therapy (cART) shortly after blood draw. PLHIV comprised two subgroups: 24 controllers with stable disease (stable CD4 T cell counts, stable HIV viral loads and no occurrence of HIV associated or AIDS defining diseases in follow-up visits), CD4 T cell counts greater than 500 cells/μl and average viral loads lower than 5,000 copies/ml (some exhibited slightly higher viral load at blood draw but not at earlier and later visits) in the absence of antiretroviral treatment and 44 progressors not meeting these criteria. Blood was collected by venipuncture. Protocols were approved by the ethics committees of the University of Hamburg and the LMU Munich. ## Analysis of gene expression by RT-qPCR RT-qPCR was carried out using freshly isolated peripheral blood mononuclear cells (PBMC) as described elsewhere. In short, PBMC were obtained using Ficoll- density gradient centrifugation. RNA was extracted from 10^6 PBMCs using a Trizol based protocol. cDNA was synthesized using Superscipt III (Invitrogen) with random hexamer primers as per manufacturer’s instruction. The RT-qPCR was carried out using the SYBR Green Mastermix (ThermoFisher) and specific primers. Samples were measured on the Rotor-Gene Q (Qiagen). Relative mRNA expression was computed with the R/Bioconductor/EasyqpcR software using the qBase algorithm and normalizing on the housekeeping genes *HPRT1* and *HMBS*. ## Surface expression of DPP4 on lymphocyte subsets Cryopreserved PBMC were analyzed by flowcytometry on a LSRFortessa (BD) using the viability dye ZombieNIR and the fluorochrome-conjugated antibodies FITC-CD3, PECy7-CD4, PerCP-CD8, BV510-CD19, BV421-CD38 and PE-CD26/DPP4 from BioLegend. Data was analyzed with R/Bioconductor as outlined by O’Neill et al. (gating hierarchy and exemplary gates). ## Statistical analysis Statistical analysis was performed using the R programming language (version 3.6). Differences in medians were tested using the Mann-Whitney U test. If the test hypothesis was one-sided, a one-sided test was chosen. Correlations were identified computing Spearman’s rho. P-values were adjusted for multiple comparisons using the Benjamini-Hochberg correction, if applicable and referred to as q-values. # Results ## Sex differences in microarray data B and T cells account for a significant number of leukocytes. A sex bias in their gene expression might be reflected in the overall gene expression in the peripheral blood. To identify sex biased genes, sex differences in gene expression were computed for three microarray datasets obtained from NCBI/GEO. Of 24,449 investigated genes, only genes were selected which were highly significantly sex biased (*p* ≤ 0.01) in at least two of the three datasets and significantly sex-biased (*p* ≤ 0.05) in all datasets. Some genes were inconsistently sex biased comparing the three datasets (i.e. higher expression in women in one dataset, higher expression in men, in the other) and were removed. Overall, this yielded 31 sex biased genes. GO category analysis was used to examine the general context of the 31 sex biased genes. The strongest overlaps of the 31 sex biased genes with GO categories were found for the categories “*GO_IMMUNE_RESPONSE”* (q = 0.0001), “*GO_IMMUNE_SYSTEM_PROCESS”* (q = 0.0001) and “*GO_DEFENSE_RESPONSE”* (q = 0.0002). These categories strongly related to the immune system and the immune response. Altogether, 11 of 31 genes were associated with any process related to the development of functioning of the immune system. ## Literature mining of sex biased genes PubMed was searched for publications about HIV, B and T cells and the respective abstracts were downloaded. This produced three sets encompassing a total of 221,326 abstracts. The 31 sex biased genes were analyzed for occurrences in each set of abstracts. Six genes occurred at least once in abstracts about HIV and B or T cells. The three genes with the lowest average *p-*values comparing the three microarray datasets (*SOCS3*, *DPP4* and *FCGR1A*) were selected for analysis by RT-qPCR. ## Sex bias in the expression of *DPP4*, *FCGR1A* and *SOCS3* in healthy individuals Gene expression of *DPP4*, *FCGR1A* and *SOCS3* was analyzed in PBMC obtained from HI by RT-qPCR. In accordance with the microarray data, it was hypothesized that mRNA-expression was higher in women. Indeed, relative mRNA expression of *DPP4*, *FCGR1A* and *SOCS3* was higher in women compared to men (*p* = 0.00029, *p* = 0.003 and *p* = 0.001, respectively;). ## DPP4+ T cells in healthy individuals The sex bias in gene expression measured by RT-qPCR was statistically most significant for *DPP4*. DPP4, also known as CD26, occurs integrated into the cellular membrane of many tissues or as soluble form in virtually all body- fluids. On B and T cells, engagement of DPP4 promotes a co-stimulatory signal and its surface expression is associated with B and T cell activation. Therefore, DPP4 expression was analyzed on B and T cells of HI by flowcytometry. Since a higher *DPP4*-mRNA expression was observed in women in both RT-qPCR and microarray data, we hypothesized, that DPP4 would be expressed at higher rates on B or T cells obtained from healthy women compared to men. In fact, higher frequencies of CD4+DPP4+ and CD8+DPP4+ T cells were observed in women (90.9% \[88.6% - 91.5%\] vs. 85.76% \[82.8%, 90.7%\]; *p* = 0.027 and 61.5% \[58.7%, 65.2%\] vs. 55.5% \[44.9%, 56.8%\]; *p* = 0.039). For B cells, no sex bias was found. ## DPP4+ T cells in HIV-1 infection Overall, lower frequencies of CD4+DPP4+ T cells were observed in PLHIV compared to healthy controls. Female progressors exhibited lower CD4+DPP4+ frequencies than female controllers (74.4% \[51.6%, 81.7%\] vs. 84% \[75%, 88.1%\], *p* = 0.01;), while no such difference was found in male controllers. Accordingly, significantly higher frequencies of CD4+DPP4+ T cells were observed in female controllers compared to male controllers (84% \[75%, 88.1%\] and 75.3% \[71.1%, 81.8%\], respectively; *p* = 0.037;), while no sex bias was observed in progressors. The percentages of CD8+ T cells expressing DPP4 were much lower in PLHIV compared to HI. Overall, this was more pronounced than in CD4+DPP4+ T cells. Even though higher frequencies of CD8+DPP4+ T cells were observed in female HI compared to male HI, no sex differences were seen in this population among PLHIV. ## DPP4+ T cells and markers of disease progression Since lower frequencies of CD4+DPP4+ T cells were observed in progressors, the relationship of CD4+DPP4+ T cells with markers of disease progression was assessed in PLHIV. Plasma viral load, absolute CD4 T cell counts and CD8+CD38+ T cells are associated with disease progression. Overall, frequencies of CD4+DPP4+ T cells correlated with absolute CD4 T cell counts (r = 0.3; *q* = 0.026), while no statistically significant correlations were observed with viral load or T cell activation. When examining these relationships separately for both sexes, strong positive correlations were observed for CD4+DPP4+ T cells with CD4 T cell counts in women (Spearman’s r = 0.50, *p* = 0.0032), but not in men (r = 0.038, *p* = 0.83). This association remained significant after adjusting for multiple comparisons (*q* = 0.019). Conversely, a trend towards a negative correlation of viral load and DPP4+CD4+ T cells was observed in women (r = -0.3, *p* = 0.087) but not in men (r = -0.099, *p* = 0.57). CD4+DPP4+ T cell frequencies did not correlate with frequencies of CD8+CD38+ T cells. Intriguingly, CD4+DPP4+ T cell frequencies correlated positively with frequencies of CD4+CD38+ T cells in women (r = 0.49, *p* = 0.0036), but not in men (r = 0.13, *p* = 0.46). # Discussion In 2018, about 38 million people lived with HIV, approximately half of whom were women. To date, few studies examined sex differences in the immune response against HIV-1. T cells play a crucial role in HIV-1 pathogenesis. Here, we observed a yet unknown sex bias in DPP4+CD4+ T cells. As previously reported, a significant number of sex biased genes of the peripheral blood is associated with the immune system and expressed at higher levels in women. These genes include autosomal as well as X- and Y-linked genes. Accordingly, GO category analysis of the 31 sex biased genes, which were identified in this study, were immunity associated: 11 of the 31 genes were related to the development or functioning of the immune system. In addition, most of these genes were expressed at higher rates in women compared to men. Integration of microarray data with literature mining identified six of the 31 sex biased genes, which were relevant in HIV pathogenesis, and linked to B and T cell biology. The three most significantly sex biased genes were investigated by RT-qPCR confirming the sex bias observed in the microarray data. While no sex bias had been previously reported for *FCGR1A* and *SOCS3*, Pérez-Durillo et al. observed higher plasma levels of the soluble protein dipeptidyl peptidase 4 (DPP4) in women compared to men. In our data, *DPP4* exhibited the most significant sex bias in mRNA expression of the three genes as tested by RT-qPCR. The gene *DPP4* is located on long arm of chromosome 2. It encodes the protein dipeptidyl peptidase 4 (DPP4), also known as CD26 or adenosine deaminase complexing protein 2 (ADCP2). DPP4 exhibits peptidase activity and occurs integrated into the cellular membrane of many tissues and cell types or as soluble form in virtually all body-fluids. Multiple studies support the role of *DPP4* in immunity through peptidase activity and as a cell-surface receptor. Therefore, we investigated DPP4 cell surface expression on B and T cells by flowcytometry. We observed higher frequencies of DPP4 on T cells obtained from healthy women compared to men. On T cells, DPP4 expression is associated with T cell activation. Generally, T cell activation is stronger in women. Higher DPP4 expression on T cells of women might reflect stronger T cell activation in women. Data on the role of DPP4 in HIV-1 infection is conflicting. DPP4 expression on T cells has been linked to susceptibility of T cells to viral entry. Cleavage of C-C motif chemokine 5 (CCL5) by DPP4 increases the anti-viral effect of CCL5, while cleavage of stromal cell-derived factor-1 (SDF-1) mitigates its antiviral activity. The HIV proteins tat and gp120 inhibit DPP4 function. Previous clinical studies found that inhibition of DPP4 by specific DPP4 inhibitors resulted in slightly increased rates of infectious diseases and might impair T cell response and function, while T cell frequencies were not affected. Interestingly, high DPP4 expression might protect from contracting HIV infection. Female sex workers exposed to HIV, who were not infected with HIV, expressed higher levels of DPP4+ T cells compared to healthy female controls. Together, these observations suggest that DPP4 promotes a robust anti-viral immune response. In PLHIV, lower frequencies of DPP4+ T cells occur very early during HIV infection and are not reversed by cART. In accordance with these observations, we found lower frequencies of CD4+ DPP4+ and CD8+DPP4+ T cells in PLHIV compared to HI. DPP4 is highly expressed on Th17 T cells. Possibly, low frequencies of DPP4 T cells in HIV-1 infection are be due to the early loss of Th17 T cells in HIV infection. Another conceivable cause of reduced DPP4+ T cell frequencies might lie in the early infection and destruction of memory/helper T cells expressing the CD4+CD45RO+CD26+ phenotype. However, both theories cannot fully explain the reduction DPP4+ T cells within the CD8+ subset. Intriguingly, we found that CD4+DPP4+ T cells were sex biased in HI and controllers, but not in progressors. Previous findings showed that sex differences in plasma viral load were less pronounced in advanced HIV-1 infection. Our data suggest that this might be reflected in the absence of sex bias in CD4+DPP4+ T cells among progressors contrasting the sex bias which can be observed among HI and healthy individuals. For CD8+DPP4+ T cells, no sex bias was observed among PLHIV, while HI exhibited higher frequencies in women compared to men. Possibly this is due to the strong reduction of CD8+DPP4+ T cell percentages in HIV-1 infection, which was more pronounced for CD8+ T cells than for CD4+ T cells. In consequence, the effect size of a sex bias in percentages of CD8+DPP4+ T cells might be smaller or absent and therefore might have escaped detection in our study. Given the decrease of CD4+DPP4+ T cells in progressive HIV-1 infection, we assessed the correlation of DPP4+CD4+ T cells with markers of disease progression. A strong correlation for CD4 counts and a trend towards an inverse correlation with viral load were observed in female, but not in male PLHIV. No association could be identified for CD8+ CD38+ T cells. Intriguingly, DPP4+ and CD38+ T cells within the CD4+ subset correlated positively in women, but not in men. Taken together, these results suggest that high CD4+DPP4+ T cell frequencies correlate with more favorable prognostic surrogate markers of HIV-1 infection in women but not in men. Recent findings showed that lower levels of soluble DPP4 were associated with a poorer prognosis in HIV-1 infection. Since DPP4 is involved in T cell function and homeostasis, lower levels of CD4+DPP4+ T cells might reflect the progressive immune deterioration in HIV infection. Our data suggest that the loss of DPP4 is associated to and might be involved in this process and that its dynamics differ comparing women and men. In addition, our results suggest, that an evaluation of DPP4 as prognostic marker needs to take its sex bias into account. An important limitation of the present study is its genetic diversity. We assessed sex differences in DPP4 on the mRNA and protein levels in a cohort of 101 individuals. All except one participant were of Caucasian race. Therefore, the present work is limited to a narrow genetic background. HIV infection is, however, a disease afflicting more non-Caucasian individuals. Future studies should include a genetically more diverse study population. In summary, we identified sex differences in the pathobiology of T cells in HIV-1 infection using a data-driven approach. Our observations illustrate that DPP4 is a component of T cell biology involved in sex differences in chronic HIV-1 infection. This opens a new line of research on sex differences in HIV-1 pathogenesis and might have important implications for the use of DPP4 as a prognostic parameter. # Supporting information We are grateful to our patients and healthy donors who contributed to this study. [^1]: The authors have declared that no competing interests exist.

# Introduction Risk evaluation and control are important components of healthcare operations. Ideally, providers would like to be able to predict health risks early in hospital admissions to take subsequent controlling actions. Different methods and techniques have been developed for this purpose, one of which is the early warning system. An early warning system is a measurement tool to assess patient health risks objectively and to quickly determine the degree of illness. Several major early warning systems exist and each type offers a slight variation in risk- assessment parameters. The Modified Early Warning Score (MEWS), the focus in our study, is a commonly used triage tool to quickly determine the severity of a hospitalized patient’s illness. The MEWS system utilizes five major vital signs—systolic blood pressure, pulse rate, respiratory rate, temperature, and the AVPU score (A: alert, V: responding to voice, P: responding to painful stimuli, U: unresponsive)—to assign an aggregate number to each patient, which comprises the patient’s health risk. Obtaining a MEWS involves assigning a number between 0 and 3 to each of the 5 vital signs. The sum of the 5 numbers yields the patient’s total MEWS score—between 0 and 15. A total score of four or higher prompts a nurse to call the patient’s physician or the hospital’s rapid response team (for more information, see reference). Other early warning systems follow a similar logic. The Standardized Early Warning System (SEWS) adds oxygen saturation level (SpO2%) to the five MEWS vital signs to detect a patient’s deterioration. The Decision-Tree Early Warning Score (DTEWS) is a decision-tree analysis based on a database of vital signs. The National Early Warning Score (NEWS) utilizes seven variables to identify deteriorating patients: respiratory rate, oxygen saturation, any supplemental oxygen, temperature, systolic blood pressure, heart rate, and level of consciousness. The main use of early warning scores is to provide early warnings to health providers to spur quick, preventive reactions. Some experts have argued that these measures in fact have much more to offer, such as helping to predict patients’ length of stay (LOS) in hospitals or health outcomes—e.g., the chance of in-hospital mortality. Given the importance of LOS and mortality in assessing healthcare quality, resource allocations, and costs, the hope is that early warning scores can make it possible to explain and improve hospital utilization and outcomes. While a warning system can be an efficient and rapid way of reducing or preventing life-threatening events, there is mixed evidence regarding the effectiveness of early warning scores. On the one hand, MEWS can predict an increased risk of death or admission to an intensive care unit (ICU) or high dependency unit (HDU). MEWS can also be used to identify patients who need hospital admission as well as those at higher risk of in-hospital death. The proportion of patients admitted and who subsequently died in the hospital was significantly higher for higher MEWS values. SEWS have also been correlated with a patient’s LOS. On the other hand, some studies have suggested that further work is needed to derive and validate early warning scores, and that scores that utilize inappropriate parameters and cut-off points should be replaced with those with higher diagnostic accuracy. A potential limitation to these and many other similar studies is a lack of detailed administrative-level data on patient and physician characteristics. Some of these studies focus on assessing simple correlations between two variables without controlling for variation across patients or across physicians. Thus, whether early warning scores influence physicians’ diagnoses and risk assessments remains an open question. In other words, we do not know if physicians are actually using these measures and if such measures are preferred over early subjective assessments. In this study, we assess whether a patient’s admission MEWS and vital signs are associated with LOS and in-hospital mortality. We also compare and contrast MEWS as an objective measure with physicians’ subjective-risk assessments early after admission. # Methodology We collected detailed data for 1,021 randomly selected patients admitted between January and September 2014 to an over-500-bed medical center (Lewis Gale Medical Center) in the state of Virginia. Our study was approved by the Lewis Gale Medical Center Education Department as patient records were anonymized and de- identified prior to analysis and the dataset is considered exempt by Virginia Tech’s Institutional Review Board. Patient records were accessed through the electronic patient management systems (MEDITECH and Crimson), which contain full patient demographic data along with individual records and medical documents including various vital signs, MEWS, clinicians’ early assessments of severity level and mortality risk, and patient outcomes such as hospital LOS and disposition conditions. Physicians’ identifiers were included in the data. Newborns and patients admitted to the ICU were excluded from our study. The data is available in file. shows the descriptive statistics for some of the main variables in our analysis that are stratified by dead versus survivors. Severity level and mortality risk are categorical variables between 1 and 4, where 1 represents the lowest severity level or mortality risk. In addition, we controlled for gender, weight, height, BMI, physician, and additional physiological measurements including oxygen saturation level and diastolic blood pressure (DBP). The variable physician is a nominal variable assigned to each attending physician who makes the subjective measures. AVPU is a categorical and nominal variable with four values (A, P, V, and U). Table A1 in reports the variables and correlations between any pairs of the variables. We found no serious correlation between the variables other than between two subjective assessments of physicians—that is, physicians’ assessment of mortality risk and severity level, which was expected. We investigated associations between different subjective and objective measures and our outcome variables. Our analysis included a range of stepwise regressions. In this study, and in the interest of parsimony, we focused on two main models: an ordinary least square, fixed-effect model to explain LOS, and a logistic regression model to explain mortality; both models were based on a wide range of independent and control variables. In addition, we analyzed effects of vital signs and early-risk indicators on physicians’ early subjective-risk assessments. # Results summarizes the results of our stepwise regression analysis for LOS in which different independent variables were controlled in each step. In addition to the presented variables, patient demographics were controlled. Although the first column in shows that the MEWS’s coefficient is significant, the low value of R² (0.02) depicts that MEWS is only slightly better than a random guess for explaining LOS, This means that the whole model only predicted 2% of LOS variation. In models M2–M6, however, MEWS is no longer significant. This is not unexpected, and one may argue that it does not speak against the usefulness of MEWS. We point out two major observations: (1) the model with only MEWS as a control variable has a very low predictive power (*R*²) even though MEWS is significant. This resonates with the common statistical argument that “significance” is different than “meaningfulness”: MEWS is significant in M1, but model M1 is only slightly better than a random guess; (2) as shown in the results of M2–M6, not all variables within MEWS (i.e., vital signs) are significant. In other words, better warning scores can be constructed based on weighted scores of some but not all of the vital signs. In models M5 and M6, vital signs lose their significance. In this step-wise analysis, R² increases from 0.02 in M1 to 0.33 in model 6. Finally, M7 shows the results of a model with the subjective measures and MEWS (but not the components of MEWS) in which MEWS is not a significant variable. In sum, as shown in the table, in more complete models, none of the MEWS or vital signs at the time of admission is associated with LOS. However, physicians’ subjective assessments represented by their assessment of mortality risk and severity level are significantly associated with LOS. We later analyze how physicians make subjective assessments. We now look at the mortality outcome variable. shows the results of our mortality analysis with logistic regressions and reports the odds ratios. In model M1, MEWS is positively associated with mortality. In models M2–M5, we find that in controlling for more variables, MEWS loses its statistical significance and that the direction of association is negative in M3–M5. Physicians’ subjective assessments seem to have a higher explanatory power of mortality. Specifically, physicians’ subjective assessment of mortality risk is significantly and positively associated with the actual mortality. This finding leads to the conclusion that patients with higher mortality risk values at the time of admission are more likely to die. In another model, we added “physician” as a control variable in our analysis, but it was not significantly associated with mortality. The results show that LOS is significantly and negatively associated with mortality; that is, the probability of mortality decreases the longer a patient stays in the hospital. It should be noted that, this can also be due to a reverse causality effect meaning that people who survive tend to stay longer in the hospital. As previously stated, no serious correlation exists between any pairs of variables in our analysis other than the correlation between the two subjective measures (Table A1). In short, although both analyses show that using MEWS as the only independent variable can somewhat explain LOS and death, these models are weak and only slightly better than random guesses. If more variables are added, it better explains LOS and death. Both of the models point to the importance of physicians’ subjective assessments. It is important to look at the effects of different variables, including vital signs, demographic characteristics, and MEWS on physicians’ subjective assessments of severity level and mortality risks. The results are shown in Tables and, which demonstrate that adding vital signs make a stronger model. In other words, it seems that physicians look at some of the vital signs (especially pulse rate, SBP, and AVPU) rather than the aggregate measure of MEWS in assigning risk measures. # Discussion Briefly, our analysis shows that LOS is more associated with physicians’ early subjective assessments. In this analysis, MEWS only weakly explains LOS and death in models in which we exclude any vital sign and demographics. Physicians’ risk assessments are more influenced by some of the risk indicators than the MEWS system and physicians’ early assessment of risk determines the duration of a hospital stay. Our results contradict findings from some previous studies that stressed the predictive power of MEWS. We believe this is because we controlled for a much greater number of variables in our analysis, and that we have a better control of in-hospital processes such as physicians’ subjective assessments, which trigger the intensity of healthcare services. Our study hypothesizes that humans’ (physicians’) early judgments are more associated with both LOS and outcomes than several of the physiological measures. The negative association between LOS and mortality as well as the positive association between physicians’ assessment of mortality risk and LOS indicate that, like all settings in which human decision making plays an important role, early-risk measures can be more useful in guiding subsequent actions—i.e., how physicians react to initial vital signs and how such reactions may prevent catastrophic events such as mortality. Our hypothesis resonates with the results of a previous study that concludes that physicians’ and nurses’ assessments of risk of mortality is accurate. While understanding the human component and that subjective assessments are important, improving the accuracy of those measures is still a necessity. Our study focused on one specific hospital. The goal here was to ensure internal validity by focusing on a specific setting, gathering as many datapoints as possible, and being aware of all potential processes that may influence the data. Our knowledge of the hospital, the number of variables, and datapoints helped to validate our study of this specific hospital. While we do not see a major issue that may limit generalization of our study, we would like to invite other researchers to be cautious about generalizing our insights. We suggest similar studies in other hospital settings, with different governance structures, different patient demographics, and different types of physician training. In addition, we suggest more studies on effects of risk factors on patients’ perceptions and decisions in visiting emergency departments. Developing effective warning scores is critical for risk analysis and effective healthcare management. Our study shows that current early warnings are not effective in explaining LOS and mortality. Major feedback loop(s) exist that affect how physicians react to vital signs as well as their own subjective measures that can later compensate for large early-risk values. This comes down to the question of the purpose of early warning scores. If the scores are for predicting risk, the goal should be developing measures that provide a quick estimation of risk. However, for outcome prediction, we need more sophisticated models that embed how physicians react to different measures of risk, processes, resources, and technologies in hospitals. To conclude, we invite more studies to develop scores that can raise physician attention and improve early- assessment accuracy as well as measures that can better predict outcomes. # Supporting Information [^1]: The authors have declared that no competing interests exist. [^2]: **Conceptualization:** NA NG JC. **Data curation:** NA. **Formal analysis:** NA NG JC. **Funding acquisition:** NG JC. **Investigation:** NA NG. **Methodology:** NA NG. **Project administration:** NA NG JC. **Resources:** NA NG JC. **Software:** NA NG. **Supervision:** NG JC. **Validation:** NA NG JC. **Visualization:** NA NG. **Writing – original draft:** NA NG. **Writing – review & editing:** NA NG JC.

# Introduction Though routine vaccination is extremely beneficial for children, one of the reasons for non-compliance of children is the adverse effect of the previous vaccination. Various side effects in the form of local (skin indurations, swelling, rash, pain, or erythema at injection site) and systemic reactions (fever, joint or muscle pain, vomiting, diarrhea, fainting, seizures, or other central nervous system effects) occur commonly after diphtheria, tetanus toxoids and pertussis (DTP) vaccination. Again, these reactions are more common after vaccination with whole cell pertusis component vaccine (DTwP) than with acellular pertusis component vaccine (DTaP). When the reactions occur, they usually occur within 24–48 hours following vaccination, are usually mild and self limited, but can result in discomfort in the child. It is a common practice for many health providers to suggest that an antipyretic be given preventively at the time of vaccine administration. If the reactogenicity of these vaccines are decreased in the general population, parental anxiety could be relieved to some extent. But there have been different schools of thought regarding prophylactic antipyretic administration. A systematic review conducted way back in 2007 concluded that parents be counseled to monitor vaccine-related adverse reactions and to treat them if and when they occur. This review summarized the findings pertaining only to DTP vaccination, and not to other childhood vaccinations. Recent clinical trials have found that although febrile reactions were significantly decreased by prophylactic antipyretics, antibody responses to several vaccine antigens were reduced. Meanwhile, the American Academy of Pediatrics (AAP) continues to say that either prophylactic or therapeutic use of antipyretics should not be withheld. Therefore, the current systematic review was planned to bridge this gap of information and provide any recommendation on the use of prophylactic antipyretics post-vaccination in children based on the available evidence. # Methods The protocol was registered with PROSPERO (Registration number: CRD42014009717). ## Types of studies Randomized controlled trials (RCTs) ## Types of participants Children of both sex and ≤6 year age undergoing routine immunization were included. Children suffering from chronic debilitating diseases, severe malnutrition (weight for height \<3SD), and underlying immunodeficiency were excluded because of unpredictability of the antibody response after immunizations. ## Types of interventions The intervention commenced either before, or after the child had received any of the routine childhood vaccinations, and consisted of prophylactic or preventive administration of antipyretics (either paracetamol or ibuprofen or both) or placebo/no prophylactic antipyretics. All formulation, dose and schedule of administration of antipyretics were considered. ## Types of outcome measures ### Primary outcome measures \(1\) Febrile reactions ≥38.0°C (100.4°F) in the first 24–48 hrs of primary and booster vaccinations \(2\) Antibody response rate \[measured by geometric mean concentration (GMC)\] after primary (2, 3, and 4 or 3, 4, and 5 months) and booster vaccinations (12–15 months, and 40–48 months) ### Secondary outcome measures \(1\) High febrile reactions ≥39.0°C in the first 24–48 hrs of primary and booster vaccinations \(2\) Local symptoms (pain, redness, and swelling at the injection site) after primary and booster vaccinations \(3\) Systemic symptoms (temperature, irritability/fussiness, drowsiness, diarrhea, vomiting, and loss of appetite) after primary and booster vaccinations \(4\) Nasopharyngeal carriage (NPC) rate of the organisms (*S. pneumoniae*, *H. influenzae*, and others) The same temperature cutoff was used to define the severity of fever in almost all the trials. All routes of temperature (oral, rectal, and axillary) measurements were considered. Pain was graded as none, mild (light reaction to touch), moderate (protesting in response to touch or pain with limb movement), or severe (child resists limb movement or keeps limb immobile). Seroprotection: defined as an antibody concentration ≥0.1 IU/mL for diphtheria and tetanus, 0.15 µg/mL for *H. influenzae* type b, and 10 mIU/mL for Hepatitis B. Seropositivity: defined as, 5 ELISA U/mL for antibodies to acellular pertussis antigens; anti-pneumococcal serotypes 1, 4, 5, 6B, 7F, 9V, 14, 18C, 19F and 23F antibody concentrations ≥0.2 µg/mL (for PCV10); anti-polio type 1, type 2 and type 3 titres ≥8. Booster vaccine response to PT, FHA and PRN, one month after the administration of the booster dose of DTPa combined vaccine was defined as appearance of antibodies in subjects who are seronegative (that is, with concentrations \<5 ELU/mL) just before booster dose and at least two-fold increase of prevaccination antibody concentrations in those who are seropositive (that is, with concentrations ≥5 ELU/mL) just before booster dose. For comparison purpose, an acceptable decreased immunogenicity of all the mentioned vaccines is that the final antibody concentrations should not be below the above mentioned seroprotective/seropositive titers after primary or booster vaccination series. ## Search methods for identification of studies We searched the Cochrane Central Register of Controlled Trials (CENTRAL, The Cochrane Library, Issue 3, March 2014), which contains the Cochrane Acute Respiratory Infection (ARI) Group and the Cochrane Infectious Diseases Group Specialized Registers, Medline/Ovid (1970 – March 2014), Pubmed (1970 – March 2014), and Embase (1988 – December 2013). For these database searching, a combination of following search terms were adopted: acetaminophen, paracetamol, ibuprofen, analgesics, antipyretics, adverse reactions, vaccination, immunization, DTwP, diphtheria tetanus–toxoid, whole-cell pertussis, DTaP, acellular pertussis, *Streptococcus pneumoniae*, *Haemophilus influenzae* type B, inactivated poliovirus, IPV, pneumococcal 7-valent conjugate, pneumococcal 10-valent conjugate, pneumococcal 13-valent conjugate, PCV, measles, mumps, rubella, MMR, meningococcal conjugate, varicella zoster, hepatitis A, hepatitis B, rotavirus, influenza, or pneumococcal polysaccharide. To identify RCTs, which results had remained unpublished; we searched the NIH clinical trial register ([www.clinicaltrials.gov](http://www.clinicaltrials.gov)). Trials that focused on the therapeutic effects of antipyretics post-vaccination were excluded from the analysis. Articles obtained from this search were cross-referenced and bibliographies were checked for all relevant information. No language restrictions were applied. The search details are given in. ## Data extraction Data extraction was done using a data extraction form that was designed and pilot tested *a priori*. Two authors independently extracted data from the included studies, including year, setting (country, setting, type of participants, vaccination schedule followed, type of vaccines administered), exposure/intervention (type of antipyretic, dose and schedule of administration, protocol deviation, type of placebo), results (outcome measures, effect, significance), and sources of funding/support. Disagreements in extracted data were resolved through discussion. ## Assessment of risk of bias in included studies Two review authors independently assessed the methodological quality of the selected trials by using methodological quality assessment forms. We undertook quality assessment of the trials using the criteria outlined in the *Cochrane Handbook for Systematic Reviews of Interventions*. Any disagreements between the two review authors were resolved by discussion with the third author. Trials were assessed with respect to the extent to which investigators minimised the potential for bias to occur and addressed other issues in relation to methodological quality. Publication bias that might affect the cumulative evidence was also assessed. ## Study descriptions Information in relation to methodological quality, characteristics of participants, interventions and outcome measures of each trial is provided in,. ## Data synthesis including assessment of heterogeneity The data from various studies were pooled and expressed as, odds ratio (OR) with 95% confidence interval (CI) for dichotomous data, and mean difference (MD) with 95% CI for continuous data. A p-value of \<0.05 was considered significant. Assessment of heterogeneity was done by I-squared (I²) statistics. If there was a high level of heterogeneity (\>50%), we tried to explore this by subgroup analysis if there were adequate number of trials. A fixed effects model was initially conducted. If significant heterogeneity existed among trials (\>50%), potential sources of heterogeneity were considered, and where appropriate, a random effects model was used. *RevMan (Review Manager) version 5.2* was used for all the analyses. # Results ## Description of studies Of 2579 citations retrieved, full text of 26 articles were assessed for eligibility. Out of these, a total of 14 articles were excluded for the following reasons: non RCTs (n = 11), adult participants (n = 03). Finally, 13 trials including 5077 children were included in the review ,. The included trials were conducted in both developed (USA = 4, Czech Republic  = 3, Canada  = 1, Germany  = 1, Turkey  = 1, Finland  = 1, and Spain  = 1) and developing countries (India  = 1). One trial used ibuprofen, two used both paracetamol and ibuprofen, and others used only paracetamol,,. The trials were heterogeneous regarding the dosage schedule of intervention, the age of the enrolled children, type of vaccine used, and the outcomes measured. Children \>1 months (not neonates) were included in the studies. Isolated DTwP vaccine was used in six trials, isolated DTaP in one trial, and rest others used combination vaccine. ## Risk of bias in included studies All the included trials had moderate to high risk of bias because of the following reasons: open or single-blind nature, small sample size, and other sources of bias. ## Effect of prophylactic Paracetamol (PCM) ### Primary outcome measures \(1\) Febrile reactions ≥38.0°C (100.4°F) in the first 24–48 hrs: Compared to the no prophylactic PCM group, there was a significant reduction in the febrile reactions of ≥38.0°C (100.4°F) in the first 24–48 hrs in the prophylactic PCM group, both after primary \[OR, 0.35; 95%CI, 0.26–0.48\] and booster \[OR, 0.60; 95%CI, 0.39–0.93\] vaccinations. However, because of a high degree of heterogeneity (\>50%), these results should be interpreted with caution. \(2\) Antibody response rate (measured by GMCs) after primary vaccination (3, 4, and 5 months age): There was significant difference in the GMC of the anti- pneumococcal IgG antibody between the prophylactic PCM group and no prophylactic PCM group, for all the vaccine serotypes: serotype 1 \[MD −0.53 (95%CI, −0.71 to −0.35)\], serotype 4 \[MD −0.8 (95%CI, −1.08 to −0.52)\], serotype 5 \[MD −0.62 (95%CI, −0.87 to −0.37)\], serotype 6B \[MD −0.2 (95%CI, −0.29 to −0.11)\], serotype 7F \[MD −0.59 (95%CI, −0.83 to −0.35)\], serotype 9V \[MD −0.46 (95%CI, −0.65 to −0.27)\], serotype 14 \[MD −1.28 (95%CI, −1.79 to −0.77)\], serotype 18C \[MD −1.47 (95%CI, −1.82 to −1.12)\], serotype 19F \[MD −2.13 (95%CI, −2.93 to −1.33)\], and serotype 23F \[MD −0.27 (95%CI, −0.43 to −0.11)\]. Regarding other vaccinations, there was significant difference in the GMC of the anti-PRP \[MD −1.99 (95%CI, −2.76 to −1.22)\], anti-diphtheria \[MD −0.89 (95%CI, −1.27 to −0.51)\], anti-tetanus \[MD −1.04 (95%CI, −1.34 to −0.74)\], anti-pertactin \[MD −27.9 (95%CI, −38.65 to −17.15)\] between the prophylactic PCM group and no prophylactic PCM group. The GMC of anti-PT, anti-FHA, anti-HBs, and anti-polio (type 1,2,3) did not show any significant difference between the prophylactic PCM group and no prophylactic PCM group. Though the GMC of all pneumococcal vaccines serotypes and some other vaccines decreased after prophylactic PCM, still the level of GMC in the prophylactic PCM group was well above the seroprotection level. \(3\) Antibody response rate (measured by GMCs) after first booster vaccination (12–15 months age): There was significant difference in the GMC of the anti- pneumococcal IgG antibody between the prophylactic PCM group and no prophylactic PCM group, for all the vaccine serotypes: serotype 1 \[MD −0.96 (95%CI, −1.37 to −0.55)\], serotype 4 \[MD −1.22 (95%CI, −1.84 to −0.6)\], serotype 5 \[MD −1.38 (95%CI, −1.91 to −0.85)\], serotype 6B \[MD −1.11 (95%CI, −1.5 to −0.72)\], serotype 7F \[MD −1.24 (95%CI, −1.81 to −0.67)\], serotype 9V \[MD −1.55 (95%CI, −2.17 to −0.93)\], serotype 14 \[MD −1.38 (95%CI, −2.28 to −0.48)\], serotype 18C \[MD −2.03 (95%CI, −2.99 to −1.07)\], serotype 19F \[MD −1.55 (95%CI, −3.08 to −0.02)\], and serotype 23F \[MD −0.93 (95%CI, −1.5 to −0.36)\]. Regarding other vaccinations, there was significant difference in the GMC of the anti- diphtheria \[MD −2.2 (95%CI, −3.82 to −0.58)\], anti-tetanus \[MD −2.2 (95%CI, −3.25 to −1.15)\] between the prophylactic PCM group and no prophylactic PCM group. The GMC of anti-PRP, anti-PT, anti-FHA, anti-pertactin, anti-HBs, and anti-polio (type 1,2,3) did not show any significant difference between the prophylactic PCM group and no prophylactic PCM group. Though the GMC of all pneumococcal vaccines serotypes and some other vaccines decreased after prophylactic PCM, still the level of GMC in the prophylactic PCM group was well above the seroprotection level. \(4\) Antibody response rate (measured by GMCs) after second booster vaccination (40–48 months age): There was significant difference in the GMC of the anti- pneumococcal IgG antibody between the prophylactic PCM and no prophylactic PCM group, for the following vaccine serotypes: serotype 1 \[MD −4.27 (95%CI, −6.75 to −1.79)\], serotype 4 \[MD −4.78 (95%CI, −8.16 to −1.4)\], serotype 5 \[MD −3.69 (95%CI, −6.67 to −0.71)\], serotype 7F \[MD −2.92 (95%CI, −4.74 to −1.1)\], serotype 9V \[MD −4.59 (95%CI, −7.4 to −1.78)\], serotype 14 \[MD −6.7 (95%CI, −13.35 to −0.05)\], serotype 18C \[MD −12.54 (95%CI, −22.1 to −2.98)\]. The GMC of the anti-pneumococcal IgG antibody for serotypes 6B, 19F, and 23F did not show statistically significant difference between the prophylactic PCM and no prophylactic PCM group. No study reported this outcome for other vaccinations. ### Secondary outcome measures \(1\) High febrile reactions ≥39.0°C in the first 24–48 hrs: compared to the placebo group, there was a significant reduction in the high febrile reactions of ≥39.0°C in the first 24–48 hrs in the prophylactic PCM group after primary \[OR, 0.31; 95%CI, 0.18–0.52\], but not booster \[OR, 0.63; 95%CI, 0.35–1.11\] vaccinations. \(2\) Pain of all grades: compared to the no prophylactic PCM group, there was a significant reduction in the pain of all grades in the prophylactic PCM group, both after primary \[OR, 0.57; 95%CI, 0.47–0.7\] and booster \[OR, 0.64; 95%CI, 0.48–0.84\] vaccinations. \(3\) Pain of moderate to severe grade: compared to the no prophylactic PCM group, there was a significant reduction in the pain of moderate to severe grade in the prophylactic PCM group after primary \[OR, 0.39; 95%CI, 0.26–0.58\], but not booster \[OR, 0.59; 95%CI, 0.24–1.45\] vaccinations. \(4\) Local redness: compared to the no prophylactic PCM group, there was a significant reduction in the local redness in the prophylactic PCM group after primary \[OR, 0.81; 95%CI, 0.68–0.95\], but not booster \[OR, 0.93; 95%CI, 0.73–1.18\] vaccinations. \(5\) Local swelling/induration: compared to the no prophylactic PCM group, there was a significant reduction in the local swelling/induration in the prophylactic PCM group after primary \[OR, 0.78; 95%CI, 0.66–0.92\], but not booster \[OR, 0.90; 95%CI, 0.68–1.19\] vaccinations. \(6\) Persistent cry: compared to the no prophylactic PCM group, there was a significant reduction in the rate of persistent cry in the prophylactic PCM group, both after primary \[OR, 0.55; 95%CI, 0.39–0.77\] and booster \[OR, 0.44; 95%CI, 0.22–0.87\] vaccinations. \(7\) Irritability/fussiness: compared to the no prophylactic PCM group, there was a significant reduction in the irritability/fussiness in the prophylactic PCM group, both after primary \[OR, 0.36; 95%CI, 0.29–0.45\] and booster \[OR, 0.66; 95%CI, 0.48–0.91\] vaccinations. \(8\) Drowsiness: compared to the no prophylactic PCM group, there was a significant reduction in the drowsiness in the prophylactic PCM group after primary \[OR, 0.82; 95%CI, 0.70–0.96\], but not booster \[OR, 0.99; 95%CI, 0.76–1.3\] vaccinations. \(9\) Anorexia/loss of appetite: compared to the no prophylactic PCM group, there was a significant reduction in the anorexia/loss of appetite in the prophylactic PCM group after primary \[OR, 0.61; 95%CI, 0.49–0.77\], but not booster \[OR, 0.85; 95%CI, 0.64–1.14\] vaccinations. \(10\) Vomiting: There was no significant difference between the prophylactic PCM and the no prophylactic PCM group regarding the reduction of vomiting. \(11\) Diarrhea: There was no significant difference between the prophylactic PCM and the no prophylactic PCM group regarding the reduction of diarrhea. \(12\) Any severe symptom: compared to the no prophylactic PCM group, there was a significant reduction in any severe symptom in the prophylactic PCM group after booster \[OR, 0.38; 95%CI, 0.20–0.71\], but not primary \[OR, 0.81; 95%CI, 0.58–1.12\] vaccinations. \(13\) Nasopharyngeal carriage (NPC) rate of the organisms (*S. pneumoniae*, *H. influenzae*, and others) There was no significant difference either in the pneumococcal carriage rate (any serotype, vaccine serotypes, or any cross-reactive serotype) or in the H. influenza carriage rate between the prophylactic PCM and the no prophylactic PCM group. The significant finding of post-booster non-typeable *H. influenzae* carriage rate \[OR, 0.61; 95%CI, 0.39–0.95\] might be due to chance or inadequate randomization. \(14\) Days of parental work loss There was no significant difference between the prophylactic PCM and no prophylactic PCM group for the days of parental work loss. ## Effect of prophylactic Ibuprofen (IB) ### Primary outcome measures \(1\) Febrile reactions ≥38.0°C (100.4°F) in the first 24–48 hrs: there was no significant difference between the prophylactic IB and no prophylactic IB groups regarding the reduction of febrile reactions ≥38.0°C (100.4°F) in the first 24–48 hrs of primary and booster vaccinations. ### Secondary outcome measures \(1\) High febrile reactions ≥39.0°C in the first 24–48 hrs: there was no significant difference between the prophylactic IB and no prophylactic IB group regarding the reduction of febrile reactions ≥39.0°C in the first 24–48 hrs of primary vaccination. \(2\) Pain all grades: compared to the prophylactic IB group, there was a significant increase in the pain of all grades in the no prophylactic IB group after primary \[OR, 1.52; 95%CI, 1.13–2.04\], but not booster \[OR, 0.97; 95%CI, 0.55–1.7\] vaccinations. \(3\) Pain (moderate to severe): compared to the prophylactic IB group, there was a significant increase in the moderate to severe pain in the no prophylactic IB group after primary \[OR, 1.73; 95%CI, 1.1–2.72\], but not booster \[OR, 0.95; 95%CI, 0.41–2.24\] vaccinations. \(4\) Local redness: There was no significant difference between the prophylactic IB and no prophylactic IB group regarding the reduction of local redness after primary and booster vaccinations. \(5\) Swelling/induration: compared to the prophylactic IB group, there was a significant increase in the swelling/induration in the no prophylactic IB group after primary \[OR, 1.44; 95%CI, 1.06–1.94\] vaccination. \(6\) Prolonged cry: There was no significant difference between the prophylactic IB and no prophylactic IB group regarding prolonged cry after primary vaccination. \(7\) Irritability/fussiness: There was no significant difference between the prophylactic IB and no prophylactic IB group regarding irritability/fussiness after primary vaccination. \(8\) Drowsiness: compared to the prophylactic IB group, there was a significant increase in drowsiness in the no prophylactic IB group after primary \[OR, 1.36; 95%CI, 1.00–1.86\] vaccination. \(9\) Anorexia/loss of appetite: There was no significant difference between the prophylactic IB and no prophylactic IB group regarding anorexia/loss of appetite after primary vaccination. \(10\) Vomiting: There was no significant difference between the prophylactic IB and no prophylactic IB group regarding vomiting after primary vaccination. \(11\) Diarrhea: There was no significant difference between the prophylactic IB and no prophylactic IB group regarding diarrhea after primary vaccination. ## Effect of prophylactic PCM and prophylactic IB ### Primary outcome measure \(1\) Antibody response rate (measured by GMCs) after primary vaccination (2, 3, 4, and 12 month age): This was reported in one trial \[presented as conference abstract\]. The trial employed 5 groups, and the results were as follows. Pneumococcal anticapsular IgG GMCs were significantly lower (p\<0.0125) in G3 (received paracetamol at vaccination and thereafter) versus G5 (no antipyretic) for 5 of 13 serotypes after the primary series. Pertussis FHA and tetanus IgG GMC was significantly lower among G4 (received ibuprofen at vaccination and thereafter) versus G5 (no antipyretic) after the primary series. No differences were observed for any antigens after the toddler dose. The trial concluded that prophylactic PCM may interfere with primary series immune response to pneumococcal antigens. Prophylactic IB did not interfere with pneumococcal responses, but may reduce response to pertussis FHA and tetanus antigens. These effects were not observed following the toddler dose. The clinical significance of these findings is unclear. ## Publication bias To assess whether there was a bias in the published literature, funnel plot was constructed using the OR and 1/SE values obtained from studies measuring the primary outcome (febrile reactions of ≥38.0°C in the first 24–48 hrs of PCM administration). In the absence of a publication bias, such a plot is expected to have a shape resembling an inverted funnel. From the asymmetry of funnel plot generated, the possibility of publication bias in the analysis cannot be ruled out. # Discussion ## Summary of evidence Prophylactic antipyretic administration significantly reduced the febrile reactions of ≥38.0°C after primary and booster vaccinations. Though there were statistically significant differences in the antibody responses between the two groups (being lower in the prophylactic PCM group), the prophylactic PCM group had what would be considered protective levels of antibodies (GMCs) to all of the antigens given after the primary and booster vaccinations. There was a significant reduction in the local and systemic symptoms after primary, but not booster vaccinations (except for any severe symptom, that had a significant reduction after booster but not primary vaccinations). The present review does not find a strong evidence to support the conclusion of a well conducted RCT that questioned the administration of prophylactic PCM during administration of childhood vaccines. This RCT had concluded that although febrile reactions significantly decreased, prophylactic administration of PCM at the time of vaccination should not be routinely recommended since antibody responses to several vaccine antigens were reduced. However, since the antibody response (GMC) was not reduced below seroprotection level, it is unlikely that prophylactic PCM would have any detrimental effect for individual child concerned. The same has been endorsed by AAP in their guidelines. Regarding the new trial studying the effect of PCM and IB simultaneously, the results are more complicated, as it found differential effect of the antipyretics on the vaccine antigen responses. The present review finds a benefit in favour of prophylactic antipyretic administration on both local and systemic symptoms post-vaccination, although the analyses included trials using mostly DTwP (6 trials) instead of DTaP (3 trials), the later being less reactive. The results of the RCT that has sparked the debate about the beneficial role of prophylactic antipyretic though cannot be ignored, but cannot be accepted with foolproof at the same time. This is because of the following four points. First, there is only a small decrease in the GMC of vaccine antibody titers that may be of statistically significant but the clinical/epidemiological relevance is not clear. The latter is supported by the fact that, in spite of being a common practice for administration of prophylactic antipyretics after immunizations for decades, there have been significant reductions in invasive disease due to *S. pneumoniae* and *H. influenzae* type b serotypes. Second, the follow up study to the above RCT has shown that regardless of the administration of prophylactic PCM, there was no effect on the nasopharyngeal carriage rate post-booster vaccination. Third, the development of fever or increase in the temperature post-vaccination due to the release of endogenous cytokines (IL 1, TNF α), has been considered as a marker of immune response to respected vaccines. Fourth, the potential interference between different vaccines when co-administered with or without antipyretics should also be taken into consideration. For example, 30–60% lower anti-HBs GMTs occur when co-administered with HPV vaccines and that without antipyretics, which might further diminish the magnitude of the immune response. It has also been seen that the acellular pertussis vaccine is much less immunogenic than the whole cell, and PCV13 develops lower IgG concentrations than PCV7 to the common serotypes. If this is already the case, adding prophylactic PCM that could lower the immune response even lower, could be a problem. If there is already herd immunity, maybe a small decrease in efficacy at the individual level will take a long time to be noticed, and would raise the need for better surveillance programs for vaccine-preventable diseases in all countries. Because of these later two findings, there is concern that prophylactic antipyretic might decrease the post-vaccination immune response further. Besides these, the findings of another RCT reporting about the infant sleep after immunization and relation of acetaminophen (paracetamol) use need mention here. This RCT found that paracetamol use post-immunization (not prophylactic) was associated with increase in the infant sleep duration. As sleep deprivation before or after has been associated with decreased antibody formation post- immunization in adults, this study postulates that use of acetaminophen post- immunization might facilitate the immune response. But this study neither studied the effect of prophylactic antipyretic nor measured the antibody response. ## Limitations Only two trials (from the same country) studied the antibody response (one trial) and carriage rate (one trial) as a result the data could not be pooled. Studies used different doses/schedules of antipyretic administration resulting in significant heterogeneity in the pooled result. The age of the participants or timing of administration also markedly differed among the studies. Only one study from developing country (India) made it difficulty in generalizing the present review findings. ## Further area of research Future trials should focus on the timing (before, with or after) and route (oral or rectal) of administration of paracetamol as well as on the subgroup of infants (term or preterm) for any correlation with the immune response. As there was no trial examining the prophylactic effect of ibuprofen on post-vaccination antibody response, future trials should focus on this. Any post-vaccination decrease in antibody titer noted in future studies should be correlated with the natural history of that particular disease. The mechanism underlying the decrease in immune/antibody response should also be explored. Immune response to varicella, hepatitis A, measles, MMR, and flu vaccine should also be studied, if feasible. Trials should also be conducted in developing countries where over- the-counter use of antipyretics (including prophylactic) are common. Other confounding factors that might affect the antibody response (e.g., infant sleep post-immunization) should also be studied. # Conclusions Though prophylactic antipyretic administration leads to relief of the local and systemic symptoms after primary vaccinations, there is a reduction in antibody responses to some vaccine antigens without any effect on the nasopharyngeal carriage rates of *S. pneumoniae* & *H. influenza* serotypes. Future trials and surveillance programs should also aim at assessing the effectiveness of programs where prophylactic administration of PCM is given. The timing of administration of antipyretics should be discussed with the parents after explaining the benefits & risks. # Supporting Information [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: RRD IP SSN. Performed the experiments: RRD IP SSN. Analyzed the data: RRD SSN. Contributed reagents/materials/analysis tools: RRD IP. Contributed to the writing of the manuscript: RRD IP SSN.

# Introduction The first generation non-nucleoside reverse transcriptase inhibitors (NNRTIs) nevirapine (NVP) or efavirenz (EFV) are frequently used in combination with two nucleoside reverse transcriptase inhibitors (NRTIs) as first-line antiretroviral therapy (ART) in both high- and low-middle income countries. The second generation NNRTI etravirine (ETR) is approved for treatment-experienced HIV-1 infected adults with resistance to NVP and/or EFV, although access to ETR in low-middle income countries (LMIC) is limited. ETR can be used in such patients due to its limited cross-resistance pattern relative to first generation of NNRTI. However, the pattern of resistance-associated mutations (RAMs) has to be considered and several constellations of mutations cause indeed cross- resistance. A large number of ETR RAMs have been revealed, including the major mutations L100I, K101E/P, E138A/G/K/Q, Y181C/I/V, Y188L, G190A/S/E, and M230L. The genotypic resistance tests (GRT) used in clinical care detect RAMs that are present in \>20% of the viral population. More sensitive techniques, such as allele-specific PCR (ASPCR) – and ultra-deep sequencing, are not yet used in the clinical routine. It has however been reported that minor resistant quasispecies may influence the outcome of ART. One alternative to GRT is phenotypic tests which have several disadvantages in the clinical setting. However, new methods for reverse transcriptase (RT) extraction and sensitive RT assays have made analyses of drug susceptibility profiles of retroviral RTs recovered directly from plasma virions feasible. Drug susceptibility testing on RT offers advantages compared to traditional phenotypic tests, such as short turnover time, lower cost, a need of only serology laboratory equipment, and they might therefore be useful in LMIC. This method has successfully been used to describe drug resistance to NVP and also recently to ETR. In contrast to genotypic assays, no complex interpretations are required. The aim of the study was to thoroughly evaluate clinical samples in our newly adapted RT-based assay for assessment of resistance to ETR and to compare the obtained phenotype with the genotype obtained with standard direct sequencing. Also, we performed ultra-deep pyrosequencing (UDPS) to identify any minor sequence variants in the HIV-1 RT gene of cases where there was a discrepancy between the major genotype and the RT phenotype. # Methods ## Patients Altogether, 25 plasma EDTA samples of 20 HIV-1 infected patients were retrospectively included from the HIV cohort at Department of Infectious Diseases, Karolinska University Hospital, Stockholm, Sweden. Of these, 15 treatment-experienced patients were randomly selected among subjects with ART failure. All of these patients had been treated or were treated with nevirapine or efavirenz. In patients 3, 5, 8, 9 and 11 , who did not respond to treatment two consecutive samples (a and b) were drawn for follow up of the virological response. For eight samples (no 3b, 5a, 5b, 8a, 8b, 10, 11a, 18), the failing regimen contained an NNRTI. For 12 samples (no 1, 3a, 6, 7, 9a, 9b, 11b, 12, 14, 15, 17, 20), the NNRTI had been stopped earlier and the failing regimen contained now antiretroviral drugs from other categories. Four treatment naïve patients (no 2, 13, 16, 19) were included since they had been infected with NNRTI-resistant strains. GRT had been performed within the clinical care by routine direct sequencing on samples from all of these individuals and one or more NNRTI mutations had been found in all of them, except in patient 7. One NNRTI-naïve patient (no 4), who had a K103R mutation, was chosen as a negative control since this mutation is reported not to affect ETR susceptibility. All the work presented in the manuscript was conducted in accordance with the Declaration of Helsinki. Ethical permission was obtained from the Regional Ethical Committee at Karolinska Institutet, Dnr. 2005/1167-31/3, Dnr. 2005/772-31/4 and written informed consent was obtained from each individual. ## Chemicals ETR (4-\[\[6-amino-5-bromo-2\[(4-cyanophenyl)amino\]-4-pyrimidinyl\]oxy\]-3.5- dimethylbenzonitrile) was a generous gift from Tibotec, Mechelen, Belgium. The drug was dissolved in dimethylsulfoxide (Sigma, St. Louis, MO). ## Phenotypic RT-based analysis of resistance to ETR Lysates containing intraviral RT were recovered from plasma using ExaVir Load, version 3 (Cavidi, Uppsala, Sweden). Briefly, virions from up to 1 ml plasma were bound to an ion exchange gel, washed, and the RT was extracted by lysing the purified virions. The RT was used to catalyse the incorporation of bromodeoxyuridine triphosphate (BrdUTP) into DNA via a poly-rA (ribodenylic acid template) plate. The amount of BrdU monophosphate (BrdUMP) incorporated into DNA was detected with an alkaline phosphatase (Ap) conjugated anti-BrdU monoclonal antibody. The Ap substrate para-nitrophenyl-diphosphate was used for colorimetric detection. The susceptibility of the RT activity to ETR was measured using ExaVir Drug susceptibility assay, as previously described. Briefly, this assay measures the ability of RT to catalyse the above reaction in the presence of a 2.5 fold step dilution of ETR. The RT activity at each drug concentration was recalculated into percentage of RT activity in the absence of drug. The IC₅₀ values were calculated using the median effect equation. The analysis was performed blindly in relation to the genotype results and to type of treatment. We have earlier described low IC₅₀ values (mean ±SD: 2.5±1.0 µM) for RT activity in the presence of ETR in treatment-naïve patients and for reference recombinant RTs. Also, the reproducibility analysis of the assay showed an inter-assay variation (CVs) of 9.4 and 11.1% and that the IC₅₀ values for ETR (mean ± SD: 1.6±0.03 mM and 3.1±0.04) were not influenced by the RT amount within the 40–828 fg/ml RT range. In the present study, two reference recombinant RTs, BH10-wild-type and its mutant form L100I, were analysed together with patient samples in each assay. The results of the phenotypic tests were evaluated independently by two persons (E.A, C.FR.K). ## Scoring of ETR susceptibility Three genotypic scoring systems were used to predict ETR susceptibility (Stanford University, Monogram Weighted Score, Tibotec Weighted Genotype Score). The Monogram and Tibotec systems use a score calculated from a list of mutations, each with their own weight factor. The Stanford scores were obtained using <http://hivdbstanford.edu/cgi-bin/Marvel.cgi>. The correlation between Monogram/Tibotec/Stanford scorings and IC₅₀ determined by our phenotypic RT assay was assessed by Spearman's rank test using software in GraphPad Prism (San Diego, California, USA). ## Direct HIV-1 sequencing Sequencing of the HIV *pol* gene had earlier been performed within the clinical routine at the Department of Clinical Microbiology, Karolinska University Hospital, Stockholm. Samples were analysed using ViroSeq HIV-1 Genotyping System v2.0 (Abbott Diagnostics, Illinois, USA) according to manufacturer's protocol on an ABI PRISM 310 Genetic analyzer (Applied Biosystems). ## Ultra deep HIV-1 sequencing HIV-1 RNA was extracted from 1 mL of plasma from nine samples (QIAamp Viral RNA Mini Kit, QIAGEN, Valencia, CA) after centrifugation at 9,000×*g* for 1 h. The 3′ end of the reverse transcriptase region was reverse transcribed and amplified using SuperScript III and Platinum Taq High Fidelity (Life Technologies, Paisley, UK). Primer used were Forward primer (HXB2: 2508–2528) 5′-GGA AGA AAT CTG TTG ACT CAG-3′ and Reverse primer (HXB2: 3441–3459) 5′- GAA GCA GAG CTA GAA CTG G-3′. Cycling conditions were: 30 min at 52°C for reverse transcription;2 min at 94°C for initial denaturation; 20 cycles of 2 min at 94°C, 30 sec at 50°C and 1 min 30 sec at 68°C; and a final polymerisation step of 5 min at 68°C. Amplicon libraries were generated from first-round PCR products. These amplicons incorporated adapters A and B, and also identifiers used in parallel sample sequencing needed for bidirectional 454 sequencing. Nested PCR conditions were a first denaturation step of 2 min at 94°C; 20 cycles of 2 min at 94°C, 30 sec at 50°C and 45 sec at 68°C; and a final polymerisation step of 3 min at 68°C using Platinum Taq High Fidelity (Life Technologies, Paisley, UK). A set of 5 overlapped amplicons were designed to cover the protein region previously amplified. Amplicon 1: Forward (2541–2566) 5′-TTA AAT TTT CCC ATT AGT CCT ATT GA-3′ and Reverse (2873–2890) 5′- ACT GGA TGT GGG TGA TGC-3′; Amplicon 2: Forward (2642–2667) 5′-AAA AGC ATT AGT AGA AAT TTG TAC AG-3′ and Reverse (2929-2951) 5′-ATA CTG CAT TTA CCA TAC CTA GT-3′; Amplicon 3: Forward (2780–2802) 5′-CAG AGA ACT TAA TAA GAG AAC TC-3′ and Reverse (3155–3174) 5′-AGA GGA ACT GAG ACA ACA TC-3′; Amplicon 4: Forward (2819–2839) 5′-TCA ATT AGG AAT ACC ACA TCC-3′, Reverse (3243–3265) 5′-TAT GAA CTC CAT CCT GAT AAA TG-3′; Amplicon 5: Forward (3018–3038) 5′- CCA GCA ATA TTC CAA AGT AGC-3′ and Reverse (3402–3422) 5′- GGA ACC AAA GCA CTA ACA GAA-3′. Nested PCR products were purified using AMPure Magnetic Beads (Beckman Coulter, Inc, Brea, CA) to eliminate primer-dimers. Concentration and quality of purified PCR products was determined using fluorometry (PicoGreen, Life Technologies, Paisley, UK) and spectrophotometry (Lab on a Chip, Agilent Technologies, Foster City, CA). Equimolar amplicon pools were made to perform emPCR, adding a ratio of 1∶1 between molecules and 454-beads. Sequencing platform used was Genome System Junior (Life Sequencing/Roche). 454 analyses were performed with the 454/Roche proprietary AVA software v2.7 after checking for cross-sample and pNL4-3 contamination. The threshold level for detection of minority variants was set at 0.5%. Access to the viral sequences of the patients can be obtained after request to the author. The dataset have been deposited in the NIH/SRA repository with the accession number PRJNA246769. # Results ## Phenotypic assay results in relation to mutational patterns The two reference recombinant RTs; BH10-wild-type and its mutant form L100I, showed the expected IC₅₀ values of 1.0±0.18 µM and 12.0±2.3 µM, respectively. RT was isolated from 20 plasma samples from 15 patients with treatment failure. Based on the IC₅₀ value of BH10-wild-type, which was included in every run, the IC₅₀ value and the fold-change in susceptibility of each isolate were determined. A fold-change of \<5 was considered as sensitive, between 5–15 as decreased susceptibility, and above 15 as resistant. In ten samples (3a, 4, 7, 9a, 11b, 12, 15, 16, 17, 18) with the lowest IC₅₀ values and a fold change of \<5, which were considered sensitive by our phenotypic assay, (mean ±SD: 3.1±1.3; range: 0.7–4.5 µM), there was a good concordance with the GRT. Thus, sequence analysis showed no mutations or non-ETR RAMs (3a: A98S + V179D; 4: K103R; 9a: A98S + K103N + V108I + K238T; 11b: V179I; 12, 15, 16 and 17: K103N; 18: V90I + K103N). Six samples had IC₅₀ values corresponding to a 5–15 fold decrease in susceptibility for ETR (range: 6.4–13.6 µM; 1: 11.8 µM, 9b: 9.2 µM, 10: 13.6 µM, 11a: 7.3 µM, 19: 6.4 µM, 20: 7.2 µM). In all of them (1: A98S + V106A; 9b: A98S + K103N + V108I; 10: K103N + V108I; 11a: K103N + V179I; 19, 20: K103N) only non- ETR RAMs were found. Four samples from three patients had resistant HIV strains which showed highly increased IC₅₀ values and a \>15 fold decrease in susceptibility (5a: 45.8 µM; 5b: 71.4 µM; 13: 68.8 µM 14: 29.1 µM), while direct sequencing showed mutations which are not known to be predictive for decreased sensitivity for ETR (5a and 5b: K101I/R + V106G; 13: K103N + G190A; 14: A98S + K103N + E138A). A further five samples (2, 3b, 6, 8a, 8b) had a \>15 fold decrease in susceptibility and strongly increased IC₅₀ (\>100 uM). The mutational patterns were, to varying degrees, predictive of a decreased sensitivity to ETR, including Y181C (3b, 8a, 8b), V901 + L100I + K103N (6), and A98S + E138E/Q + K238T (2). ## Ultra-deep pyrosequencing (UDPS) UDPS was performed on nine samples (1, 2, 5b, 10, 11a, 13, 14, 19, 20) for which there seemed to be a discrepancy between the major genotype and the RT phenotype. The median nucleotide coverage per sample ranged from 1499 to 2506 sequences. Relative to UDPS, direct sequencing detected all mutations corresponding to \>20% but no mutations \<20% of the viral population. Altogether eleven mutations were detected by UDPS, but not by direct sequencing, ranging from 0.54% to 19.56%. In four samples (1, 11a, 13, 14), UDPS detected additional minor variants associated with decreased ETR susceptibility (1: Y181C; 11a: G190A; 13: L100I; 14: K101E + T238K). In three samples (2, 5b, 20), identical mutations were found with the two sequencing techniques. Thus, the UDPS did not reveal any further minor variants that could explain the increased IC₅₀ values of 100 µM, 71.4 µM and 7.2 µM, respectively. In two samples (10 and 19), UDPS showed additional minor mutations not known to be associated with decreased ETR susceptibility (10: V179I; 19; V106I). ## Correlation between RT-based phenotypic sensitivity and genotype scoring based on direct sequencing There was a significant correlation between the Stanford scoring system and the phenotypic assay results (r = 0.70 p\<0.0001) assessed by Spearman's rank test. The Monogram and Tibotec scoring systems also correlated with the IC₅₀ values (r = 0.65 p\<0.0005; r = 0.64 p\<0.0005, respectively). ## Resistance results in patients who had stopped NNRTI Twelve samples (1, 3a, 6, 7, 9a, 9b, 11b, 12, 14, 15, 17, 20) of eleven patients were drawn after that an NNRTI containing regimen had been terminated. In patients 1 and 20, NVP and EFV had been stopped six and eight weeks earlier, respectively. The IC₅₀ values were slightly increased (11.8 µM and 7.2 µM, respectively) but the Stanford scoring predicted full sensitivity (A98S + V106A; K103N, respectively). However, in patient 1 UDPS revealed additional minor mutations (K103T+ V108I + V179I + Y181C) which possibly could explain this difference. In patients 6 and 14, a clear decreased phenotypic sensitivity against ETR (6: \>100 µM; 14: 28.8 µM) was found despite that \>1 year and \>3 years, respectively, had passed since the cessation of an NNRTI. Genotyping further supported this finding for patient 6 (direct sequencing: V90I + L100I + K103N) and patient 14 (UDPS: A98S + K101E + K103N + E138A + K238T). In four patients, several years had passed since their NNRTI treatment was terminated (7, 11b, 15: \>5.5 years; 9a: 3.5 years). In patient 12, 18 weeks had passed. No phenotypic resistance was found in these patients although non-ETR RAMs persisted (11b: V179I; 12 and 15: K103N; 9a: A98S + K103N + V108I + K238T). However, a second sample (9b) of patient 9 drawn four months later showed a slightly decreased phenotypic sensitivity (9.2 µM) with the same mutational pattern with a concomitant increase in plasma viral load. UDPS could not be performed due to lack of plasma. ## Resistance results in patients with ongoing NNRTI failure Eight samples (3b, 5a, 5b, 8a, 8b, 10, 11a, 18) of six patients were drawn during failure of an ongoing NNRTI containing regimen. In four samples (3b, 8a, 8b, 18), a concordance was seen between the pheno- and genotypes. Thus, patient 3b exhibited a high ETR resistance with both methods (IC₅₀ \>100 µM; genotype: V90I + A98S + V179D + Y181C). In both samples of patient 8, phenotypic resistance (\>100 µM) and the mutations K103N + Y181C were found. Patient 18 had a sensitive phenotype (4.1 µM) and was devoid of mutations other than K103N + V90I after 10 weeks failure. In four samples, discordance was seen between the RT-based phenotype and the genotype. In patient 5, two samples were drawn with three months interval during failure with EFV-containing regimen. An increasing IC₅₀ of 45.8 µM and 71.4 µM, respectively, was seen despite that direct sequencing as well as UDPS showed only mutations (K101I/R + V106G) which are not known to be ETR- associated. Patient 10 exhibited an increased IC₅₀ (13.6 µM), but the identified mutations (direct sequencing: K103N + V108I; UDPS: K103N + V108I + V179I) predicted an ETR sensitive virus. In patient 11 a sample drawn early after failure showed a slightly increased IC₅₀ (7.3 µM), while direct sequencing showed K103N + V179I. However UDPS identified additionally G190A giving support for that resistance was developing. ## Effects of ETR on patients with transmitted NNRTI resistance Four patients (no 2, 13, 16, 19), who had been infected with an NNRTI resistant strain, were also analysed. Patients 2 and 13 had a strongly increased IC₅₀ values (\>100 µM; 68.8 µM) despite that the sequence analysis predicted only a slightly decreased sensitivity (2: K103N + E138Q, K238T; 13: K103N + G190A). The remaining two samples had the K103N only, and a low 1.7 µM (16) and a slightly increased 6.4 µM (19) IC₅₀ value, respectively, was found. Patient 4 who had a K103R mutation was also sensitive with the phenotypic assay. # Discussion In high-income countries, there has been a considerable decline of HIV-1 drug resistance during the last decade. In contrast, transmitted and acquired HIV drug resistance is increasing in low-income countries, especially resistance to the first generation NNRTIs, NVP and EFV. The second generation NNRTI, etravirine (ETR), is approved for treatment-experienced patients and cross- resistance to NVP and EFV is considered to be limited. We have earlier shown that a simple and affordable RT-based phenotypic resistance test can be used for detection of ETR resistance, including cross-resistance to other NNRTIs in clinical samples. In this study, we further expanded the evaluation of the method in a larger clinical cohort including patients with past or ongoing failure on the first generation NNRTIs by comparing the results with those of direct sequencing and UDPS. We found that most of strains from patients with ongoing or past failure to a first generation NNRTI had various degrees of decreased ETR susceptibility by our phenotypic test, despite that the mutational pattern based on direct sequencing did not always predict ETR resistance. For our assay, we previously demonstrated low IC₅₀ values (mean ±SD: 2.5±1.0 µM) in treatment- naïve patients, a good reproducibility and an independence on the amount of plasma HIV RNA for the outcome of the test. It can be argued that to describe a more precise laboratory cut-off for decreased susceptibility an extensive evaluation has to be done and that the assay may overestimate the presence of decreased susceptibility to ETR. However, in the present experiments ETR gained expected changes in IC₅₀ values for all recombinant HIV-1 RTs equivalent to previous reports, one NNRTI naïve patient (no 4) had a low IC₅₀ value, and the test could plausibly describe the lack of impact of K103N on the IC₅₀ value of ETR in the majority of samples. Thus, it is clear that the method can discriminate between mutants with or without ETR-relevant mutations. One possible explanation for the discrepancy between the phenotypic and the genotypic test results obtained by direct sequencing is resistance in minor viral variants. Therefore, UDPS was used in nine samples in which a discrepancy was found. There was a concordance between the direct sequencing and the UDPS for mutations consisting of \>20% of the viral population. Also, eleven additional RAMs were found by UDPS, in all cases \<20% of the viral population, which is well in line with earlier results on the detection levels of direct sequencing. Importantly, in four patients ETR RAMs were found in the minor quasispecies. In at least three of them, the proportion of mutated virus was so high that it is possible that these mutations may have contributed to the phenotypic resistance explaining the discrepancy with the genotype obtained by direct sequencing. Thus, it is possible that our phenotypic method is more sensitive than direct sequencing in identifying minor quasispecies with RAMs in the RT. The degree and pattern of an increased sensitivity as well as the clinical relevance remains to be determined. Also, the clinical treatment history was concordant with our phenotypic results, which corresponded to a true decreased susceptibility for ETR. Thus, during early ART failure and before the NNRTI was stopped, an increase of the IC₅₀ was seen in four of eight samples despite that a sensitive virus was predicted by the genotype. Also, in two patients who stopped NNRTI some weeks before the sampling, an increased IC₅₀ but not ETR-resistance mutations was found. For subjects who had stopped the 1^st generation NNRTI-containing regimen for one or more years, a good concordance between the methods was seen. In addition, in two subjects (7 and 16) with very poor adherence, no phenotypic resistance was identified and only K103N in one of them. These patient histories and the UDPS comparison indicate that our phenotypic method may detect resistance to ETR despite that the direct sequencing predicts a sensitive phenotype. The cause of the discrepancies between the phenotype and the genotype results is not known presently. The mutations predictive for cross-resistance to ETR- resistance have been mainly identified in clinical studies using direct sequencing and it cannot be excluded that not-yet described mutations exist which may influence the ETR sensitivity. The effects of the defined mutations may also vary in different genetic environments. It seems less likely that the replicative capacity of the virus influenced the results although the ratio of relative RT activity to infectious titer has been reported to group according to virus tropism. Thus, there was no correlation between the viral load in plasma, the CD4 T-cell counts and the IC₅₀ of the phenotypic test (data not shown). Also, we have previously shown that the RT amount can vary within a range of 40-828 fg RT/ml without affecting the determination and reproducibility of IC₅₀ values. It should be remembered that our phenotypic method uses lysates originating from intact virions. In contrast, the genotypic and phenotypic methods based on recombinant viruses use HIV RNA which partly may represent defective virus. Actually, it has been reported that endogenous RT activity, not p24 content or viral RNA load, is the best surrogate measure of infectious HIV-1 titer in both cell-free supernatants and viruses purified on sucrose cushions. It can therefore be speculated that the viral population which is studied with our phenotypic assay is more representative for the ongoing viral replication. The clinical utility of the phenotypic method remains to be established. It is clear that there is a strong correlation between the results of our phenotypic method and the predicted antiviral scores according to three genotypic scoring systems. A high IC₅₀ was found in all samples with key-mutations, Y181C and L100I, which are associated to ETR resistance. However, in a substantial number of samples there was a phenotypically decreased susceptibility but no known ETR-RAMs. The clinical relevance of these findings and clinical cut off of our method should be studied on larger patient- populations. However, in view of the high potency of newer anti-HIV drugs and the low failure rate of todays ART, such a study is likely to demand samples from a larger international study. In the present perspective, it is believed that the RT-based phenotypic method for detection of ETR resistance in plasma is possible it could be used as an alternative for studying resistance to first and second generation NNRTI. While waiting for studies on its clinical usefulness, it could also be useful for studies on the kinetics of NNRTI resistance during ART failure. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: EA PN AS. Performed the experiments: EA MC MNJ RP. Analyzed the data: EA PN AS CFRK MC MNJ RP. Contributed reagents/materials/analysis tools: CFRK AS. Wrote the paper: EA AS CMS PN CFRK. Performed RT-based phenotypic assay and analyzed the data: MC MNJ RP. Performed Ultra-deep sequencing and analyzed the data: MC MNJ RP.

# Introduction Globally, suicide has been recognized as a severe public health concern associated with significant disability, psychosocial impairment, and medical illness. Approximately 800,000 people take their own lives each year. Furthermore, suicide rates among adolescents and young adults, especially on university campuses, have been increasing at an alarming rate. Significant precursors to dying by suicide are nonlethal suicidal behaviors—such as suicide ideation and suicide attempts. A report from the USA in 2014 claimed that an estimated 108,000 full-time college students had attempted suicide. Statistics show the prevalence of suicide ideation or suicide plans among college students to range between 5.4–38% worldwide. In China, suicide is the leading cause of death for young adults. The suicide rate in China has been decreasing but remains the second leading cause of death for this age group. University students constitute a unique group due to their relatively high intelligence, aspirations and self-esteem. However, they are in a developmental period, transitioning from adolescence to adulthood, and they must adapt to new living conditions as well as new social and academic challenges. Many of them develop mental disorders, some of which are serious. When confronted with adverse life events, a tense campus climate and psychological stress can exacerbate the stress response, and extreme reactions such as suicidal behaviors are likely to occur. Previous studies have revealed that an increased risk of suicidal behavior in university students is correlated with several identifiable factors, such as personality traits, substance abuse, negative life events (such as academic difficulties), social disconnectedness (poor interpersonal networks), conflict in their families and medical illness, which can include mental disorders or physical disorders (such as a stroke). In a systematic review investigating the relationship between suicide and stroke, stroke as a medical disorder was identified as a significant risk factor for suicidality (including both suicidal ideation and suicidal behavior), especially among depressed patients. The existence of previous mood disorders, a prior history of stroke, and cognitive impairment were found to be important risk factors for suicide. In other studies with undergraduate students, a population group that normally has fewer physical health problems, variables such as hopelessness, anxiety, and depression appear to be the more important risk factors. Subjective well-being factors, such as those related to meaning in life and limited coping abilities, are also potential predictors of suicidality. The present study was designed to explore the association between a number of psychological factors and suicidal behaviors among university students in China, which could inform prevention strategies by better understanding the key risk factors and protective factors that are relevant to university students. Specifically, the main study purpose was to examine whether risk factors including hopelessness, depression, anxiety, stress, coping styles, and protective factors (including orientation to happiness, meaning in life) were significantly associated with suicidal thoughts and behaviors in our sample group. # Methods and material ## Subjects and procedure This study was conducted in November 2016 in Jinan, the capital city of Shandong province, China. Shandong province lies to the south of Beijing and has approximately 100 million residents, of whom there are approximately 7.23 million people living in Jinan. Two large public medical-related universities in Jinan were selected, and two large faculty colleges from each university were sampled for this study. A stratified-clustered-random sampling approach was then undertaken to choose participants from three or four classes in each grade (i.e., freshman, sophomore, junior and senior levels). All students in the sampled classes were invited to complete the self-report questionnaires. It took each respondent approximately 15 minutes to complete the questionnaire during class breaks. The study was approved by the institutional review boards and by the Ethics Committee at Shandong University School of Public Health. Of the 2,197 questionnaires distributed, 2,074 responses were received from undergraduate students, resulting in a response rate of 94.4%. Among the respondents, 706 were males (mean age = 19.86, SD = 1.50 years) and 1,368 were females (mean age = 19.75, SD = 1.33 years). In China, females are the dominant participants in higher education, and it is therefore common for the proportion of female students in universities to exceed that of male students. The mean age of all respondents was 19.79 (SD = 1.39 years). The sample consisted of 574 (27.7%) freshmen, 521 (25.1%) sophomores, 619 (29.8%) juniors, and 360 (17.4%) seniors. Regarding ethnicity demographics, 1,939 (93.5%) were Han Chinese and 135 (6.5%) were from ethnic minorities. ## Measures The Suicidal Behaviors Questionnaire-Revised (SBQ-R) comprises four items, each assessing a different suicidal behavior: lifetime suicide ideation and suicide attempts; frequency of suicide ideation over the past twelve months; threats of suicide attempt; and likelihood of suicidal behavior in the future. The total score, ranging from 3 to 18, could be used to evaluate the suicide risk for undergraduate students. The Chinese version of the SBQ-R was translated by our research team. In this study, the Cronbach alpha value of the SBQ-R was 0.67. The Beck Hopelessness Scale (BHS) is a 20-item instrument measuring one’s negative attitudes about the future. It was modified from yes/no to a 5-point Likert scale ranging from 1 (strongly agree) to 5 (strongly disagree) in a study with Chinese respondents. The Chinese version of the BHS had satisfactory reliability and validity in adolescents, similar to the findings of this study (α = 0.90) based on the 5-point Likert scale. The Orientation to Happiness Questionnaire (OTH) has 12 items that evaluate two different strategies for pursuing happiness (6 items per strategy): a life of pleasure (e.g., “In choosing what to do, I always take into account whether it will be pleasurable.”) and a life of meaning (e.g., “My life serves a higher purpose.”). Using the Chinese version of the OTH, responses were given on a 5-point Likert scale ranging from 1 (very much unlike me) through 5 (very much like me), and higher subscale scores demonstrate a stronger endorsement of the corresponding orientation. The Cronbach alpha value of the OTH was 0.84. Meaning in life was measured by the Chinese version of the Meaning in Life Questionnaire (MLQ). It is made up of two five-item subscales: the presence of meaning in life (MLQ-P) and a search for meaning in life (MLQ-S). All ten items are rated from 1 (absolutely untrue) to 7 (absolutely true). The item scores are then summed, and higher scores indicate a stronger correspondence to meaning in life. In this study, the Cronbach alpha value of the MLQ was 0.88 (0.78 and 0.88 for the MLQ-P and MLQ-S respectively), showing satisfactory internal consistency. The Depression Anxiety Stress Scales (DASS) is a 42-item questionnaire including three subscales to measure three different negative emotional states, namely, depression, anxiety and stress. A revised version of the DASS, the DASS-21, was used in this study; it has seven items for each subscale and has been widely used for the identification and assessment of the corresponding emotional symptoms. The score range for each item varies from 0 (did not apply to me at all) to 3 (applied to me very much or most of the time). A lower score reflects better mental health. Studies have shown the DASS-21 to be a reliable and valid inventory and suitable for studies in Chinese college students. In this study, the Cronbach alpha values of the DASS were 0.84, 0.82, and 0.79 for the depression, stress and anxiety subscales, respectively. The Brief COPE Scale (BCS) is a multidimensional scale, presenting 14 subscales, each with two items measuring a strategy that one would take under some certain kind of stressor. The fourteen coping strategies included in the Brief COPE are self-distraction, active coping, denial, use of instrumental support, substance use, use of emotional support, positive reframing, behavioral disengagement, venting, planning, humor, acceptance, religion, and self-blame. It has been translated into various languages and used in different countries. In this study, an eleven-factor model was adopted to measure situational and dispositional coping responses. ## Statistical analysis Statistical analyses were performed using IBM SPSS Statistics for Windows, Version 24.0 (IBM corp., and Armonk, New York, US). Cronbach’s alpha was used to measure the reliability of scales. Multiple linear regression with stepwise selection method was used to explore significant predictors of suicidal behaviors. Age, gender, and all psychological variables, including hopelessness, depression, anxiety, stress, coping styles, orientation to happiness, and meaning in life were simultaneously entered into the multiple regression model to predict suicidal behaviors with those weak predictors being removed stepwise to result in a final model containing all significant predictors. Furthermore, the variance inflation factor (VIF) and tolerance were calculated to determine the presence of multicollinearity. All relevant data are in Supporting Information. # Results The mean scores and standard deviations for each of the scales are shown in, along with Cronbach’s alphas for each scale. The high Cronbach’s alphas indicated satisfactory reliability for the instruments. The scale scores were subjected to multiple regression, which shows the significant components for the prediction of SBQ. Collinearity diagnosis showed a satisfactory variance inflation factor (VIF\<5) and tolerance in multiple regression. The strongest predictors of suicidal behavior were hopelessness, depression, stress and some coping styles like self-distraction and self-blame. As a protective factor, the presence of meaning in life showed the potential to prevent suicide. This was also consistent with the Pearson correlations, the highest of which were associated with hopelessness, depression and stress. Moreover, we found that age and gender were statistically significant for regression beta, but orientation to happiness had few contributions to suicidality. # Discussion The main findings of the present study were the following: the major risk factors for suicidal behavior were shown to be hopelessness, depression, stress and negative coping styles; furthermore, perceiving meaning in life was shown to be a protective factor for suicidal behavior. The linear regression indicated that hopelessness, depression, stress and the presence of meaning in life were significant predictors of suicidal behavior. It has been well documented in suicidology literature that the greater the severity of depression and the higher the levels of hopelessness and stress, the greater is the likelihood of suicidal behavior in college students. In Furr’s research, approximately 53% of students stated that they had experienced depressive symptoms since entering into college, increasing the likelihood of suicidal ideation or behaviors. Lester found that undergraduate students living on campus were more depressed than those living with parents or off campus. In the present study, the Depression subscale assessed dysphoria, devaluation of life, self- deprecation, lack of interest/involvement, anhedonia, and inertia. Consistent with most previous studies, depression was a predictive factor for suicidal behavior among Chinese university students. Hopelessness has been shown to be a risk factor for suicidal behavior, both nonlethal and lethal behavior types. Individuals experiencing hopelessness have a pessimistic attitude about the future and react poorly to stressors, and hopelessness acts as a mediator between psychological distress and suicidal behaviors. Another well-documented predictor of suicidal behavior in university students is stress. University students face a barrage of stress through homework, projects and internships and thus have little time to join extracurricular activities for relaxation. They also must adjust to new living arrangements, many of which involve being away from home. This present study found that having meaning in life was a protective factor for suicidal behavior, which was consistent with previous research. Having meaning in life may be defined as the pursuit and accomplishment of worthwhile goals and the accompanying satisfaction. A sense of meaning contributes to the foundation of overall happiness and predicts positive traits and psychological strengths associated with fulfillment in life. People without meaning in life tend to experience emptiness and have less motivation to continue to live. Thus, it is important to guide students to establish clear and reasonable life goals, to face and accept frustrations and setbacks in life, to release negative emotions and to optimize meaningful life situations in order to lower the likelihood of future suicide. Previous research has shown that negative coping styles were associated with suicidal behavior. Only some coping skills were found to have a weak relationship with suicidal behavior, and one reason for that finding may be poorly validated measures of coping skills. Furthermore, students with a high suicide risk tended to adopt passive coping skills when facing difficulties due to the lack of proper coping mechanisms to transfer their emotions. The findings of the present study, based on the availability of useful measures to assess suicidal risk in university students in China, suggest that suicidal behaviors are significantly associated with several risk and protective factors. The data highlight the need to provide effective psychological outreach programs and suicide prevention measures and interventions for this population, taking into consideration the risk factors and protective factors of this age group. University students, young adults who have recently enrolled in institutions of higher education, might not yet have developed a mature perspective on life and death or have stable outlooks on suicide. These programs and interventions should be conducted as soon as the students commence undergraduate programs and should be implemented on campus to reach more university students. In designing these preventive programs and interventions, it is important to screen students for risks and protective factors; to design good counseling techniques and content to address the multifacet sources that can increase risk factors; and to identify if an individual student possesses certain protective factor(s) so he or she can be placed in psychotherapy or counseling. For example, counseling and guidance for university students could draw on important protective factors, namely, meaning in life, to prevent suicidal behavior. There are some limitations to this study. First, a cross-sectional study is insufficient to draw firm conclusions on causality, which is better studied using a longitudinal design that follows a cohort of individuals. Second, the samples were recruited from a limited region, and most of the selected subjects were medical students. Therefore, the results may not be representative of all university students. Third, there may have been potential biases associated with unmeasured confounding variables. # Supporting information We are thankful for all the teachers and staff from Shandong University School of Public Health and Medicine and Shandong University of Traditional Chinese Medicine for providing logistic support as well as the participants for this study. [^1]: The authors have declared that no competing interests exist.

# Introduction The extent of conformational change associated with allosteric regulation ranges from movement of entire domains to minimal changes in the protein backbone. While cases involving significant changes in protein structure are often nearly self-explanatory at the structural level, subtle structural alterations are more delicate to capture and interpret. They may involve alterations in pre-existing equilibria of functionally distinct conformational states and complex ligand- binding mechanisms involving conformational selection or induced fit. Moreover, it is increasingly recognized that structural characterization of small conformational changes may be hampered by experimental restrictions such as the conditions used to grow protein crystals which are frequently far from physiological and enforce a particular conformational state on the studied protein. The cysteine peptidase cathepsin K is becoming established as a model enzyme for regulation of papain-like peptidases via sites distant from the active site. However, lack of significant conformational change in X-ray structures has thus far hindered our understanding of this system. The enzyme itself is a critical enzyme in bone homeostasis and has potent collagenase activity. Its targeting is considered one of the most promising approaches for future treatment of osteoporosis. Unfortunately, the first generation of orthosteric inhibitors had limited success. The most successful compound odanacatib concluded phase 3 clinical trials but was terminated recently due to side effects. With regards to allostery, cathepsin K is the first member of the papain-like family in which this mode of regulation was characterized. Glycosaminoglycans (GAGs) were the first known allosteric effectors of cathepsin K with presumably major biological roles in the regulation of its collagenase activity. Chondroitin-4-sulfate (C4S) specifically is a major enhancer of collagen degradation by cathepsin K. Current evidence suggests that the biologically active forms involve oligomeric cathepsin K/C4S complexes with varying stoichiometry and collagenase activity. Thus far, three binding sites for C4S are known on cathepsin K. Other GAGs were also attributed roles in collagenolysis regulation with effects ranging from activation to inhibition. GAGs were also shown to increase the activity of cathepsin K on synthetic substrates and protein substrates other than collagen, e.g. elastin, and are thus allosteric effectors of cathepsin K in general. In our work, we have been investigating allosteric mechanisms by combining computational and experimental approaches. Using the statistical coupling analysis we identified the protein sector of papain-like peptidases, an evolutionarily conserved network of residues that transmits allosteric communication. Based on these results we identified a novel allosteric site on cathepsin K which was later shown to partially overlap with a C4S-binding site. Up-to-date we have characterized two effectors that bind at this site (NSC13345 and NSC94914, respectively, both from the US NCI Developmental Therapeutics Program). Both had inhibitory activity on the hydrolysis of synthetic and protein substrates, but different activity profiles, presumably due to their different modes of interaction with the allosteric site. Despite these data, the acceptance of cathepsin K as a model for allosteric regulation in papain-like peptidases has been hindered due to the absence of conformational change in X-ray structures of cathepsin K in complexes with allosteric effectors and orthosteric inhibitors. Herein we address this issue by a computational approach based on the significant amount of available structural data. Up-to-date over 50 X-ray structures of cathepsin K alone and over 400 structures of papain-like peptidases are available from the Protein Data Bank, hence a large enough sample is available for thorough analysis. We investigate the conformational space of the papain-like family, the position of cathepsin K within it, we examine the conformational variability of the molecule alone and in complexes with known effectors and simulate the binding of a peptide substrate into the active site in the presence of different effectors. Based on these results we provide a molecular model for allosteric regulation of cathepsin K and provide an outlook of similar mechanisms in other related enzymes for which experimental data are not yet available. # Results and discussion ## Conformational space of papain-like peptidases As the starting point of this work, we aimed to obtain a perspective of the total conformational variability within the papain-like family. For this purpose we investigated its conformational space using principal component analysis (PCA) on the complete ensemble of X-ray structures currently available from the Protein Data Bank (see for a complete list of PDB entries). As the major focus of this work, the cathepsin K molecule was used as reference for data analysis and presentation. shows the distribution of secondary structure elements along the sequence of human cathepsin K with designations that will be used throughout the manuscript. The three consecutively numbered loops will also be discussed in detail in the continuation. In the first step of the analysis, the invariant core, which comprises residues with highly conserved spatial positions at the family level, was calculated and mapped on the structure of human cathepsin K, as shown in (invariant core residues are colored blue). The enzyme itself has the typical papain-like endopeptidase fold consisting of L- and R-subdomains according to their relative positions in the standard orientation (middle representation). The active site cleft is positioned between both subdomains at the top of the molecule. Each subdomain contributes one residue to the catalytic diad Cys-His and loops lining the cleft at the respective sides. The L-subdomain contains a central α-helix (α2) which runs vertically across the whole molecule and two shorter α-helices (α3 and α4), whereas the R-subdomain contains a central antiparallel four- stranded β-sheet with a barrel-like organization (β-strands 1 through 4) and two short α-helices (α5 and α6). Both subdomains are connected via a two-stranded antiparallel β-sheet at the front (β-strands 1' and 2') and the N-terminal loop at the back of the molecule which contains a short α-helical segment (α1). The invariant core of the molecule consists of the N-terminal portion of the helix α2, the four-stranded β-sheet, a portion of the loop connecting both subdomains on the back side and a single residue (position 6 in cathepsin K) immediately preceding helix α1. Function-wise, the invariant core in cathepsin K comprises the active site and the C4S-binding site on the back side, and lies beneath the allosteric site at the bottom right side which is bound by both GAGs and small molecule effectors. Within the active site cleft, the invariant core comprises the S1 site (according to the nomenclature by Schechter and Berger) as well as the primed sites, but only part of the S2 site, which is the primary specificity determinant in papain-like endopeptidases. Sites preceding the S2 site are not formed by spatially conserved residues, hence this part is more diverse between family members and potentially more flexible within individual molecules. For the purposes of PCA, the ensemble was superposed onto the invariant core prior to analysis. The results of PCA showed that most of the conformational variability (approx. 64%) is contained within the first two principal components (PC1 and PC2) which account for 50.4% and 13.5% of sample variance, respectively, enabling its representation in a straightforward two-dimensional diagram. The plot of PC1 versus PC2 shows the existence of four groups within the papain-like family that are distinguished by the positions of cleavage sites within protein substrates. The exopeptidases cathepsin B (a peptidyl dipeptidase with additional pH-dependent endoproteolytic activity), cathepsin X (a carboxypeptidase) and dipeptidyl-peptidase I (also known as cathepsin C) each form separate groups, whereas papain-like endopeptidases form the fourth group. This basic division reflects the divergence of these groups from the common ancestor in early eukaryotes. The endopeptidase group also includes the aminopeptidase cathepsin H which, based on proenzyme homology, is more closely related to the endopeptidases than other exopeptidases. Addition of PC3 (7% of sample variance) results in separation of bacterial and eukaryotic endopeptidases, but apart from this does not provide significant additional information. Therefore, the plot of PC1 versus PC2 will be used for analytical purposes in the continuation. The endopeptidase group itself shows well- structured hierarchy based roughly on the phylogenetic relationships of the proteins. Bacterial proteases form two distinctive groups. The lower one is formed by xyllelain, a papain-like peptidase from pathogenic strains of the plant pathogen *Xyllela fastidiosa*, whereas the upper group is formed by the *Clostridium* peptidase Cwp84, an S layer multidomain protein. Distinctive groups are also formed by endopeptidases from unicellular eukaryotes (mostly *Trypanosoma* and *Plasmodium* species), plants and animals. Within the animal group, which contains mostly mammalian enzymes, cathepsin H lies separate from the rest, whereas the endopeptidases are clustered together, with the exception of two structures of procathepsin L2 from the mealworm *Tenebrio molitor* which cluster with cathepsin H. Within the endpopeptidase group, cathepsins L, S and K form partially overlapping clusters and cathepsin V is clustered together with cathepsin L. Cathepsin F is located outside of the main clusters, as expected due to its distant evolutionary relation, but nevertheless overlaps with some cathepsin L and S structures, thus reflecting a similar overall conformation of the molecule. Further subdivision can be observed within the plant and protist groups, but will not be further investigated herein. Structural variability within the family, as identified by PCA, is shown in. The first shows root mean square fluctuations (RMS fluctuations, RMSF) along the sequence alignment of the ensemble and the second shows three-dimensional representations of the variance explained by the first two PCs. The most variable regions include the long left-side loop connecting helix α4 and strand β1' (loop 2), strands β1' and β2', and the C-terminal part of the loop connecting strands β1 and β2 (loop 3) which lines the right side of sites S1 and S2. RMSF values for these regions are between 2 Å and 5 Å. Structural superposition of various papain-like peptidases shows that these fluctuations are the result of structural differences between family members. For additional insight into their structural diversity, three-dimensional structures of representative family members are shown in and their superpositions with cathepsin K in. It must also be noted that the analysis in is based only on 178 non-gap positions in the sequence alignment of the ensemble and does not account for additional structural elements found only in individual members of the family. Interestingly, the corresponding regions with high fluctuations are also not part of the previously identified protein sector of papain-like peptidases. Thus, they appear to represent elements that evolve independently from the rest of the molecule. The results of PCA also showed dispersion of PC values calculated for the same protein, indicating structural variabilility of the associated regions within individual proteins. To analyze the variability of cathepsin K, we examined RMS fluctuations in 55 non-redundant cathepsin K structures and created a superposition of four structures with extreme values of PC1 and PC2, respectively. In general, little variability was observed in in the molecule. Loop 2 (residues 91 through 105 in cathepsin K) was the most variable part, however, maximal RMSF values were only about 1 Å. This is well below the 2 Å threshold, which is a common criterion for conformational change. Nevertheless, this region appears to be functionally important in cathepsin K, as it contains two predicted allosteric sites that are presumably bound by several different effectors and is also involved in the proposed oligomerization of cathepsin K. Other parts that were identified as flexible at the family level showed only minor fluctuations, which could also be considered as artefacts from the modeling of loop regions in the X-ray structures. Interestingly, structural superposition also showed a different conformation of loop 1 in one of the molecules. This loop lines the left side of sites S1, S2 and S3 and showed comparably little variance at the family level. The observed conformation belongs to the structure of procathepsin K ( shows the structure available under PDB accession code 1BY8) and was thus far not observed in mature cathepsin K. Nevertheless, it indicates that loop 1 can adopt different conformations. Thus, taken together, the results of PCA indicate that the loops surrounding the non-primed sites of the active site (loops 1 and 3) are flexible, which may be important for the functional properties of cathepsin K and other papain-like peptidases. However, at least in the case of cathepsin K, no convincing evidence of conformational variability could be obtained from X-ray data. Lack of observable change could be due to crystal contacts or to crystallization conditions that usually involve high salt concentrations, to which cathepsin K is known to be sensitive. Since most structures include inhibitors bound into the active site, these could also influence the conformation of the latter. For further insight into the conformational fluctuations in cathepsin K, the ensemble of its X-ray structures was also analyzed by normal mode analysis. Fluctuations corresponding to the first five (non-trivial) normal modes are illustrated in and the corresponding coordinates are supplied as. Essentially, residues that comprise the invariant core show little fluctuation, whereas the rest of the molecule is more flexible. Normal mode 1 shows coordinated movement that comprised most of the molecule. Most importantly, it comprises loops surrounding the active site as well as structural elements comprising the allosteric sites (see for reference) indicating that intramolecular movements are coupled between these sites. ## Conformational space of cathepsin K investigated by molecular dynamics (MD) simulations Results in the previous section showed that conformational variability exists in the family of papain-like peptidases, but only limited variability could be observed in cathepsin K by crystallographic analyses. To investigate the conformational space not reachable by X-ray crystallography, we performed MD simulations of cathepsin K alone and in complexes with substrate or allosteric effectors. For comparison, the analysis was also extended to other animal endopeptidases in their unbound states. The calculated trajectories were analyzed using PCA and mapped onto the conformational space defined by the X-ray ensemble in the previous section. Overall, trajectories contained between 100 and 240 ns of total simulation time (see section MD simulations for details) and each run was repeated multiple times to minimize the impact of random events. As expected, MD conformers of cathepsin K sampled a significantly greater portion of the conformational space than observable by X-ray crystallography. Its conformational space spans all X-ray conformers of animal endopeptidases as well as cathepsin H, but does not extend into the conformational space of its non-animal (i.e. plant) counterparts. To filter out low-populated states and for ease of comparison, the 5th and 95th percentiles were calculated for each PC and are represented graphically as a rectangle. The latter, therefore, excludes conformers with extreme values of either or both principal components and shows that the region of conformational space which includes between 80% and 90% of the »common« conformers is significantly smaller than the entire sampled conformational space. Nevertheless, in the case of cathepsin K, it still encompasses the entire X-ray ensemble of vertebrate endopeptidases, and, conversely, X-ray conformers of cathepsin K still represent only a small portion of the area. Due to their structural and functional similarity to cathepsin K, all human papain-like endopeptidases (cathepsins L, S, V and F) were investigated by MD simulations in the same manner. Similar to the former, all enzymes sampled a significantly greater portion of conformational space in comparison to their X-ray structures. Comparison of the 5th/95th percentile boundaries of both PCs showed significant overlaps of the conformational spaces between the evolutionarily closely related cathepsins K, L, S and V, indicating that all vertebrate cathepsin L-like enzymes adopt similar conformations. In contrast, the conformational space of the evolutionarily distant cathepsin F was distinctively different from these peptidases. In the continuation, we analyzed MD simulations of cathepsin K in complexes with different ligands. These included the octapeptide substrate AGLKEDDA bound into the active site to simulate the bound mode of the enzyme, allosteric effectors NSC13345 and NSC94914, and C4S bound to each of the three known binding sites on cathepsin K. The particular substrate was chosen based on individual site preferences of human cathepsin K as determined by proteomic analyses. Cathepsin K/C4S complexes were investigated at 1:1 stoichiometry. The proposed oligomeric forms observed in X-ray structures were not included since their structure- function relationship is not fully understood. The results showed that MD conformers from these simulations were constrained to approximately the same conformational space as ligand-free cathepsin K. However, examining the 5th/95th percentile boundaries of PC values showed notable differences between free enzyme and its complexes with NSC13345 and NSC94914, respectively. These results indicate a redistribution of populations within the total conformational space accessible to cathepsin K. A similar effect was observed for C4S bound to the third binding site (complex 3), but not for the remaining simulations. Altogether, the observed differences were not drastic and the common conformers (i.e. rectangles) still overlapped significantly. Nonetheless, the observed differences are, for example, comparable to those between cathepsins K and S, i.e. two different peptidases. It is also of note that effectors NSC13345 and NSC9414 both caused redistribution of conformers, but in different directions in conformational space. Thus, despite binding to the same site, they exhibited different effects on protein conformation. Moreover, binding of C4S to the same site (complex 2) did not significantly affect the conformational space of cathepsin K. Altoghether, these results highlight our previous findings that allosteric effects triggered by the effectors strongly depend on their mode of interaction with this binding site. For structural evaluation of the observed differences in conformational space, RMS fluctuations were first calculated for each simulation. As expected, fluctuations in ligand-free cathepsin K were higher than those determined for the crystal structures. However, RMSF values at most positions were still well below 2 Å and the mean RMSF value for the whole protein was about 0.8 Å, indicating the presence of a single conformation. The presence of substrate or C4S did not have significant effects on fluctuations in cathepsin K, whereas both small molecule effectors significantly increased the flexibility of the protein (mean RMSF values of 1.2 Å and 1.3 Å for NSC13345 and NSC94914, respectively), mostly due to increased flexibility of loop regions. Interestingly, the strongest effect was observed for loop 2, despite its location on the opposite side of the molecule with regards to the ligand-binding site. For direct comparison between the simulations and to X-ray data, average MD conformers were calculated from each trajectory. Their positions within the conformational space are shown in and mirror the observations made in the previous paragraphs. Most average conformers lie outside of the conformational space accessible in crystallographic experiments, and complexes with both small molecule effectors lie farthest from ligand-free enzyme. Superposition of average MD conformers showed a potentiation of the differences observed by superposition of X-ray structures (see for comparison), but significant conformational changes were not observed. Highest degree of variability was associated with loops 1 and 2, and, to a smaller extent, loop 3. In addition to the comparison of average MD conformers, we also attempted to identify individual conformations in the simulations of cathepsin K in complexes with NSC13345 and NSC94914, respectively, which had RMSF values significantly higher than 2 Å at multiple locations. In each case, three major clusters were identified. They contained 53% (cluster 1), 33% (cluster 2) and 6% (cluster 3) of all frames in the case of the NSC13345 complex and 59% (cluster 1), 28% (cluster 2) and 7% (cluster 3) in the case of the NSC94914 complex. Superpositions of representative frames from each cluster and average MD conformers are shown in for the NSC13345 and NSC94914 complexes, respectively. In general, the clusters differ among each other mostly by the conformations of loops 1, 2 and 3, and the differences are more pronounced in the NSC94914 complex. In both cases, however, the observed differences were small and the clusters were very sensitive to changes in calculation parameters. Therefore, average MD conformers were used for structural analysis in the continuation. Taken together, MD simulations showed that loop regions, especially loops marked 1, 2 and 3, are the only regions of significant conformational variability in cathepsin K. Therefore, the observable allosteric effects are likely to be transmitted via these loops. ## Conformational variability of the active site Ultimately, the action of effectors results in structural change of the active site and altered enzyme activity. The results in the previous sections indicate that only minor conformational changes occur in cathepsin K, however, these can be sufficient to modify enzyme activity, especially when keeping in mind that modulation of cathepsin K usually involves a few-fold decrease or increase in enzyme activity and not full inhibition or activation. We investigated the variability of the entire active site region in detail by comparing the distributions of interatomic distances that define its geometry between the MD simulations. Significant differences were observed only around sites S1 and S2 that are formed by loops 1 and 3, which were both identified as flexible regions in the previous sections. The remaining interatomic distances along the active site cleft did not vary significantly between simulations. Specifically, distances between residue pairs Gly65-Asn161 and Gly66-Leu160 were identified as the critical determinants. These two pairs form the narrowest part of the active site cleft and line both sides of the S2 site, the primary specificity determinant of papain-like endopeptidases. To avoid ambiguity, the pairs will be referred to as the proximal and distal pairs, respectively, based on their proximity to the catalytic diad and the region will be referred to as the S2-S1 cleft. Of the involved residues, Asn161 is the most flexible. Main chain rotations result in its Cβ atom being oriented either away or towards the cleft, whereas rotation about the Cα-Cβ bond results in varying positions of its terminal amide group. For the purpose of comparison between simulations, we determined interatomic distances between the carbonyl O atom or the Cβ atom of Asn161 and the Cα atom of Gly65 which is the closest atom on the other side of the cleft. Analogously, we examined the distance between carbonyl oxygens of residues Gly66 and Leu160. The distributions of interatomic distances in MD simulations are shown in and. Data for the investigated ensemble of X-ray structures are also included for comparison. Taken together, the results showed that binding of substrate reduced the fluctuation of the enzyme and stabilized a particular conformation which can be assumed to be the catalytically productive conformation. Interatomic distances between each atom pair in the cathepsin K/substrate complex were similar to those determined for the X-ray ensemble. Free enzyme adopted a conformation in which the proximal pair was closer together and therefore the S2-S1 cleft was narrower than in the cathepsin K/substrate complex. The distal pair was distributed between two distinct populations which did not deviate considerably from the distances determined for the cathepsin K/substrate complex. In the presence of the inhibitors NSC13345 and NSC94914, multiple populations were observed, mirroring the increased flexibility of cathepsin K in the presence of these effectors. Distances determined for the former were overall similar to free enzyme, whereas the latter also induced an additional population with a significantly narrower S2-S1 cleft, akin to cluster 3 in. In contrast, C4S stabilized conformations with interatomic distances between each pair similar or greater than in the cathepsin K/substrate complex regardless of the site to which C4S was bound. Experimental data had collectively shown, that all effectors predominantly affect *K*_m with only minor variations observed in the catalytical rates of the enzyme. To assess whether the observed differences in the shape of S2-S1 cleft suffice to affect the substrate affinity of cathepsin K, we docked the substrate AGLKEDDA into the active sites of average MD conformers calculated from each simulation. It must be noted that cathepsin K was treated as a rigid molecule, therefore the effects of induced fit of the substrate were not taken into account in the docking experiments. The pose of AGLKEDDA bound in a substrate-like manner is shown in. It was produced by re-docking the substrate to the average MD conformer of cathepsin K from the simulation of the cathepsin K/substrate complex. Free cathepsin K adopted a narrower conformation in the S2-S1 cleft, as discussed above. As a consequence, docking experiments produced binding poses in which the non-primed part of the substrate failed to bind into the active site in a substrate-like manner. Most notably, Leu2 of the substrate was prevented from binding into the S2 pocket. Taking into account that docking was performed with a rigid receptor molecule, these results support the binding of substrate to the enzyme via an induced fit mechanism where the non-primed sites of the enzyme must move further apart in order to accomodate the substrate. We have, in fact, previously observed two separate, catalytically competent, conformations of cathepsin K at neutral pH *in vitro*, which appear to be analogous to the conformations observed in the computational experiments herein. The tense (T) state, akin to the conformation of free cathepsin K herein, was stabilized by low salt buffers in the absence of substrate, had lower substrate affinity and was less susceptible to inhibition by irreversible and reversible inhibitors, whereas the relaxed (R) state with higher substrate affinity was stabilized by high salt concentration, similar to the conditions used to produce protein crystals. Moreover, a slow hysteretic transition from the T to the R state was observed when the enzyme was stimulated with low concentrations of substrate. The structural basis for the experimentally observed transitions thus appears to be the herein identified conformational change involving the non-primed sites of the enzyme. In a similar fashion, docking of the substrate AGLKEDDA to average MD conformers of cathepsin K in complexes with allosteric modifiers was performed and analyzed. Small molecule effectors NSC13345 and NSC94914 stabilized conformations with a narrow S2-S1 cleft similar to the free enzyme. As a result the substrate failed to bind into the S2 pocket. Both effectors can thus be predicted to stabilize the T state. This is in agreement with experimental data which have shown that both effectors act predominantly by reducing the substrate affinity of cathepsin K. In contrast, C4S stabilized conformations with a wider S2-S1 cleft. Accordingly, the substrate could be readily docked into the active site in a manner analogous to the cathepsin K/substrate complex. This indicates that C4S stabilizes the R state, as observed experimentally at physiological plasma pH. ## Other cathepsins Based on the data presented in the previous sections, we also sought to examine the variability of other human cathepsin endopeptidases (cathepsins L, V, S and F) and its potential effects on their activity. Unfortunately, no allosteric effectors of family members other than cathepsin K have yet been characterized at the structural level. However, there have, been several indications of their existence. GAGs, in particular, have been shown to modulate the activity and stability of not only cathepsin K but also of cathepsins S and B as well as several non-animal endopeptidases including papain, cruzipain from *Trypanosoma cruzi*, rhodesain from *Trypanosoma brucei* and cysteine peptidase B from *Leishmania mexicana*. Moreover, DNA has been shown to act in a fashion similar to GAGs and regulate the susceptibility of cathepsin V towards inhibition by serpins. Identification of mechanisms of active site adaptation similar to cathepsin K would indicate that papain-like peptidases share a common mechanism of allosteric regulation that is exploited by GAGs and could potentially also be targeted for drug design. As above, we perfomed MD simulations followed by analysis of interatomic distances and docking of substrate molecules to X-ray and average MD structures. Appropriate substrates for each enzyme were constructed according to their specificity profiles. The examined endopeptidases showed varying extent of conformational change in the loops lining the non-primed sites. The width of the S2-S1 cleft was determined from the distributions of interatomic distances between the proximal and distal atom pairs equivalent to those in cathepsin K. The results are shown graphically in and the numerical values are given in. Similar to cathepsin K, all free enyzmes showed tendencies towards a narrower S2-S1 cleft in comparison to initial X-ray structures. The effect was most apparent in cathepsin F and could be attributed predominantly to conformational changes in loop 1. In contrast, cathepsins L, V and S showed conformational changes predominantly in loop 3. Judging from the superposition of X-ray and average MD conformers, cathepsin S exhibited the least conformational change. It did, however, exhibit the widest interval of proximal pair distances, indicating that the S2-S1 cleft in cathepsin S is somewhat more flexible than in its homologs. These observations were mirrored in the results of docking experiments in which we compared the poses of substrate molecules docked into the active sites of X-ray and average MD conformers of each peptidase, where X-ray conformers represent the bound state, since they were derived from X-ray structures of enzyme/inhibitor complexes, and average MD conformers represent the free state. The results are shown in. The docking poses of the substrate AAVGGTHA bound to cathepsin S were virtually identical, as expected due to the lack of observed conformational change. Less expectedly, similar docking poses were also obtained for the substrate AGFGGHHA bound to cathepsin L, indicating that narrowing of the S2-S1 cleft does not necessarily adversely affect substrate affinity. In contrast, cathepsins V and F, which had average MD structures with pronouncedly tighter S2-S1 clefts akin to cathepsin K (see for comparison), exhibited similar steric hindrance in adopting their respective substrates (AGFGGHHA and AGLKAAAA, respectively) into the active site cleft. Taken together, these results indicate that other papain-like endopeptidases exhibit similar active site dynamics as cathepsin K and thus may be regulated allosterically in a similar manner. # Conclusion Allosteric regulation is slowly surfacing as a common regulatory strategy in papain-like peptidases. Our understanding at the structural level has so far been restricted to the identity of allosteric sites and a limited number of effectors, whereas knowledge on how these effectors work has been limited due to the absence of obvious conformational change observable in X-ray structures. Herein we have made significant progress in our comprehension of this mode of action. We now have strong indications that allosteric regulation in cathepsin K, the primarily investigated papain-like peptidase from the allosteric perspective, proceeds via stabilization of conformational states with different accesibility of the S2 site, the primary specificity determinant, to the substrates. Conformational plasticity is an inherent property of cathepsin K and our data indicate that similar mechanisms also operate in other cysteine cathepsin peptidases. Altogether this analysis provides an important basis for experimental investigation of allostery in this family. # Methods ## Structural data collection and visualization All coordinates used for analyses and simulations were retrieved from the RCSB Protein Data Bank ([www.rcsb.org](http://www.rcsb.org)). A complete list of entries used in this work is given in. UCSF Chimera was used for superposition and other coordinate manipulations, as well as for visualization of all molecular structures presented in the manuscript. ## MD simulations MD simulations were performed using NAMD version 2.11 with CUDA acceleration on a single node-desktop computer equipped with a quad-core processor and a CUDA- capable graphics card. The starting coordinates for all simulations were retrieved from the RCSB Protein Data Bank (PDB) under the following accession numbers: 1ATK and 1U9V for human cathepsin K, 3C9E and 4N8W for human cathepsin K in complexes with C4S, 5J94 for human cathepsin K mutant Cys25Ser in complex with NSC13345, 5JA7 for human cathepsin K mutant Cys25Ser in complex with NSC94914, 1MHW for human cathepsin L, 1FH0 for human cathepsin V, 2OP3 for human cathepsin S and 1M6D for human cathepsin F, respectively. CHARMM 27 force field parameters were used for the proteins and the ligands were parametrized using the SwissParam web server. The simulations were run in explicit solvent under periodic boundary conditions at a constant temperature of 298 K and with default values of interaction parameters. The systems were minimized for 200 steps and equilibrated for 200 ps. The trajectories collected at a frequency of 10 ps per frame. The lenghts of individual runs were between 20 ns and 150 ns. In the case of ligand dissociation from the binding site, which occured occasionally with all ligands, only frames containing ligand bound to the binding site were retained. Final trajectories were pooled from three runs. Total simulation times collected were 200 ns for ligand-free cathepsin K, 100 ns for cathepsin K/substrate and cathepsin K/C4S complexes, and for cathepsins L, V, S and F, 125 ns for the cathepsin K/NSC94914 complex and 240 ns for the cathepsin K/NSC13345 complex. The Wordom program was used to calculate average structures (average MD conformers)from the trajectories, to calculate RMSF and per-frame distances between designated atom pairs, as described in the text, and to cluster the trajectories of simulations of cathepsin K/NSC13345 and cathepsin K/NSC94914 complexes. Clustering was performed using the quality threshold (qt) method and root mean square distances at a cut-off radius of 1.2 Å ## Principal component analysis PCA was performed with the Bio3D package on the ensemble of structures collected in. If necessary, the PDB files were manually edited to replace non-standard residue codes (e.g. derivatives of the active site Cys25 residue) with standard three-letter codes. The sequences were aligned using the *pdbaln* command and the alignment then manually refined (supplied as). The ensemble was filtered to remove conformationally redundant structures using the *filter*.*rmsd* command with a cut-off value of 0.1 Å. Based on the alignment, the invariant protein core was determined using the *core*.*find* command, all sequences superposed onto the core and then analyzed with PCA. The *mktrj* command was used to create trajectories representing the conformational variability along each principal component, which were analyzed using UCSF Chimera. The *rmsf* command was used to calculate the RMSF of the ensemble after its superposition onto the invariant core. Trajectories from MD simulations described in the previous section were superposed onto the invariant core with the *fit*.*xyz* command and then each frame projected onto the principal components of the X-ray ensemble using the *project*.*pca* command. ## Normal mode analysis Normal mode analysis was performed using the Bio3d package on an ensemble of non-redundant 55 structures of cathepsin K. The ensemble was obtained after filtering the sample of all cathepsin K structures with the *filter*.*rmsd* command using a cut-off value of 0.1 Å to remove conformationally redundant strutures. The first five normal modes were converted to trajectories with the *mktrj* command and visualized in UCSF Chimera. ## Docking of the substrate AGLKEDDA into the active site of cathepsin K As the ligand molecule, the octapeptide Ala-Gly-Leu-Lys-Glu-Asp-Asp-Ala (AGLKEDDA) was constructed *de novo* using UCSF Chimera and the torsion angles constrained to an extended parallel β-strand conformation. The molecule was then prepared for docking using AutoDock Tools. Polar hydrogens were added to the molecule, all single bonds defined as rotatable and peptide bonds designated as non-rotatable. As the receptor, the cathepsin K structure retrieved from the Protein Data Bank under accession number 1ATK was used. All non-protein atoms were removed and non-polar hydrogens added. AutoDock Vina was used to dock the ligand into the active site of the receptor, which was treated as a rigid body. 20 docked poses were calculated at an exhaustiveness level of 20. Model selection was performed manually by screening solutions according to known data on the binding of substrates into the active site of papain-like endopeptidases (see refs.), i.e. extended conformation of the peptide chain with all residues interacting with the enzyme, residue Leu2 bound in the S2 pocket and the scissile bond between Lys3 and Asp4 positioned above the catalytic residue Cys25. The same procedure was used to dock the substrate AGLKEDDA into the active sites of average MD conformers of cathepsin K derived from MD trajectories. Since structure averaging over MD trajectories results in distortion of flexible side chains, correct geometry of the receptors was first restored by energy minimization with NAMD. Minimization was performed until a negative total energy of the protein was reached, which was between 200 and 300 steps. Superposition of the minimized and initial structures showed that the minimization did not cause significant changes in the coordinates of backbone atoms. ## Docking of substrate molecules to other papain-like endopeptidases Other substrate molecules were docked into the active site of respective enzymes by the same procedure described in the previous section. The constructed substrates were Ala-Gly-Phe-Gly-Gly-His-His-Ala (AGFGGHHA) for cathepsins L and V, Ala-Ala-Val-Gly-Gly-Thr-His-Ala (AAVGGTHA) for cathepsin S and Ala-Gly-Leu- Lys-Ala-Ala-Ala-Ala (AGLKAAAA) for cathepsin F. The PDB entries used to retrieve the macromolecular coordinates were 1MHW, 1FH0, 2OP3 and 1M6D for cathepsins L, V, S and F, respectively. As above, docking was performed using the initial X-ray conformers and average structures calculated from MD trajectories. # Supporting information [^1]: The author has declared that no competing interests exist.

# Introduction As an important concept in nonlinear science, chaos is characterized by unstable dynamic behavior with sensitive dependence on initial conditions and includes infinite unstable periodic motions. The control and synchronization of chaotic systems have gained considerable attention in recent years. Furthermore, several researchers have recently directed increasing interest toward the chaotic behavior of fractional-order dynamical systems. However, considerable research on the behavior of fractional-order dynamical systems requires the parameters of the fractional-order chaotic system to be determined in advance. Unfortunately, these parameters are usually unknown. In the past decades, numerous methods were proposed to solve the parameter estimation of integer-order chaotic systems. Comparatively, little attention has been devoted to the parameter estimation of fractional-order chaotic systems. In this study, we preliminarily focus on the parameter estimation problem of fractional-order chaotic systems; those problem can be formulated as a multidimensional optimization problem. To solve this multidimensional optimization problem, a novel algorithm called quantum parallel particle swarm optimization (QPPSO) is proposed. The parallel characteristic of quantum computing is used to improve the ergodicity of QPPSO. Moreover, quantum states are updated by a new quantum evolution equation, which is constituted by the current rotation angle, individual optimal quantum rotation angle, and global optimal quantum rotation angle. The performance of QPPSO in solving the parameter estimation problem of fractional-order chaotic systems is investigated through a comparison with those of other evolutionary optimization algorithms. The remainder of this paper is organized as follows. A brief review of relevant work on the chaotic behavior of fractional-order dynamical systems and the parameter estimation of chaotic systems is presented in Section 2. Parameter estimation for fractional-order chaotic systems from the viewpoint of optimization is formulated in Section 3. A discussion on QPPSO after a brief introduction of the quantum parallelism character is made in Section 4. Numerical simulation results based on several typical fractional-order chaotic systems and comparisons with a few existing approaches are provided in Section 5. Finally, the conclusions and a brief summary of the results are presented in Section 6. # Literature Review Considerable research has been done on the chaotic behavior of fractional-order dynamical systems. The chaotic behavior of the fractional-order Lorenz system was studied in, in which the authors determined that the system with *Σ* \< 3, where *Σ* is defined as the sum of the orders of all involved derivatives, can exhibit chaotic behavior. In, the chaos and hyperchaos of fractional-order Rössler equations were studied. In this research, the authors showed that chaos can exist in the fractional-order Rössler equation with an order as low as 2.4 and that hyperchaos can exist in the same equation with an order as low as 3.8. In, the chaotic behavior of the fractional-order Chen system was examined. The authors determined that chaos exists in the fractional-order Chen system with an order less than 3. In, Lu numerically investigated the chaotic behavior of the fractional-order Lü system. A remarkable finding is that the lowest order for this system to exhibit chaos is 0.3; this system is therefore the lowest-order chaotic system among all chaotic systems reported in the literature. More fractional-order systems with chaotic behavior are discussed in detail in. However, these studies do not focus much on estimating the parameters of fractional-order chaotic systems. To date, much work has been made on the parameter estimation of integer-order chaotic systems. Dai et al. used the genetic algorithm (GA) to estimate the parameters of the Lorenz chaotic system. Li et al. utilized a chaotic ant swarm algorithm to identify the parameters of the logistic iteration and Lorenz systems. In, differential evolution was used to identify the parameters of the Lorenz system. In, the particle swarm optimization (PSO) algorithm was applied to solve the parameter estimation of chaotic systems. Numerical simulation shows that PSO is a feasible approach for the parameter identification of integer- order chaotic systems. Although some progress has been made in the parameter estimation of integer-order chaotic systems, the parameter estimation of fractional-order chaotic systems is more complicated than that of integer-order chaotic systems. Therefore, we put forward QPPSO to solve the complicated problem of parameter estimation of fractional-order chaotic systems. Moreover, QPPSO can also be applied to many other aspects such as 3-d object retrieval and recognition, hyperspectral image classification and visual-codebook compression. # Problem Description Consider the following *n*-dimensional fractional-order chaotic system: $$D^{q}X = F\left( {X,X_{0},\theta} \right),$$ where *X* = (*x*₁, *x*₂, …, *x_n*)^*T* ∈ *Rⁿ* denotes the *n*-dimensional state vector of the original system, *X*₀ represents the system initial state, *q* = (*q*₁, *q*₂, …, *q_n*)^*T* ∈ *Rⁿ* is a set of fractional order of the original system, and *θ* = (*θ*₁, *θ*₂, …, *θ*_D)^T ∈ *R^D* is the value of the original system parameters. Suppose that the structure of the system is determined in advance, therefore the estimated system can be described as follows: $$D^{\hat{q}}\hat{X} = F\left( {\hat{X},X_{0},\hat{\theta}} \right),$$ where $\hat{X} = \left( {{\hat{x}}_{1},{\hat{x}}_{2},\ldots,{\hat{x}}_{n}} \right)^{T} \in R^{n}$ expresses the *n*-dimensional state vector of the estimated system, $\hat{q} = \left( {{\hat{q}}_{1},{\hat{q}}_{2},\ldots,{\hat{q}}_{n}} \right)^{T}$ is the estimated order of the system, and $\hat{\theta} = \left( {\hat{\theta}}_{1},{\hat{\theta}}_{2},\ldots,{\hat{\theta}}_{D} \right)^{T}$ is a set of estimated parameters. Essentially, the parameter estimation of the fractional-order chaotic system problem considered here involves searching for the optimal parameters <u>$\hat{q}$</u> and $\hat{\theta}$, the performance index shown in is minimized. $$\min J = \frac{1}{M}{\sum\limits_{k = 1}^{M}\left\| {X_{k} - {\hat{X}}_{k}} \right\|}^{2},$$ where *M* denotes the length of data used for parameter estimation. *X_k* and ${\hat{X}}_{k}$ (*k* = 1, 2..., *M*)denote the state of the original and the estimated systems at time *k*, respectively. Evidently, the parameter estimation for the fractional-order chaotic system can be considered asa multi-dimensional continuous optimization problem, where *q* and *θ* are the decision variables, and *J* is the optimization goal. The principle of parameter estimation for fractional-order chaotic systems in the optimization sense is shown in. Because of the unstable dynamic behavior of fractional-order chaotic systems, accurate parameters are difficult to obtain. Moreover, traditional optimization methods are difficult to derive in global optimal parameters as many local optima in the landscape of *J*. Therefore, a novel algorithm called QPPSO is proposed to solve the parameter estimation of fractional-order chaotic systems. # QPPSO ## Quantum parallelism Quantum parallelism, which enables quantum computers to simultaneously evaluate a function *f*(*x*) for many different values of *x*, is a fundamental feature of numerous quantum algorithms. Consider the circuit shown in. The circuit applies *U_f* to an input that is not in the computational basis. The data register is prepared in the superposition $\left( {\left| 0 \right\rangle + \left| 1 \right\rangle} \right)/\sqrt{2}$, which is created with a Hadamard gate acting on $\left| 0 \right\rangle$. The state $\left| \psi \right\rangle$ is then calculated with *U_f* as follows: $$\left| \psi \right\rangle = U_{f}H\left| {0,0} \right\rangle = U_{f}\frac{1}{\sqrt{2}}\left( {\left| {0,0} \right\rangle + \left| {1,0} \right\rangle} \right) = \frac{1}{\sqrt{2}}\left( {\left| {0,f(0)} \right\rangle + \left| {1,f(1)} \right\rangle} \right)$$ where $\left| \cdot \right\rangle$ is called the Dirac notation and it is the standard notation for states in quantum mechanics, *H* is the Hadamard gate, and *U*_f is the quantum circuit that takes inputs, such as $\left| x,y \right\rangle$, to $\left| {x,y \oplus f(x)} \right\rangle$. In this study, a single *f*(*x*) circuit is used to simultaneously evaluate the function for multiple values of *x* by utilizing the capability of a quantum computer to superposition different states. This procedure can easily be generalized to functions on an arbitrary number of bits by using a general operation known as the Hadamard transform, or sometimes, the Walsh–Hadamard transform. The result of initially conducting the Hadamard transform on *m* Q-bits in $\left| 0 \right\rangle$ state is expressed as follows: $$\begin{array}{cl} \left| \psi \right\rangle & {= H^{\otimes m}\left| 0 \right\rangle^{\otimes m} = H \otimes H \otimes \cdots \otimes H\left| {00\cdots 0} \right\rangle} \\ & {= \frac{1}{\sqrt{2}}\left( {\left| 0 \right\rangle + \left| 1 \right\rangle} \right) \otimes \frac{1}{\sqrt{2}}\left( {\left| 0 \right\rangle + \left| 1 \right\rangle} \right) \otimes \cdots \otimes \frac{1}{\sqrt{2}}\left( {\left| 0 \right\rangle + \left| 1 \right\rangle} \right)} \\ & {= \frac{1}{\sqrt{2^{m}}}\left( {\left| {00\cdots 0} \right\rangle + \left| {00\cdots 1} \right\rangle} \right) + \cdots + \left( \left| {11\cdots 1} \right\rangle \right)} \\ & {= \frac{1}{\sqrt{2^{m}}}{\sum\limits_{x = 0}^{2^{m} - 1}\left| x \right\rangle}} \\ \end{array},$$ where $H^{\otimes m}$ is the *m* times inner product of Hadamard, and $\left| 0 \right\rangle^{\otimes m}$ is the *m* times inner product of $\left| 0 \right\rangle$. In, the sum is over all possible values of *x*, that is, the Hadamard transform produces an equal superposition of all computational basis states. Moreover, the Hadamard transform efficiently performs this process and thus produces a superposition of 2^*m* states by using *m* gates only. ## Quantum parallel particle swarm optimization algorithm In this section, we introduce the QPPSO optimization algorithm as follows: ### Quantum encoding The Q-bit is the smallest unit of information, which can be expressed as follows: $${\left| \varphi \right\rangle = \cos\theta\left| 0 \right\rangle + \sin\theta\left| 1 \right\rangle}.$$ Therefore, the Q-bit can be coded as $\begin{bmatrix} {\cos\theta} \\ {\sin\theta} \\ \end{bmatrix}$, where cos*θ* or sin*θ* just represents a probability amplitude. However, in QPPSO, cos*θ* or sin*θ* is no longer the probability amplitude but is a certain value. The quantum encoding of QPPSO is shown in. In, an arbitrary *x*_j can be expressed as a string of *m* Q-bits shown as follows: $$\left| x_{j} \right\rangle = \left\lbrack \begin{array}{lllll} {\cos\theta_{1j}} & \cdots & {\cos\theta_{ij}} & \cdots & {\cos\theta_{mj}} \\ {\sin\theta_{1j}} & \cdots & {\sin\theta_{ij}} & \cdots & {\sin\theta_{mj}} \\ \end{array} \right\rbrack$$ where *θ_ij* = 2*π* × *rand*; *i* = 1, 2, …, *m*; *j* = 1, 2, …, *n*. *rand* is a random number between zero and one. Therefore, the tensor product of $\left| x_{i} \right\rangle$ with itself can be expressed as follows: $$\begin{array}{l} {\left| \text{A}_{j} \right\rangle = \left| x_{1j} \right\rangle \otimes \left| x_{2j} \right\rangle\cdots \otimes \left| x_{mj} \right\rangle} \\ {= \left\lbrack \begin{array}{l} {\cos\theta_{1j}} \\ {\sin\theta_{1j}} \\ \end{array} \right\rbrack \otimes \left\lbrack \begin{array}{l} {\cos\theta_{2j}} \\ {\sin\theta_{2j}} \\ \end{array} \right\rbrack\cdots \otimes \left\lbrack \begin{array}{l} {\cos\theta_{mj}} \\ {\sin\theta_{mj}} \\ \end{array} \right\rbrack} \\ {= \left\lbrack \begin{array}{l} {\cos\theta_{1j} \times \cos\theta_{2j} \times \cdots \times \cos\theta_{mj}} \\ {\cos\theta_{1j} \times \cos\theta_{2j} \times \cdots \times \sin\theta_{mj}} \\ \cdots \\ {\sin\theta_{1j} \times \sin\theta_{2j} \times \cdots \times \sin\theta_{mj}} \\ \end{array} \right\rbrack\text{=}\left\lbrack \begin{array}{l} A_{j1} \\ A_{j2} \\ \cdots \\ A_{j2^{m}} \\ \end{array} \right\rbrack} \\ \end{array}$$ For an *n*-dimensional space, each basis in the space has a value range. For simplicity, all the bases are set a value among (*a_j*, *b_j*). The division of the solution space is shown in. In, *x_jh* can be expressed as follows: $${x_{jh} = \frac{A_{{}^{jh}} + 1}{2}\frac{b_{j} - a_{j}}{2^{m}}h},$$ where *h* = 1, 2, …, 2^*m*. Based on subpopulation parallel computing, the algorithm running rate is exponentially accelerated. Moreover, individuals belong to different subspaces, so premature phenomena can be efficiently prevented. ### Quantum state update In quantum space, the equations in traditional PSO are inapplicable because of the uncertainty relation between the coordinate and the momentum. To restrain the behavior of particles in quantum space, a formula that consists of the current rotation angle, the individual optimal quantum rotation angle, and the global optimal quantum rotation angle is proposed and expressed as follows: $${\theta_{ijk}\left( {t + 1} \right) = \frac{c_{1}r_{1}\theta_{P_{ijk}}(t) + c_{2}r_{2}\theta_{G_{i}{}_{j}}(t)}{\left( {c_{1}r_{1} + c_{2}r_{2}} \right)} \pm w \cdot \ln\left\lbrack {{1/u_{ijk}}(t)} \right\rbrack\left| {\frac{1}{L}{\sum\limits_{k = 1}^{L}{\theta_{P_{ijk}}(t) - \theta_{ijk}(t)}}} \right|},$$ where *k* = 1, 2, …, *L, L* is the population size of QPPSO, *c*₁ and *c*₂ are constants, *w* is the inertia weight, *r*₁, *r*₂, and *u_ijk*(*t*) are the random numbers between zero and one, and *t* is the current iteration number. ### Structure of QPPSO The procedure of the QPPSO algorithm is summarized in the following section, and the flowchart of QPPSO is shown in. 1. Step 1: Initialize the parameter sets of the algorithm. The population is initialized by quantum encoding in, and the solution space is decomposed by. 2. Step 2: The objective values of all particles are evaluated. We set the *p_best* of each particle and its objective value equal to the current position of the particle and to its objective value, and then set *g_best* and its objective value equal to the position of the particle and to the objective value of the best initial particle. 3. Step 3: The position of every particle is updated by. 4. Step 4: The objective values of all particles are evaluated. 5. Step 5: The current objective value of each particle is compared with the objective value of its *p_best*. If the current value is better, then *p_best* and its objective value are updated with the current position. 6. Step 6: The best particle of the current population with the best objective value is determined. If the objective value is better than that of *g_best*, then *g_best* and its objective value are updated with the position and objective value of the current best particle. 7. Step 7: If the stopping criteria are met, we output *g_best* and its objective value; otherwise, we go back to Step 3. # Simulation and Comparisons ## Typical fractional-order chaotic systems In this section, numerical simulation and comparisons are conducted on the basis of some typical fractional-order chaotic systems, including the fractional-order Chen system, fractional-order Lorenz system, fractional-order Rössler system, and fractional-order Lü system. 11. The fractional-order Chen chaotic system can be expressed as follows: $$\left\{ \begin{array}{l} {\frac{d^{q_{1}}x_{1}}{\text{d}t^{q_{1}}} = a \cdot \left( {x_{2} - x_{1}} \right)} \\ {\frac{d^{q_{2}}x_{2}}{\text{d}t^{q_{2}}} = \left( {c - a} \right) \cdot x_{1} - x_{1} \cdot x_{3} + c \cdot x_{2}} \\ {\frac{d^{q_{3}}x_{3}}{\text{d}t^{q_{3}}} = x_{1} \cdot x_{2} - b \cdot x_{3}} \\ \end{array} \right.,$$ where *a, b, c, q*₁, *q*₂, and *q*₃ are the unknown constant parameters of the fractional-order chaotic systems that should be estimated. When *a* = 35, *b* = 3, *c* = 28, *q*₁ = 0.93, *q*₂ = 0.9, and *q*₃ = 0.88, this system exhibits a chaotic dynamical behavior. 12. The fractional-order Lorenz chaotic system is expressed as follows: $$\left\{ \begin{array}{l} {\frac{d^{q_{1}}x_{1}}{\text{d}t^{q_{1}}} = a \cdot \left( {x_{2} - x_{1}} \right)} \\ {\frac{d^{q_{2}}x_{2}}{\text{d}t^{q_{2}}} = \left( {c - a} \right) \cdot x_{1} - x_{1} \cdot x_{3} + c \cdot x_{2}} \\ {\frac{d^{q_{3}}x_{3}}{\text{d}t^{q_{3}}} = x_{1} \cdot x_{2} - b \cdot x_{3}} \\ \end{array} \right..$$ The system is in a chaotic state when *a* = 10, *b* = 28, *c* = 8/3, *q*₁ = 0.993, *q*₂ = 0.993, and *q*₃ = 0.993. 13. The fractional-order Rössler chaotic system can be expressed as follows: $$\left\{ \begin{array}{l} {\frac{d^{q_{1}}x_{1}}{\text{d}t^{q_{1}}} = - \left( {x_{2} + x_{3}} \right)} \\ {\frac{d^{q_{2}}x_{2}}{\text{d}t^{q_{2}}} = x_{1} + a \cdot x_{2}} \\ {\frac{d^{q_{3}}x_{3}}{\text{d}t^{q_{3}}} = b + x_{3}\left( {x_{1} - c} \right)} \\ \end{array} \right..$$ The system is in a chaotic state when *a* = 0.5, *b* = 0.2, *c* = 10, *q*₁ = 0.9, *q*₂ = 0.85, and *q*₃ = 0.95. 14. The fractional-order Lü chaotic system can be expressed as follows: $$\left\{ \begin{array}{l} {\frac{d^{q_{1}}x_{1}}{\text{d}t^{q_{1}}} = a\left( {x_{2} - x_{1}} \right)} \\ {\frac{d^{q_{2}}x_{2}}{\text{d}t^{q_{2}}} = - x_{1} \cdot x_{3} + c \cdot x_{2}} \\ {\frac{d^{q_{3}}x_{3}}{\text{d}t^{q_{3}}} = x_{1} \cdot x_{2} - b \cdot x_{3}} \\ \end{array} \right..$$ The system is in a chaotic state when *a* = 36, *b* = 3, *c* = 20, *q*₁ = 0.985, *q*₂ = 0.99, and *q*₃ = 0.98. For the previously discussed systems, we solve all parameters by using the numerical algorithm derived from the G-L definition of fractional derivatives to obtain the state variables *x, y, z*. Numerical results show that these systems are chaotic, and their chaotic behavior is shown in. ## Simulations on the systems In our simulation, all previously discussed systems freely evolve from random initial states. After a period of the transient process, a state vector is selected as the initial state *X*₀ for parameter estimation. The sampling time is *h* = 0.01, and the total number of states to calculate *J* is set as 100. We compare our QPPSO with GA and PSO. For a fair comparison, the maximum generation number and the searching range of the parameters are the same in all algorithms. That is, the maximum generation number is set as 1,000, and the population size is set as 80. The searching spaces of the parameters are shown in. The other settings for the parameters of the algorithms are as follows. For the GA method, the crossover rate is set as 0.9, and the mutation probability is set as *Pm* = 0.1. For the PSO method, according to Clerc’s stagnation analysis, the inertia weight is set as *ω* = 1/(2 × log(2)), and two acceleration coefficients are set as *c*₁ = *c*₂ = 0.5 + log(2). The probability threshold for random topology is set as *Pr* = 1 − (1 − 1/*s*)³, where *s* is the population size, and *Pr* is used to determine the proportion of local informants to the entire population. For QPPSO, we also set *c*₁ = *c*₂ = 0.5 + log(2) and the inertia weight *ω* = 1/(2 × log(2)). shows the mean of the objective function values, the standard deviation, and the best objective value of 50 independent runs. shows the estimation values for the parameters of the chaotic systems. The results shown in the tables indicate that QPPSO has better performance than the GA and PSO methods in the parameter estimation of fractional-order chaotic systems. This conclusion can also be obtained from the convergence curves of the objective function for the different methods displayed in. In this figure, the logarithmic scale is used for the y-axis for convenience in plotting the data. For the fractional-order Chen and Lü chaotic systems, the converging speed for QPPSO is slower than those for the GA and PSO methods in the initial period of evolution. However, QPPSO converges more quickly than the GA and PSO methods during the later period of evolution. The estimation values for the parameters in QPPSO are more accurate than those for the parameters in the GA and PSO methods. For the fractional-order Lorenz and Rössler chaotic systems, the converging speed for QPPSO is faster than those for the GA and PSO methods during all periods of evolution. also shows that the estimation values for the parameters in QPPSO are more accurate than those for the parameters in the GA and PSO methods. shows one typical run of the tuning trajectories of the parameters of fractional-order chaotic systems with respect to the number of generations by the QPPSO method. Less than 500 iterations are needed for the parameters to reach the steady state and converge to the actual parameters. This result shows the effectiveness and feasibility of using QPPSO to estimate the parameters of fractional-order chaotic systems. From the above results, we can conclude that the calculation method of the introduced quantum computing parallel characteristic can improve the ergodicity of the algorithm and then increase the ability of global convergence of the algorithm. Moreover, the quantum states updated by can make the convergence speed of QPPSO faster than that of PSO because the speed limited equation is no longer suitable in quantum space. The above results also show the effectiveness and efficiency of QPPSO. # Conclusion This study proposed a new QPPSO algorithm that was applied to solve the unknown parameter estimation of the fractional-order chaotic system. The parallel characteristic of quantum computing is used in QPPSO. This characteristic causes the calculation of each generation to exponentially increase. The behavior of particles in quantum space is restrained by the quantum evolution equation, which consists of the current rotation angle, individual optimal quantum rotation angle, and global optimal quantum rotation angle. Numerical simulation based on several typical fractional-order systems and comparisons with some typical existing algorithms demonstrate the effectiveness and efficiency of the proposed algorithm. Future work should develop QPPSO-based approaches more effective and adaptive than current ones and apply the algorithm to other systems. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: YH. Performed the experiments: FG. Analyzed the data: Y. Li. Wrote the paper: Y. Liu.

# Introduction Ulcerative colitis (UC), a primary type of inflammatory bowel disease (IBD), is a chronic bowel disorder characterized by recurrent uncontrolled inflammation of the intestinal mucosa. UC was perceived as a Western disease in the past, based on the high incidence and prevalence reported in industrialized countries (especially northern Europe and North America). However, the incidence and prevalence of UC has continued to increase in various regions around the world in recent decades; numerous developing countries traditionally considered as areas of low incidence are undergoing a dramatic rise in disease incidence. Not surprisingly, UC is currently emerging as a global disease. Increasing prevalence may partly account for the heavy health burden UC brings about; another crucial contributor could be patient requirements for lifetime care due to a lack of a guaranteed curative therapeutic regimen. Conventional medical treatment of UC involves the administration of high-dose steroids and immunomodulators. Recently, significant advances have provided substantial insight into the etiology and pathogenesis of UC. The role of intestinal microbes in particular was shown to be of greater significance than previously considered. Thus, novel therapies geared towards modulating the intestinal flora has become a research focus, a representative of which was the extensive use of probiotics in IBD, including UC. The beneficial effects of probiotics in the induction of remission and the maintenance of UC have been examined in numerous animal experiments as well as in clinical trials. As one of the most extensively commercially exploited probiotic bacteria, the effectiveness and safety of several *Lactobacillus* strains for supplementation have been evaluated in several studies. However, to date, the understanding of these strains is still far from being sufficient for appropriate application in UC patients. The possible therapeutic efficacy of certain *Lactobacillus* strains in UC has yet to be assessed or identified further. Thus, to identify more putative *Lactobacillus* species that may be of therapeutic value for UC, we undertook a study to investigate the changes of different *Lactobacillus* species in UC patients. Because species-specific primers can only be developed if there are known gene sequences containing highly variable regions of the universal *Lactobacillus* genes, we merely analyzed the changes in 11 common *Lactobacillus* species in UC patients compared with healthy controls using published *Lactobacillus* species-specific gene sequences. Based on this analysis, three *Lactobacillus* strains from three species were selected to assess the therapeutic effects on DSS-induced experimental colitis in BALB/c mice. # Materials and Methods ## Participants Eligible patients were recruited from the Gastroenterology Department of The Second Xiangya Hospital between April 2008 and May 2012. Forty-five active UC patients (26 males and 19 females; average age, 34 years) based on widely accepted diagnostic criteria, and 45 population-based healthy controls (30 males and 15 females; average age, 28 years) were enrolled in the study. Participants were excluded if they had concurrent bacillary dysentery, ischemic enterocolitis, enterophthisis, or the use of antibiotics within 4 weeks prior to the sampling. All human fecal samples were collected with written informed consent from patients prior to participation in the study. The protocols for the collection and analysis of the samples were approved by the Ethical Committee of the Second Xiangya Hospital of Central South University, in accordance with the current revision of the Helsinki Declaration. ## Collection and storage of fecal samples Fresh fecal samples were collected from 45 patients and 45 healthy controls. All samples were placed in sterile, anaerobic ice boxes and immediately stored at –80°C until processing within 2 h. ## DNA extraction from fecal samples DNA extraction was performed following the manufacturer’s instructions (QIAamp DNA Stool Mini Kit; Qiagen, Hilden, Germany). Ninety samples were processed by a single individual. The concentration of the DNA extracts was assessed by measuring the optical density at wavelengths of 230 nm, 260 nm, and 280 nm after dilution with deionized water. The purity of total DNA extracted was checked by electrophoresis on a 1% agarose gel. ## PCR screening of *Lactobacillus* species in which the detection rate significantly differed between UC patients and healthy controls The primers used in the investigation are listed in. BLAST searches were performed to determine the specificity of the primers. For DNA extracts from each fecal sample, PCR amplification reactions (in a total volume of 20 μL) were performed on a PCR thermal cycler (PC-808; Astec, Fukuoka, Japan). Each reaction was performed in a 20 μL reaction volume containing 2 μL of template DNA, 2 μL of 25 mM MgCl₂, 0.5 μL of 10 mM dNTPs, 5 μL of 10×PCR buffer, 10 pmol of each primer, and 1.0 U of Taq DNA polymerase (Promega, Madison, WI, USA). PCR was performed using the following parameters: an initial denaturation at 95°C for 3 min, followed by 36 cycles of 94°C for 30 s; 56°C for 40 s; 72°C for 40 s; and a final elongation period at 72°C for 5 min. The amplifications were confirmed by standard electrophoresis of the PCR products (5 μL) using 2% agarose gels (Sigma–Aldrich, St. Louis, MO, USA) prepared in 0.5× TBE buffer (45 mM Tris base, 45 mM boric acid, and 1 mM EDTA \[pH 8.0\]) and visualized by silver staining. A *Lactobacillus* species was considered to be detectable in the fecal samples of participants when primed PCR products were obtained. Thus, we screened out the *Lactobacillus* species that had significantly different detection rates between the patients and controls. ## Quantitative real-time PCR (qPCR) analysis of the *Lactobacillus* species in UC patients and healthy controls The qPCR reaction was performed using an Applied Biosystems 7300 Real-Time system (Applied Biosystems, Foster City, CA, USA). Quantification assays and data analyses were performed using SDS version 1.2.3 software (Applied Biosystems). A 50 μL PCR reaction contained 5.0 μL of template DNA (50 ng), a 2.5 μL mixture of screened primers (10 pmol/L for each), 25.0 μL of 2× GoTaq qPCR Master Mix, 0.5 μL of 100× CXR reference dye, and sterile ddH₂O. The qPCR reaction conditions were as follows: DNA polymerase activation at 95°C for 5 min, followed by 40 cycles of DNA melting at 95°C for 15 s and annealing at 56°C for 60 s. The specificity of the qPCR reaction was confirmed by melting- curve analysis. The data were analyzed using the 2^-ΔΔCt method. As reported in several studies, we used total bacteria as the endogenous control to normalize the data. The threshold cycle (Ct) indicates the fractional number at which the amount of amplified target reaches a fixed threshold. ΔCt was calculated as the difference between the Ct value of the primers specific for *Lactobacillus* species and the Ct value of the primers for total bacteria. ΔΔCt is defined as the difference between the ΔCt value of UC patients and the ΔCt value of the control group. The fold change of expression of each Lactobacillus species in UC patients relative to that of the healthy controls was calculated using log10 RQ where RQ is 2^-ΔΔCt. Log₁₀RQ correlates directly to up- regulation (positive value) and down-regulation (negative value). A value of 1 in log₁₀RQ (relative quantification) represents a 10-fold increase in expression. Similarly, a value of -1 represents 10-fold decrease in expression compared to control. ## Animal trial protocol Sixty female (8-week-old) BALB/c mice weighing 20.0 ± 2.0 g were purchased from Hunan SJA Laboratory Animal Co. Ltd., Changsha, China. Each group consisted of ten mice. All mice were housed at a room temperature of 20–22°C, 50 ± 10% humidity, and a 12 h diurnal light/dark cycle. Throughout the trial period, the mice were fed standard rat chow and water *ad libitum*. All animal experimental procedures were reviewed and approved by the Institutional Animal Care and Use Committee of Central South University. Three accessible standard *Lactobacillus* strains that were previously isolated and identified by our research institutions (preserved in the China Centre for Type Culture Collection) were selected as candidate *Lactobacillus* strains for further experimental effect verification. The candidate *Lactobacillus* strains were *L*. *fermentum* CCTCC M206110, *L*. *plantarum* NCIMB8826, and *L*. *crispatus* CCTCC M206119. The animals were randomly divided into three experimental groups (DSS-treated mice given 0.4 mL of 3.0×10⁸ colony forming units/mL of *L*. *fermentum* CCTCC M206110, *L*. *crispatus* CCTCC M206119, or *L*. *plantarum* NCIMB8826), three control groups (a healthy control group, a negative \[normal saline\] control group, and a positive \[mesalazine; 6 mg/20 g\] control group). Candidate strains of three *Lactobacillus* species, saline, or mesalazine were administered once daily via gavage from days 1–9. Between days 3 and 9, experimental colitis was induced via the consumption of 5% dextran sulfate sodium (DSS; Mw 40 kDa, MP Biomedicals Inc., Solon, OH, USA) in the drinking water. Food, water/DSS consumption, body weight, fecal and urine output, and survival condition were monitored daily. The disease activity index (DAI) was determined by scoring the change in body weight, occult blood, and gross bleeding, as described previously. On day 9, all mice were sacrificed by cervical dislocation. Then, the colons were removed and the colon length was measured. The distal colon was then collected and fixed in 10% buffered formalin for hematoxylin and eosin staining. Other parts of the colon were preserved in liquid nitrogen. ## Histologic analysis Nine randomly selected fields (magnification 200×) were inspected in each section by two pathologists blinded to the treatment protocol. Grading of intestinal inflammation was determined as follows. For each category of the score (inflammation, depth of lesions, and destruction of crypt), the points were multiplied by a factor of involvement of the visible epithelium. The sum of the two category scores (inflammation and depth of lesions) were added to the inflammatory damage score (range, 0–24). The score of the destruction of the crypt category represents the crypt damage score (range, 0–16). ## Statistical analysis All data are presented as the mean ± standard deviation (SD). Statistical package for social sciences (SPSS) software (version 17.0; SPSS Inc., Chicago, IL, USA) was used for data management and statistical analysis. The detection rate of the different *Lactobacillus* species in UC patients and healthy controls was compared using a chi-square test. Weight, colon length, DAI scores, and histologic scores of all of the mouse groups were analyzed using an ANOVA test. Post hoc t-tests were conducted when the ANOVA values were significant. A *P*-value \< 0.05 was considered to be statistically significant. # Results ## The detection rate of 11 *Lactobacillus* species in UC patients and healthy controls Using DNA extracts from fecal samples of 45 patients and 45 healthy controls, PCR amplification reactions were performed. The detection rates of the 11 *Lactobacillus* species in 45 patients and 45 healthy controls are listed in. Eight of the 11 *Lactobacillus* species were detectable in \>50% of the healthy controls, and 5 of the 11 *Lactobacillus* species were detectable in \>90% of the healthy controls. Two of the 11 *Lactobacillus* species (*L*. *johnsonii* and *L*. *casei*) were detectable in all of the healthy controls. Compared with the healthy controls, the detection rate of 4 of the 11 *Lactobacillus* species decreased significantly in the UC patients (*P* \< 0.05). The four *Lactobacillus* species were *L*. *crispatus* (86.7% vs. 53.3%, *P* = 0.004), *L*. *fermentum* (95.6% vs. 35.6%, *P* = 0.001), *L*. *gasseri* (100.0% vs. 37.8%, *P* = 0.002), and *L*. *salivarius* (75.6% vs. 26.7%, *P* = 0.041). The detection rate of 2 of the 11 *Lactobacillus* species increased significantly in UC patients (*P* \< 0.05). The two *Lactobacillus* species were *L*. *delbrueckii* (40.0% vs. 88.9%, *P* = 0.005) and *L*. *paracasei* (20.0% vs. 64.4%, *P* = 0.028). ## Quantitative analysis of the *Lactobacillus* species in UC patients and healthy controls The results of quantitative analysis of the 11 *Lactobacillus* species in UC patients and healthy controls are shown in. Compared with the healthy controls, the number of *L*. *gallinarum*, *L*. *fermentum*, *L*. *gasseri*, *L*. *salivarius*, *L*. *acidophilus*, *L*. *casei*, *L*. *paracasei*, and *L*. *plantarum* decreased significantly in UC patients (*P* \< 0.05), while the number of *L*. *crispatus*, *L*. *delbrueckii*, and *L*. *reuteri* increased significantly (*P* \< 0.05). ## Effect of different *Lactobacillus* strains on weight loss in DSS-induced colitis in BALB/c mice Four days after inducing colitis, deaths began to occur due to gastrointestinal bleeding. The survival statuses of the animals at the end of the experiments are shown in. All groups exhibited significant weight loss during DSS treatment compared to the healthy control group, which was given water and a standard diet (*P \<* 0.05), especially the negative control group. The positive control group had significantly less weight loss compared with the negative control group (-0.71 ± 0.28 g vs. -1.84 ± 0.72 g, *P* \< 0.05). The weight loss in the *L*. *fermentum* CCTCC M206110 group (-0.82 ± 0.39 g vs. -1.84 ± 0.72 g, *P* \< 0.05) and the *L*. *plantarum* NCIMB8826 group (-1.19 ± 0.55 g vs. -1.84 ± 0.72 g, *P* \< 0.05) was also significantly less than the negative control group; however, the weight loss was significantly higher in the *L*. *crispatus* CCTCC M206119 group than in the negative control group (-2.26 ± 0.51 g vs. -1.84 ± 0.72 g, *P* \< 0.05;). ## Effect of different *Lactobacillus* strains on colon length in DSS-induced colitis in BALB/c mice All groups exhibited significant colon length reduction during DSS treatment compared to the healthy control group, which was given water and a standard diet (*P \<* 0.05). The mesalazine group had significantly improved colon length shortening compared with the negative control group (9.18 ± 1.39 cm vs. 6.71 ± 1.47 cm, *P* \< 0.05). In addition, the *L*. *fermentum* (9.22 ± 1.69 cm vs. 6.71 ± 1.47 cm, *P* \< 0.05) and *L*. *plantarum* groups (7.95 ± 1.19 cm vs. 6.71 ± 1.47 cm, *P* \< 0.05) both had significantly reduced shortening of the colon length compared with the negative control group. Furthermore, the *L*. *fermentum* group had more improvement in colon length shortening than the *L*. *plantarum* group (9.22 ± 1.69 cm vs. 7.95 ± 1.19 cm, *P* \< 0.05); however, the results indicated that *L*. *crispatus* exacerbated the colon length shortening in comparison with the negative control group (5.25 ± 1.19 cm vs. 6.71 ± 1.47 cm, *P* \< 0.05;). ## Effect of different *Lactobacillus* strains on DAI in DSS-induced colitis in BALB/c mice All groups exhibited a significantly higher DAI score during DSS treatment compared to the healthy control group, which was given water and a standard diet (*P \<* 0.05;). Compared with the negative control group, mesalazine treatment resulted in a significant decrease in the DAI score (2.77 ± 0.63 vs. 3.57 ± 0.32, *P* \< 0.05). Moreover, the DAI scores of animals in the *L*. *fermentum* group were also significantly less than the negative control group (2.90 ± 0.22 vs. 3.57 ± 0.32, *P* \< 0.05); however, the results indicated no significant difference in the DAI in the *L*. *plantarum* group or *L*. *crispatus* group compared with the negative control group (*P* \> 0.05;). ## Effect of different *Lactobacillus* strains on histologic changes in DSS-induced colitis in BALB/c mice The colon sections of the healthy control group showed intact mucosae with glands secreting abundant mucin. The colon sections of DSS-treated animals showed loss of the epithelial layer, goblet cell depletion, neutrophil infiltration, and distortion/destruction of the crypt architecture compared with the healthy controls; however, when the mice were co-treated with *L*. *fermentum*, the level of inflammatory infiltration was significantly lower compared with the negative control (9.40 ± 3.52 vs. 16.1 ± 4.48, *P* \< 0.05), as was the extent of crypt structure destruction (6.80 ± 2.49 vs. 10.70 ± 3.16, *P* \< 0.05). The results indicated no significant difference in the severity of inflammatory infiltration or crypt destruction between the *L*. *plantarum* group and the negative group (*P* \> 0.05). Notably, more severe inflammatory infiltration was observed in the *L*. *crispatus* group compared with the negative controls (20.50 ± 3.37 vs. 16.1 ± 4.48, *P* \< 0.05), with no significant difference in the severity of crypt destruction. # Discussion Therapies aimed at modifying and manipulating the gut flora have been implicated in the pathogenesis of UC through the use of probiotics and have received increased attention. The application of *Lactobacillus* to UC is based on the evidence indicating the reduction of the fecal *Lactobacillus* count in UC patients. In the current study, we demonstrated that the change in *Lactobacillus* in UC patients was more complicated at the species level. Not all *Lactobacillus* species were decreased in UC patients. The results of quantitative analysis of the 11 *Lactobacillus* species indicated that some *Lactobacillus* species increased significantly in UC patients compared with healthy controls, such as *L*. *crispatus*. Furthermore, the therapeutic effects of *Lactobacillus* strains were correlated with changes in the concentrations in the distal colon. *Lactobacillus* strains of different species could exert completely opposite effects in UC. Based on our study, *L*. *crispatus* numbers increased significantly in UC patients compared with healthy controls. The animal experiments further revealed that *L*. *crispatus* CCTCC M206119 aggravated weight loss and colonic histologic damage in the mouse colitis model. Previous studies on the effects of *L*. *crispatus* in induced colitis have reported conflicting results. Castagliuolo et al. have demonstrated that *L*. *crispatus* M247 reduced the severity of DSS colitis in a dose-dependent fashion, while Ulinski and Aoun revealed that the *L*. *crispatus* CCTCC M206119 strain is involved in the exacerbation of intestinal inflammation in DSS-colitis mice. Thus, we should be more cautious in administering *Lactobacillus* to UC patients in clinical practice. The results obtained in the present study reveal that *L*. *fermentum* CCTCC M206110 appeared to be effective at attenuating DSS-induced colitis in BALB/c mice, given that the important clinical parameters, such as mortality, weight loss, DAI scores, and colonic histologic damage were improved in mice receiving *L*. *fermentum* compared with the controls. A previous study demonstrated that mice with colitis treated with *L*. *fermentum* had an improved survival rate, DAI score, and colonic mucosa histologic scoring. The study further indicated that *L*. *fermentum* attenuated trinitrobenzenesulfonic acid-induced colitis, which was associated with an increase in intestinal superoxide dismutase activity and a reduction in oxidative stress, nuclear factor κB (NF-κB) activity, and cytokine production. Our study tested the effect of *L*. *fermentum* CCTCC M206110 in DSS-induced colitis because of a previously noted reduction in *L*. *fermentum* species in UC patient feces. The present study indicates that *L*. *fermentum* exhibited beneficial anti-oxidative and anti- inflammatory properties in the mouse colitis model. Thus, a *L*. *fermentum* probiotic could be recommended as potential adjuvant therapy in combination with olsalazine to achieve a more efficacious treatment for UC. Nevertheless, because the associated clinical data are rather limited and far from convincing, the concomitant use of *L*. *fermentum* warrants well-designed, large, randomized, placebo-controlled trials to investigate the unresolved issues related to efficacy, dose, duration of use, and single or multistrain formulation. In the current study, we failed to observe an obvious attenuation effect of *L*. *plantarum* NCIMB8826 in DSS-induced colitis in BALB/c mice. The results showed that *L*. *plantarum* NCIMB8826 improved weight loss and colon length, but with no significant influence on the DAI or colonic histologic damage in the mouse colitis model. During the last decade, the therapeutic effect of *L*. *plantarum* strains on ameliorating experimental colitis in mouse models has been reported. Additionally, the anti-inflammatory and immunomodulatory activities of *L*. *plantarum* were evaluated, including reducing the production of pro-inflammatory cytokines (tumor necrosis factor (TNF)-α, interleukin (IL)-1, and IL-6) and inhibiting the Toll-like receptor (TLR)4-linked NF-κB and mitogen-activated protein kinase (MAPK) signaling pathways. Furthermore, *L*. *plantarum* exhibited antioxidant properties by significantly decreasing lipid peroxidation (TBARS) and nitric oxide production and increasing the glutathione concentration. It is arbitrary and inexact to draw the conclusion that *L*. *plantarum* is ineffective in DSS-induced colitis based on the present study, given the strong possibility of a false-negative result due to an insufficient sample size. Thus, the potential probiotic effect of *L*. *plantarum* on UC must be further assessed. Hence, it is of great importance to establish the mechanism(s) underlying the beneficial interaction between the colonic epithelium, intestinal immunity, and the microbiota for the extensive but evidence-based application of probiotics in UC. Based on the previous studies, the gut microbiota plays a role in shaping the mucosal immune system. *Clostridia* and *Bacteroides* have been proven to attenuate intestinal inflammation in mice through the induction of CD4⁺FOXP3⁺ regulatory T lymphocytes. Other members of the microbiota can attenuate mucosal inflammation by antagonizing the activation of NF-κB or modifying the cytokine status. In conclusion, the change in *Lactobacillus* in UC patients was more complicated at the species level. Not all *Lactobacillus* species are beneficial for UC patients. Administration of *L*. *crispatus* CCTCC M206119 supplement aggravated DSS-induced colitis, while *L*. *fermentum* CCTCC M206110 proved to be effective at attenuating DSS-induced colitis. *L*. *plantarum* NCIMB8826 showed no obvious attenuation effect in DSS-induced colitis. Thus, the potential probiotic effect of *L*. *plantarum* species on UC has yet to be assessed. Future studies should aim to determine the mechanisms underlying the interactions between *Lactobacillus*, the colonic epithelium and intestinal immunity, which may be essential for the extensive but evidence-based application of probiotics in UC. # Supporting Information [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: YC LG FL. Performed the experiments: YC DL XL. Analyzed the data: YC LG. Contributed reagents/materials/analysis tools: LG CO DL XL. Wrote the paper: LG YC HW CO. [^3]: Current address: Department of Biological Sciences, Missouri University of Science and Technology, Rolla, MO, United States of America

# Introduction The cornea is a transparent tissue on the surface of the eye with refractive properties for bending light rays. The development of the vertebrate cornea involves inductive interactions between surface ectodermal and mesenchymal tissues. At embryonic day 8.5 to 9.0 (E8.5–9.0), a thickened region of the head ectoderm, defined as the lens placode, gives rise to both the lens and the presumptive corneal epithelium. The primitive corneal epithelium forms after the lens vesicle detaches from the overlying surface ectoderm. At around E12.0–12.5, the perioptic mesenchyme (mostly neural crest cells) migrates into the space between the lens and the primitive corneal epithelium. At E14.5–15.5 in the mouse eye, the posterior mesenchymal cells closest to the lens differentiate into a thin layer of corneal endothelium, and the anterior chamber subsequently forms between the lens and cornea. The mesenchymal cells between the corneal epithelium and endothelium begin to differentiate into keratocytes and form corneal stroma. The corneal epithelium continues to differentiate after birth and, upon eyelid opening at two weeks of age, the corneal epithelium expands from two cell layers to a self-renewing, stratified epithelium comprising eight to 10 cell layers. The fully developed cornea is composed of three layers derived from two embryonic germ tissues: a stratified corneal epithelium with surface ectoderm origin on the outer surface, expressing the keratin 3 and 12 (K3/K12) pair; the stromal layer underneath, sparsely populated by keratocytes composed of highly aligned collagen, and the inner surface of the cornea, covered by a single-layer endothelium. Corneal injury and disease can lead to opacification, neovascularization, fibrosis and defective wound healing. These pathological conditions together constitute the second leading cause of blindness worldwide. Understanding the inductive factors and signals that regulate corneal cell proliferation and differentiation has important implications for the development of therapeutic approaches for controlling corneal repair and homeostasis and preventing blindness. Several lines of evidence support the integral role of fibroblast growth factors (FGFs) in corneal cell proliferation and differentiation. As many as 22 FGFs have been identified in vertebrates. FGF signaling is activated through binding of the growth factor to its cell surface receptors to stimulate receptor dimerization and activation of receptor tyrosine kinases, ultimately leading to activation of various downstream signal transduction cascades. Four fibroblast growth factor receptor (FGFR) genes (*FGFR1* to *FGFR4*) have been cloned and identified in mammals. Additionally, multiple FGFR isoforms, differing in structure and ligand affinity, can be generated through alternative splicing of primary transcripts. For example, two FGFR2 variants, FGFR2IIIb and FGFR2IIIc, are generated by alternative splicing at the second half of Ig domain III of the *FGFR2* locus. During corneal development, FGF-7 and FGF-10 are secreted by corneal mesenchymal cells and both can bind with affinity to FGF receptor 2 (FGFR2-IIIb) isoform, which is expressed mainly in limbal and central corneal epithelium. These expression patterns imply that FGFR2-signaling may promote limbal stem cell proliferation and participate in modulation of corneal epithelium renewal and homeostasis. In vitro functional studies have shown that FGF-7 enhances the growth and proliferation of cultured corneal epithelial cells but does not significantly affect motility. Topical application of FGF-7 was shown in vivo and in vitro to accelerate corneal epithelial wound healing. In an investigation of the role of FGFR activation in corneal development, transgenic mice overexpressing FGF-7 or FGF-10 in the developing lens (starting as early as E11.5) exhibited hyperproliferative corneal epithelial cells that subsequently were induced to alter their cell fate from corneal epithelium to lacrimal gland epithelium. In another study of transgenic mice, overexpression of FGF-3, another member in the FGF family also capable of activating FGFR2IIIb, was found to stimulate epithelial-to-glandular transformation in the developing cornea of the transgenic mice. However, when excess FGF-7 was induced in the corneal epithelium of young mice, the main phenotype was hyperplasia in the epithelial layer, without alteration in cell fate. The corneal epithelium increased in thickness from 6 or 7 cell layers to more than 20 cell layers, with extended K14 expression from the basal to suprabasal to superficial layers. Phenotypic variations caused by excessive FGF-7 were found in the eyes of embryos and young pups, which may be explained by the age-dependent differences of FGFR2-activated signaling network in developing corneal epithelium and the plasticity of progenitor cells. However, these gain-of-function studies have not defined the normal biological role of FGFR2 in corneal development. The function of FGFR2 in the development of ocular surface ectodermal tissues, including the lens and the lacrimal glands, has been investigated using the *Fgfr2* conditional knockout mice (referred as *Fgfr2*^*CKO*) driven by a surface ectodermal Cre line, the *Le-Cre*. These studies revealed that the FGFR2-activated Ras-ERK signaling pathway is essential for cell survival and cell cycle exit during ocular lens development and for induction of the lacrimal glands. Although FGFR2 is known to be expressed in the corneal epithelium, the developmental changes in the cornea of *Fgfr2* conditional knockout mice have not been investigated in detail. In this study, we demonstrate that FGFR2 is required for corneal epithelial cell proliferation at the stage shortly after the lens vesicle detaches from the surface ectoderm. In contrast to its role in the lens, FGFR2 is not essential for corneal epithelial cell survival. Furthermore, we demonstrate that FGFR2 plays an essential role in maintaining the Pax6 expression, independent of ERK-signaling, in corneal epithelium. In the absence of Pax6, differentiation and maturation of corneal epithelium is inhibited in *Fgfr2*^*CKO* mice. We also found that the abnormal development of corneal epithelium in *Fgfr2*^*CKO* mice does not affect the migration and proliferation of the corneal mesenchymal cells, but prevents these cells from differentiating into mature keratocytes, suggesting that loss of FGFR2 interferes with the signaling interactions between the corneal epithelium and underlying mesenchyme. # Materials and Methods ## Mice Mice carrying the *Fgfr2* flox alleles and the *Le-Cre* transgenic mice were obtained from Dr. Michael Robinson (Miami University, Oxford, OH, USA), with permission from Drs. David Ornitz and Ruth Ashery-Padan, respectively. The ERK1/2 double deletion mice (*LeCre- Mapk1*^*fl/f**l;Mapk3*^*-/-*) were described previously. To generate *Fgfr2* conditional knockout mice, *Le-Cre* mice were bred to *Fgfr2*^*flox/flox* mice and the heterozygous offspring *Le- Cre;Fgfr2*^*flox/+* mice were then crossed with *Fgfr2*^*flox/flox* to make *Le-Cre;Fgfr2*^*flox/flox* (referred *as Fgfr2*^*CKO*) mice. The *Le-Cre* mice in all experiments were heterozygous for the transgene. The *Fgfr2*^*loxP/loxP* mice are referred to as *wild type* (*WT*). Because *Le-Cre* hemizygous transgenic mice were shown to develop eye abnormalities on some genetic backgrounds, we recently crossed the *Fgfr2*^*CKO* mice with C57BL6J mice to examine potential defects in the cornea of *Le-Cre;Fgfr2*^*flox/+* mice. Animal use was in accordance with the Association of Research in Vision and Ophthalmology (ARVO) Statement for the Use of Animals in Ophthalmic and Vision Research, and all experimental procedures were approved by the Animal Care and Use Committee of the University of Missouri-Columbia. ## Histology, immunohistochemistry and immunofluorescence Embryonic and newborn mouse heads were fixed in 4% paraformaldehyde for 2 hours or overnight and processed for histological analysis by hematoxylin and eosin (H&E) staining, as described previously. For immunohistochemistry and immunofluorescence, the following primary antibodies were used: anti-Ki67 (M7249, Dako, Carpinteria, CA, USA); anti-keratin-14 (K14) (PBR-159P) and anti- Pax6 (PBR-278P), both from Covance Inc, Princeton, NJ, USA; anti-N-cadherin (33–3900, Zymed, Camarillo, CA, USA), and anti-phospho-histone H3 (sc-8656-R, Santa Cruz Biotechnology, Santa Cruz, CA, USA). Anti-keratocan and anti- keratin-12 (K12) antibodies were gifts from Dr. Chia-yang Liu at University of Cincinnati (Cincinnati, OH, USA). For immunofluorescence, Alexa Fluor conjugated secondary antibodies were purchased from Invitrogen (Carlsbad, CA, USA) and cell nuclei were counterstained with 4′,6-diamidino-2-phenylindole (DAPI). The signal-enhancement TSA kit (NEL741B001KT, PerkinElmer, Boston, MA, USA) was used for immunofluorescence against Ki67, K12 and N-cadherin. For immunohistochemistry, biotinylated secondary antibodies were from Vector Laboratories (Burlingame, CA, USA) and color was developed by using 3, 3′-diaminobenzidine as a substrate (D4293, Sigma, St Louis, MO, USA). Sections were counterstained for cell nuclei by hematoxylin. ## Brdu incorporation and TUNEL assays 5-Bromo-2′-deoxyuridine (BrdU) was administered intraperitoneally into pregnant mice at a concentration of 0.1μg/gm body weight and labeled for 1 hour prior to embryo isolation. BrdU immunohistochemistry was performed as previously described. Terminal deoxynucleotidyl transferse dUTP nick end labeling (TUNEL) assay was performed with in situ apoptosis detection kit (S7165, Millipore, Billerica, MA, USA), following the manufacturer’s instructions. ## Statistical analysis Quantification of cell proliferation was performed by determining the fraction of labeled nuclei over the total number of nuclei present in a given section. A minimum of 3 embryos for each genotype were analyzed at a given time point. Data are expressed as mean ± SEM and p-values were calculated using Mann-Whitney Test (p\<0.05 was considered significant). # Results ## Conditional deletion of *Fgfr2* in ocular surface ectoderm affects corneal development In eye development, the surface ectoderm gives rise to both corneal epithelium and the lens. The essential role of FGFR2 in lens development has been shown in the *Fgfr2* conditional deletion mice by a surface ectoderm driver *Le-Cre*. However, the function of FGFR2 in corneal epithelial development has not been assessed. We first examined corneal development in the eyes of *Fgfr2*^*CKO* mice by histology (H&E staining). We found that at E12.5 the presumptive corneal epithelium in *Fgfr2*^*CKO* eyes looked either similar to or slightly thinner than the corneal epithelium in *WT* control eyes. In both genotypes, the ocular mesenchymal cells had migrated into the space between the lens and the corneal epithelial layer. At E13.5, however, developmental abnormalities were noted in *Fgfr2*^*CKO* corneas. The corneal epithelial layer in *Fgfr2*^*CKO* eyes was significantly thinner than normal. The cell density in corneal epithelium was reduced and cells were absent in some areas (indicated by the arrowhead). The corneal defects in *Fgfr2*^*CKO* eyes progressed beyond the epithelial layer. The *WT* corneal epithelium consisted of basal cuboidal and superficial flattened cells, whereas the *Fgfr2*^*CKO* corneal epithelium consisted entirely of flattened cells. In E16.5 *Fgfr2*^*CKO* eyes, corneal stroma was thinner but cells were more densely packed, with intensified eosin staining, when compared to the age-matched *WT* corneal stroma. The anterior chamber was formed in both genotypes. As previously reported, eyelid fusion did not occur in *FGFR2*^*CKO* eyes. Additional defects also occurred in anterior segments development of *Fgfr2*^*CKO* eyes, including abnormal accumulation of extra cells and tissues in the anterior chamber and a defective corneal endothelial layer. These data suggest that in addition to normal lens development, FGFR2 function is also required for normal corneal development. ## FGFR2 is not required for corneal epithelial cell survival The lens has been shown to undergo the loss of FGFR2-induced apoptosis during development. To investigate whether the reduced cell density in corneal epithelium of FGFR2CKO mice is caused by apoptosis, TUNEL assay was performed on eyes at different ages ( through). We found a drastic increase in apoptotic cells in Fgfr2CKO lenses, a result consistent with the previous finding. The corneal epithelial layer in both WT and Fgfr2CKO eyes, however, did not demonstrate an increase in apoptotic cells, suggesting that FGFR2 is not required for corneal epithelial cell survival. TUNEL-positive cells were also detected in other ocular tissues, including the corneal mesenchymal cells and the hyaloid vascular cells in both WT and Fgfr2CKO eyes, but no significant difference was noted between the two genotypes. ## FGFR2 is essential for corneal epithelial cell proliferation during early-stage development We investigated whether the reduced cell number in *Fgfr2*^*CKO* corneal epithelium was affected by defects in cell proliferation. Immunofluorescence of Ki67, a cell proliferation marker expressed in active cycling cells, was performed on E13.5 eyes, the age when the corneal epithelial defect becomes apparent. We found that Ki67-positive cells were markedly reduced in both the central and peripheral regions of the corneal epithelial layer in *Fgfr2*^*CKO* eyes when compared to the littermate *WT* eyes ( through ). In contrast to the differences between *Fgfr2*^*CKO* and *WT* Ki67-positive cells in corneal epithelium, no noticeable difference was found between the two genotypes in the number of Ki67-positive cells in the lens epithelium and corneal stroma. To further analyze the effect of FGFR2 deficiency on cell cycle progression, BrdU incorporation assay and expression of phospho-histone H3 (p-H3) were performed on E13.5 eyes ( through). BrdU is a marker for cells in the S-phase, whereas p-H3 is a marker for G2-M phase cells. Consistent with the results of Ki67 expression, both BrdU and p-H3–labeled cells were significantly reduced in the corneal epithelial layer of E13.5 *Fgfr2*^*CKO* eyes when compared to *WT* eyes. To quantify the labeling index, the ocular surface ectodermal layer was divided into central and peripheral areas, as illustrated in. In the central area, *WT* BrdU and p-H3 indices were 0.45±0.06 and 0.15±0.03, respectively, whereas they were 0.24±0.08 and 0.05±0.03, respectively, in *Fgfr2*^*CKO* cornea (p\<0.05), a statistically significant decrease for both cell cycle markers when compared to control *WT* eyes. For the peripheral area, BrdU and p-H3 indices were obtained from the temporal side where lacrimal gland budding occurs at the fornix. Similar to the central area, indices were significantly decreased in *Fgfr2*^*CKO* eyes (0.28±0.15 for BrdU and 0.07±0.07 for p-H3 in *Fgfr2*^*CKO* eyes vs. 0.52±0.1 and 0.20±0.04, respectively, in *WT* eyes). BrdU index in the central area was calculated to determine whether *Fgfr2* deletion in surface ectodermal tissues affects cell proliferation in corneal mesenchyme migrated. There was no statistical difference between *WT* and *FGFR2*^*CKO* mice (0.46±0.12 vs. 0.40±0.06), suggesting that cell proliferation in corneal stromal cells is not affected by deletion of *Fgfr2* in corneal epithelium. Taken together, our data suggest that after lens vesicle detachment, FGFR2 is required for cell proliferation in the overlying surface ectoderm, which later differentiates into the corneal epithelium, lacrimal glands and conjunctival epithelium. ## FGFR2 is important for differentiation of corneal epithelial cells and for maintaining Pax6 expression in these cells At the stages between E16.5 to P0, the *WT* corneal epithelium consists of two cell layer, so as in the *Fgfr2*^*CKO* cornea. However, the cell shape is changed from cuboidal to flattened in the mutant epithelium, suggesting differentiation could be affected by *Fgfr2* deletion. The hallmark of corneal epithelial differentiation and maturation is the expression of keratin 12 (K12) by E14.5. K12 is not expressed in the limbal and conjunctival epithelium. In contrast to K12, K14 is expressed in all ocular surface epithelial tissues. To investigate whether loss of FGFR2 in the corneal epithelium affects cell differentiation, K12 and K14 immunofluorescence was examined in E16.5 *WT* and *Fgfr2*^*CKO* eyes. The results showed that although the corneal epithelial layer was significantly thinner in *Fgfr2*^*CKO* eyes, K14 was expressed in this cell layer in *Fgfr2*^*CKO* eyes, as well as in *WT* eyes. However, K12 expression was not detected in *Fgfr2*^*CKO*corneal epithelium but was in *WT* corneal epithelium. This result suggests that loss of FGFR2 affects corneal epithelial cell differentiation and maturation. Transcription factor Pax6 is known to be essential for K12 expression and corneal development. Pax6 is expressed in several ocular cell types, including the corneal and limbal epithelia, where gene expression is maintained throughout life. In E12.5 *WT* and *Fgfr2*^*CKO* eyes, Pax6 was expressed in developing corneal epithelium as well as in the lens and retina. The expression patterns were similar in *WT* and *Fgfr2*^*CKO* eyes. In E16.5 *WT* cornea, Pax6 expression was also found in conjunctival epithelium of the eyelid. In contrast, Pax6 immunofluorescence was almost abolished in *Fgfr2*^*CKO* cornea and a low-level expression was detected in a few cells (labeled by arrows). This finding suggests that FGFR2 plays a critical role in maintaining Pax6 expression. Loss of Pax6 expression could result in the absence of K12 expression and account for the abnormal differentiation of corneal epithelium in *Fgfr2*^*CKO* mice. ERK is one of the major downstream effector of FGFR activation. To investigate whether FGFR2′s role of maintaining Pax6 level requires ERK activity, we examined the Pax6 expression in the E16.5 cornea of ERK1/2 double conditional deletion mice. We found that, in contrast to the *Fgfr2*^*CKO* cornea, Pax6 expression was maintained normally in the ERK1/2-deficient corneal epithelial cells, suggesting that FGFR2 controls the Pax6 level through an ERK- independent signaling pathway. In the ERK1/2-deficient cornea, the stromal layer looked abnormal and corneal endothelium was absent, probably due to the severe defective and degenerative lens in these mice. ## Abnormal differentiation of corneal mesenchymal cells in *Fgfr2*^*CKO* mice Cre expression in the *Le-Cre* mice is limited to the ocular surface ectodermal tissues, however, developmental defects were seen in the mesenchyme-derived tissues, such as the corneal stroma. We assessed whether differentiation of corneal mesenchymal cells was affected by defective corneal epithelium in *Fgfr2*^*CKO* mice. Keratocan is a cornea-specific keratan sulfate proteoglycan, and is considered a phenotypic marker for keratocytes. Keratocan expression was found in the stroma of *WT* eyes but not in *Fgfr2*^*CKO* eyes. In contrast, expression of N-cadherin, a marker for corneal endothelial cells, was seen in eyes of both genotypes. Taken together, the data suggest that abnormal differentiation of the corneal epithelial cells affects the normal differentiation of corneal keratocytes but not endothelial cells in *Fgfr2*^*CKO* mice. Throughout the study, we have used *Fgfr2*^*flox/flox* mice as a control (referred as *WT*) to compare with *Le-Cre;Fgfr2*^*flox/flox* mice (or *Fgfr2*^*CKO*). Recently it was reported that hemizygous *Le- Cre* transgenic mice can develop severe eye defects on some genetic background. We examined the expression of Pax6 and keratocan in the corneas of *Le- Cre;Fgfr2*^*flox/+* heterozygous mice at postnatal day 5 (P5). We found no significant changes in the expression of these proteins in corneas between *Fgfr2*^*flox/+* and *Le-Cre;Fgfr2*^*flox/+* mice. This result suggests that the changes we have demonstrated in this study are unlikely caused by the expression of *Le-Cre* transgene in corneal epithelium. # Discussion ## FGFR2 is required for corneal epithelial cell proliferation during early eye development In early steps of eye development, FGF-7 and FGF-10 are expressed in the perioptic mesenchyme, some of these cells migrate into the space between lens and corneal epithelial layer to form the presumptive corneal stroma. Both FGF-7 and FGF-10 are ligands for FGFR2. In the previous “gain-of-function” studies, overexpression of FGF-7 and FGF-10 driven by the lens-specific αA-crystallin promoter resulted in suppression of corneal epithelial cell fate and induction of ectopic lacrimal gland formation at the corneal surface of transgenic mice. These results suggest that tight regulation of FGFR-signaling in corneal epithelium plays a critical role in controlling cell fate determination and commitment. In this study, we used a surface ectodermal Cre driver (*Le-Cre*) and demonstrate that loss of FGFR2 in corneal epithelial cells causes significant decrease in cell proliferation without affecting the epithelial cell fate and cell survival. Combining the ligand expression patterns with our data from this “loss-of-function” study, we propose that cell proliferation in the prospective corneal epithelium is regulated by FGFR2-signaling through epithelial-mesenchymal interaction in a paracrine fashion. The mature corneal epithelium is a stratified tissue that continuously renews itself. The mitogens for later stage and older corneas are likely secreted by the lacrimal gland and distributed via the tears over the ocular surface. For example, epidermal growth factor (EGF) and transforming growth factor α (TGFα) exist as a component of human tears. EGF and TGF-α share a common EGF receptor (EGFR). Both growth factors are capable to stimulate corneal epithelial cell proliferation in vitro and in vivo. Other growth factors, including hepatocyte growth factor (HGF), FGF-7, insulin-like growth factor (IGF)-1 and IGF-2 had all been proved to stimulate corneal epithelia cells proliferation in a dose- dependent manner in vitro. HGF and FGF-7 are expressed in stromal keratocytes and are highly upregulated following epithelial injury. The levels of FGF-7 and FGFR2 transcripts were highest in limbal fibroblasts and epithelial cells respectively in the periphery, while the expression of HGF and its receptor is higher in central cornea, suggesting a regional specificity of these two growth factor-signaling in control of cell proliferation. In our study, we also demonstrate that the cell proliferation indices in the peripheral corneal epithelium are higher than that in the central region in both *WT* and *Fgfr2*^*CKO*. The role of FGFR2 in later developmental stage and in mature cornea can be investigated using an inducible Cre line. ## Corneal epithelial cell fate is specified but cannot commit for further differentiation and maturation in *Fgfr2*^*CKO* mice We demonstrate that cytokeratin K12 (K12), a marker for corneal epithelial cell differentiation, is not expressed in *Fgfr2*^*CKO* eyes, suggesting the FGFR2-signaling is also required for corneal epithelial cell differentiation and maturation. However, K14 is expressed in both *WT* and *Fgfr2*^*CKO* corneal epithelium, suggesting that corneal epithelial cell fate is specified but remains in a less differentiated form. One potential mechanism for the differentiation defect in *Fgfr2*^*CKO* mice might be due to the loss of Pax6 expression in corneal epithelium. We show that Pax6 is expressed in E12.5 corneal epithelial layer of *Fgfr2*^*CKO* eyes but was almost absent at later stage (E16.5), suggesting that FGFR2-signaling activity plays an important role in maintaining Pax6 level in corneal epithelium. Similar observation was reported by Faber et al that, when a dominant negative FGFR1 is expressed in the lens, Pax6 expression levels were reduced. However, the underlying mechanism of FGFR-signaling in maintaining Pax6 level might be different between these two surface ectodermal derived tissues. For example, FGFR-ERK signaling plays a major role in lens development, whereas ERK activity (judged by phosphorylated ERK level) was hardly detectable in the developing corneal epithelium and loss of ERK1/2 did not affect the Pax6 levels in these cells. The downstream signal transduction pathways of FGFR2 in corneal epithelial cells require further investigation. Pax6 is known to be essential for normal corneal morphogenesis. During development, Pax6 is expressed in the surface ectoderm prior to and during corneal epithelial differentiation, and is maintained in the adult corneal epithelium including the limbal region where the stem/progenitor cell population exists. Pax6 functions as a co-activating factor for K12 expression. K12 expression was reduced in Pax6^+/- mouse cornea. Thus, loss of Pax6 in *Fgfr2*^*CKO* cornea can directly affect activation of genes, such as K12, which are crucial for corneal epithelial cell differentiation and maturation. Additionally cellular adhesion was also compromised in the Pax6^+/- mutant corneal epithelium. In postnatal and adult Pax6^+/- mouse cornea, the epithelial layer is thinner owing to a reduction in the number of cell layers, despite a tenfold increase in the proliferative index and no change in TUNEL labeling. Because proliferation of limbal and corneal epithelial cells in Pax6^+/- mice was not reduced, it confirms that decrease of cell proliferation in corneal epithelium of *Fgfr2*^*CKO* mice is a direct result of FGFR2-deficiency not due to the loss of Pax6. It is worthwhile to mention that during normal development FGFR-signaling level in corneal epithelial cells must be under tight control to ensure normal differentiation. As mentioned in the “Introduction”, FGF-3, FGF-7 and FGF-10 are the ligands of FGFR2. Overexpression of any of these FGFs from the lens at early developmental stage can alter the corneal epithelial cell fate, initially induce these cells to over-proliferate and then differentiate into secretory cell types (e.g. lacrimal and Harderian glands). Expressing an active form of Ras in corneal epithelium can also inhibit K12 expression. The gain- and loss-of- function studies all indicate that proper corneal epithelium differentiation and maturation depend on the tightly controlled FGFR-signaling activity in these cells. ## Interaction between corneal epithelium and underlying mesenchyme is critical for cell fate commitment and differentiation in both tissues When the lens vesicle and primitive corneal epithelium are completely separated, the space between them is filled by invading corneal mesenchymal cells which are mostly of neural crest origin. In mouse, the posterior mesenchyme cells closest to the lens differentiate into the corneal endothelium and subsequently the anterior chamber is formed. The mesenchyme cells between the corneal epithelium and endothelium start to differentiate into stromal keratocytes. One of the differentiation markers for keratocytes is keratocan, a type of keratan sulfate- containing proteoglycans (KSPGs) uniquely abundant in the corneal stroma. In *Fgfr2*^*CKO* corneal stroma, keratocan was not expressed, even though the mesenchymal cell proliferation was not affected, suggesting that differentiation of keratocytes was disrupted by *Fgfr2* deletion in the corneal epithelium. While more molecular markers for keratocyte differentiation need to be examined, we suspect that the stromal cell differentiation defect could result from loss of Pax6 expression in the corneal epithelial layer. In *Pax6*^*+/-* mutant embryos, corneal epithelium was abnormal in K12 expression and was thinner than *WT* littermates, and corneal stroma appeared irregular, hypercellular, and thickened. However, the abnormalities seen in *Fgfr2*^*CKO* stroma did not completely resemble the defects in *Pax6*^*+/-* cornea, suggesting that in addition to lower level of Pax6, other factors as a result of FGFR2-signaling in corneal epithelium also contribute to stromal cell differentiation in normal corneal development. ## FGFR2 plays a different role in corneal epithelium and lens development During vertebrate eye development, lens and corneal epithelium arise from the same surface ectoderm and share several common features, for example, they are both transparent and require sustained expression of Pax6. At E11.5–12.0, the lens vesicle detaches from the overlying surface ectoderm which subsequently forms the prospective corneal epithelium. Our study demonstrates that FGFR2 plays a different role in these two ectodermal tissues with the same origin. In the lens, FGFR2 is required for cell survival and cell cycle withdrawal during fiber differentiation, but it is dispensable for cell proliferation. In contrast, FGFR2 is essential for cell proliferation in corneal epithelium but is dispensable for cell survival. The different response is likely caused by different downstream signal transduction pathways activated by FGFR2. Previous study by Burgess et al showed that activation of Ras, a downstream effector of FGF-signaling, triggers different sets of downstream targets in lens and in corneal epithelium. Constitutive activation of Ras initially increased cell proliferation in both lens and corneal epithelial cells, correlating with increased levels of cyclin D1 and D2 expression in both cell types. This initial increase was sustained in the corneal epithelium, but not in the lens. Instead, cell cycle inhibitors, p27^kip1 and p57^kip2, were upregulated in the lens followed by hyperproliferation, probably through upregulation of transcription factor Prox1. Furthermore, the downstream effectors of FGFR2-Ras signaling also differ between lens and corneal epithelial cells. For example, the phospho-ERK1/2 level was increased in the lens, but not in the corneal epithelial cells, in response to Ras activation. When we examined the pERK1/2 levels in E12.5 cornea by immunofluorescence, we found that the pERK1/2 level was hardly detectable in central corneal epithelium in either *WT* or *Fgfr2*^*CKO* eyes. Double deletion of *Mapk1* and *Mapk3* (encoding for ERK2 and ERK1 respectively) in the surface ectoderm did not cause any visible changes in early steps of central corneal development although the cell numbers appeared to be reduced in the peripheral area near conjunctival fornix (data not shown), suggesting that role of FGFR2 in central corneal development is not controlled by ERK-signaling. This result is also consistent with the previous report in Ras-overexpression transgenic mice. We conclude that FGFR2-signaling activates ERK-dependent and ERK-independent pathways in the lens and in central corneal epithelial cells respectively. The signaling effectors and targets of FGFR2 in corneal epithelial cells have not been well defined and needs to be further investigated. In summary, we demonstrate that FGFR2-activated ERK-independent signal is essential for corneal epithelial cell proliferation during early eye development. In later stage, FGFR2 is also critical for corneal epithelial cell differentiation and maturation, one potential mechanism is to maintain Pax6 expression in corneal epithelium. In contrast to the ocular lens, FGFR2 is dispensable for survival of corneal epithelial cells. Our study implies that keratocyte differentiation in corneal stroma depends on some inductive signals released from the overlying corneal epithelium. More studies are needed to investigate the downstream signaling events of FGFR2 in corneal epithelium and signals/factors from the corneal epithelial cells responsible for keratocyte differentiation. # Supporting Information We thank Drs Ruth Ashery-Padan and David Ornitz for permission to use the *Le- Cre and Ffgr2-flox* mice, respectively; Dr Winston Kao for anti-K12 and anti- keratocan antibodies. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: LL LWR. Performed the experiments: JZ DU. Analyzed the data: JZ DU LWR. Wrote the paper: JZ LWR. [^3]: Current address: Yenepoya Research Centre, Yenepoya University, Mangalore, Karnataka, India.

# Introduction Zika virus (ZIKV) is a mosquito-borne flavivirus (family *Flaviviridae*), transmitted by *Aedes* mosquitoes. It was first identified in a monkey in the Zika forest of Uganda in 1947. In the following 60 years, ZIKV was repeatedly shown to circulate in mosquitoes in Africa and Asia, and was also isolated from humans with asymptomatic to mild infection. ZIKV gained global attention after its introduction into Brazil in 2013 and subsequent rapid spread in the Americas starting in May 2015. As of November 2016, autochthonous transmission of ZIKV was reported from 47 countries and territories in South and Latin America; and an association between ZIKV infection and neurological complications including Guillain-Barré syndrome and congenital defects in children born to women infected by ZIKV during pregnancy could be demonstrated. The first larger ZIKV outbreaks were reported from the Pacific region. In a study from a ZIKV-outbreak on Yap Islands (Federate States of Micronesia) in 2007, 49 confirmed and 59 probable cases of symptomatic ZIKV disease were identified and a seroprevalence of 73% anti-ZIKV IgM antibody positives in over three-year olds was observed. In 2013, French Polynesia was hit by a ZIKV epidemic with approximately 28,000 cases corresponding to 11% of the population. During this outbreak, Guillain-Barré syndrome and microcephaly were first shown to be associated with ZIKV infection. Seroprevalence for anti-ZIKV antibodies in French Polynesia before the 2013 outbreak was estimated to be 0.8% based on screening banked blood from 593 blood donors. There are two known ZIKV lineages, one African and one Asian. The reported outbreaks in the Pacific as well as the Americas were caused by the Asian linage, suggesting an Eastward spread from Southeast Asia into the Pacific and the Americas. Although Madagascar was described as potential ecological environment for ZIKV transmission, there are no reports from Madagascar on the presence of ZIKV after clinical or serological examinations. However, ZIKV infection may remain unrecognized as it usually presents with only mild disease or asymptomatic infection. Direct evidence of ZIKV infection by PCR is only possible in the acute phase of infection. Different serological methods are used to detect antibodies against ZIKV in serum including IgG and IgM ELISA, indirect immunofluorescence assays (IIFA) and virus neutralization tests (VNT). All these methods bear risks of cross-reactivity with other flaviviruses, most notable dengue virus (DENV). VNT have a higher specificity since they directly test the activity of neutralizing antibodies in serum towards live ZIKV. A previous study demonstrated high specificity of ELISAs for the detection of anti-ZIKV antibodies, when using serum samples from patients with different flavivirus infections. However, recently, false positive results were described with serum samples from malaria patients. In this seroprevalence study we investigated the presence of ZIKV antibodies in archived plasma samples that were collected in Madagascar from pregnant women in 2010, and for which serology for CHIKV, DENV and PCR for malaria had been performed previously in order to assess if ZIKV was circulating on the island at that time. To confirm the Zika-ELISA results, IIFA and VNT were performed. # Methods ## Plasma sample set Plasma samples stem from cross-sectional surveys that were carried out between May and July 2010 in pregnancy follow-up services in six different locations of Madagascar. Venous EDTA blood samples were collected for a study on malaria parasitemia in pregnant women and for investigation of a previous outbreak of a arboviral infection hat had taken place around 3–4 months before sampling at the Eastern Coast of Madagascar in the surroundings of the cities of Mananjary and Manakara. Supernatant plasma was stored at -20°C for serological analysis. Plasma samples were collected at six different locations aiming for 200 plasma samples from each location. Two of the locations were at the East Coast at sea level (Mananjary and Manakara). Further four locations were highland locations on different elevation levels at 466m (Ifanadiana), 860m (Tsiroanomandidiy), 920m (Moramanga) and 1280m (Ambositra). ## Ethical approval Blood was collected from all pregnant women presenting for routine pregnancy screenings to the local health centre who gave consent to their participation in the study. All participants provided written informed consent to participate in the study. Ethical clearance for malaria and antiviral antibody testing was obtained from the “Comité d’ethique de la Vice Primature Chargée de la Santé Publique”. ## Enzyme Linked Immunosorbent Assays (ELISA)-testing For the detection of IgG or IgM antibodies to ZIKV we used an anti-ZIKV ELISA (IgG/IgM) assay (Euroimmun, Lübeck, Germany), which uses the recombinant ZIKV- non-structural protein 1 (NS1) to minimize cross-reactivity to other flaviviruses. The measurement against a manufacturer-provided calibrator led to a ratio of extinction value for the sample to extinction value for the calibration solution. The manufacturer suggests interpreting the resulting ratios by classification into three categories: 1. Extinction Ratio \< 0.8:     Negative 2. Extinction Ratio ≥ 0.8 - ≤1.1:  Borderline 3. Extinction Ratio \> 1.1:     Positive ## Indirect immunofluorescence assays (IIFA) IIFA for ZIKV was performed with ZIKV strain MR766-infected Vero E6 cells. In brief, infected Vero cells were spread onto slides, air dried, and fixed in acetone. Plasma samples were serially diluted in phosphate-buffered saline (PBS) starting with an initial dilution of 1:10, added to the cells, and incubated for 90 min at 37°C. After washing with PBS, slides were incubated with fluoresceine isothiocyanate-labeled rabbit anti-human IgG and IgM antibodies at 37°C for 25 min. IgG titers or IgM titers of 1:20 or more were considered positive. ## Detection of antibodies against chikungunya virus (CHIKV) and dengue virus (DENV) Anti-CHIKV-IgG und Anti-DENV-IgG were also detected using immunofluorescence analogous to the procedure described above. ## Viral neutralization tests (VNT) For the VNT serial twofold dilutions from 1:20 to 1:2,560 were prepared for each plasma, and a volume of 50μL from each dilution was transferred to microtiter plates. A volume of 50μL containing 50 PFU of ZIKV strain MR766 was added to each well except for the controls, which contained phosphate-buffered saline (PBS). The plates were incubated at 37°C for 1h. Then, 100μL of the plasma/virus mixture were transferred to a 96-well plate of Vero cells containing approximately 2x10E5 cells/well in EMEM enriched with 1% fetal bovine serum, 1% penicillin-streptomycin, 1% \[200mM\] L-glutamine, 1% kanamycin, and 3% amphotericin B and incubated at 37°C in the presence of 5% CO₂. After six days, the microtiter plates were fixed in formaldehyde, stained with crystal violet, and the presence/absence of cytopathic effect was read under an inverted microscope. ## Sample size considerations for assessing ZIKV circulation in Madagascar in 2010: Sensitivity and specificity of the diagnostic test and certainty of conclusions As our aim was to assess if ZIKV circulated in the human population of Madagascar in 2010 without having evidence for clinical manifestations or an outbreak of ZIKV, we assumed a low seroprevalence. Sample size considerations were focused on the coastal samples as the probability for the presence of ZIKV was expected to be higher at the coast. We could perform the ELISA on 433 samples from the coast. For the EUROIMMUN ELISA, the specificity, based on samples from 1,015 uninfected individuals, was estimated to be 99.8%. The probability for a single test of a negative sample to be positive is 1-specificity and for a specificity of 99.8%, this would be 0.2%. Using a binomial distribution, we thus get a probability of 58.0% to find at least one (false) positive sample in 433 tests. <img src="info:doi/10.1371/journal.pone.0176708.e001" id="pone.0176708.e001g" /> P ( x ≥ 1 ) = 1 − P ( x = 0 ) = 1 − \[ ( 433 0 ) 0.002 0 0.998 433 \] When generalizing this formula to $$P\left( {x \geq k} \right) = 1 - {\sum\limits_{k}^{n}{\begin{pmatrix} n \\ k \\ \end{pmatrix}b^{k}a^{n - k}}}$$ *k* = *minimum number of positives, n* = *sample size, a* = *specificity, b* = *1-specificity* one can calculate the probability for getting different numbers of false positives (among samples that in reality are all negative). As calculating cumulative probabilities by hand is quite cumbersome we used and online tool for our assumptions (<http://stattrek.com/online-calculator/binomial.aspx>). For at least two positive samples among 433 tests this probability would be 21.5% for at least three positive samples 5.7% and for at least four positive samples 1.2%. These are the probabilities to falsely claim ZIKV occurrences based on the respective number of positively tested samples under the assumption that the test specificity was indeed 99.8%. Prevalence estimates can be statistically corrected for the imperfection (specificity \<1) of a diagnostic test. ## Molecular detection of *Plasmodium* species After DNA extraction from peripheral venous blood-EDTA/8 M urea (v/v) using the QIAamp^® DNA Blood mini Kit according to the manufacturer’s protocol (Qiagen), infection with *Plasmodium* parasite was determined by using a real- time PCR methodology. *Plasmodium falciparum* was detected with primers and probes targeting the *cox1* gene and *P*. *ovale*, *P*. *malariae* and *P*. *vivax* with primers and probes targeting the *18S rDNA*. On a LightCycler 480 (Roche), the PCR was carried out in a total volume of 20μL containing 5μL of extracted genomic DNA or plasmids including the gene targeted for each species as positive control. The mixture to detect *P*. *falciparum* contained 0.2 mM dNTPs, 3.5 mM MgCl₂, 0.4 μM of forward primer, 0.5 μM of reverse primer, 0.15 μM each of the donor and acceptor probes, 0.25 g/l bovine serum albumin (BSA) and 0.5 U Taq polymerase (Platinum, Invitrogen). The multiplex PCR to detect *P*. *malariae*, *P*. *ovale* and *P*. *vivax* used three forward primers; each primer is specific of one species and 1 reverse primer for the three species. The mixture differed from the first one by primer concentration: 0.7 μM of forward primer for *P*. *malariae*, 0.5 μM of each forward primer for *P*. *ovale* and *P*. *vivax*, 0.4 μM of reverse primer. For both PCRs, the following PCR program was used: 2 min at 95°C; 10 touchdown repeated cycles of 5 sec at 95°C, 10 sec at 63°C to 58°C, 7 sec at 72°C; and 45 repeated cycle of 5 sec at 95°C, 10 sec at 58°C, 7 sec at 72°C. Melting analysis was performed by denaturing for 30 sec at 95°C and cooling for 2 min to 45°C followed by heating at the rate of 0.1°C/s from 45°C to 75°C. # Results ## Coastal locations Plasma samples from the two coastal areas (Mananjary and Manakar) from 433 pregnant women were investigated. Each sample was analyzed by ELISA in two independent experiments. For 421 samples the ELISA result was negative. For the remaining 12 samples, the ratios were borderline or positive. In case of doubt, a third ELISA was performed (see for details). All anti-ZIKV IgM tests for the coastal locations were negative (no signal or ratio \<0.8). Among the four samples from the coastal locations with a positive anti-ZIKV IgG, two were positive for *P*. *falciparum* (50%), compared to 73 (17%) of the remaining 429 anti-ZIKV-antibody negatives samples (prevalence ratio 2.9, 95% CI 1.1–8.0). Of the other two ZIKV-positive samples, one showed a positive IgG against DENV, and one was negative for both *P*. *falciparum* and DENV. Further testing by IIFA revealed that three of the four suspected samples, M0159, M0068 and M0784, were positive for ZIKV-specific IgG and IgM antibodies. However, when tested by VNT, no neutralizing antibodies against the African ZIKV strain MR766 were detected in the three tested sera. ## Highland locations Plasma samples from 783 pregnant women from the four highland cities (Ifanadiana, Moramanga, Ambositra, Tsiroanomandidiy) were investigated. For 779 samples the ELISA was negative. For the remaining four samples, the ratios were borderline or positive. In contrast to the coastal samples, two of the IgG- positive samples were also positive for IgM. Among the two samples from the highland locations with a positive anti-ZIKV IgG and IgM, two were positive for *P*. *falciparum* (100%), compared to 79 (10%) of the 781 anti-ZIKV-antibody negatives (prevalence ratio 9.9, 95% CI 8.0–12.2). The two samples that were positive both in the anti-ZIKV IgG ELISA and the *P*. *falciparum* PCR were also the only samples in the dataset that were clearly positive in the anti-ZIKV IgM ELISA. Confirmatory testing by IIFA was negative for both suspected samples. # Discussion The study results lead to two statements: First, the presence of ZIKV antibodies in pregnant women from Madagascar in 2010 is unlikely. Second, malaria parasites may interfere with anti-ZIKV ELISA tests and may lead to false positive test results. Outbreaks of CHIKV and DENV infections, other arboviral diseases, occurred during the last 10 years in Madagascar. In contrast, ZIKV infections have not been described clinically or serologically in Madagascar so far. A limitation of our study is that only six of the 1,216 ZIKV IgG-ELISA tests performed in our study were positive. While this is a small number and limits statistical interpretability, we cannot conclude on the presence of ZIKV virus in the human population of Madagascar before 2010, without considering the possibility of the positive tests being “false positives”. At the coast we found four positive samples among 433. Based on our binomial assumptions, the probability of having at least four (false) positive samples among 433 is only 1.2% when assuming a specificity of 99.8%. We would conclude that at least some of our positive ELISA test results were probably true positives. However, we have to assume that malaria infection interferes with the anti-ZIKV ELISA test, so its specificity is lowered in our malaria-endemic study population. Although prevalence ratios for the association between ZIKA-test positivity and malaria infections were relatively high (10 in the highlands and 2.9 at the coast), we always have to keep in mind that they contained only two anti-ZIKV positive samples in the highlands and four at the coast. Notably, the frequency of malaria-PCR positivity was higher in the few ZIKV- ELISA positives than for ZIKV-ELISA negatives, hence malaria may have led to (false) ZIKV-ELISA positives. However, not all ZIKV-ELISA positive samples were also malaria positive, indeed two anti-ZIKV IgG positive samples were malaria negative. When using the specificity without considering malaria (99.8%), the probability for obtaining at least two false positives among 433 samples is 22%. Cross-reactivity was expected towards flaviviruses that exhibit a high degree of sequence and structural homology. Especially the envelope (E) glycoprotein is very similar among flaviviruses, whereas the secreted NS1 protein is more conserved and ZIKV-specific. Van Esbroeck *et al*. confirmed the high ZIKV specificity claimed by the EUROIMMUN assay, using plasma of 10 PCR-confirmed dengue patients, which all tested negative for anti-ZIKV IgGs. However, in the same study it was also reported that 14 (41%) out of 34 PCR-positive malaria infections tested positive for ZIKV IgM and IgG, although ZIKV infection was excluded for 11 of these patients by VNT. Three of six IIFA tests performed on ELISA-positive samples were positive for ZIKV-specific IgG and IgM antibodies, but no neutralizing antibodies against the African ZIKV strain MR766 were observed. Anti-ZIKV ELISA IgM-positive tests were only found in the two IgG-positive individuals from the highlands, both also carrying malaria parasites. This study suggests that the association between anti-ZIKV IgG ELISA positivity and *P*. *falciparum* PCR -positivity was stronger in highland regions with a lower *P*. *falciparum* endemicity than highly endemic coastal regions. In summary, it is important to consider the potential interference between ZIKV ELISA and *P*. *falciparum* infection as previously reported. As ZIKV virus often occurs in malaria-endemic regions, this shortcoming of the ZIKV ELISA has to be taken into account. # Supporting information We thank all pregnant women for participation in this study and the staff of the health centers in Mananjary, Manakara, Ifanadiana, Ambositra, Tsiroanomnadidy and Moramanga for their hospitality and support. Many thanks to Mr Klaus Jürries for sketching the map of Madagascar. [^1]: The authors have declared that no competing interests exist. [^2]: **Conceptualization:** NS EM. **Data curation:** NS. **Formal analysis:** NS. **Funding acquisition:** NS JSC. **Investigation:** NS EM OMA NR RR. **Methodology:** DW EM SJ JSC. **Project administration:** NS EM RR. **Resources:** RR JM. **Supervision:** NS EM JS-C. **Validation:** NS EM. **Visualization:** NS. **Writing – original draft:** NS EM. **Writing – review & editing:** NS EM DW OMA DD SJ DT NR JM RR JSC.

# Introduction An estimated 250 million episodes of malaria led to nearly one million deaths in 2008, the brunt of which was borne by young children and infants in sub-Saharan Africa. In addition to its impact on the health of individuals, malaria places considerable costs on households, communities and nations. Intermittent preventive treatment in infants (IPTi) and children (IPTc) have received attention in recent years as potential interventions to reduce malaria morbidity and mortality. Both follow the same strategy: to deliver a full course of an anti-malarial treatment to a population at risk at specified time points whether or not they are known to be infected. Both aim to retain the benefits of chemoprophylaxis whilst avoiding the acceleration of drug resistance or impairing the development of acquired immunity. The two interventions differ in their target age group and delivery system. By targeting infants under 12 months, IPTi is able to benefit from the existing delivery strategy of the Expanded Programme on Immunization (EPI). The delivery of IPTi involves training health workers to administer a dose of an antimalarial drug during routine scheduled visits in health facilities and to document this using modified EPI monitoring tools. IPTc has targeted mainly children up to the age of five years – but also older age groups and school children –. In children under five years, IPTc has no established delivery system but studies have used community health workers and outreach clinics to provide doses for the target age-group. Studies have mostly focused on IPTc as a seasonal intervention in settings with seasonal transmission. Both IPTi and IPTc have been found to reduce clinical incidence. Several clinical trials in different settings have shown IPTi to be effective against malaria to varying degrees. A pooled analysis of data from six completed trials of IPTi with sulfadoxine-pyrimethamine (SP) estimated a 30% (95% CI 20%, 39%) protective efficacy (PE) against clinical malaria and 38% (13%, 56%) PE against hospital admissions with malaria parasites. Studies using drugs other than SP found that efficacious, long-lasting drugs had a greater PE than shorter-acting drugs or drugs with high levels of resistance. Seasonal IPTc with combinations of SP, artesunate (AS) and amodiaquine (AQ) has been seen to reduce the incidence of clinical malaria in children under the age of five years in settings with a short malaria transmission season, and where the transmission season is longer. A pooled analysis of IPTc trials estimated a 75% (64%, 83%) protective efficacy against malaria episodes during the intervention period. The trials were not designed to detect an impact on mortality due to the very large sample sizes required. There have been costing and cost-effectiveness studies alongside many of the IPT trials,. In nearly all of the studies where IPT was efficacious, it was highly cost-effective. In sites where IPTi had a significant effect, the cost per malaria episode averted for IPTi-SP ranged from US\$ 1.36 to 4.03 based on trial specific data. For IPTi using more expensive antimalarials, the cost per treated episode averted ranged from US\$4.62 using AQAS to US\$ 18.56 using mefloquine (MQ). For seasonal IPTc, a trial in Ghana estimated the costs per episode averted for three different regimens administered over the six month transmission period. Bimonthly SP cost \$105 (\$75, \$157) per treated episode averted, bimonthly ASAQ was \$212 (\$127, \$399) and monthly ASAQ was US\$68 (\$62, \$75). The estimates for district scale-up fell to \$28, \$60 and \$22 respectively. In addition, where efficacious, IPT reduced health system costs and showed significant savings to households from malaria episodes averted. The variations in efficacy and cost-effectiveness between trials stem not only from the choice of drug but also from the different setting and trial characteristics such as transmission intensity, timing of delivery, local costs and use of other interventions. This raises questions that the trials were not designed to answer such as the effects of the different characteristics, the impact on severe malaria and mortality, and the limits beyond which IPT is no longer cost-effective. It is not feasible to carry out a large number of large field trials of different combinations to determine the impact of each factor on different malariological outcomes. Where data cannot be collected, mathematical modelling can be used to provide predictions. In this paper, we use a comprehensive model of *Plasmodium falciparum* epidemiology and economics to investigate the influence of different variables on the effects and the cost-effectiveness of IPT in preventing disability adjusted life years (DALYs): target age group and delivery channel, seasonal or year-round delivery, transmission intensity and seasonality of the setting, the timing of the first IPT dose for seasonal delivery, different coverage levels of treatment for malaria fevers and the impact of drug resistance. We also predict the impact of IPT on transmission intensity. # Methods ## The simulation model We use a dynamic, individual-based, stochastic simulation of malaria epidemiology which has been described elsewhere. Briefly, there is a simulated population of humans who are updated at each five-day time step via model components representing new infections, parasite densities, acquired immunity, uncomplicated and severe episodes, direct and indirect malaria mortality, infectivity to mosquitoes and case management. This study does not include simulation of anaemia. The course of parasite densities over an infection are described by averaged empirical data. Immunity to asexual parasites is derived from a combination of cumulative exposure to both inoculations and parasite densities, and maternal immunity. The inclusion of acquired immunity allows us to model potential effects of IPT on immunity through loss of exposure and the inclusion of infectivity captures potential effects on transmission intensity. The probability of a clinical attack of malaria depends on the current parasite density and a pyrogenic threshold. The pyrogenic threshold responds dynamically to recent parasite load, increasing or saturating with exposure to parasites and decaying with time, and thus is individual-and time- specific. Severe malaria can arise in two ways, either as a result of overwhelming parasite densities or through uncomplicated malaria with concurrent non-malaria co-morbidity. Mortality can be either direct (following severe malaria) or indirect (uncomplicated malaria in conjunction with co-morbidity, or during the neonatal period as a result of maternal infection). Thus the model does not assume a fixed case fatality rate for malaria episodes, but makes a number of intermediary assumptions to model pathways from an acute episode to death. The parameter values for each of the components of the model were estimated by fitting to data from a total of 61 malaria field studies of different aspects of malaria epidemiology and are reported elsewhere. The model has been validated using age-specific results from six IPTi trials with SP. It has subsequently been validated against trials of IPTc. ## Simulation strategy We simulate seasonal and non-seasonal delivery for both IPTi and IPTc to allow us to separate the effects of seasonal delivery from the combination of the intervention target age-group and delivery channel. We simulate two contrasting seasonal patterns (constant and highly seasonal transmission) and two IPT drugs (SP and ASAQ). These four factors have two levels each making a set of 16 baseline intervention scenarios. For each of these scenarios, we then investigated the impact of varying levels of drug resistance, transmission intensity, the timing of seasonal implementation and the proportion of malaria fevers which are effectively treated. We also simulate the impact of either widening the target IPTc age group or shifting it into older ages. We simulate 10 seeds for each scenario each of a population of 100,000 individuals aged up to 90 years over ten years from the introduction of the IPT programme. ## Delivery frequency and modality of IPTc and IPTi in the baseline scenarios For IPTi, the EPI visits were assumed to be scheduled at 3, 4 and 9 months of age. For seasonal delivery, only infants who were presented for their EPI visits during the short transmission season received the doses and so no infant would receive all three doses. The baseline target IPTc age group was 3 months to 5 years. We model IPTc delivery either every two months throughout the year or as three IPTc doses at monthly intervals during the malaria season. IPTc was assumed to be delivered, and costed, via community health workers (CHWs) who were individuals in the community given a small financial incentive to deliver IPTc. Studies in The Gambia and Ghana found that CHWs were able to reach more children under five years than outreach services linked to EPI. The timing of the doses relative to the start of the season for seasonal delivery is shown in. In the baseline scenarios, we assumed that IPTi coverage was 95% per dose (86% for 3 doses) and that IPTc coverage was 84% per dose for CHW. ASAQ includes 3 tablets to be taken on 3 consecutive days: we assumed that compliance with the tablets given to carers to administer at home was 100%. Compliance has been reported to be 88% to 99% in trial settings. ## IPT drugs We simulate two drugs, ASAQ and SP, both of which have previously been chosen for IPT trials. They demonstrate different characteristics: SP is a cheap, long- acting drug with high levels of resistance whereas ASAQ is a more expensive, shorter-acting drug currently meeting less drug resistance. We recognise that there are other potential IPT drug candidates that we have not included, nevertheless, the contrasting characteristics of these two drugs demonstrate the substantive effects. A simple model component for antimalarial drug action was adapted from Hastings and Watkins and incorporated into the simulations. Briefly, the ability of SP and ASAQ to both clear existing infections and to prevent new infections becoming established depends on the genotype of the infection. Infections are assigned genotypes randomly according to assumed frequencies. Hastings and Watkins quantify the chances of failing treatment with correct dosing of SP for wildtype infections and infections with single, double and triple dihydrofolate reductase (*dhfr*) mutations conferring resistance at 0, 0, 0, 50% respectively while periods of prophylactic effects are 52, 12, 12, 2 days. We round these to the five-day timesteps used by the model. For ASAQ, we assume that all infections are cleared and that for AQ-sensitive infections, the prophylactic period is assumed to be 15 days and for AQ-resistant infections, 0 days. ## Intervention costs The costs were based on economic evaluations conducted in a range of IPTi and IPTc sites. Similar cost categories and methodological costing approaches were used for both interventions covering district costs associated with community sensitization, behaviour change and communication, drug distribution and administration, training and supervision. The costs were identified from components of trial budgets and primary data on resource use. Care was taken to exclude costs relating specifically to research or to a trial environment. We costed IPTc only in the baseline target age group of 3 months to 5 years since while schools may be used for some age groups, the delivery mechanisms for others are unclear. Costs of incentivising the CHW who delivers the IPTc drugs to an assumed 250 children were included in administration costs to reflect an allowance of approximately US\$10 a month during the months of administration. The costs of the IPT drugs were based on prices presented on the International Drug Price Indicators List : SP was assumed to cost \$0.02 per dose and ASAQ \$0.36 per course of three tablets. Remainder fractions of tablets were assumed to be wastage. The cost of the intervention to households was found to be negligible and therefore excluded. The cost per dose of IPT drugs, delivery and administration remained unchanged for each dose irrespective of the number of doses given. The costs of training, sensitization and a minimum level of supervision were assumed to be fixed over the course of the year and therefore the unit cost per dose of IPT year-round was less than that of seasonal delivery. This assumption was based on trial activity and discussions with implementers about how IPT would be delivered if introduced as part of routine activity. Whether the doses were given throughout the year or concentrated in three months, a one-off training would be held each year for those involved in delivering IPT. Sensitization activities would involve the same resources even though the message about IPT frequency would be different. Supervision was assumed to be semi-fixed in that a minimum would be required and so it would be slightly more intense for seasonal than year-round IPT. The costs were calculated in US\$2006 to be comparable with previous costs for case management, all costs were then inflated to US\$2009 using US dollar inflation rates. ## Potential cost savings of IPT The simulations include direct malaria treatment costs to both the providers and households. We do not include indirect costs such as potential earnings forgone by the carers. The health system adopted is based on a previously used model with artemisinin combination therapy as the first-line treatment with low rates of access. The case management costs assumed have been previously published. ## Cost-effectiveness The approach adopted follows previous work on modelling the cost-effectiveness of malaria vaccines, and follows standard practices. The primary epidemiological outcome was the number of DALYs averted since they are a comparable, summary measure of the burden. One DALY represents a year of healthy life lost. Years of life lived with disability were calculated on the basis of the duration of disability and the disability weights for the different malaria attributable disease conditions obtained from the Global Burden of Disease study. DALYs were calculated assuming age-specific life expectancies typical for an East African setting with low malaria transmission, and with no age weighting to follow standard cost-effectiveness practices. Future costs and health gains are discounted at 3%. The cost-effectiveness ratios are to be interpreted as incremental cost-effectiveness ratios (ICERs) of implementing the interventions in the simulated scenarios relative to a do nothing scenario which corresponds to maintaining only case management. Recognising that the selection of cost-effectiveness thresholds in published literature is subjective, we refer to the conservative cut off point of US\$ 223 per DALY averted to reflect a cost-effective intervention, and US\$ 37 per DALY averted to reflect a *highly* cost-effective intervention. These thresholds are based on US\$ 150 and US\$ 25 thresholds suggested by the World Bank in 1993, and inflated to their 2009 equivalent. # Results Both IPTc and IPTi were cost-effective in a wide range of simulated settings. ## Seasonal and year-round delivery in seasonal and perennial transmission settings The effect of seasonal delivery depends on the seasonal pattern of transmission. In the constant transmission setting ( top row), year-round delivery averts a greater number of DALYs than seasonal delivery for both IPTi and IPTc ( a and c). Year-round delivery is also more cost-effective than seasonal delivery, substantially so for IPTc whereas for IPTi the difference is less pronounced. As well as the different unit costs, the different spacing of doses in combination with the assumed prophylactic periods contribute to year-round IPTc delivery being more cost-effective. For IPTi, the numbers of DALYs averted are proportional to the numbers of doses administered. In the highly seasonal setting (bottom row), the numbers of DALYs averted are much closer for seasonal and year-round delivery since there are few episodes outside the main transmission season. In this case, seasonal delivery is more cost-effective than year-round for both IPTi and IPTc due to the lower total costs. ## Transmission intensity and target age group We focus on the lower end of the range of transmission because interest lies in determining where IPT is not cost-effective. For all of the IPT strategies, the predicted number of DALYs averted is low at low transmission intensities and increases up to a plateau at moderate levels decreasing slightly at very high transmission intensities. This slight decrease in DALYs averted at high transmission intensities occurs despite increasing predicted total numbers of DALYs in the scenarios with no IPT. The predictions suggest that IPTi and IPTc are cost-effective, although not highly cost-effective, at the low transmission intensities simulated. With zero transmission, IPT would clearly not be cost- effective. A clear lower limit below which IPT is not cost-effective is not obvious: care should be taken with interpretation since the model was created using data mainly from medium and high transmission settings and so there is greater uncertainty in predictions for low transmission intensity settings. The predictions do however indicate an approximately log-linear relationship. Since the age-distribution of episodes is affected by transmission intensity, the relative impact at a low transmission intensity was compared for IPTc and IPTi for uncomplicated episodes, severe episodes and deaths. IPTc averts more clinical events than IPTi in the simulated scenarios, and the ratio of clinical events averted by IPTc to IPTi is greater at an EIR of 1 than at an EIR of 21. The size of the ratio, however, depends on the outcome and the seasonality of the setting. The predicted number of DALYs averted increased as the target age group for IPTc was widened to include older children for all transmission levels simulated (top row), although the number of DALYs averted per IPT dose decreased. Since the predictions cover the first ten years of an IPT programme they will not include potential rebound effects in those receiving IPT for long periods of time. Increasing the target ages of a five-year wide age-band for IPTc ( bottom row) lead to a decrease in DALYs averted with increasing age for settings with an EIR of 6 and 21. The scenario with an EIR of 1, however, increased gently to a maximum at 20 years. This prediction is highly uncertain, and is driven in the model by the increased surface area of the individual leading to a higher number of mosquito bites. Nevertheless, the predictions suggest that IPT would be more beneficial in older age groups at very low transmission intensities. ## Choice of IPT drug In all of our simulated scenarios, SP averted a greater number of DALYs and was more cost-effective than ASAQ. This is driven by the longer prophylactic period of SP, the lower costs and the low levels of drug resistance we have assumed. ## Timing of first dose in seasonal settings The cost-effectiveness of seasonal delivery of IPTi and IPTc in a highly seasonal setting is sensitive to the timing of the first dose. Too early or too late and at least part of the treatment and prophylactic actions of the drug are wasted. There is some leeway however and SP was less sensitive to timing than ASAQ due to the longer prophylactic period. If the three-month delivery period was begun very early, very few DALYs were averted and the corresponding costs per DALY were subject to considerable stochasticity: these values could not be included on the figures. Whilst the extent of mistiming is unlikely to be three months in practice, these predictions show that badly timed implementation can push IPT over the cost-effectiveness threshold. ## Proportions of malaria fevers treated As treatment coverage increases the number of DALYs averted by IPT decreases and the costs per DALY increase. This is driven by a reduction in the total DALY burden: the prompt treatment prevented severe malaria and deaths and cleared infections which could later produce symptoms and the high treatment coverage of all age groups lead to a small reduction in transmission. There was no apparent synergy between health system coverage and IPT. At very high treatment coverage levels using ASAQ, some scenarios are no longer cost-effective. However, this only occurred at treatment levels which would be unrealistic even for very good health-systems. ## Drug Resistance We simulated the effect of varying levels of drug resistance for year-round IPTi and seasonal IPTc both with SP only. In both cases, the number of DALYs averted decreased with rising drug resistance and the corresponding cost-effectiveness decreased. In all of the scenarios simulated, IPT remained cost-effective. However, if a drug has no effect whatsoever, then clearly this would not be the case. We did not simulate levels of resistance which would render SP completely ineffective. The linear pattern between DALYs averted and drug sensitivity was also observed for constant seasonality and for other transmission intensities with an annual EIR of 1 and 200 (not shown). ## Impact of IPTi and IPTc on transmission intensity We predicted a negligible impact of both IPTi and IPTc on the infectious reservoir and on transmission intensity except where the wider age groups for IPTc were simulated. # Discussion The predictions suggest that both IPTi and IPTc are cost-effective in the majority of scenarios simulated, even with the conservative thresholds we have used. In general, IPTc averted a greater number of DALYs than IPTi, but the costs were greater and consequently the costs per DALY averted were greater. A greater number of DALYs are averted for both IPTc and IPTi by year-round compared to seasonal delivery in perennial transmission settings, but similar numbers of DALYs are averted in seasonal settings since there are few episodes outside the main transmission season. Seasonal delivery is more cost-effective in seasonal settings, and year-round in constant transmission settings. However the difference is more pronounced for IPTc than IPTi due to the different proportions of fixed costs and also different assumed drug spacing during the transmission season. Cost-effectiveness was predicted to decrease with decreasing transmission, badly timed seasonal delivery in a seasonal setting, short-acting and more expensive drugs, increased frequencies of drug resistance and increasing levels of treatment coverage. A greater number of DALYs were averted as the target age groups were widened to include older children for IPTc in all simulated transmission settings, although the number of DALYs averted per IPT dose fell slightly. The number of DALYs averted decreased as the target ages for a five-year age-band were increased except for very low transmission intensities. This concurs with a systematic review of the age-distribution of episodes by transmission intensity in children, however the burden in adults is not well established. Our aim was not to pit IPTc and IPTi against each other since they are both interventions focusing on drug administration and differing only in target age group and delivery system. Instead, we aimed to tease out the contribution of seasonal delivery in different settings for both interventions and to investigate factors which affect their impact and cost-effectiveness. We selected a limited number of scenarios in order to focus on a manageable number of questions and to investigate the substantive effects. We recognize that there are many other potential scenarios differing in characteristics such as IPT schedule, target age groups, candidate drugs, seasonal patterns and also equity and heterogeneity in IPT coverage. The costs of an IPT programme are largely driven by the cost of the IPT drug and, in the case of IPTc, delivery since it does not benefit from the existing delivery strategy EPI. The predicted cost- effectiveness is driven by the low costs, particularly for IPTi, and the impact on DALYs, particularly for IPTc. The DALYs were dominated by the contribution from deaths. Our predicted costs per episode averted are generally lower than estimates from clinical trials. The predicted cost per uncomplicated episode averted ranges from \$1.08 to 17.59 for seasonal IPTc including all the values for transmission intensity, drug resistance and treatment coverage in comparison to \$22 to \$60 per treated episode estimated for district delivery from trial data. For year- round IPTi with SP, the predicted cost per episode averted ranges from \$0.42 to \$7.71 ( and not shown) compared to \$1.36 to \$11.93 per treated episode from trial studies. Although our predictions are not of the specific trials and so differ in many ways, the largest single reason for the difference is that we have predicted all episodes whereas the trial data refers to treated episodes only. Our model for acute episodes is fitted to data from Ndiop and Dielmo in Senegal where intensive daily active surveillance was carried out thus capturing episodes unavoidably missed by passive case detection. The predicted cost-effectiveness of IPTc and IPTi is in line with other malaria control interventions. Inflating using US dollars only to US\$ 2009, the cost per DALY averted of insecticide-treated net programmes is estimated to range from \$14 to \$74, and for indoor residual spraying, US\$131–145. Case management is estimated to cost between \$11 and \$31 depending on the treatment drug and for IPTp estimates vary from \$42 to approximately \$422 per DALY averted. The question of where the boundaries of IPT cost-effectiveness lie has been raised, particularly for transmission intensity. Our predictions show that, as transmission intensity decreases to low levels, the number of DALYs averted decreases and the corresponding cost per DALY increases. Unfortunately, while the log-linear nature of the relationship is apparent from our predictions, a boundary where IPTc and IPTi are no longer cost-effective is not. Even at low transmission intensities, IPT is predicted to be cost-effective although not highly cost-effective according to the World Bank thresholds however IPT would clearly not be cost-effective if there was no transmission. Caution must be taken with interpreting the predictions for settings with low transmission intensities since the model was created used data from mainly medium and high transmission settings and does not take into account variables such as heterogeneity in transmission or decay of immunity which may have strong effects in low transmission settings. In addition, children under five years old are not predicted to be the optimum target group at very low transmission intensities and the combination of other factors is important, for example higher frequencies of drug resistance would increase the levels of transmission intensity at which IPT is no longer cost-effective. Both IPTi and IPTc in children under five years of age were predicted to have negligible effects on transmission in this study. This agrees with previous predictions of IPT in these target age groups, although simulations of IPTc in children aged 5 to 18 years suggested a reduction in transmission. These results indicate that a wider target age range including older children would be necessary to reduce transmission, such as is being considered for mass screen and treat. An increased incidence of clinical episodes following the end of the prophylactic period was observed in some IPTi trials, but not others. In the seasonal IPTc trials which followed the participants up over the following season, no rebound effects were observed in individual trials , but a meta- analysis indicated a small increase in the incidence of clinical episodes. A previous study of year-round IPTi predicted a modest increase in susceptibility following the prophylactic period which was outweighed by the cumulative benefits. The model would also predict this same pattern for IPTc doses in older children, although to a much lesser extent if there is a seasonal break in transmission. Monthly SP may be akin to chemoprophylaxis after which rebounds have been observed. In practice, whether IPT equates to chemoprophylaxis would depend on coverage, the timing of delivery and levels of drug resistance. Limitations of the model components are discussed elsewhere,. Some assumptions are especially relevant to this study. The predictions are likely to be sensitive to parameters relating to age-patterns and outcomes of severe disease in the first years of life. The predictions of indirect malaria mortality, and to a lesser extent, severe episodes rely on age-dependent co-morbidity functions. In a trial setting with access to good health care, the age-pattern of co-morbidity may be quite different to that assumed by our models, which were fitted to datasets from other settings. This would affect age-dependent comparisons such as between IPTc and IPTi, and post-intervention effects. We acknowledge that a limitation of this study is the lack of a full sensitivity analysis. We are currently developing an interface which will facilitate extensive probability sensitivity analyses. Additionally, an ensemble of models with alternative assumptions where uncertainty exists would provide information on model sensitivity. The model component for the action of antimalarial drugs was compatible with our within-host model. The drugs are assumed to act on the infection, either clearing or sparing it. This model would be unable to account for observed effects such as density-dependent cure rate or effects of acquired immunity. More sophisticated models for within-host parasite dynamics and drug concentrations are in preparation. We used DALYs as an aggregate measure to minimise the number of predictions presented. They are an imperfect measure and depend on value judgements for the disability weighting, discounting and age-weighting and on the life table used. We followed standard practices, and calculated the DALYs with no age weighting recognizing that there is a lack of consensus on this issue. The simulations assume a low coverage of case-management and no other interventions. This is a common approach which measures the ‘full’ impact of an intervention and offers consistency when comparing ICERs. However, we recognise decision makers may already have a variety of interventions in operation and want to know the incremental benefit of changing their existing status quo. Model components for other interventions are under development and these predictions for IPT contribute to a growing database of the likely effectiveness of different malaria control strategies generated using a common simulation platform. When considering IPTc or IPTi for a specific location, both the local characteristics and issues other than epidemiological impact and cost- effectiveness should be considered. This study does not address issues of affordability nor of safety, development of drug resistance, first line treatment drug choice, sustainability or malaria species other than *P. falciparum*. In conclusion, modelling can extend the information available to policy-makers by providing predictions of the likely impact and cost-effectiveness for settings, for outcomes and for multiple strategies where, for practical reasons, trials cannot be carried out. Our predictions indicate that both IPTi and IPTc can be cost-effective interventions in a range of settings. This cost- effectiveness is driven by low delivery costs and the predicted impact on mortality. Both IPTi and IPTc could be rendered cost-ineffective by very low transmission, mis-timed seasonal delivery, ineffective drugs, very high treatment coverage or combinations of these factors. Seasonal delivery is more cost-effective in seasonal settings and year-round in constant transmission settings, the difference is more pronounced for IPTc than IPTi. Predictions suggested that the optimum target age group for IPT in settings with a very low transmission intensity would include children over five years. We are grateful for comments from Rick Steketee, Alan Brooks, the members of the Technical Advisory Group (Michelle Gatton, Azra Ghani, Simon Hay, Steve Lindsay, Kevin Marsh, David Schellenberg) and two anonymous reviewers. We thank the many volunteers who donated computing power via [malariacontrol.net](http://malariacontrol.net/). We are grateful to Alain Studer, Diggory Hardy, Aurelio Di Pasquale and Michael Tarantino for software development. [^1]: Developed the software architecture: NM. Conceived and designed the experiments: AR LC. Performed the experiments: AR NM. Wrote the paper: AR NM ES TS LC. [^2]: The authors have declared that no competing interests exist.

# Introduction Sex determination (SD) establishes the sexual fate of an organism and initiates the gonad differentiation process (reviews:). A variety of signals, including genetic, environmental or even social cues, were found to be sex determinants in vertebrates (see reviews). The most extensively studied mode of genetic SD is chromosomal sex determination (CSD) as in mammalian and avian species, for example. In this system, sex is determined by a primary switch located on one or both members of a well- differentiated sex chromosomal pair (see e.g.,). Since in mammals, including humans, sex is determined by CSD, the vast majority of our knowledge on the molecular regulation of vertebrate sex is based on data collected from such systems. In the other type of genetic sex determination system, called polygenic (multigenic or multifactorial) sex determination (PGSD), the genes with strong influence on sex determination and/or gonad differentiation are distributed throughout the genome and the combination of their alleles determines the sex of the individual. This form of sex determination has not been studied extensively at the experimental level (for exceptions, see e.g.): European seabass and a handful of cichlid species from Lake Malawi are the only fish species that were shown to utilise this system to date. Sex can also be determined by signals from the environment and there are several environmental effects known to influence sex of an organism. Temperature is one of the most commonly studied environmental cues for sex determination. In many reptiles, sex is determined by environmental temperature during the thermosensitive periods of embryo development or egg incubation (for reviews see). Examples of other environmental cues that also have an influence on sex include pH in guppy and social interactions in some reef fishes. Over the past decades, zebrafish (*Danio rerio*) has become an important laboratory model organism for many areas of research (for examples see e.g.:). Despite being a popular model for developmental biology and biomedical research, very little is known about its sexual development (for review see). Moreover, most of the current knowledge on zebrafish sexual development is related to its gonad differentiation (for reviews see) while the mode of its sex determination is still disputed. Most cytogenetic studies showed that the zebrafish has chromosomes of similar size and morphology. This lack of distinct morphological differences together with poor karyotype banding pattern resulted in difficulties with accurately assigning chromosomal pairs (for review see). Therefore, it is not easy to search for sex chromosomes based on size differences, their distinct trademark in most mammalian and avian species. An alternative approach to cytogenetic approaches would be to search for differences between the two sexes at the level of the whole genome. PCR-based methods such as random amplified polymorphic DNA (RAPD;) and amplified fragment length polymorphism (AFLP;) have been used successfully for identification of sex markers in fishes and other vertebrates. Earlier, we have developed a new PCR-based mass genotyping technique called fluorescent motif enhanced polymorphism (FluoMEP;) that combines the advantages of RAPD and AFLP. In this study, we have utilized this technology to search for sex-linked DNA markers in the genome of three fish species, including zebrafish. Another molecular tool used for comparing the male and female zebrafish genome in this study is array comparative genome hybridization (aCGH). This method allows for the detection of differences, called copy number variations (CNV), between two complex DNA samples on a genome-wide scale. The method is based on hybridization of two samples onto a ‘tiling array’ that contains probes scanning through the whole genome at regular intervals. Originally, aCGH was developed for the analysis of chromosomes aberrations in cancer cells. Over the years, this method has also been utilized for various purposes such as studying evolution, , understanding the impact of CNV on transcriptome and isolation of molecular markers. The aim of this study was to perform a detailed analysis on zebrafish sex determination, by combining the power of traditional and molecular technologies. Through analysis of sex ratios in a large number of families, we show that i) sex ratios vary among different families; ii) parental genotypes have a major effect on the sex ratio; and iii) one of the two sexes can be depleted through systematic selection in a few generations. Moreover, PCR-based screens and aCGH performed by a custom-designed tiling array were both unable to find general differences between the genome of the two sexes in two different zebrafish strains. The above data all point towards a genetic mechanism of sex determination and the lack of a chromosomal sex determination system in the zebrafish. We, therefore, propose that zebrafish sex determination is polygenic. # Materials and Methods ## Fish stocks and tail fin samples Experiments performed at Temasek Life Sciences Laboratory were approved by Temasek Life Sciences Laboratory Institutional Animal Care and Use Committee (approval ID: TLL(F)-10-001) and performed according to its guidelines. Experiments performed at Max-Planck-Institut für Entwicklungsbiologie were registered at Regierungspräsidium Tübingen (approval ID 35/9185.46) and carried out according to the Protection of Animals Act (Tierschutzgesetz) and its guidelines. Zebrafish (*Danio rerio*) of the AB strain, Tübingen strain and a wild type strain, called Toh, purchased from a local aquarium shop were used in this study. All zebrafish were kept in AHAB (Aquatic Habitats, Apopka, FL, USA) recirculation systems according to standard protocols, with the exception of the population density study in which fish were raised in an Aqua Schwartz system. Guppy (*Poecilia reticulata*) fin clips from visually sexed individuals were kind gifts from Dr. Rob Brooks (UNSW, Sydney, Australia). Visually sexed rosy barb (*Puntius conchonius*) individuals were purchased from a local fish trading company (Qian Hu Fish Farm, Singapore). Their tail fin samples were collected under anesthesia and stored in absolute ethanol at −20°C until use. ## Fish husbandry Adult zebrafish were kept as mixed sex groups in 2.75 L tanks at a density of \<10 individuals per liter. Breeding was carried out in meshed-bottom mouse cages of one liter volume placed into a second cage containing egg water. Breeding pairs were set up at the previous evening in the presence of artificial plants and eggs were collected before noon the next day. Pairs that were reluctant to yield eggs were given a slight cold shock by adding ice-cold egg water (about 20% of the tank volume) 1–2 hours after the start of breeding period. Ripe females that failed to produce eggs with two different males were gently squeezed to aid the removal of eggs, if any, potentially ‘stuck’ in their body and set up for repeated mating one week later. Fertilized eggs were collected from the bottom of the cage, rinsed on a tea filter and transferred into plastic trays with egg water containing methylene blue. Survivals were recorded at 24 and 48 days post fertilization (dpf). Batches with survival below 50% during this period were discarded and their parents were crossed again later. Embryos were transferred onto the AHAB system before hatching and they were grown there at the following densities (unless indicated otherwise): \<100/L for embryos, \<80/L for larvae, \<20/L for juveniles and \<10/L for young adults. ## Sexing zebrafish Zebrafish were sexed visually, based on the following two criteria (unless otherwise noted): i) general body shape; and ii) the presence of ‘genital papilla’ (or cloacal protrusion;) in females (observed on unstressed fish kept in water). Individuals with intermediate body shape and poorly observable papilla were gently squeezed and checked for eggs or sperm. In absence of either, individuals were culled, dissected under a stereo microscope and their gonad was analyzed. Those individuals with unclear sex were not included in the calculation of sex ratio. Only 7 out of the 62 families sexed had such individuals and their ratio was typically less than 5%. ## Selection experiment for increased sex bias We have performed a multi-generation selection experiment in order to increase sex bias. Based on the sex ratio of the offspring we have chosen five lines to be used for selection against males or females. Pairwise full-sib crosses were performed according to a multifactorial design. The offspring were sexed at about 3 months of age and family sex ratio was recorded. From crosses that produced highly skewed sex ratio, usually three robust males and three robust females were chosen as brooders to produce the next generation. ## Population density effects on sex ratio Rearing density experiments were performed using the Tübingen strain. Embryos and larva were raised at 29°C in petri dishes until 5 dpf at which time they were transferred to 1.5 L of fish water at varying densities. At 10 dpf, 600 ml of fish water was added to each tank to facilitate counting of larvae. At approximately 14 dpf, larval were placed into circulating water, resulting in 2 L of water per tank. From 5 dpf to about 14 dpf, larva were fed powdered fry food two times daily after which time the food source was changed to freshly hatched *Artemia nauplia*. Two experiments were performed to test the effect of rearing density on sex ratios. In experiment 1, a total of 44 populations were analyzed spanning a period of about four months. In experiment 2, 30 populations were analyzed spanning about six and a half months. Embryos were collected from pairwise matings on day 0. In experiment 1, embryos from 2 or 3 crosses were often pooled, however more than one pool was often collected per day. Embryos were then sorted into petri dishes containing 50 embryos. In the second experiment, all embryos collected on a given day were pooled before sorting into dishes containing 50 embryos each. On day 5 dpf, larvae were set out at densities of 100, 50 or 25 larvae in 1.5 L of fish water. In experiment 2, the same number of tanks per density were set up out on a given day (e.g. 2 tanks with 100 larvae, 2 tanks with 50 larvae and 2 tanks with 25 larvae) whereas in experiment 1 the number of tanks per density per day was not controlled. Overall, the difference in the design of the two experiments should have resulted in a lesser degree of genetic diversity between populations in experiment 2 compared to experiment 1. Larva and juvenile fish were counted every 10 days from 10 dpf up to 30 or 40 dpf for most. In initial experiments, little to no lethality was observed after 30 dpf thus, in experiment 2 counting ceased after 30 dpf for most populations. All fish were raised to adulthood and then sexed. For the first 30 populations, fish were sexed by dissection and observation of the gonad and subsequent populations were sexed based on coloration. ## FluoMEP assay Genomic DNA (gDNA) samples were extracted from tail fins by digesting them at 55°C overnight in 800 µl of SET buffer (0.5% SDS, 50 mM EDTA, 10 mM Tris/Cl pH 8.0, 200 mM NaCl) and 250 µg/ml Proteinase K (Roche Diagnostics, Indianapolis, IN, USA). Then standard phenol chloroform extraction was performed and the gDNA pellet was dissolved in 100 µL of 1× TE. Pooled male and female samples were generated by combining equal quantity of individual gDNA samples (nine individuals of rosy barb, eight individuals of guppy and four individuals of zebrafish for each sex). FluoMEP screening was carried out as described previously. Bulk segregant analysis was performed using the male and female pooled gDNA samples to screen for potentially sex-linked markers. Potential markers were then subjected to an additional round of analysis on the individuals that formed the pooled samples for confirmation. ## Array comparative genomic hybridization Four families of zebrafish were used for aCGH and each family consisted of the parents, two male offspring and two female offspring individuals. Two of the families were from the AB strain and the other two were from the Toh strain. Genomic DNA samples for aCGH were extracted from tail fins using DNeasy Blood and Tissue kit (Qiagen) according to the manufacturer's instructions with slight modification. Instead of 2 hours incubation, tail fin samples were incubated overnight at 55°C in lysis buffer AL and proteinase K with slow shaking (70 rpm). The quality of extracted gDNA samples was checked on Nanodrop 1000 (Thermo Scientific, Waltham, MA, USA) and by agarose gel electrophoresis. Individual samples were labelled with NimbleGen Dual-Color DNA Labelling Kit (Roche NimbleGen, Madison, WI, USA) and hybridization was carried out according to the manufacturer's instructions with MAUI hybridization system (BioMicro Systems, Salt Lake City, UT, USA). The oligo array was custom-designed by NimbleGen (Roche NimbleGen) based on zebrafish Zv7 (danre5) genome assembly. During the course of this study Zv8 (danre6) was released, all probes were re- mapped onto the new assembly for data analysis. Each array contained 120 thousand probes (55–70mers) with median spacing of about 10 kb. As preliminary tests have confirmed the accuracy of our procedure, no technical replicates were used for reasons of cost-efficiency. The array was scanned at 5 µm resolution with Axon GenePix 4000B Microarray scanner (Molecular Devices, Sunnyvale, CA, USA). Raw fluorescent intensity data was retrieved by NimbleScan software (Roche NimbleGen) then imported into Partek Genomic Suite software (Partek Incorporated, St. Louis, MO, USA) for analysis. For copy number detection the genomic segmentation algorithm was used. A minimum of 5 markers were specified, *P* value threshold set at 0.001 and signal-to- noise ratio set at 0.3. All data was collected according to MIAME guidelines and deposited in NCBI GEO (GSE34338). ## Validation of aCGH results For validation, we used the same gDNA samples that were analyzed by aCGH, plus an additional male and female offspring per family were included. All PCR primers (see for primer sequences) were designed to target a sub-region of the copy number variable region (CNVR) using Primer3 version 0.4.0. Quantifast Probe PCR kit (Qiagen) was used for PCR validation of CNVR2 and CNVR5. The amplified products were then analysed by 2% agarose gel. A single-copy exon (DrSC23) was used as reference. For validation of CNVR3, real time quantitative PCR was carried out as described previously using MyIQ real-time PCR detection system (Biorad, Hercules, CA, USA) with IQ Sybr Green Supermix (Biorad). Samples were normalized against two reference loci (DrSC19 and DrSC23), both of which were found to be a single-copy exon. Relative quantification was calculated to estimate gain or loss of copy number with reference to paternal gDNA sample. # Results ## Wide-ranging sex ratios among zebrafish families The classical method to determine if a species is using chromosomal sex determination system is to analyze the sex ratio among many families. In the presence of strong CSD, the sex ratio is expected to be close to 50%. In order to elucidate whether CSD is the main sex determination system in zebrafish, the sex ratios of 62 families were analyzed. The percentage of males among the families analyzed ranged from 4.8% to 97.3% with median of 51% (std. dev. ±22.6%;). Such a wide-ranging sex ratio among the families would be highly unusual for a predominantly sex chromosomal system. ## Skewed family sex ratios are very likely due to genetic factors In order to investigate the potential reason for skewed family sex ratios repeated single pair crossing was carried out. Nineteen breeding pairs were crossed twice on different occasions and their offspring were raised at similar, but not identical conditions (i.e. ambient water temperature ranging between 27–29°C, variable densities and amount feed). Based on the sex ratio of the first set of clutches, the breeding pairs were divided into three groups: female-biased group (ten pairs; less than 40% males in their offspring), unbiased group (three pairs; 40–60% males in their offspring) and male-biased group (six pairs; more than 60% males in their offspring). In the female-biased group, the difference in mean male percentages for the first and second batches was 1.3%. The biggest difference produced by a female-biased pair was 15.4% (mating pair 16). Three pairs were assigned to the unbiased group and the mean difference between their 1^st and 2^nd cross was 6%. In the male-biased group, mating pair 6 showed an unusually big, 25.2% drop in the sex ratio (from 82.9% to 57.7% males) that was 1.6 fold higher than the second highest change and 3.7 fold higher than the mean of the rest. We decided to remove this pair from the comparison and used data for the remaining five pairs only, where the mean difference in the male percentages for the first and second batches was 0.2%. The biggest difference produced by a male-biased pair was 15.1% (mating pair 3). Overall, we observed very similar offspring sex ratios between the first and second crosses from the same breeding pair indicated by the high R² value of 0.8985. The fact that sex ratios of different batches of offspring from the same breeding pair were very similar suggests that sex in zebrafish is heritable, whereas wide-ranging sex ratios across the families point towards a complex genetic trait. ## Enhancement and maintenance of sex-biased lines through multiple generations by full-sib selective breeding This experiment was carried out with the aim to determine if sex-biased ratios in lines can be maintained or increased by selecting for breeding pairs that produced brood with highly skewed sex ratios through several generations. A total of 5 lines were established and followed through two to four generations. Several lines were split into sub-lines that were later split further depending on the sex ratios resulting from the multifactorial crosses. Altogether, offspring from 26 fourth generation families were grown to maturity and sexed. In two families, we managed to generate an all-male offspring in the F3 generation, whereas our efforts to generate all-female offspring were unsuccessful. Here, we describe two male-biased families that we managed to maintain for two generations through selection from a single line and split in the third generation. The F₃ mean sex ratios of the two families were 96.8% (family D8F3_1) and 93.3% (family D8F3_2) males. We then analyzed the family sex ratio variation for each generation by calculating coefficients of variation (CV). It was observed that after selection the family sex ratio CV decreased at least two-folds when compared to the F₀ generation. To verify that the decrease was due to selection pressure, we performed a control experiment with the D4F3 family by doing a mass cross for the F₂ generation without any selection. The control family sex ratio CV for the F₀ generation was 21.96% and after selection the F₁ generation family sex ratio CV was 7.2%, about three-fold lower. However, after F₂ mass cross the family sex ratio CV of F₃ increased to 25.36%. The “bouncing back" of F₃ family sex ratio CV to a level similar to F₀ indicates that selection pressure was indeed maintaining the highly skewed family sex ratio. This indicates that zebrafish sex is a genetic trait and the fact that we were able to keep highly skewed sex ratios - and even eliminate one of the two sexes in some cases - suggests the absence of a strong effect by sex chromosomes on sex determination. ## FluoMEP assay identified sex-linked DNA markers from guppy and rosy barb, but not from zebrafish FluoMEP assay was used to search for DNA markers tightly associated with sex from three different fish species' genomes. The first species was the guppy (*Poecilia reticulata*) which has XX/XY sex chromosomes. Altogether, 144 different primer combinations utilizing the same common motif primer were tested. They yielded three male-specific sex markers , that showed 100% agreement with phenotypic sex in eight individuals tested (data not shown). Next, we screened the genome of rosy barb (*Puntius conchonius*) that is also known to have XX/XY sex chromosomes , with 386 primer combinations (based on two common primers) and obtained two male-specific sex markers. When tested on eight individuals, the sexing efficiency of the two markers was also 100% (data not shown). These results demonstrated that FluoMEP is able to isolate sex-linked DNA markers from fish genomes with substantial differences between the male and female genomes. In order to search for sex-linked DNA markers in zebrafish, we used a total of 258 FluoMEP primer combinations (based on 29 common primers) to screen pooled male and female zebrafish genomic DNA samples. However, no sex-linked DNA marker was found suggesting that there are no substantial differences between zebrafish male and female genomes. ## No universal sex-linked CNV was detected in four zebrafish families by aCGH We continued our investigation for sex-linked differences at the genome level by aCGH. We used a custom-designed oligonucleotide microarray containing 120,000 probes covering the assembled zebrafish genome (Zv7). By testing samples from two families each of the AB and Toh strains, a total of 255 CNV regions (CNVRs) were detected. Among them, 64 CNVRs were present in both strains, 105 were unique to the Toh strain and 86 were present only in the AB strain. Five CNVRs were common to all the four families screened. As we expected that a sex- determining chromosomal region would be present in all strains, we analyzed the five common CNVRs on individuals, but none of them turned out to be inherited in a sex-linked pattern. Additional five CNVRs showed apparent family-specific sex linkage and were further analyzed by PCR-based methods with additional two offspring individuals (one male and one female) from the same families. Multiple primers were designed for the first two CNVRs, but failed to yield a PCR product, presumably due to differences between the Zv7 genome assembly used for the probe design and Zv8 used for the analysis of results. Two of the remaining three CNVRs were found not to be sex-linked by PCR, while the last one was found not to be sex-specific by real time quantitative PCR (qPCR;). In fact, none of the additional offspring individuals analyzed did show sex-linked inheritance pattern for any of these three markers. Therefore, we concluded that no family-specific, sex- linked CNVR was identified from the four zebrafish families analyzed. ## Rearing density has a limited effect on sex ratios in zebrafish Density has been shown to affect sex ratios in some fish species (see for review). To ask whether rearing density influences sex determination in zebrafish, we raised groups of zebrafish at three different densities from 5 dpf to adulthood and assayed the sex ratios of the resulting adults. The three groups had starting densities of 100, 50, and 25 larvae per 1.5 liters of water, respectively. Two independent experiments were performed that varied slightly in their design yet resulted in similar outcomes with respect to relative sex ratios across different rearing densities. However, in each experiment, we observed wide-ranging sex ratios at all the tested starting densities. This profile was similar to that observed in the breeding experiment. We found that zebrafish reared at high density had approximately twenty per cent more males on average than those raised at middle or low densities indicating a modest effect of large differences in rearing density on sex determination in zebrafish. As the time window in which sex determination occurs is not well defined, we wanted to account for potential changes in population densities due to larval or juvenile death in the above experiments. The number of fish was counted every 10 days beginning on day 10 post fertilization. The highest degree of lethality was typically between 10 and 20 dpf, which corresponded to the period at which both food and water regimes were altered. After 20 dpf, limited loss was observed in most populations and after 30 dpf most individuals survived. Despite some larval lethality, the average percentage of dead fish in each density group from experiment 2 did not differ significantly indicating that larval death did not contribute to the observed higher percentage of males in the high density group. # Discussion ## Molecular and breeding data suggest a genetic sex determination system without a predominant sex chromosome in zebrafish Although zebrafish has become one of the prime vertebrate models for developmental biology, its sex determination mechanism is still unknown. Therefore, the primary aim of this project was to find out more about the sex determination of this species. The first question we asked was: does zebrafish use a chromosomal sex determination system? So far, several cytogenetic analyses were performed on zebrafish karyotypes to search for a size-heteromorphic chromosomal pair, which is a hallmark of CSD with highly differentiated sex chromosomes. However, the accurate assignment of chromosomal pairs is hampered by lack of substantial size differences among the zebrafish chromosomes and their poor staining. Ten teams reported the lack of a heteromorphic chromosomal pair in the zebrafish karyotype, while only two publications described the presence of such a pair. Researchers have also tried to look for sex chromosomes in the zebrafish genome by searching for sex bivalent synaptonemal complexes and performing comparative genomic hybridization (CGH) between male and female gDNAs. Negative results from both latter studies – together with the vast majority of cytogenetic data - suggest that zebrafish does not possess heteromorphic sex chromosomes. Our breeding data also indicate the absence of chromosomal sex determination in zebrafish, whereby the inheritance of a particular chromosome would be the predominant determiner of sex. We observed variable family sex ratios from 62 clutches of offspring from different breeding pairs. Broods from species that have strong chromosomal sex determination system typically exhibit a narrow range of family sex ratios that do not divert substantially from 1∶1 (male to female; e.g. Nile tilapia and rainbow trout). Moreover, we were able to obtain several strongly male-biased zebrafish families by selective crossing of brooders that produced higher proportion of male offspring over a few generations. In the chromosomal sex determination system, the chance for the occurrence of such sex-biased families would be very low, because the ratio of male and female would tend to ‘bounce back’ close to 1∶1 in the next generation. These breeding data also indicate that zebrafish sex determination is unlikely to be based primarily on sex chromosomes. To further prove the absence of chromosomal sex determination system, we screened the zebrafish genome for sex-linked differences with molecular tools. The first experiments we performed were a series of comparative FluoMEP assays. After screening through 258 FluoMEP primer combinations, no confirmed sex marker was obtained from zebrafish. On the other hand, sex markers for guppy and rosy barb were detected by using the same method. The latter data prove that the FluoMEP assay is suitable for isolating sex-linked markers from genomes known to contain heteromorphic sex chromosomes. The fact that we were unable to obtain sex-linked markers from zebrafish with the same method provides an additional indication that no substantial differences exist between the male and female genomes. Even if there are sex chromosomes in zebrafish, they will be likely showing limited differences at sequence level and therefore undergo recombination with each other along the majority of their length. In this case, the identification of such sex chromosomes through the analysis of pools generated based on phenotypes would be extremely difficult. Next, we performed array comparative genomic hybridization (aCGH) on four families of zebrafish. Through the analysis of 120 thousand genomic locations, a total of 255 CNVRs were observed and most showed a pattern of Mendelian inheritance. However, no universal sex-linked CNVR was found among the four families of zebrafish tested. The aCGH results suggest that the possibility for highly differentiated, heteromorphic sex chromosomes in the zebrafish genome is quite low. The caveat of our current aCGH approach is that the assembly of zebrafish genome (Zv8) on the basis of which the probes were analyzed managed to assemble only about 89% of the total sequences obtained. Therefore, there is still a possibility that there are sex-linked CNVRs “hiding" in the remaining 11% of the genome. Furthermore, the probes present on the custom-made oligo array have a median spacing of 10 kb intervals and by setting the window of detection to 5 probes per window allows for a resolution of around 50 kb. This means that any genomic difference with less than 50 kb in length will not be picked up by our aCGH approach. However, we argue that the size difference for most active sex chromosomal pairs will likely exceed 50 kb in length, as in case of the medaka, the only known SD region described from teleosts so far. Recently a high resolution zebrafish CNV map was published by analyzing 80 genomes with 1.4 kb probe spacing CNV array. Although it has higher resolution than our CNV array the study did not performed comparative analysis of the male and female genome. Considering the combined data of the FluoMEP and aCGH approaches, the majority of the (assembled) zebrafish genome was probed for sex-linked sequences in this study. Although we still cannot completely rule out the presence of a sex chromosomal pair, it is unlikely that a single predominant sex-determining region exists. As our breeding data also do not support the presence of a sex chromosome, we provide a strong case against CSD in zebrafish. Very recently, a genome-wide association study was performed for the identification of sex determining regions with a SNP array containing over 5,300 features. The authors reported two regions on two separate chromosomes (Chr5 and Chr16) accounting for 16% variance of the trait, providing a direct experimental evidence for a polygenic sex determination system in the zebrafish. These data further strengthen the notion that zebrafish sex is not determined by a sex chromosomal pair. Our data and results from vast majority of the above studies contradict a recent suggestion that zebrafish has a female dominant (ZZ/ZW) sex determination system. Although the results described in that publication seem to support the possibility of a ZZ/ZW sex chromosome system, their data do not conclusively demonstrate that this mode of sex determination is actually in place. Attempts to identify the genetic factor(s) regulating sex or the proposed sex chromosomes were not made in their study. ## Zebrafish sex is determined genetically Since molecular and breeding studies failed to identify heteromorphic sex chromosomes or their effect, we next sought to find out if genetic factors are involved in zebrafish sex determination. We performed repeated single pair mating in which 19 randomly selected breeding pairs were bred twice. The environmental factors such as ambient temperature, amount of food given and rearing density were not tightly controlled. Even so, broods derived from the same breeding pair did not exhibit major sex ratio differences between repeated crossings of 18 out of 19 breeding pairs tested. This indicates that the wide- ranging sex ratios normally observed are most likely due to the parental genotypes. In addition, we showed that sex ratio variation decreases substantially under selective pressure, a strong indication that sex is a genetic trait. Another interesting phenomenon we observed was that after three generations of selection we were able to obtain two all-male families while attempts to produce all-female families were unsuccessful. We do not have an explanation for this difference and we propose that further investigations are needed to elucidate the underlying reasons. Nevertheless, our data show that zebrafish uses primarily genetic sex determination system. As we have also demonstrated that CSD is not likely the mode of sex determination in zebrafish, we propose that a PGSD is in place. Based on our data and the recent aforementioned association study, we propose that the number of genes contributing to the sex determination process might be far more than just a handful. ## Polygenic sex determination might be more common among vertebrates than expected The vast majority of our knowledge about vertebrate sex determination was obtained from species using sex chromosomal systems. On the other hand, over 90% of the fish species analyzed through karyotyping does not show the presence of differentiated sex chromosomes (see for review). Recently, it was proposed that multiple parallel sex determining pathways are likely to operate in species with CSD and this mode of SD could be extended to species with PGSD as well. In such scenario, both systems might have a more similar regulation than expected, differing only in the location of the factors: all would map onto the sex chromosomes in CSD, whereas in PGSD some (or all) of them would be located on the autosomes. Therefore, analysis of zebrafish and other fish species utilizing the PGSD system could be important for basic research and potentially useful for aquaculture projects as well. ## Environmental factors have limited influence on zebrafish sex ratio Temperature is the most commonly studied environmental cue for sex determination. It is utilized by many reptile species, and some fish species. In animals with temperature-based sex determination (TSD), substantial fluctuations in the environmental temperature will likely cause significant changes in the offspring sex ratio. Two papers reported that the temperature at natural habitat of zebrafish ranges from 26 to 38°C. However, it is believed that 26 to 29°C is the temperature range for normal zebrafish development and rearing them within this range did not result in significant sex ratio changes. It was also observed that exposure to increased temperature (35–37°C) either during early development (5–48 hpf) or between 17–27 dpf resulted in male-biased sex ratio. On the other hand, at our laboratories we observed high mortality if zebrafish larvae were grown at 37°C from the beginning. Therefore, temperature is unlikely to be the primary signal for zebrafish sex determination, but might exert secondary effects on its sexual development. Rearing density is another environmental cue known to influence sex ratio of some fish species such as the American eel. The exact underlying mechanisms of how rearing density directs sexual development are still unknown. We have tested the effect of rearing density on zebrafish sex, and found a substantial increase of males at high density (100 individuals per 1.5 liters of water). In another study, slow growth rate as a result of limited food supply - usually experienced at high rearing density - had been suggested to influence zebrafish sex differentiation leading to higher percentage of males. Nonetheless, we think that rearing density is unlikely to be the primary determinant for zebrafish sex, as we observed wide ranging sex ratios at all three densities tested. A strong determinant should produce broods of very similar sex ratio. In addition, the response to these environmental factors seems to differ between families indicating that influence of rearing density on sex ratio is most likely conferred by the genotype of the fish. Another environmental factor that is known to have an effect on zebrafish sex ratio is oxygen level. It was found that under hypoxic conditions there was a reduction of estrogen synthesis leading to an increase of androgen to estrogen ratio which favors male development. However, the decreased oxygen level had only limited effect on the sex ratio of zebrafish leading to higher percentage of males (12.5% differences). This is unlikely the cause of wide ranging sex ratio observed in the zebrafish. Published data and our results both seem to suggest that non-extreme environmental factors do not have a major effect on zebrafish sex ratio. No drastic change in sex ratio upon environmental effects experienced at the natural surroundings of the species was observed in any of the studies. Furthermore, response to environmental factors varies among the treatment groups. This indicates that the underlying genotype of each individual is directing sexual development in response to environmental stimulus. ## Conclusions For this study, we performed classical breeding experiments together with large- scale genomic analyses to show that zebrafish sex is determined genetically with no sign of a chromosomal sex determination system. The characteristics of sex ratios observed in zebrafish were as follows: i) wide variation among different families; ii) strong influence from parental genotypes; and iii) the ability to eliminate one of the sexes by selection. All these features point toward a species without a predominant chromosomal sex determination system. Our in-depth investigation by molecular tools (i.e. FluoMEP and aCGH) also failed to identify any difference between the male and female genomes. Several studies, which investigated environmental impacts on zebrafish sex ratio, found them either to result in limited change or show strong effects outside of the physiological range of the species. This cannot account for the wide-ranging sex ratios among families; hence we reckon that zebrafish does not use a primary environmental sex determination system. Taken together, the above data indicate either the lack of sex chromosomes in zebrafish or the presence of very weak ones that are frequently over-ridden by strong modifier genes. In our opinion, these two situations are principally the same, as there are several genes distributed throughout the genome with major effects on sexual development in both; therefore we propose that zebrafish sex determination should be considered polygenic. Earlier, others have indicated the possibility of a polygenic sex determination system for zebrafish based on a single set of experiment each (see e.g. &). Our study adds data obtained by four different methods that all point to a polygenic sex determination system, creating a tipping point in this argument. # Supporting Information We thank Prof Christiane Nüsslein-Volhard for her support of the project, Dr Rob Brooks for guppy samples, Dr Alex Chang and Qian Hu Fish Farm for helping to obtain the rosy barb samples, Miss Rashmi Sukumaran for developing the FluoMEP marker screening program, TLL fish facility for taking care of the zebrafish and TLL DNA sequencing facility for performing the FluoMEP electrophoretic analysis. [^1]: Conceived and designed the experiments: WCL RB KRS LO. Performed the experiments: WCL RB ZL RS KRS LO. Analyzed the data: WCL RB ZL KRS. Wrote the paper: WCL KRS LO. [^2]: Current address: Department of Molecular Biology, Nijmegen Center for Molecular Life Sciences, Nijmegen, Gelderland, The Netherlands [^3]: Current address: Molecular Genetics and Development Division, Prince Henry's Institute of Medical Research, Melbourne, Victoria, Australia [^4]: Current address: Department of Orthopedic Research Laboratories, Children's Hospital Boston, Boston, Massachusetts, United States of America [^5]: The corresponding author has served as an Academic Editor of PLoS ONE since 2006. This does not alter the authors' adherence to all the PLoS ONE policies on sharing data and materials. All other authors have declared that no competing interests exist.

# Introduction Tropical lowland rainforests shrink worldwide rapidly due to intensive ongoing deforestation. In Sumatra (Indonesia), lowland rainforest was cut massively in the 1970s and 1980s and transformed into rubber and oil palm plantations, leaving only few remnants of natural forest, which are predominantly located in national parks. Beside monocultures of oil palm (*Elaeis guineensis*), and rubber (*Hevea brasiliensis*), an extensive management system of rubber trees planted into rainforest, also called “jungle rubber”, was established in the early 20^th century. The understory vegetation of such land-use systems is usually rapidly colonized by herbaceous weeds. Alien species rapidly establish populations and may influence the native flora (“invasiveness” sensu), but also native species can colonize novel anthropogeneous habitats in which they were not present before (“colonizers” sensu). Displacement of native biota, change of ecosystems, environmental disturbance and hybridization with native species are the major threats of invasive plants to the maintenance of tropical ecosystems. Invasion history is an important factor influencing population genetic structure and diversity. Many studies compared genetic diversity between the native and the invasive range, but often failed to find differences between native and invasive areas. In the invaded area, species often show reduced genetic variation, which is in general referred to genetic bottlenecks after colonization because of founder effects and geographical isolation from source populations, especially after long distance dispersal. However, multiple introductions can rapidly increase genetic diversity. After the initial introduction phase and a lag period of genetic adjustment, a phase of spread will rapidly expand the distribution range of the alien species. Subsequently, isolation-by-distance mechanisms may initiate geographical differentiation within the invaded area. Genetic diversity in the invaded area can enhance the adaptive potential to a novel environment. Introduced plants will be exposed to novel stress situations, and will be under selection pressure on adaptive features. Indeed, niche shifts have been documented for many invasive plant species. This adaptation process is also expected to become apparent only after a certain residence time and distribution over larger areas. Standing genetic variation, or novel mutations enable plants to adapt to novel ecological niches and to establish in habitats of the invaded area. Under this aspect, it is expected that populations differentiate genetically according to ecological conditions, because gene loci under selection would differentiate via beneficial mutations. Adaptation may relate to local natural conditions of the invaded area (e.g., climate, soil and bedrock type, or geomorphological structure). On the other hand, the type of plantation and the management of land-use systems can change ecological conditions. For instance, carbon content of soil and degree of erosion differ between land-use systems in Sumatra. Despite the fact that invasive species in land-use systems are widely distributed in the tropics, the information on genetic structure and adaptive potentials of herbaceous alien plants in tropical land-use systems remains insufficient. Polyploidy is another important factor influencing genetic diversity and distribution of invasive plants. Many authors stress the positive effects of polyploidy for invasions (e.g., increased heterozygosity, vigor, life span, seed longevity, and seedling survival;), and indeed polyploids are more frequent than diploids in the invasive plant floras of the temperate zones. However, the information on ploidy levels of tropical invasive species is too leaky to make general conclusions. Uniparental reproduction is another potential factor which makes both asexual and selfing plants pre-adapted to invasions. Apomictic species and selfers can found populations via single individuals and are therefore potentially better colonizers than related outcrossers; this colonization ability is most efficient after long distance-dispersal of seeds (Baker’s law;). Vegetative propagation, in contrast, remains in terrestrial biota usually spatially restricted, but is also an efficient local reproductive strategy of invasive plants. Uniparental reproduction, however, would result in reduced genetic diversity or even clonality, and could reduce adaptive potentials to novel environments. We investigate the model system *Centotheca lappacea* (L.) Desv. (subfamily Panicoideae, family Poaceae), a perennial grass with 30–100 cm long erect culms. Native to west tropical Africa, tropical to temperate Asia, Australia, and the Pacific islands, it is widely used as a forage grass. Although *Centotheca lappacea* was reported from natural rainforests of Thailand or Malaysia, this species grows in Indonesia mainly as a weed in clearings, forest edges and paths, road sides, waste places, cocoa, oil palm and rubber plantations. The species is the most frequent grass in the understory of oil palm and rubber plantations in the investigated regions of the Jambi province in Sumatra, but was observed only once in natural forests in the National Parks of the Jambi region (author’s team, pers. obs.). In contrast to other invasive grasses, the species colonizes not only intensively used monocultures, but also the more natural “jungle rubber” systems. Other than in Sumatra, *C*. *lappacea* was also reported from Kalimantan to dominate oil palm plantations, being used for testing herbicides or from non-natural shrub vegetation in Vietnam. Some authors regard the species in reserved areas as “invasive alien”. However, the species is not yet recorded in the Global Invasive Species Database, and its actual invasive potential needs to be investigated. The first documentation of the species in herbaria in Sumatra dates back to the mid of the 19^th century, but it was regularly collected only from the 1920s onwards (Global Biodiversity Information Facility, GBIF, as of December 2014). This time period coincides with the establishment of rubber (from 1904 onwards) and oil palm plantations in Sumatra (from 1913 onwards;). Thus, after a residence time of c. 100 years, genetic bottlenecks probably have been overcome and first signatures of genetic differentiation and adaptation should be apparent in the introduced area. However, the factors for the wide distribution and abundance of this species were so far unknown, and no population genetic study is yet available for this species. With respect to abundance of *C*. *lappacea* in intensively managed landscapes and disturbed habitats, the species might take advantage of selfing or an asexual reproductive mode, enhancing colonization abilities and invasiveness—a phenomenon well documented for other plant species. Despite reports for a chromosome number 2n = 24, Levy (2002) suggested that the species might be polyploid. Although many related widespread panicoid grasses reproduce asexually via apomixis, no study of mode of reproduction was so far available for *C*. *lappacea*, and the genus was not recorded in the Apomixis database ([www.apomixis.uni-goettingen.de/](http://www.apomixis.uni-goettingen.de/)). The aims of this study are 1) to analyze whether population genetic diversity and degree of divergence in the invaded range would fit to measures of established populations, 2) to test a hypothesis of geographical substructure of populations, 3) to identify candidate loci under natural selection as potential signatures of local adaptation to landscape heterogeneity and/or management types; 4) to assess ploidy level and reproductive strategies of *Centotheca lappacea* as putative factors enhancing invasiveness, and 5) to discuss the general invasive potential of the species. # Material and Methods ## Ethic statement The study was conducted using samples collected based on Collection Permit No. 3924/IT3/PL/2013 recommended by the Indonesian Institute of Sciences (LIPI) and issued by the Ministry of Forestry (PHKA). ## Plant material The sampling sites were established in the frame of the project Collaborative Research Centre CRC990 “Ecological and Socioeconomic Functions of Tropical Lowland Rainforest Transformation Systems (Sumatra, Indonesia)”comprising four replicate sampling sites (50 x 50 m) for each land-use/transformation system (oil palm, rubber plantation, jungle rubber) and for natural forests in two landscapes of Jambi province, Bukit Duabelas and Harapan (<http://www.uni- goettingen.de/en/310995.html>;). In the framework of recording the complete understory vegetation by the CRC990 and investigating the ten most abundant invasive herbaceous species present in these plots, we identified *Centotheca lappacea* as the most abundant grass (according to presence/absence of the species). The species occurred in all 24 land-use plots (in five plots with less than five individuals), but only in one of eight natural forest plots. Leaves, flower bud fixations and seeds of *Centotheca lappacea* were collected during 2013 (May–September) from 19 spatially separated sites, treated here as distinct populations. Individual plants were sampled at a minimum distance of 5–10 m to each other to avoid the sampling of putative clones. We further had to restrict the sampling to flowering or fruiting plants for assessment of mode of reproduction. Our target was to sample at least 10 plants per plot, but the restrictions of a minimum-distance and of flowering/fruiting plants resulted in less than 10 samples in some of the plots. Altogether we collected a total of 173 individuals in 19 land-use plots. ## DNA extraction and AFLP fingerprinting Genomic DNA was extracted from (silica gel dried) leaf material using the DNeasy 96 Plant kit (Qiagen, Hilden, Germany) following manufacturer´s instructions. The total of 173 samples were analyzed using standard AFLP protocols with a slight modification and two selective AFLP primer combinations. The restriction enzymes EcoRI and MseI were incubated with 4μl DNA solution simultaneously overnight. For preselective amplification reaction the primers E01/M03 (A-3’/G-3’) were used and 4μl restriction-ligation solution was added to the PCR reaction. The PCR product was diluted in water and processed with two combinations of selective primers: E35/M63, E32/M66. EcoRI/MseI primer combination E35/M63 contains the three nucleotides ACA/GAA as the specific extension and E32/M66 the nucleotides AAC/GAT at the 3’end of the AFLP primer. Prior to fragment analysis in Genetic Analyzer 3130, ABI PRISM, the PCR product was diluted in water and 12μl HiDi formamide. As a size standard GENSCAN 500 ROX was added to the solution, while the fragments were labeled with the fluorescent FAM marker. Fragment scoring was carried out in the program GeneMarkerV2.4.2 (© SoftGenetics, LLC) and fragments of 50–500 bp in length were scored. AFLP multi- locus genotypes were scored using '0' for the absence and '1' for the presence of a locus. Reproducibility of AFLP profiles was assessed by repeating independent DNA extractions and AFLP amplifications. We repeated 50% of randomly chosen individuals. Only fragments which could be unambiguously recognized as present or absent in the replicated individuals were scored by applying a scoring threshold of 10% of the highest peak’s intensity within the locus under consideration. Samples with poor DNA quality were not considered for the analysis. The final binary matrix comprised 173 individuals from 19 populations scored at 170 loci. ## Estimation of genetic diversity Genetic diversity at population level was estimated as follows: 1) number of private bands and number of polymorphic loci per population were calculated using FAMD 1.31, 2) amount of AFLP haplotypes per population was counted using Arlequin 3.5, 3) proportion of polymorphic loci (PLP) at the 5%-level and Nei's gene diversity (H_j; assuming Hardy—Weinberg equilibrium/HWE) were computed with AFLP-SURV 1.0 and 4) average panmictic heterozygosity (H_s; no HWE assumption) was computed with Hickory 1.1. Measures of genetic diversity across populations were compared between Bukit Duabelas and Harapan employing a non-parametric t-test (1000 permutations) in the software PAST 2.17c. Similarity among individual genotypes was visualized via Neighbor- joining tree (incl. bootstrap support values inferred from 1000 replicates) based on standard Simple matching coefficient (SMC) using FAMD 1.31. We compared the overall genetic diversity within each land-use system based on Nei's gene diversity (H_j;) and panmictic heterozygosity (H_s;). Differences of land-use median values were tested by a Kruskal-Wallis test and Mann-Whitney pairwise comparisons (uncorrected and Bonferroni corrected *P*-values) in PAST 2.17c. In the same program, we tested the diversity indices for correlation using Spearman´s ρ coefficient. To check, whether unequal population sizes inflate allelic diversity estimates, we used AFLPdiv and computed band richness within each population after rarefaction to 5 (the smallest population size;). ## Genetic differentiation and loci under selection To analyze genetic differentiation, genetic variation was partitioned and tested by Analysis of Molecular Variance (AMOVA) using Arlequin 3.5. First, total genetic diversity was partitioned among populations (*F*_*ST*) between Bukit Duabelas and Harapan regions. Second, we conducted another AMOVA to infer differentiation among (*F*_*CT*) and within (*F*_*SC*) land- use systems. To detect candidate loci under natural selection departing from a neutral model, we analyzed the dataset of 170 loci by using BayeScan 2.1 and parameters set as default. Nine loci with a threshold probability ≥ 0.99 of being under diversifying selection (decisive evidence,) were identified by the BayeScan algorithm. To get insights into their contribution to geographical differentiation, we analyzed these loci by Principal component analysis (PCA) using the program PAST 2.17c. In order to support results from BayeScan, the divergence outlier analysis (DOA) was conducted with the program Mcheza. Genetic structure of populations from Bukit Duabelas and Harapan was further analyzed using STRUCTURE 2.3.4. The Bayesian analysis based on all 170 loci was performed applying an admixture model, a burn-in of 500,000 generations and a subsequent run length of 700,000 generations, testing values of K (assumed number of genetic populations) between 1 and 10 with 20 replicates per K value. To evaluate the fit of different clustering scenarios (best K) we analyzed the mean log probability, L(K), and the change in the log probability, ΔK. Alternatively to the Bayesian inference, a method for quick mapping of admixture without source samples was employed in the clustering program FLOCK. In contrast to Structure, instead of employing the MCMC algorithm, FLOCK uses an iterative method based on allele frequencies. The datasets from Bukit Duabelas and Harapan were each tested for isolation-by-distance (IBD) by correlating pairwise *F*_*ST* values (Arlequin 3.5) with geographic distances (in km) employing a Mantel test in PASSaGE; the significance was tested after 1000 permutations. Geographic distances inferred from GPS coordinates were computed in Geographic Distance Matrix Generator v1.2.3. ## Ploidy level To assess the chromosome number on a representative subset of plants, a set of ten seeds per three different land-use systems were germinated in soil inside climate growth chambers at 25°C. Three root tips in active growth from each of eight cultivated plants were pretreated with a saturated aqueous solution of α-bromonaphthalene for three hours at room temperature. Selected root tips were fixed for 12–24 hours in three absolute ethanol: one glacial acetic acid and then conserved in 70% ethanol at 4°C. Most of the pre-treated materials were directly hydrolyzed with 1N HCl at 60°C for 10 min and stained with basic fuchsine. Feulgen staining following methods by was performed for chromosome counting. Meristem cells were macerated in a drop of 2% aceto-orcein and then squashed. Cells in mitotic stages were observed under a Leica DM 5500B Microscope in 1000x magnification (Leica Microsystems GmbH, Wetzlar, Germany), the total chromosomes in a cell were counted to define the number of chromosomes and the ploidy level. Representative images were taken with a DFC450C camera (Leica Microsystems GmbH, Wetzlar, Germany). To assess the ploidy level of the whole sampling, flow cytometry was performed on 164 plants by using 0.5 cm² of dry leaves from field collections. As a reference we used fresh leaf tissue from a plant for which previously chromosomes were counted. A single leaf was chopped with a razor blade in a petri dish containing 200 μL Nuclei extraction buffer (Otto 1;). The resulting suspension was filtered through 30-μm mesh Cell Tric disposable filter (Sysmex Partec GmbH, Münster, Germany), and stained with 800 μL staining buffer (Otto 2;). On the next step of analysis we pooled two leaves from different individuals to make the analysis faster. Samples were analyzed using a Partec PA II Flow Cytometer (Sysmex Partec GmbH, Münster, Germany) with the UV-detector operating at 355 nm. Ploidy levels of the leaf tissues were estimated by comparing the sample peak to the standard peak. Approximately 3000 nuclei were measured per sample. Data analysis was performed using PA II’s Partec FloMax software. The mean values of DNA content of the leaves were established to infer the ploidy level of the sample. The coefficient of variation for each sampled peak was around 5% or less. A regression analysis for histogram data of dried leaves was applied to confirm that the mean values of G1 and G2 peaks had a linear correlation (*P* \< 0.05). All statistical analyses were carried out with Statistica software version 10.; StatSoft, Inc. (2011). ## Cyto-embryological analysis To test for sexual versus apomictic embryo sac development, a total of 338 spikelets from 16 different individuals was analyzed by microscopic observation of megasporogenesis and embryo sac development following methods described in Young et al.. Inflorescences that were previously fixed in FAA and stored in ethanol 70% were dehydrated in 100% ethanol for 30 minutes. Afterwards, they were incubated in 300 μl of upgrading series of methyl salicylate (Merck KGaA, Darmstadt, Germany) diluted in ethanol (25%, 50%, 70%, 85%, and 100%) for 30 minutes in each steps. Spikelets were dissected to prepare the ovules and anthers, then ovules and anthers were amounted in methyl salicylate on glass slides. The stages of ovule and anther development were analyzed by using a Leica DM5500B microscope with Nomarski DIC optics in 400x magnification (Leica Microsystems GmbH, Wetzlar, Germany). Images were taken by a DFC450C camera (Leica Microsystems GmbH, Wetzlar, Germany). ## Flow cytometry seed analysis To reconstruct sexual vs. apomictic pathways of seed formation, Flow Cytometric Seed Screen (FCSS) protocols described by and were used to analyze 310 mature seeds. A single seed from each individual was chopped with a razor blade in a petri dish containing 200 μL Nuclei extraction buffer (kit Cystain UV precise P, Sysmex Partec). The resulting suspension was filtered through 30-μm mesh Cell Tric disposable filter (Sysmex Partec GmbH, Münster, Germany), and stained with 800 μL staining buffer (kit Cystain UV precise P, Sysmex Partec). All samples were incubated for 30 sec to 60 sec on ice before measurement. In the next step of analysis we pooled five seeds from the same individual and measured DNA content as described above for leaves. The mean values of DNA content for embryo and endosperm of each measurement were calculated to infer the ploidy levels of embryo to endosperm. The coefficient of variation for each sampled peak was around 5% or less. The rationale for using FCSS is based on the different ratios of embryo to endosperm relative DNA content within seeds as a consequence of fertilization of unreduced (apomictic) versus reduced (sexual) embryo sacs. In the sexual seed, double fertilization leads to the formation of a 2n (2C) embryo and a 3n (3C) endosperm (2: 3 ratio). In the apomictic pathway, fertilization occurs either only in the central cell of the unreduced embryo sac and produces a 2n (2C) parthenogenetic embryo and 5n (5C) or 6n (6C) pseudogamous endosperm (1n or 2n from sperm plus 4n from central cell; 2: 5 or 1: 3 ratio). Alternatively, apomixis can also be fertilization-independent (2n embryo and 4n endosperm; 1: 2 ratio). For flow cytometry data, the mean, minimum and maximum of the histogram peak values, and Pearson product-moment correlation for the mean values were calculated. ## Self-compatibility tests Pollen-pistil self-compatibility was tested by bagging *C*. *lappacea* inflorescences of genotypes from different collection sites. Selected inflorescences were trapped in sulphite paper bags 2–5 days before flowering and harvested between 45 and 60 days after anthesis. Presence of full caryopses was checked with a Leica S6E stereoscope (Leica Microsystems GmbH, Wetzlar, Germany). # Results ## AFLP data ### Genetic diversity and AMOVA In total, 173 individuals of *Centotheca lappacea* from 19 populations were scored and analyzed. Two AFLP primer combinations resulted in 170 clearly scorable and reproducible fragments sized from 53–443 bp, and 74.71% of them were polymorphic. Measures of population genetic diversity (i.e., PLP, H_j, H_s; see values) were intercorrelated and revealed no significant difference between Bukit Duabelas and Harapan (PLP: t = 0.941, *P* = 0.317; H_j: t = 1.634, *P* = 0.128; H_s: t = 0.923, *P* = 0.370). However, rubber land-use systems differed significantly from both jungle rubber and oil palm by lower means of panmictic heterozygosity (H_s; ). Considering all 19 populations, genetic diversity was mostly distributed within populations rather than among populations (overall *F*_*ST* = 0.173). The genetic variation was partitioned to 6.53% between Bukit Duabelas and Harapan and to 16.16% among populations within both regions (AMOVA;). Accordingly, the individuals did not cluster separately with respect to regions in the Neighbor-joining tree. These analyses further separated individual genotypes without any indication of clonality. Populations within land-use systems were moderately differentiated in both Bukit Duabelas (11.72%) and Harapan (19.70%) (AMOVA;), but no significant differentiation among land-use systems was found within regions. For Bukit Duabelas, we obtained a weak but significant positive correlation (*r* = 0.37, *P* = 0.045) between geographic distances and genetic divergence among populations. In Harapan, no significant isolation-by-distance was observed. ### Loci under selection and genetic structure Analysis in BayeScan identified nine non-neutral loci exceeding the threshold for decisive evidence of selection (posterior probability ≥ 0.99), all of them with a diversifying effect. Six out of these nine loci were concordantly depicted by an alternative search algorithm implemented in Mcheza. A principal component analysis (PCA) based on all nine loci from BayeScan shows a tendency of a separation between Bukit Duabelas and Harapan individuals along the first ordination axis (PC1, 30% of variation;). In particular, the loci '228', '246', '135', '99', depicted by both programs, separate between the both regions. The model-based clustering (Structure analysis) recognized five distinct genetic partitions (best K = 5) with pronounced admixture. According to both direct log- likelihood posterior probabilities and ΔK, this assessment was unambiguous and 34.12% of individuals were assigned to a particular partition with a posterior probability ≥ 0.90, supporting strong admixture between the partitions. The absolute number of genetic partitions depicted by Structure was in congruence with an independent estimate given by non-Bayesian method implemented in FLOCK, suggesting the best K ≥ 4, and rejecting the second-best result for ΔK (K = 2) in the STRUCTURE analysis. Four of the five genetic partitions occurred in all 19 populations and differed only in their proportions between Bukit Duabelas and Harapan. One partition occurred in all except for seven Harapan populations (dark grey color). The five genetic partitions were present in all three land- use systems, exhibiting only moderate proportional differences (most pronounced in partition “3”, which is most frequent in rubber and less frequent in jungle rubber;, red color). ## Ploidy level and breeding system of *Centotheca lappacea* The chromosome number of *C*. *lappacea* was 2n = 2x = 24 in all samples investigated. Ploidy levels of *C*. *lappacea* were screened by flow cytometry of leaf material from field collections using fresh leaf material of a plant with known chromosome number as standard. The fresh leaves of 21 diploid samples had a mean value of the first peak fluorescence intensity (G1) 100.37 ± 8.93, with a coefficient of variation (CV%) value less than 5%. DNA content of dried field-collected leaves of a total of 93 samples representing 123 individuals indicated the same ploidy level compared to those of the standard fresh leaves, with a mean value of the G1 peak of 92.01±7.24 (a slightly lower DNA content measurement in silica-gel dried leaves compared to fresh materials is a regular observation in flow cytometric studies and is due to DNA degradation. All samples were confirmed as diploid. There was no indication of polyploidy. Microscopic analysis of ovules indicated obligate sexual reproduction without any evidence for apomixis. *Centotheca lappacea* develops after meiosis a tetrad of megaspores, but only the chalazal (functional) megaspore develops into a 7 celled and 8-nucleated *Polygonum*-type embryo sac (not shown;). No aposporous initials were observed in the ovules. Formation of full caryopses was only observed in open pollinated inflorescences, while in bagging experiments only empty spikelets were recovered from a total of nine inflorescences (and around 5,200 spikelets) from six genotypes. Flow cytometric analysis on the seeds revealed consistently a 2: 3 ratio of embryo to endosperm ploidy , suggesting that they originated sexually from double fertilization of an haploid egg cell and two reduced polar nuclei by haploid sperm nuclei. Altogether the data indicate that *C*. *lappacea* is a sexually reproducing species, and sexual seeds are formed via a *Polygonum*-type ovule development and allogamy due to the presence of a pollen-stigma incompatibility system. # Discussion We provide the first study on putative factors linked with invasiveness of the tropical grass *Centotheca lappacea*. Here, we focused on population genetic diversity, genetic structure, mode of reproduction, and ploidy level in three different agroforest systems in Sumatra. Our population genetic study suggests that genetic diversity is in the invasive grass *C*. *lappacea* in Sumatra mostly distributed within populations (c. 83% of total variation). The *F*_*ST* value (0.173), estimating genetic differentiation among populations, is lower than grand means of plant populations in early successional stages (0.37) and more similar to late successional stages (0.23; values for dominant markers). A low genetic differentiation (*F*_*ST* = 0.182, AFLP data) was also reported for sexual *Solidago canadensis* populations in the invasive range in China, where the species was introduced c. 85 years ago. A scenario of post-colonization selection for a single genotype, as reported for the invasive asexual plant *Chromolaena odorata*, cannot be inferred from our data. Since we do not have any information on genetic diversity from the native areas of *C*. *lappacea*, we cannot reconstruct colonization history with our data alone. However, the distribution of diversity within populations would fit better to a scenario of established outcrossing populations rather than to recently founded populations. Herbarium collections indicate that the species appeared in Indonesia in the 2^nd half of the 19^th century and was regularly sampled from the 1910s and 1920s onwards, which coincides with the onset of jungle rubber and oil palm land-use systems in Sumatra). Accordingly, the species may have already overcome initial genetic bottlenecks and is at present in an advanced phase of establishment and adaptation to environmental heterogeneity in the introduced area. Individuals do neither cluster in populations nor in geographical regions, and strong admixture exists between the genetic partitions that occur in both regions. This lack of geographical structure resembles the situation in other genetically diverse invasive weeds. Population genetic structure of *C*. *lappacea* is in the whole area partitioned into five gene pools, which all occur in both landscapes (Harapan and Bukit Duabelas), but with unequal proportions within the areas. The populations in Harapan were more homogeneous and showed no isolation-by-distance, while in Bukit Duabelas the populations showed a weak but significant geographical substructure, which may relate to a higher landscape heterogeneity in the hillside region of Bukit Duabelas. Altogether, the populations in the Jambi region appear still like a big “metapopulation” with only a weak geographical substructuring and continuous gene flow over the whole area. Interestingly, populations from rubber plantations in both landscapes exhibited a significantly lower gene diversity / heterozygosity compared to jungle rubber and oil palm. Otherwise there was no indication of a genetic differentiation according to land-use systems. We identified nine non-neutral AFLP loci under diversifying selection as identified by BayeScan, which represent 5.3% of all loci investigated here. This result resembles findings on a comparable AFLP dataset of the invasive species *Mikania micrantha*, in which 2.9% of loci were identified as outliers and interpreted as signatures for local adaptation. Our study may have revealed candidate loci for a potential adaptation of *C*. *lappacea* after the time of spread, as predicted by theory. The outlier loci showed a tendency to differentiate populations according to the two landscapes in the Principal Component Analysis. We hypothesize that candidate loci under selection in *Centotheca lappacea* may reflect the first fine-tuned signatures of regional adaptation to different ecological conditions in the two landscapes. Bukit Duabelas represents a more heterogeneous, hillside landscape, while Harapan represents a rather uniform plain landscape, which may have effects on abiotic ecological conditions within the plots. Moreover, soils in Bukit Duabelas are dominated by clay acrisol, while in Harapan they are dominated by sandy loam. Hence, results of our study may provide a starting point for more detailed genetic studies on local adaptation. The predominant distribution of genetic diversity within populations fits to an obligate sexual, outcrossing breeding system. Sexual ovule development was observed by microscopic analysis, and obligate sexual seed formation was confirmed by flow cytometric seed screening. Bagging of inflorescences revealed no seed set and hence confirmed self-incompatibility. These results are surprising as other invasive Panicoid grasses in Sumatra were reported to be apomictic (e.g., *Paspalum conjugatum* and *Pennisetum polystachion*). According to field collections of the author’s team and individual genotypes observed in our AFLP data, also vegetative propagation of *Centotheca lappacea* can be ruled out as a factor for invasiveness. Since grasses are wind-pollinated, *C*. *lappacea* had obviously no pollinator limitation in the invaded area, and successfully established populations via outcrossing. Sexual reproduction may have rapidly restored genetic diversity after initial colonization phases with recurrent gene flow within and among populations, and natural selection can act efficiently on a diverse gene pool. Polyploidy and related genomic features, in contrast, can be ruled out as a factor for the invasion success of the species. Whether polyploidy is in tropical areas less relevant for invasion success than in temperate regions (e.g.,), needs to be studied. Gene flow among populations of the species in land-use systems may have been enhanced by efficient seed dispersal mechanisms via spikelets with adherent bristles. Diaspores adhering on clothing of farm-workers can easily be carried over longer distances by the use of cars or motorbikes in the area of plantations. The importance of human-mediated transportation of diaspores over roads, and its effects on population genetic structure, was also recognized in the invasive plant *Flaveria bidentis*. Also a good germinability of seeds was observed in our cultivation experiments. These features may strongly contribute to the distribution success of a plant species. Our study provides for the first time data on occurrence, population genetic structure, and mode of reproduction of *C*. *lappacea* in land-use systems in Sumatra. These basic data can be useful for estimating the potential of the species to invade natural rainforests. The species does occur in jungle rubber, it is also elsewhere shade-tolerant and thus probably pre-adapted to the colonization of natural rainforest. Shade tolerance is an important factor for invasiveness, as light is in tropical forest understory the most limiting resource for plant growth. From a population genetic view, *C*. *lappacea* successfully has established populations in agroforest systems, and probably can grow in natural forests as well. However, in the Jambi region the species was observed in only one of the eight investigated natural rainforest plots of National Parks. Within natural forest ecosystems, the scarcity of human visits, and the lack of human dispersal routes along roads and footpaths are probably major limiting factors for spread. As an outcrosser, the species cannot establish populations with a single erratic diaspore which might occasionally be dropped in the understory. In so far, the present potential of the species to invade natural, undisturbed forests may remain limited as long as no human activities would open opportunities to regularly transport diaspores into the forest. However, in the land-use systems, the species has gained a wide distribution and abundance. The species is not regarded as an extremely noxious weed by farmers, and therefore, no special eradication measures are taken beside the general weed management in plantations; pers. obs. of the author’s team. Hence, the species already has many potential source populations and a high propagule pressure exists in close vicinity of rainforest reservations. Although the invasive potential is at the moment regarded as moderate, the species may become rapidly invasive with human disturbance and continued deforestation of lowland rainforest systems in Sumatra. # Supporting Information [^1]: Elvira Hörandl (EH) is a PLOS ONE Editorial Board member. This does not alter the authors' adherence to all the PLOS ONE policies on sharing data and materials. [^2]: Conceived and designed the experiments: EH NO DH RF NB. Performed the experiments: NB FU DH. Analyzed the data: LH EH NB FU DH NO. Contributed reagents/materials/analysis tools: NO ST NB BV. Wrote the paper: LH EH NB DH FU NO ST BV RF.

# Introduction Natural killer (NK) cells are specialized immune cells that eliminate pathogen infected and tumorigenic cells. Target cell killing is mediated by the secretion of perforin and granzymes, which are stored within the secretory lysosomes of NK cells. The recognition of a target cell induces the formation of an activating immunological synapse at the contact site of the two cells. Secretory lysosomes are polarized towards the immunological synapse, where they fuse with the plasma membrane releasing their cytotoxic contents. The pore forming protein perforin then facilitates the entry of pro-apoptopic granzymes into the target cell cytoplasm resulting in cell death. NK cell cytotoxicity is severely impaired in the hematological disorder familial hemophagocytic lymphohistiocytosis (FHL). Subtypes 4 (FHL-4) and 5 (FHL-5) are caused by the mutation of genes encoding syntaxin 11 and Munc18-2 respectively. Analysis of NK cells isolated from subjects with FHL-4 and FHL-5 revealed a defect in secretory lysosome exocytosis ,. In these cells recognition of a target cell causes secretory lysosomes to polarize to the activating immunological synapse, but they are unable to fuse with the NK cell plasma membrane and cannot release their contents,. Syntaxin 11 is a soluble N-ethylmaleimide (NEM)-sensitive factor attachment protein receptor (SNARE), a class of proteins that catalyse membrane fusion reactions by forming trans-SNARE complexes that bridge opposing membranes. It is abundant in the immune system and is expressed by B lymphocytes, cytotoxic T lymphocytes (CTLs), dendritic cells, mast cells, monocytes, macrophages, NK cells and neutrophils. In addition to a role in secretory lysosome exocytosis in NK cells, syntaxin 11 has been reported to be required for the exocytosis of secretory organelles in CTLs, neutrophils and platelets, whereas in macrophages it inhibits phagocytosis and regulates late endosome-lysosome fusion. Despite its role in exocytosis of secretory lysosomes by NK cells, syntaxin 11 does not co-localize with secretory lysosomes in resting NK cells, but it is polarized to the immunological synapse when NK cells are activated by conjugation to target cells. Furthermore, syntaxin 11 interacts with Munc18-2, a member of the Sec-1/Munc18-like (SM) family of proteins whose members regulate SNARE-mediated membrane fusion reactions. SM proteins also chaperone syntaxins, regulating the level and localization of these SNAREs. This chaperone function is evident in FHL-5, in which mutations in Munc18-2 result in a pronounced reduction in the level of syntaxin 11. In contrast, how mutations associated with FHL-4 result in the loss of function of syntaxin 11 in NK cells is poorly understood. Herein we dissect the molecular basis for FHL-4 by examining how disease- associated mutations affect the interaction of syntaxin 11 with other proteins and cellular membranes. FHL-4 deletion and frameshift mutations result in the abrogation of secretory lysosome exocytosis and the consequent loss of NK cell cytotoxicity. We show that these FHL-4 mutations have differential effects on SNARE binding by syntaxin 11, but the FHL-4 mutant proteins retain a Munc18-2 binding site. Moreover, syntaxin 11 is S-acylated in NK cells and this is dependent on the C-terminal cysteine rich region, which is deleted in all of the FHL-4 mutants characterized. This posttranslational modification is required for the membrane association of syntaxin 11 and for the polarization of this protein to the immunological synapse. We also show that syntaxin 11 recruits Munc18-2 to intracellular membranes and that this is dependent on the cysteine rich region of syntaxin 11. Together these findings demonstrate an important role for S-acylation in the function of syntaxin 11 in NK cells. # Materials and Methods ## Antibodies The following antibodies were used: rabbit anti-syntaxin 11 (Proteintech Group Inc), mouse polyclonal anti-syntaxin 11 (Abcam), rabbit anti-synaptosomal- associated protein, 23 kDa (SNAP23, Synaptic Systems), mouse anti-glyceraldehyde 3-phosphate dehydrogenase (GAPDH, clone 6C5, Abcam), rabbit anti-calreticulin (Calbiochem), mouse anti-Myc epitope tag (clone 9E10, Sigma Aldrich) and rabbit anti-calnexin (Stressgen). ## DNA constructs ### (i) Syntaxin 11 green fluorescent (GFP) fusions A construct encoding a GFP fusion of the wild type human syntaxin 11 (pEGFP-C3-syntaxin 11) was generated by PCR through extension of the truncated syntaxin 11 open reading frame in pCMV-Tag3a-syntaxin 11 (a kind gift from Dr R. Prekeris) using the oligonucleotides 5′syn11 and 3′Syn11, the resultant PCR product was then inserted into pEGFP-C3 (Clontech) at the *Xho* I and *Eco* RI sites. A GFP fusion of the syntaxin 11-Q268X mutant was generated by PCR using pEGFP-C3-syntaxin 11 as a template and the oligonucleotides 5′Syn11 and 3′Q268X, the resultant PCR product was cloned into pEGFP-C3 at the *Xho* I and *Eco* RI sites. The GFP fusion of the N-terminal deletion syntaxin 11ΔN24 was cloned by amplifying pEGFP-C3-syntaxin 11 with the oligonucleotides 5′Syn11ΔN24 and 3′Syn11 and cloning the resultant product into pEGFP-C3. The cysteine mutant syntaxin 11C5A was cloned by amplifying pEGFP-C3-syntaxin 11 with the primers 5′Syn11 and 3′Syn11C5A and cloning the resultant product into the *Xho* I and *Eco* RI sites of pEGFP-C3. Oligonucleotide sequences are provided in. ### (ii) Syntaxin 11 Glutathione S-transferase (GST) fusions A GST fusion construct of the wild type human syntaxin 11 (pGEX4T.1-syntaxin 11) was cloned by PCR amplification of the syntaxin 11 sequence using pEGFP-C3-syntaxin 11 as a temple and the oligonucleotides 5′Syn11GST and 3′Syn11GST. The resultant PCR product was inserted into the *Xho* I and *Eco* RI sites of pGEX4T.1 (GE Life Sciences). The cloning of a GST fusion of syntaxin 11 Q268X was identical to that of the wild type sequence except that the oligonucleotide 3′Syn11Q268XGST was used instead of 3′Syn11GST. Constructs encoding GST fusions of the syntaxin 11 mutants L194fsX2, V124fsX60, T37fsX62 and E25X were generated using the QuikChange site-directed mutagenesis kit (Stratagene) as per the manufacturer's instructions using pGEX4T.1-syntaxin 11 as a template and the sense oligonucleotides L194, V124, T37 and E25 respectively; the antisense oligonucleotides were the exact complement. The GST fusion of the mutant Q230fsX125, was cloned using a two-step PCR procedure. The syntaxin 11 cDNA sequence (MGC clone 5176646) was used as a template and amplified with either 5′Syn11GST and Q230FsR or with 3′Q230GST and Q230FsR oligonucleotides. The two PCR products were gel purified, mixed together and used as the template in a second stage PCR reaction with the oligonucleotides 5′Syn11GST and 3′Q230GST, the resultant PCR product was cloned into pGEX4T.1 at the *Xho* I and *Eco* RI sites. Oligonucleotide sequences are provided in. ### (iii) Munc18-2 constructs To generate a construct encoding human Munc18-2 tagged at the N-terminus with a Myc-epitope, a two step PCR procedure was used. The primers Munc18NT-F1 and Munc18NT R1b were used in step 1 to amplify the Munc18-2 human cDNA clone (MGC clone 71251). The gel purified PCR product was then used as the template in a second PCR reaction and amplified with the primers Munc18NT-F2 and Munc18NT R2. The resultant PCR product was then cloned into pcDNA3.1Pac using *Eco* R1 and *Bam* H1 to generate pCDNA3.1Pac-Myc-Munc18-2. To generate a mCherry fusion, the Munc18-2 open reading frame was cut out of pcDNA3.1Pac-Myc-Munc18-2 and inserted into the pmCherry-C2 vector (Clontech) using *Eco* R1 and *Bam* H1. Oligonucleotide sequences are provided in. ## Culture and transfection of cells HeLa-M and YTS cells were cultured as described previously. 721.221 cell lines were maintained in RPMI 1640 media supplemented with 10% (v/v) fetal bovine serum (FBS). HeLa-M cells were transfected with Lipofectamine 2000 (Invitogen) as per the manufacturer's instructions. YTS NK cells were transfected by nucleofection (Amaxa). Briefly, 5×10⁶ YTS cells were resuspended in 100 µl nucleofector solution R, to which 5 µg plasmid DNA was added. Cells were nucleofected using program X-01 and mixed immediately with 500 µl RPMI 1640 supplemented with 20% (v/v) FBS. After 5 min at 37°C, 5% CO₂, cells were plated out in 2 ml of growth medium at 37°C, 5% CO₂, an additional 2 ml of growth medium was added 1 h post nucleofection. ## Immunoprecipitation 10⁷ YTS cells per immunoprecipitation were incubated in the presence or absence of 100 mM NEM for 30 min prior to lysis. Cells were lysed into NET buffer \[50 mM Tris pH 7.5, 150 mM NaCl, 1 mM ethylenediaminetetraacetic acid (EDTA), 0.1% Nonidet-P40, 0.25% gelatin and complete protease inhibitors (Roche Applied Science)\] for 30 min on ice before the lysate was centrifuged at 13,000 g for 10 min at 4°C in a microcentrifuge. The supernatant was precleared by incubation for 1 h with protein G sepharose (Sigma Aldrich) and centrifuged at 13,000 g for 1 min. Mouse syntaxin 11 antibody was added to the supernatant and incubated on ice for 2 h, protein G-sepharose was then added and the samples rotated for 2 h at 4°C. The beads were washed 4 times with NET buffer and bound proteins eluted by boiling in Laemmli sample buffer. ## GST Pulldowns BL21(DE3) pLys was transformed with each GST expression plasmid, cultured in 100 ml Luria-Bertani medium containing 100 µg/ml ampicillin (100 µg/ml) and grown at 37°C with shaking at 200 rpm to an OD₆₀₀ of 0.6. The culture was then induced with 1 mM isopropyl-β-d-thiogalactopyranoside and grown for a further 4 h. The cells were harvested by centrifugation and resuspended into 2 ml STEP buffer \[10 mM Tris pH8.0, 150 mM NaCl 1 mM EDTA, 1 mM dithiothreitol, 1 mM phenylmethanesulfonyl fluoride\] with 0.25% (w/v) sarkosyl on ice. The bacterial suspension was then sonicated on ice and Triton X-100 added to a final concentration of 0.5% (v/v). Cellular debris was pelleted by centrifugation in a microcentrifuge for 10 min at 13,000 g. GST fusion proteins were purified from the supernatant with glutathione-Sepharose 4B affinity beads as specified by the manufacturer (Pharmacia Biotech). 5×10⁶ HeLa-M cells transfected with pcDNA3.1-Myc-Munc18-2 were used per pulldown. 48 h post transfection, the cells were lysed on ice in STEP buffer with 0.5% Triton X-100 and centrifuged at 13,000 g in a microcentrifuge for 10 min. The supernatant was added to 50 µg of each GST-fusion bound to beads and incubated at 4°C with rotation for 2 h. The beads were washed 4 times with STEP buffer, with 0.5% Triton X-100, and bound proteins eluted by boiling in Laemmli sample buffer. ## Flow cytometric analysis of syntaxin 11 expression HeLa-M cells were transfected with GFP-syntaxin 11 expression constructs and co- transfected with either the control plasmid pcDNA3.1Pac or pcDNA3.1Pac- MycMunc18-2. Each transfection was performed in triplicate. 48 h post- transfection the cells were washed once with phosphate buffered saline 0.1% bovine serum albumin and resuspended into PBS 0.1% BSA. The cells were analysed on a BD-LSRFortessa flow cytometer (Becton Dickinson), the cell population was gated to exclude debris, 10,000 gated events were analyzed and the mean fluorescence value of the total cell population was determined. ## Membrane Fractionation 10⁷ YTS cells were resuspended into homogenization buffer \[25 mM 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid pH 7.5, 250 mM sucrose, 1 mM EDTA, complete protease inhibitors), homogenized using a ball bearing homogeniser (Isobiotec) with a 10 µm clearance and centrifuged at 400 g for 5 min at 4°C in a microcentrifuge. The resultant postnuclear supernatant was centrifuged at 50,000 g for 30 min at 4°C in an S100-AT3 rotor (Sorvall) to separate the cytosol (supernatant) from the membrane fraction (pellet). ## Live Cell Confocal Microscopy 5 h post-transfection, YTS cells were centrifuged on Lymphoprep (Axis Shield) at 850 g for 20 min in a bench top centrifuge to remove cellular debris, washed in cell media and allowed to recover for 1 h by incubation in growth medium at 37°C. To visualize secretory lysosomes the YTS cells were preloaded with 10 nM LysoTracker Red or LysoTracker Green (Molecular Probes). Cells were washed in phenol red free RPMI 1640 media supplemented with 10% FBS, 5×10⁵ cells were resuspended into 2 ml of the same medium and incubated in 35 mm glass bottomed dishes (Ibidi). In order to visualize conjugated cells, 2.5×10⁵ YTS cells were incubated in the glass bottomed dishes at 37°C for 15 min with 2.5×10⁵ 721.221 target cells that had been preloaded with 6 µM Cell Trace Far Red (Molecular Probes). Cells were imaged using a LSM700 Zeiss Confocal microscope at 63,000× magnification. ## Identification of putative sites for S-acylation Potential S-acylation sites in the syntaxin 11 protein sequence were identified using the CSS-PALM 2.04 software set at a medium threshold. ## Acyl-biotinyl exchange assay for S-acylation S-acylation was assayed using an adaptation of the procedure described by Drisdel and Green (2004). Briefly, 10⁷ YTS cells were sonicated in lysis buffer \[500 mM NaCl, 2.7 mM KCl, 10 mM Na₂HPO₄, 1.8 mM KH₂PO₄, 5 mM EDTA, 1% (v/v) Triton X-100, complete protease inhibitors\] on ice and then centrifuged at 13,000 g in a microcentrifuge. NEM was added to the supernatant to a final concentration of 25 mM and the sample incubated at 4°C for 16 h. Proteins were precipitated from the cell lysate by chloroform methanol precipitation and centrifugation for 10 min at 13,000 g in a microcentrifuge. Protein pellets were resuspended into resuspension buffer \[50 mM Tris pH 7.5, 150 mM NaCl, 2% (v/v) sodium dodecyl sulfate, 8 M urea\]. To one half of the sample, lysis buffer with 8 mM Biotin-1-Biotinamido-4-\[4'-(maleimidomethyl) cyclohexanecarboxamido\]butane (Biotin-BMCC, Pierce) was added whilst to the other half lysis buffer with 8 mM Biotin-BMCC and 10% (v/v) hydroxylamine was added. The samples were incubated for 2 hours at 4°C before precipitating and resuspending the proteins as before. The samples were then incubated with neutravidin agarose beads (Pierce) for 2 h at 4°C. Beads were washed with 500 mM NaCl, 2.7 mM KCl, 10 mM Na₂HPO₄, 1.8 mM KH₂PO₄, 0.1% (v/v) Triton X-100 and bound proteins eluted by incubation in resuspension buffer and by boiling in Laemmli sample buffer. # Results ## FHL-4 mutations have differential effects on *SNAP23* binding by syntaxin 11, but do not affect Munc18-2 binding The homozygous FHL-4 truncation (Q268X and E25X) and frameshift (Q230fsX125, L194fsX2, V124fsX60 and T37fsX62) mutations are associated with a defect in secretory lysosome exocytosis and the corresponding loss of NK cytotoxicity, but how these mutations impair the function of syntaxin 11 is poorly understood. We, therefore, studied the effect of these mutations on the interaction of syntaxin 11 with other proteins. Since the SNARE protein SNAP23 binds to syntaxin 11 in other cell types and in platelets, we examined whether this interaction also occurs in NK cells. In initial experiments SNAP23 did not co-immunopreciptate with syntaxin 11 from cell lsyates prepared from the model human NK cell line YTS. Therefore, prior to cell lysis, YTS cells were incubated with NEM to inhibit NEM-sensitive factor mediated disassembly of SNARE complexes. Syntaxin 11 antibodies immunoprecipitated syntaxin 11 irrespective of whether the cells were pre-treated with NEM, whereas SNAP23 was only co-immunopreciptated in cell lysates prepared from cells pre-treated with NEM. These data are consistent with SNAP23 being a binding partner for syntaxin 11 in NK cells. Consequently, SNAP23 was used to determine if SNARE binding was affected by FHL-4 mutations. The effect of FHL-4 mutations on the interaction of syntaxin 11 with its binding partners were then examined using GST pulldowns of cell lysates prepared from HeLa-M cells transfected with a construct encoding Myc-Munc18-2. Whilst a GST fusion of full-length syntaxin 11 bound SNAP23 from cell lysates, no binding was observed for GST alone. The Q268X truncation mutant protein has an intact SNARE domain and bound to SNAP23, whereas the Q230fsX125, L194fsX2, V124fsX60, T37fsX62 and E25X mutant proteins, which lack part or all of the SNARE domain had substantially reduced binding to SNAP23. In contrast, all of the syntaxin 11 FHL-4 mutant proteins bound to Myc-Munc18-2. This suggests that all of the mutant proteins characterized in this study retain the ability to bind to Munc18-2, due to the presence of a Munc18-2 binding site in the N-terminal 24 residues of syntaxin 11. By contrast, with the exception of Q268X, FHL-4 mutations abrogate binding to SNAP23. Intriguingly, in FHL-5 the level of the syntaxin 11 protein is also reduced, consistent with a role for Munc18-2 in stabilizing syntaxin 11 expression. We therefore investigated whether interaction with the N-terminal 24 residues of syntaxin 11 is required for the stabilization of syntaxin 11 expression by Munc18-2. As anticipated a *de novo* mutant of syntaxin 11 that lacked the N-terminal 24 residues (syntaxin 11ΔN24) did not bind to Munc18-2 in a GST pulldown experiment. We then performed a flow cytometric assay to quantify the effect of Munc18-2 on the expression of syntaxin 11. There was a significant increase in fluorescence of cells when cells transfected with a GFP fusion of syntaxin 11 (GFP-syntaxin 11) were also transfected with Myc-Munc18-2. In marked contrast Myc-Munc18-2 did not increase the level of GFP-syntaxin 11ΔN24. Immuoblotting confirmed that this increase in cell-associated GFP fluorescence in cells co-expressing Myc-Munc18-2 was due to elevated expression of GFP- syntaxin 11, whereas the level of GFP-syntaxin 11ΔN24 was unaffected by Myc- Munc18-2. Thus, the Munc18-2 dependent stabilisation of the expression of syntaxin 11 is dependent on the N-terminal 24 residues of syntaxin 11. ## Membrane association of syntaxin 11 and its polarization to the activating immunological synapse are impaired by the FHL-4 Q268X mutation The membrane association of SNAREs is critical for their function in membrane fusion reactions. Syntaxin 11 is an atypical SNARE in that it lacks a transmembrane domain, but it has a C-terminal cysteine-rich region that has been suggested as a site for S-acylation. However, in many cell types this region is not essential for membrane association. We therefore examined the role of the C-terminal cysteine-rich region, which is absent in the FHL-4 mutant protein syntaxin 11 Q268X, in the membrane association and localization of syntaxin 11 in NK cells. Centrifugation of postnuclear supernatants revealed that syntaxin 11, expressed endogenously by the YTS NK cells, fractionated into the membrane enriched fraction, as did the S-acylated protein SNAP23 and the integral membrane protein calnexin, whereas the cytosolic enzyme GAPDH was present in the cytosolic fraction. Likewise for transfected YTS cells, GFP-syntaxin 11 was present in the membrane fraction. In contrast the GFP-syntaxin 11 Q268X was in the cytosolic fraction. The level of GFP-syntaxin 11 Q268X did not exceed that of the endogenous wild type protein, indicating that the presence of the mutant protein in the cytosolic fraction was not due to overexpression. The localization of these proteins was also examined by live cell confocal microscopy. Consistent with previous reports, GFP-syntaxin 11 was localized predominantly to cytoplasmic punta in resting YTS NK cells that were distinct from acidic compartments labelled with LysoTracker dye (secretory lysosomes). In YTS NK cells activated by conjugation to the 721.221 target cells, GFP-syntaxin 11 associated with the cytoplasmic punta was concentrated in the region of the immunological synapse. In resting YTS NK cells, GFP-syntaxin 11 Q268X displayed a more diffuse cytoplasmic localisation than that of the wild type protein and although in some of the conjugated cells there was an increased concentration of GFP-syntaxin 11 Q268X in proximity to the immunological synapse, a diffuse cytoplasmic distribution of this protein was still evident. These data demonstrate a critical role for the C-terminal 21 residues in determining the localization of syntaxin 11 in both resting and activated NK cells. ## Wild type syntaxin 11, but not the FHL-4 mutant Q268X, promotes Munc18-2 recruitment to the activating immunological synapse Since syntaxin 11 binds to Munc18-2 and these proteins co-localize in NK cells, we examined whether this interaction recruits Munc18-2 to cellular membranes. In the absence of co-transfected GFP-syntaxin 11, a mCherry fusion of Munc18-2 (mCherry-Munc18-2) exhibited a diffuse cytoplasmic localisation in resting YTS NK cells that was unaltered when the YTS cells were activated by conjugation to target cells. In marked contrast, when co-expressed with GFP-syntaxin 11, both mCherry-Munc18-2 and GFP-syntaxin 11 were localized to cytoplasmic punta in resting YTS NK cells and were polarized to activating immunological synapse in conjugated cells. Taken together these data suggest that the endogenous syntaxin 11 in YTS NK cells is unable to recruit mCherry-Munc18-2 to cellular membranes, potentially because syntaxin 11 is already bound to the endogenous Munc18-2. Conversely, expression of GFP-syntaxin 11 increases the overall pool of syntaxin 11, enabling mCherry-Munc18-2 to be recruited to membranes. However, although GFP-syntaxin 11ΔN24 was localized to cytoplasmic puncta in resting YTS NK cells and polarized to the activating immunological synapse, co-expressed mCherry- Munc18-2 exhibited a diffuse cytoplasmic localization. Likewise mCherry-Munc18-2 exhibited a diffuse cytoplasmic localization in cells when co-expressed with GFP-syntaxin 11 Q268X. Thus, both the N-terminal and C-terminal regions of syntaxin 11 are required for the recruitment Munc18-2 to cytoplasmic membranes and to the activating immunological synapse. ## Syntaxin 11, but not the FHL-4 Q268X mutant, is S-acylated in NK cells S-acylation involves the covalent addition of long chain fatty acids to the thiols of cysteine residues and is often termed palmitoylation because the predominant fatty acid used is palmitate. The C-terminal cysteine rich region of syntaxin 11 has been suggested as a site for S-acylation. Correspondingly analysis of the syntaxin 11 protein sequence, using the CSS-PALM software, revealed that 5 cysteines within the C-terminal cysteine rich region of syntaxin 11 are potential sites for S-acylation. Since these cysteines are absent in the syntaxin 11 Q268X mutant protein we examined whether these residues are required for membrane association. The cysteine residues were mutated to alanine (syntaxin 11C5A) and the membrane association of this *de novo* mutant protein examined. GFP-syntaxin 11C5A was present in the cytosolic fraction after centrifugation of the postnuclear supernatant, and displayed a diffuse cytosolic localisation in both resting and conjugated YTS NK cells, demonstrating that these 5 cysteine residues are required for membrane association. To confirm that syntaxin 11 is S-acylated in NK cells acyl-biotinyl exchange was used. In this procedure hydroxylamine cleaves long chain fatty acids from proteins to reveal free cysteines that can be biotinylated, enabling these proteins to be pulled down using avidin beads. Syntaxin 11 expressed endogenously by YTS NK cells was pulled down from cell lysates treated with hydroxylamine, demonstrating that syntaxin 11 is S-acylated in YTS NK cells. GFP-syntaxin 11 was also pulled down from hydroxylamine treated cell lysates prepared from transfected YTS cells. In contrast, neither GFP-syntaxin 11C5A nor GFP-syntaxin 11 Q268X were pulled down. These data demonstrate that cysteine residues in the C-terminal region of the protein, absent in all of the FHL-4 mutants characterised herein, are required for S-acylation of syntaxin 11. Thus, S-acylation is required for the membrane association of syntaxin 11 in NK cells and FHL-4 mutations that delete the C-terminal cysteine rich region correspondingly abolish this posttranslational modification. # Discussion In this study we used disease-associated mutations to dissect the sequence requirements for the interaction of syntaxin 11 with SNAP23, Munc18-2 and cellular membranes and hence determine how FHL-4 frameshift and truncation mutations result in the loss of function of syntaxin 11 in NK cells. We show that SNAP23 is a binding partner of syntaxin 11 in NK cells. This interaction has also been reported in B cells, as well as for syntaxin 11 transfected into J774 macrophages and HeLa cells. By contrast, in RAW 264.7 macrophages transfected with syntaxin 11, SNAP23 binding was not observed and instead syntaxin 11 forms a non-classical SNARE complex that sequesters Vti1b. The SNARE binding partners, and functions, of syntaxin 11 may therefore be cell type specific. Notably, syntaxin 11 also binds to SNAP23 in platelets, in which both proteins are required for granule secretion. FHL-4 mutant proteins, in which there is a partial or complete deletion of the SNARE domain, had reduced binding to SNAP23. Conversely the syntaxin 11 Q268X mutant protein retains an intact SNARE domain and was able to bind to SNAP23 in GST pulldowns. This suggests, that in many cases of FHL-4, reduced binding to SNAP23 may be a factor in the inability of secretory lysosomes to fuse with the plasma membrane in NK cells. Syntaxin 11 also binds Munc18-2, an interaction that is dependent on an acidic region and the hydrophobic pocket of the SM protein. Whereas the recently identified FHL-4 mutant protein syntaxin 11 L58P is unable to bind to Munc18-2, GST-fusions of the FHL-4 mutant proteins characterised herein, bound to Munc18-2. These data are consistent with the retention of an accessible Munc18-2 binding site by FHL-4 frameshift and truncation mutant proteins. This region in syntaxins is known as the N-peptide and is a key site for interaction with cognate SM proteins. During the preparation of this manuscript the crystal structure of Munc18-2 was reported and consistent with our data the N-peptide of syntaxin 11 was shown to be critically important for the interaction between this SNARE and Munc18-2. SM protein binding has multiple functions, including the chaperoning of syntaxins. Indeed, the chaperone function for Munc18-2 is highlighted in FHL-5, in which mutations in Munc18-2 result in reduced levels of syntaxin 11. Our data demonstrate that the N-terminal 24 residues of syntaxin 11 are required for the stabilization of the expression of this SNARE by Munc18-2. In addition to stabilizing the expression of cognate SNAREs, the chaperone function of SM proteins can also facilitate the trafficking of SNAREs to specific intracellular compartments. Indeed, Munc18-2 expression redistributes syntaxin 3 to the plasma membrane in syntaxin 11 deficient murine CTLs. However, we observed that the deletion of the N-terminal region of syntaxin 11 prevented the recruitment of Munc18-2, but not of syntaxin 11, to intracellular membranes in resting NK cells and to the immunological synapse in conjugated NK cells. Furthermore, Munc18-2 exhibited a diffuse cytoplasmic distribution in NK cells that co-express the cytosolic syntaxin 11 Q268X mutant. Thus, our data suggest that syntaxin 11 determines the intracellular localization of Munc18-2 in NK cells. Despite having no transmembrane domain, syntaxin 11 is membrane associated in NK cells, which we show is dependent on the cysteine-rich C-terminal region. This implies that FHL-4 truncation and frameshift mutant proteins are non-functional in NK cells, at least in part, because they cannot associate with membranes, a prerequisite for SNARE proteins to promote membrane fusion reactions. The C-terminal cysteine rich region is not required for membrane association in all cell types, as deletion of this region does not prevent membrane association in macrophages, NRK cells and HeLa cells. Membrane association in the absence of the C-terminal region in other cell types may be due to interactions with membrane proteins, such as other SNAREs, and may reflect cell-type specific behaviour not observed in NK cells. In addition, we show that syntaxin 11 is S-acylated in NK cells and that this is dependent on cysteine residues in the C-terminal region of the protein. Based on these observations we predict that S-acylation will not only be abolished in the Q268X mutant protein, but in all other FHL-4 truncation and frameshift mutants of syntaxin 11. A number of other SNAREs are S-acylated, both SNAP23 and its neuronal homologue SNAP25, which lack transmembrane domains, use this posttranslational modification for membrane association. Intriguingly, syntaxins 1a, 1b, 6, 7, 8 and 12, all of which have transmembrane domains, are S-acylated –. Removal of the S-acylation site reduced the cycling of syntaxin 7 between the plasma membrane and endosomes. Thus, S-acylation of syntaxin 11 may not only be important for its interaction with membranes, but also for the trafficking of the protein in the cell. Indeed, deletion or mutation of the cysteine rich domain prevents polarization of syntaxin 11 to the immunological synapse, although this is most likely secondary to the inability of these mutant proteins to associate with membranes. In summary the loss of syntaxin 11 function in FHL-4 frameshift and deletion mutants is associated with the inability of the mutant proteins to associate with membranes, precluding not only the polarization of syntaxin 11 to the activating immunological synapse, but also that of Munc18-2. # Supporting Information We thank Rytis Prekeris for the syntaxin 11 cDNA, Gareth Howell for advice on confocal microscopy and the Hewitt lab for helpful discussions. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: AH OF NG MP EH. Performed the experiments: AH OF NG MP EH. Analyzed the data: AH OF NG MP EH. Contributed to the writing of the manuscript: AH OF NG MP EH. [^3]: Current address: School of Biochemistry, University of Bristol, Bristol, United Kingdom [^4]: Current address: Cell and Developmental Biology Programme, Center for Genomic Regulation, Barcelona, Spain