Datasets:

Daniel-P-Gonzalez

/

OCD

like 0

# Introduction Since 1994, nearly 90 countries and regions worldwide have reported one or more cases of multidrug-resistant (MDR) tuberculosis (TB). MDR TB is defined as TB caused by a strain of *M. tuberculosis* that is resistant to at least isoniazid and rifampin, two of the most important first-line drugs used to treat the disease. China had approximately 140,000 MDR TB cases in 2004, or one third of the estimated global burden of MDR TB. In some provinces of China, the prevalence of MDR TB among new cases and previously treated cases was above 10% and 30%, respectively. However, the number and percentage of MDR TB cases in Shanghai, one of the largest cities in Asia, has not previously been reported. In 2006, multiple cases of extensively drug-resistant (XDR) TB were reported in South Africa, raising concerns that XDR strains of *M. tuberculosis* are prevalent in other populations. Many of the cases reported from South Africa occurred among HIV-infected persons and there was a high case fatality rate. By November 2007, 41 countries had reported XDR TB cases. The mortality rate of XDR TB patients varies in different countries and depends on the study population and their HIV status,. More than fifty percent of HIV-uninfected XDR TB patients have been successfully treated in South Korea, but the duration of treatment is prolonged and the financial burden is high. XDR TB has not previously been formally reported in mainland China, but given the high prevalence of TB nationwide, the large number of MDR TB cases in the country, and the long-time use of second line anti-TB drugs and fluroquinolones, the existence of XDR TB in China seems inevitable. We performed a retrospective study to determine the number and percentage of TB patients in Shanghai with MDR and XDR TB, and to determine whether there is transmission of MDR and XDR strains of *M. tuberculosis* in Shanghai. # Materials and Methods ## Study population We performed a retrospective cohort study using the existing data and specimens at the Shanghai Municipal Center for Disease Control and Prevention (Shanghai CDC), from TB patients who were diagnosed in Shanghai during March 2004 through November 2007. Since 1995 in Shanghai, all suspected pulmonary TB cases detected in general hospitals or community health centers were referred to a specialized TB hospital or TB clinic for further diagnostic tests, including sputum smear examinations, culture and chest radiography. There were 31 designated district TB hospitals in Shanghai; all of the pretreatment positive cultures from patients in each hospital were sent to the Tuberculosis Reference Laboratory (TRL) at Shanghai CDC for drug susceptibility testing and species identification. Shanghai CDC also collected data on the social and demographic characteristics, treatment history, clinical characteristics, drug- susceptibility test results, and clinical outcomes of each patient. All of the investigation protocols in this study were approved by ethics committee of Fudan University. Since this was a retrospective study and all patients' information was routinely collected by Shanghai CDC for analysis and reports to the government, consent was not obtained from the patients during 2004 through 2006. We started a research project in 2007, and informed consent has been obtained from all patients since then for the information to be used in scientific studies. ## Drug susceptibility testing (DST) TRL at Shanghai CDC participated in the World Health Organization/International Union against Tuberculosis and Lung Disease Global Project on Anti-Tuberculosis Drug Resistance Surveillance. Species identification of mycobacteria was performed by conventional biochemical and PCR tests. The first-line drug susceptibility testing was routinely performed by the absolute concentration or proportion methods on isolates of *M. tuberculosis*. In the present study, we restricted our analysis to those patients whose first- line drug susceptibility tests were performed by the proportion method. The following drug concentrations were used: isoniazid (0.2 µg/ml), rifampin (40.0 µg/ml), streptomycin (4.0 µg/ml) and ethambutol (2.0 µg/ml). For any isolates that were MDR, we also used the WHO Guidelines for drug susceptibility testing for second-line anti-tuberculosis drugs for DOTS plus. We chose five second-line drugs widely used for MDR TB treatment in Shanghai and performed drug susceptibility tests using the following concentrations: ofloxacin (2.0 µg/ml), kanamycin (30.0 µg/ml), capreomycin (40.0 µg/ml), amikacin (40.0 µg/ml) and 4-aminosalicylic acid (1.0 µg/ml). To ensure the consistency and reliability of results, all second-line drug susceptibility testing was performed by a senior technician who had completed and passed the WHO's quality control evaluation. All drugs were obtained from Sigma Life Science Company (USA). ## Definitions MDR TB was defined as tuberculosis disease caused by a strain of *M. tuberculosis* that was resistant to at least isoniazid and rifampin. XDR was defined as TB with resistance to at least isoniazid, rifampin, a fluoroquinolone (e.g. mofloxacin, ofloxacin, levofloxacin, sparfloxacin, gatifloxacin, ciprofloxacin) and one of three injectable second-line drugs (capreomycin, kanamycin, and amikacin). Pre-XDR was defined as disease caused by a strain resistant to isoniazid and rifampin and either a fluoroquinolone or a second- line injectable drug, but not both. In the present study, we use the terms simple MDR, which refers to isolates that are resistant to just isoniazid and rifampin but not pre-XDR TB and XDR TB. New cases were defined as TB patients who denied having had any prior anti-TB treatment or who received anti-TB treatment for \<30 days. Previously treated cases were TB patients who reported having been treated for tuberculosis for at least 30 days or who had documented evidence of prior treatment in the case report form or surveillance database. Acquired drug resistance was defined as the isolation of drug-resistant *M. tuberculosis* from a patient who has been treated for tuberculosis for one month or longer. Primary drug resistance is the isolation of a drug-resistant strain from a patient without a history of previous treatment. Migrants were defined as individuals from other areas of China who moved to Shanghai. Residents were defined as persons with a registered permanent residence in Shanghai. ## Genotyping method We used the VNTR-7 and VNTR-16 methods to genotype the 189 clinical isolates of *M. tuberculosis*, following the protocol described by Zhang. First, VNTR-7 was performed in all isolates and the isolates with identical VNTR-7 genotyping pattern were further differentiated by VNTR-16. We also used the deletion- targeted Multiplex PCR (DTM-PCR) method to identify the Beijing genotype strains. Primers were synthesized by Invitrogen Bio Co. (China) and polymerase chain reaction (PCR) mixtures were prepared using the 2×Taq MasterMix (Tiangen Co., China). PCR products were separated using an agarose gel and were analyzed by Quantity 1 gel imaging system (Bio-Rad Co., USA). ## Clinical treatment There is presently no standard treatment strategy, such as DOTS-Plus, to guide the therapy of MDR-TB patients in Shanghai. Individualized therapies were given to MDR TB patients based on the patient's physical and financial situation, the strains' drug-susceptibility patterns and the clinicians' experience. The following drugs were used, in different combinations: two injectable second-line drugs, including capreomycin and amikacin; fluoroquinolones, including ofloxacin, levofloxacin, gatifloxacin, moxifloxacin and ciprofloxacin; a modified form of isoniazid, called prothionamide; two modified forms of rifampicin, called rifapentine and rifabutin; and 4-aminosalicylic acid. ## Statistical analysis We used the chi-square test of proportions to identify significant differences between two or more groups of patients. A *p* value\<0.05 was considered statistically significant. Odd ratios (ORs) and 95% confidence intervals (CI) were calculated to measure the association between patient characteristics and the outcome of interest. All analyses were performed using Stata statistical software (version 8.0SE, Stata Corporation, College Station, Texas, USA). # Results ## MDR TB From March 2004 through November 2007, there were 19,722 newly registered pulmonary tuberculosis patients in 31 designated district tuberculosis hospitals in Shanghai. Of these, 6,200 (31.4%) patients were culture positive for *M. tuberculosis*. We excluded 1,537 culture-positive TB patients who were diagnosed during May 2005 through February 2006 because the drug susceptibility testing of their isolates was performed using the absolute concentration method. We compared the characteristics such as sex, age and treatment history of the patients who were included and excluded from the study population and found no significant differences (data not shown). Of the remaining 4,663 TB patients, we further excluded 95 (2.0%) patients whose isolate lacked a drug susceptibility test result and 189 (4.1%) patients who were infected with Mycobacterium other than *M. tuberculosis* (MOTT). The remaining 4,379 patients with culture- confirmed *M. tuberculosis* and their drug susceptibility test results were used for analysis. Among them, 247 (5.6%, 247/4,379) TB patients had disease caused by a MDR strain of *M. tuberculosis*. ## XDR TB To determine the number and percentage of the MDR TB patients that were infected with an XDR strain of *M. tuberculosis*, we performed drug susceptibility testing for five second-line drugs on the isolates of the MDR TB patients in the study population. However, the isolates from 72 of the 247 MDR TB patients were missing or could not be recovered from the storage; isolates from 175 (70.9%) MDR TB patients were tested for susceptibility to a second-line drug. We compared the clinical characteristics of the TB patients whose isolate was tested for susceptibility to second-line drugs and the TB patients whose isolates were missing or could not be recovered, and there were no statistically significant differences between the two groups (data not shown). Among the 175 MDR patients, 109 patients (62.3%) were infected with a strain of *M. tuberculosis* that was simple MDR, 55 patients (31.4%) were infected with a strain that was pre-XDR, and 11 (6.3%) patients were infected with an XDR strain of *M. tuberculosis*. ## Characteristics of MDR, pre-XDR and XDR TB patients We compared the patient characteristics associated with MDR, pre-XDR and XDR TB, such as age, sex, treatment history and status (resident versus migrant). Most MDR patients were 30 to 59 years old, and XDR TB was more likely to occur in persons 45 to 59 years than among person who were younger or older (p = 0.0067). Sixty percent of the patients with MDR and 54.5% of the patients with XDR were new cases. ## Treatment outcomes of MDR patients Fifty-six percent of the patients with MDR TB and 9.1% of the patients with XDR TB were successfully treated. The cure rate in simple MDR TB, pre-XDR TB and XDR TB patients decreased and the mortality increased as the drug resistance increased. The cure rate of MDR TB in new cases was higher than the cure rate of MDR TB among previously treated cases. ## Genotypes of *M. tuberculosis* We genotyped one isolate from each of 175 MDR TB patients. 87.3% (165/189) of the MDR isolates and 90.9% (10/11) of the XDR isolates were Beijing genotype strains. Patients infected with *M. tuberculosis* strains that had the same VNTR pattern were assumed to be part of a chain of transmission. Four clusters were identified, each with 2 different patients. To determine whether the cluster of two patients represented a chain of transmission of *M. tuberculosis*, we sought epidemiological links between them. Patients in two clusters lived in different districts, but patients in the other two clusters lived in the same district and had been diagnosed and treated in the same hospital and no additional epidemiological links between them were detected. None of the isolates with XDR were in a genotype cluster. # Discussion The present study showed that 5.6% of the tuberculosis patients in Shanghai were infected with a strain of *M. tuberculosis* that was MDR. Furthermore, 6.3% of the MDR TB patients were XDR. Nearly 55% (6/11) of the XDR TB patients were new cases, suggesting there is transmission of highly drug-resistant strains of *M. tuberculosis* in Shanghai. Since second-line drugs have been used in Shanghai for several decades and there is no standard treatment strategy for patients with MDR and pre-XDR strains, there was concern that the prevalence of XDR TB in Shanghai would be higher. Based on our study using specimens from the 31 designated district tuberculosis hospitals, we report that XDR TB occurs in Shanghai, albeit currently with a relatively low prevalence during the study period. After XDR TB became a public concern during 2006, many countries retested their stored isolates and XDR strains were reported in 41 countries. The reported percentage of XDR TB patients among the MDR TB patients varies between countries, from 3% in the United States to 19% in Latvia, and the average rate is about 10%. The percentage of MDR and XDR TB patients that are detected depends on the study design, the sampling frame and the study population. For example, hospital based studies will likely include a greater proportion of seriously ill TB patients, including patients who are undergoing retreatment. In contrast, community based studies will likely provide a lower estimate of the percentage of total TB patients who are infected with an MDR or XDR strain of *M. tuberculosis*. More than half of the infections with MDR and XDR strains of *M. tuberculosis* in our study occurred among new TB cases, a finding which suggests that the transmission of MDR and XDR strains of *M. tuberculosis* is a serious problem in Shanghai. The prevalence of primary drug resistance may actually be higher than our estimate. Transmission of drug-resistant *M. tuberculosis* can cause primary drug resistance among individuals with no prior history of TB, as well as among individuals with a history of prior TB treatment. Previously, we showed that 84% of retreated TB patients with drug-resistant disease actually had primary drug resistance, not acquired drug resistance. Andrews, et al. also reported that most of the MDR/XDR cases in South Africa were primary drug resistance. In addition, a recent meta-analysis showed that primary drug resistance was associated with poor treatment outcomes if treatment regimens were not based on drug susceptibility test results. In communities with a high rate of primary drug resistance, approaches which assign treatment regimens based solely on the patient's history of prior treatment may amplify drug resistance. MDR and XDR patients in China are often poor and cannot afford the second-line anti-TB drugs, making it difficult to achieve treatment success. MDR and XDR patients without adequate treatment regimens will continue to be sources of infection, causing more MDR and XDR infections and TB patients in their communities. Therefore, new strategies, including rapid drug susceptibility testing techniques, are urgently needed. To identify chains of transmission of *M. tuberculosis* is arduous work, requiring a prospective population-based cohort study design, high case detection rates and additional epidemiological information. In the present study, four clusters of patients with identical genotype patterns were identified among 175 MDR TB patients, but epidemiological links between them were difficult to establish. Another limitation of the present study is that isolates of some TB patients were not available for drug susceptibility testing and genotyping, or their drug susceptibility test was performed with the absolute concentration method. However, this is a retrospective study and we were able to investigate only some of the MDR TB patients. Therefore, the recent transmission of *M. tuberculosis* based on genotyping results is likely underestimated. Overall, the treatment outcomes of MDR TB patients have been less favorable than the treatment outcomes of TB patients whose disease is caused by a pan- susceptible strain, a mono-resistant, or a poly-resistant strain of *M. tuberculosis*. Some MDR TB patients can be successfully treated, even as outpatients. The XDR TB epidemic in Tugela Ferry, South Africa, which had a high prevalence of co-infection with HIV and a high case fatality ratio, led some to label XDR TB an “untreatable” disease. However, a study from South Korea reported that more than fifty percent of HIV-uninfected XDR TB patients were successfully treated, as were 60.4% of XDR TB patients in Peru and 41.2% of XDR TB patients completed therapy in California. Our study showed that the cure rate is higher among new TB cases than among previously treated TB cases, whether they were MDR or pre-XDR patients. A previous study reported that TB patients with primary drug resistant tuberculosis had better treatment outcomes than TB patients with acquired drug resistant TB, although this is not always the case. The concept of pre-XDR TB is important for public health purposes, to allow the community to assess how many XDR TB cases are likely to emerge and to make concerted efforts to treat the individuals whose disease is caused by a pre-XDR strain of *M. tuberculosis*. Theoretically, it is difficult to gain resistance to two kinds of drugs simultaneously. In our study, 78% (43/55) of the pre-XDR isolates were resistant to a fluoroquinolone and were, therefore, one mutation away from becoming XDR. Fluoroquinolones have been used extensively in Shanghai during the past twenty years to treat drug-resistant TB patients and retreatment patients. In summary, 5.6% of the TB cases in Shanghai were infected with a MDR strain of *M. tuberculosis*, and further drug susceptibility testing showed that 6.3% of the patients whose isolate was MDR were actually XDR. More than half of the patients with MDR and XDR were new cases, suggesting that the transmission of drug-resistant strains in Shanghai is a serious problem. Currently, the rapid diagnosis and treatment of persons with TB, particularly any form of drug- resistant TB, are high priority public health interventions. Universal drug susceptibility testing is recommended for new and retreatment TB cases. We thank Dr. Peter Small for his valuable review of the manuscript. [^1]: Conceived and designed the experiments: JM QG. Performed the experiments: MZ PX XG LW. Analyzed the data: MZ XL XS KD JM QG. Contributed reagents/materials/analysis tools: XG LW. Wrote the paper: XL KD JM QG. [^2]: The authors have declared that no competing interests exist.

# Introduction Since the development of electronic cigarettes (e-cigarettes) in early 2000 and their introduction as an alternative to cigarettes, scientific evidences on their safety, smoking cessation effects, and adverse reactions have been accumulating. Nevertheless, the e-cigarette market has grown rapidly through aggressive marketing using mass media such as the Internet. In particular, e-cigarette companies are expanding the e-cigarette market primarily by targeting and marketing to adolescents and other young age groups through sponsorship of youth-oriented events, the development of e-cigarette flavors that appeal to youth, and lax youth e-cigarette purchase accessibility through online and offline stores. According to the United States National Youth Tobacco Survey (NYTS), ‘ever e-cigarette’ use in high school students doubled in 1 year from 4.7% in 2011 to 10.0% in 2012 while the estimated number of adolescent ever e-cigarettes users among ‘never cigarette smokers’ more than tripled from 79,000 in 2011 to 263,000 in 2013. McMillen *et al*. reported that the prevalence of current e-cigarette use in the 18–24 age group in the U.S. was 0.0% in 2010, but increased to 14.2% in 2013. Both the prevalence and margin of increase in the 18–24 age group were higher than for higher age groups. The 2013 Canadian Tobacco, Alcohol and Drugs Survey reported that younger age groups (15–24 years old) had higher current and ever e-cigarette use than older age groups (greater than 25 years old). These studies indicate that younger age groups have greater exposure to e-cigarettes. Even if the continued debate about the efficacy of e-cigarettes as an alternative to cigarettes is excluded, the hypothesis that e-cigarettes are a gateway drug to cigarettes in nonsmokers has been proposed as a problem associated with e-cigarettes. According to Bunnell *et al*., intention to smoke cigarettes was 1.7 times higher in e-cigarette ever users than e-cigarette never users among adolescent never smokers. Dutra *et al*. also reported that adolescent e-cigarette users showed high odds ratios (ORs) for current or ever cigarette smoking. According to Lee *et al*., the OR of adolescent current e-cigarette users for current cigarette smokers was 64.9 times higher than that of e-cigarette never users. The NYTS study also supported the hypothesis that e-cigarette ever users were more open to future cigarette smoking. Recent longitudinal studies in US and Swiss also supported that adolescent e-cigarette users were more likely to be cigarette smokers at follow-up assessments (6 or 12 month after baseline). Therefore, it is important to prevent adolescent never smokers from initiating e-cigarette use or former smokers from reinitiating cigarette smoking through e-cigarette use. To do so, it is necessary to identify the nonsmoker groups who are more likely to use e-cigarettes. Given that friends’ cigarette smoking is a strong predictor of smoking initiation in adolescents, this peer influence is also suspected to significantly influence e-cigarette use in nonsmoking adolescents. However, there are few studies on this topic. To the best of our knowledge, the only cohort study on e-cigarette use related to peer group smoking is from Germany. This study reported that the OR of ever e-cigarette use in participants with smoker friends was 2.06 times (overall, including both former smokers and never smokers) or 1.78 times (never smokers) higher than in participants with no smoker friends. In the U.S., a cross-sectional study of psychosocial factors for e-cigarette use showed that both e-cigarette and cigarette use among friends were strongly associated with current e-cigarette use in adolescents. The study showed that the ORs for current e-cigarette use increased with increasing number of friends who smoked tobacco products and ranged from 104 times (3 or 4 out of 4 closest friends smoked e-cigarettes) or 11.2 times (3 or 4 out of 4 closest friends smoked cigarettes) higher than those with no closest friends who smoked e-cigarettes or cigarettes. However, the previous two studies were limited because they did not consider the effects of prior cigarette experience on e-cigarette use. Cigarette smoking status categorized as current, former or never smoker is not only closely associated with factors related to e-cigarette use but is in itself a strong risk factor for e-cigarette use. Therefore, smoking status should be considered when analyzing factors related to e-cigarette use. It is difficult to disentangle the impact of peer cigarette smoking and cigarette smoking status on youth e-cigarette use. In order to remedy this issue, and account for the effect of cigarette smoking status, it is necessary to restrict analyses of the impact of peer smoking on e-cigarette use to non- current smokers and to stratify these analyses into never and former cigarette smoking youth. Therefore, we used nationally representative data from Korean adolescents to investigate the association between peer cigarette smoking and e-cigarette use by nonsmokers and determine whether such association are also dependent on past cigarette experience. # Materials and Methods ## Study population Data were analyzed from the 10^th Korea Youth Risk Behavior Web-based Survey (KYRBS-X) conducted in 2014 by the Korea Centers for Disease Control and Prevention. The KYRBS is a nationally representative, self-reported, and anonymous online survey that was administered to Korean students enrolled in grades 7 to 12. The KYRBS uses a stratified multistage probability sampling design to produce nationally representative statistics on health behaviors in Korean adolescents. A total of 72,060 students from 799 schools (400 middle schools and 399 high schools) completed the KYRBS-X (response rate = 97.2%). Additional details about the sampling methodology and survey procedure are available elsewhere.After excluding current smokers, the study population included 65,753 nonsmokers (7,660 former smokers and 58,093 never smokers). Here, nonsmokers were defined as adolescents who reported not smoking in the past month. This secondary data analysis was exempt from review by the Institutional Review Board of the Daegu Catholic University Medical Center (CR-15-084). ## Measures ### E-cigarette use The e-cigarette use outcome variable was evaluated in two ways for current and ever e-cigarette use. First, ever e-cigarette use was defined as a “yes” response to the following question: “Have you ever used e-cigarettes?” Second, among ever smokers, participants who selected “yes” to the question “During the past 30 days, have you used e-cigarettes?” were considered current e-cigarette users. ### Smoking-related factors Peer cigarette smoking was assessed using responses to the following question: “Do any of your closest friends smoke tobacco?” Participants were provided with four possible answers: 1) None of them, 2) Some of them, 3) Most of them, and 4) All of them. Here, because the sample size of the ‘all’ group (0.8%) was too small to be separately categorized, this group and the ‘most’ group (5.0%) were combined (most/all). Cigarette smoking status among nonsmokers was classified into former smokers or never smokers using a composite measure of the two questions: “Have you ever tried cigarette smoking, even one or two puffs?” and “During the past 30 days, how many days did you smoke, even one puff?” Those who answered “yes” to the first question and “had not smoked in the past 30 days” to the second question were classified as former smokers. Those who answered “no” to the first question were classified as never smokers. Participants with a family member such as a parent, sibling, or grandparent who currently smoked cigarettes were considered to be living with a household member that used tobacco. ### Other characteristics Other covariates were categorized into two domains: sociodemographic and lifestyle. Sociodemographic variables included sex, school type (middle school, general high school, or vocational high school), region of residence (metropolitan city, city, or province), and perceived academic performance (high, middle, or low; classified using the question, “During the past 12 months, how would you rate your academic performance?”). Lifestyle and psychosocial factors included frequency of alcohol drinking per month (never, less than 6 times, or 6 or more times), experience of drug use (yes or no; classified using the question, Have you ever taken a drug or inhaled butane gas/bond habitually or intentionally), and perceived stress level (low, middle, or high; classified using the question, “How much stress do you usually feel?”). ## Statistical analysis Multivariate logistic regression was conducted to estimate the relationship between peer cigarette smoking and e-cigarette use after adjusting for covariates including sex, school, location, perceived academic performance, alcohol intake, drug experience, perceived stress level, cigarette smoking status, and current smoking household member. In order to assess the association with peer cigarette smoking from the perspective of both simple smoking experience and current e-cigarette use, two different models were used to assess risk factors for current and ever e-cigarette use. Because cigarette smoking status is closely associated with e-cigarette use, an interaction model between peer cigarette smoking and cigarette smoking status was used to estimate the association between peer cigarette smoking and e-cigarette use according to cigarette smoking status. All analyses were performed using SPSS version 19.0 (IBM, Armonk, NY, USA) and a p-value of \< 0.05 was considered significant. Complex SPSS sampling procedures were used to accurately represent the adolescent population in Korea. # Results Among adolescent nonsmokers, 3.8% and 1.2% were ever and current e-cigarette users, respectively. Ever e-cigarette use was significantly higher in former smokers (19.9% vs. 1.6% for never smokers; p\<0.01) and in those with a current smoking household member (4.2% vs. 3.2%; p\<0.01). Current e-cigarette use was also significantly higher in former smokers (6.3% vs. 0.5% in never smokers; p\<0.01) and those with a current smoking household member (1.3% vs. 1.1%; p\<0.05). The prevalence of ever or current e-cigarette use significantly increased as the proportion of closest friend smokers increased (p\<0.01). Therefore, 19.8% or 8.7% of nonsmokers with most/all smoking closest friends were ever or current e-cigarette users, respectively. With the exception of location, all other well-known risk factors for cigarette smoking or e-cigarette use were significantly associated with e-cigarette use. After adjusting for all of the covariates, former smokers were more likely to be ever (adjusted OR = 7.89; 95% confidence interval \[CI\], 7.17–8.69) or current e-cigarette users (adjusted OR = 5.32; 95% CI, 4.52–6.27) than never smokers. A dose-response relationship between the proportion of closest friend smokers and e-cigarette use was observed. Specifically, adjusted ORs for ever e-cigarette use increased from 2.05 (95% CI, 1.82–2.30) for participants with some closest friends who smoked to 5.50 (95% CI, 4.77–6.34) for those who reported most or all of their closest friends smoked. Similarly, adjusted ORs for current e-cigarette use increased from 2.23 (95% CI, 1.77–2.81) to 7.82 (95% CI, 5.97–10.25) across these groups. The prevalence of e-cigarette use increased steadily with an increase in the proportion of smoking closest friends among both never smokers and former smokers. Under the same condition of the proportion of smoking closest friends, the prevalence of ever or current e-cigarette use was consistently higher in former smokers than never smokers. However, the rate of increase of e-cigarette use prevalence in never smokers was greater than that in former smokers. In particular, while the prevalence of ever e-cigarette use in former smokers increased 3.3 times (from 10.95% in those with no smoking closest friends to 36.61% in those with most/all closest friends), it increased 15.5 times (from 0.68% to 10.57%) in never smokers. The results from the interaction model between peer cigarette smoking and cigarette smoking status are shown in. Regardless of cigarette smoking experience, the adjusted ORs for current or ever e-cigarette use increased significantly with increasing proportion of smoking closest friends in never smokers (p\<0.001 for trend). This significant association was also present in former smokers. However, a significant interaction between the proportion of smoking closest friends and cigarette smoking status was observed in both ever and current e-cigarette use models (p\< 0.001 for interaction) and the slopes of the adjusted ORs for current or ever e-cigarette use in never smokers were more than twice as steep as those in former smokers. The adjusted ORs in interaction terms (Former ⨉ Some, Former ⨉ Most/All) were significantly lower than 1.0 in both the ever and current e-cigarette use models; this means that former smokers are less likely to be affected by the proportion of closest smoking friends compared to never smokers. The interaction effect was larger in the most/all group than the some group. Specifically, the adjusted ORs in interaction terms for Former ⨉ Most/All (ever e-cigarette use, 0.39; current e-cigarette use, 0.26) were lower than those for Former ⨉ Some (ever e-cigarette use, 0.69; current e-cigarette use, 0.54). # Discussion This study showed that the proportion of closest friends who smoked had a significant relationship with e-cigarette use in adolescent nonsmokers. Both ever and current e-cigarette use increased significantly as the proportion of closest friend smokers increased, regardless of past smoking experience. This finding is consistent with results from a German cohort study that indicated peer cigarette smoking affected lifetime e-cigarette use. Furthermore, our results suggest that in adolescents peer group cigarette smoking plays an important role in not only cigarette smoking but also in e-cigarette use. In particular, considering that the adjusted OR of peer group cigarette smoking was higher among the analyzed variables, the results were consistent with previous results that indicated peer group cigarette smoking can have a significant influence on various types of adolescent smoking such as smokeless tobacco. In a longitudinal study conducted with 12-year-old adolescents in the United States, as friend compliance (measured with the following question; “I do what my friends want me to do, even if I really don’t want to.”) increased, the use of smokeless tobacco also increased. In addition, another cross-sectional study of U.S. adolescents showed that approval and use of e-cigarettes and cigarettes among friends were strongly associated with e-cigarette use. These findings suggest that not only directly assessed peer smoking but also perceived peer influence or psychosocial factors regardless of cigarette use can influence the use of cigarette alternatives including e-cigarettes. In addition to peer group cigarette smoking acting as a direct pathway to adolescent cigarette smoking, an alternative or indirect pathway to adolescent cigarette smoking can occur via e-cigarette use. Several longitudinal studies among adolescents or young adults in US or Swiss reported recently that e-cigarette use in adolescent non-smokers or never-smokers was closely associated with both willingness to smoke and smoking initiation. Moreover, this phenomenon can be accelerated by the renormalization strategy from aggressive e-cigarette marketing. In the past several decades, efforts have been made to establish desirable social norms about smoking through denormalization strategies, which are main strategies used in global tobacco control. Tobacco industry denormalization strategies have also shown a reduction in the rate of adolescent cigarette smoking. However, the psychological barriers to cigarette use formed through denormalization strategies has been threatened by e-cigarette company renormalization strategies through various marketing techniques. As a result, there is the risk that lowered psychological barriers from renormalization strategies increase the likelihood of cigarette smoking via e-cigarette use (indirect pathway), rather than through peer group cigarette smoking (direct pathway). In particular, the present study showed that the influence of peer group cigarette smoking on e-cigarette use was over 2 times higher in never smokers than former smokers and this difference could be explained by renormalization. In other words, psychological barriers to e-cigarette use have already been lowered in former smokers as a result of their past smoking experience. Consequently, former smokers may perceive e-cigarette use as non-deviant behavior; therefore, the role of renormalization strategies by e-cigarette companies may be nonsignificant in former smokers compared to never smokers. In contrast, in never smokers, the threshold of psychological barriers that recognizes e-cigarette use as deviant behavior may be lower than that for cigarette use due to renormalization strategies. As a result, it is believed that even never smokers who were not influenced by peer group cigarette smoking to initiate cigarette smoking may react to e-cigarette use and show a stronger response to peer influence than former smokers. Specifically, having more friends who were cigarette smokers was associated with a greater response margin to e-cigarette use regardless of past cigarette experience. Because having a greater number of peer group cigarette smokers leads to lower negative perceptions about cigarette smoking, these results suggest that the interaction between peer group cigarette smoking and e-cigarette renormalization had a more substantial influence on diminishing psychological barriers. Because the present study was cross-sectional, it was unable to evaluate whether there was a causal relationship between peer group cigarette smoking and e-cigarette use. Moreover, because it posed no questions about social norms or perceptions about cigarettes or e-cigarettes, it was difficult to directly identify the cause of the differential influence of peer group cigarette smoking on e-cigarette use between former and never smokers. Thus, there is a need for continued study of this subject in the future. Despite these limitations, the present study had the following advantages. First, the study used nationally representative data from a large-scale survey including 2,369 ever e-cigarette users and 752 current e-cigarette users. Second, after excluding current smokers, analyses were performed by dividing the subjects into former and never smokers. This enabled accurate assessments of peer group cigarette smoking on e-cigarette use, including simultaneous analysis of the interaction between peer group cigarette smoking and cigarette smoking status. Finally, when comparisons were made with comparable 2011 U.S. NYTS data, it was revealed that e-cigarette use in Korean adolescents (4.7%) was more than 4 times higher than in U.S. adolescents (1.1%). Furthermore, among current smokers, the dual user rate (36.6%) was much higher than that of U.S. adolescents (10.6%). Moreover, the rate of current e-cigarette only users was higher in Korean adolescents (1.1%) than in U.S. adolescents (0.6%). Considering that e-cigarettes are widely marketed through the Internet and South Korea is globally an Internet powerhouse, South Korea is at risk for dramatic increases in future e-cigarette use. That being the case, the present study was meaningful in that it is the first to assess factors related to e-cigarette use in Korean adolescents. # Conclusion Peer group cigarette smoking had an important relationship with adolescent e-cigarette use and this relationship was greater on never smokers than former smokers. These findings give warning that peer group cigarette smoking can be combined with e-cigarette renormalization strategies to enable the expansion of the e-cigarette market by reaching adolescent never smokers who would otherwise not be interested in cigarette use. Not only did e-cigarette users have positive perceptions of e-cigarettes, they also had liberal views on future cigarette smoking and had high potential to become dual users. Therefore, it is essential to continue to study e-cigarettes to accurately assess their health hazards and use e-cigarette denormalization strategies to instill proper e-cigarette perceptions in adolescents. This will mitigate adolescents’ expanded use of e-cigarettes and prevent cigarette smoking in this group. [^1]: The authors have declared that no competing interests exist. [^2]: **Conceptualization:** SWP JHH. **Formal analysis:** JHH. **Funding acquisition:** SWP. **Methodology:** JHH. **Writing – original draft:** JHH. **Writing – review & editing:** SWP JHH.

# 1 Introduction Reaching for objects and manipulating them with our hands is one of the most natural abilities of primates and a fundamental way of interacting with our environment. We can learn to coordinate our limbs and hands elegantly, dexterously and almost effortlessly, yet implementing this level of skillfulness in artificial systems or robotic platforms still poses a challenging problem. Part of the problem of reaching for an object in space involves how we represent the location of such a target object relative to the body or the relevant part of the body. Neuroscientific studies seem to suggest that the brain represents the location of objects using a rich tapestry of coordinate frames. Investigating the reference frames of neurons performing, guiding or preparing motor actions is among the most prominent issues of current neuroscience. It is crucial for understanding how spatial information is processed and represented in the brain in order to facilitate the planning of movements within our peripersonal space. While visual information is initially represented in retino-centric coordinates, precise guiding of motor tasks such as visually-guided reaching is enabled by non-retinal coordinate maps connected to various body parts in the later stages of processing. The occipito-parietal components of the dorsal stream connect via the parieto-premotor pathway to various regions of the posterior parietal cortex (PPC) particularly to areas within the intraparietal sulcus (IPS) and the superior parietal lobe. In general, the parieto-premotor pathway seems to perform the necessary coordinate transforms required for planning and guiding motor tasks. Neurons encoding the location of objects in a head-centred, limb- centred, body-centred, hand-centred frame of reference or mixtures of the latter two have been reported. Single cell recordings in non human primates generally reveal a gradient in response properties from predominantly eye-centred encoding in early visuomotor areas to body-centred or effector-centred representations in more parieto-frontal areas, including hand-centred representations of reach vectors found in area 5d. Neurons with hand-centred receptive fields fire maximally whenever a target object is at a specific position with respect to the hand while remaining relatively invariant against head movements, saccades or even across different positions of the hand relative to the body. Hand-centred encoding is not restricted to area 5d but has been found across different parts of the PPC and premotor cortex. While in ventral premotor cortex (PMv) hand- centred representations independent from eye position have been reported, in dorsal premotor cortex (PMd) neurons encoding reference frames centered on the hand, eye or a combination of both have been found. Recently, studies have also identified and studied the role of hand-centered representations in the human brain. The aim of this paper is to present a new model of how visual neurons with hand- centred receptive fields may develop in the primate brain using biologically plausible learning mechanisms. Various computational models have been previously suggested to account for the different stages of coordinate transformations, aiming to explain some of the response properties found in neurophysiological recordings. For example, an early model developed by Zipser and Andersen showed how eye-centred visual representations could be mapped to head-centred representations by supervised learning in a backpropagation of error network. Later, Chang et al. adapted the model of Zipser and Andersen by adding a proprioceptive hand-position signal, which allowed the development of hand- centred responses in the output layer. However, these two models made use of supervised backpropagation of error learning, which is generally regarded as biologically implausible for a couple of reasons. For example, it is difficult to explain where a suitable teaching signal might come from to guide supervised learning of coordinate transformations from one reference frame (e.g. eye- centred) to another (e.g. hand-centred) in the visual system. Moreover, the propagation of the teaching signal down to earlier layers is regarded as biologically implausible because the feedback neurons would require precise knowledge of the derivatives (with respect to the target signal) of the feedforward neurons exactly at the operating time. For these reasons, we propose that a more biologically plausible approach for explaining the development of coordinate transformations in the primate visual system is to build neural network models that self-organise their synaptic connectivity through an unsupervised process of visually-guided competitive learning. However, progress with such model architectures has been fairly scarce. In an extensive approach to accounting for the neural mechanism underlying coordinate transforms, Pouget and Sejnowski modelled the response properties of neurons in PPC by basis function units that could represent multiple reference frames simultaneously. Recently, Magosso et al. proposed an unsupervised neural network model that integrates visual and tactile peri-hand representations to generate bimodal hand-centred neurons. Nevertheless, these authors did not provide a theory of how hand-centred visual neurons might arise in the primate brain because they assumed their existence in early layers of the network. Another study was recently carried out by Galeazzi et al., who modelled the visually-guided development of hand-centred visual neurons using an unsupervised, competitive learning mechanism. These authors demonstrated the development of neurons that responded selectively to visual objects presented in particular locations with respect to the hand, while responding invariantly as the hand-object configuration was shifted across different retinal locations. It was subsequently shown that this unsupervised learning model was sufficiently robust that it could still develop hand-centred visual representations even when being trained with more realistic images of the hand presented against natural scenes or when being trained on image sequences that were driven by natural human gaze changes recorded with an eyetracking system. These model simulations all relied on an invariance learning mechanism known as *trace learning*. This learning mechanism operates by binding together input patterns that tend to occur in temporal proximity. If the eyes saccade around a visual scene containing a fixed hand-object configuration during visual training, then trace learning causes the corresponding images of that hand-object configuration in different retinal locations to become associated with the same subset of output neurons. These neurons are then able to respond selectively to the particular hand-object configuration in a shift invariant manner across the different retinal locations seen during training. However, the trace learning mechanism used by Galeazzi et al. has a couple of potential limitations from the point of view of biological plausibility. First, the trace learning rule used to modify the synaptic weights needs to incorporate a memory trace of recent neuronal activity, and so it is significantly more complex than a standard Hebbian learning rule. Secondly, even more limiting is the fact that trace learning requires that different retinal views of the same hand-object configuration must be seen clustered close together in time. That is, each hand-object configuration must be thoroughly visually explored through a sequence of saccadic shifts in retinal position before encountering the next hand-object configuration. If these image sequence statistics are compromised during training, for example if different hand-object configurations are seen in quick succession, then this will sharply degrade the ability of the model to develop hand centred visual neurons. This may occur if the relative positions of the hand and nearby objects are changing rapidly, which will happen during reaching for example. This weakness of trace learning was demonstrated in the original study of Stringer et al., which showed trace learning failing when two different visual stimuli were rapidly interleaved through time. These limitations of the trace learning mechanism employed by Galeazzi et al. beg the question of whether an alternative learning mechanism might be found that can produce hand-centred visual neurons using a simpler Hebbian learning rule and without the restrictive assumption that particular hand-object configurations must be visually explored through a lengthy series of saccades before a new hand-object configuration comes into view. In this paper, we present a new unsupervised, competitive learning model of how hand-centred visual neurons may develop, which eliminates the two potential weaknesses of trace learning described above. This simplified approach utilises an invariance learning mechanism known as *continuous transformation (CT) learning*, which was successfully applied to the problem of rotation invariant visual object recognition by Stringer et al. and to guide translation invariant visual representations by Perry et al.. CT learning relies on a standard Hebbian learning rule, with no need for a memory trace of recent neuronal activity. The learning mechanism operates by binding together subsets of smoothly changing input patterns, for example, corresponding to particular hand-object configurations seen shifted across different retinal locations due to saccades. This binding is achieved by exploiting the spatial overlap that naturally occurs between similar images such as the same hand-object configuration presented in nearby retinal locations. Details of the CT learning mechanism and its performance properties are described in. Importantly, CT learning does not require that the input patterns associated together (e.g. different retinal views of the same hand-object configuration) are seen in close temporal succession. In other words, CT learning is not limited by constraints on the temporal sequencing of training images. Our simulations described below confirm that this learning mechanism is robust with respect to the order of stimulus presentation. CT learning may consequently cope with a much broader range of natural spatiotemporal image statistics, in which the relative positions of the hand and surrounding objects do not remain fixed for relatively long periods. Regarding generalization, we can further report that our model responds properly to untrained hand-object configurations given that the density of initial training across the hand-centred space was sufficiently high. Furthermore, to verify the robustness of this learning mechanism to more realistic training conditions, we increased the density of trained hand-centred locations from almost orthogonal locations in the first simulations (e.g. *left*, *top*, *right* around the hand) to a near continuum (i.e. \>10 locations). This poses a potential challenge for CT learning given its inherent learning properties. CT learning operates by simply binding together subsets of smoothly varying input patterns by exploiting the spatial overlap or similarity between successive members. However, by increasing the density of trained hand- centred locations, neighbouring locations start to show a greater degree of overlap, making it more difficult to separate different views of highly overlapping hand-object configuration. The question is then, how would CT learning behave in this scenario. First, CT learning could continue to bind together different retinal views of the same (or neighbouring) hand-object configurations to produce neurons with hand-centred receptive fields. Alternatively, CT learning could also start binding together overlapping input patterns corresponding to different hand-object configurations at the same retinal location, and therefore fail to produce hand-centred representations that respond selectively to a highly localised region of hand-centred space. Indeed, our simulations showed that the network continued to develop hand- centred neurons even as the number of trained hand-centred locations approached a continuum. In this case, individual hand-centred neurons learn to respond to a small localised region of hand-centred space across different retinal locations. Importantly, we find that the receptive fields of these hand-centred neurons are distributed uniformly through the hand-centred space. In the final results section of this paper, we provide a theoretical framework for explaining why CT learning continues to drive the development of hand- centred neurons even when the network is trained on a (near) spatial continuum of hand-centred target locations. This is achieved by applying Principle Component Analysis (PCA) to the visual stimulus sets, consisting of alternative hand-object configurations presented across different retinal locations, which are used to train the network. It is found that the main directions of variance within the training stimuli can predict whether or not the network develops hand-centred neurons using CT learning. Most importantly, this new framework carries implications of various previous work that utilised the same learning mechanism for translation or rotation invariant object recognition as well as differentiating face identity from face expression. # 2 Methods ## 2.1 VisNet The numerical simulations presented in this paper were conducted using the VisNet model of hierarchical processing in the primate visual system, which is shown in. The same VisNet architecture was used in our previous simulations of the development of hand-centred visual neurons using trace learning. However, in the simulations reported below, we implement a Hebbian learning rule in the model instead of the trace rule. VisNet is composed of a hierarchy of four competitive layers of neurons. Within each layer, the neurons compete with each other through lateral inhibitory mechanisms described below. Layer 4 is the output layer, which develops hand- centred visual neurons after training. Between successive layers of neurons there are bottom-up (feed-forward) synaptic connections, which are modified by a Hebbian learning rule during training. This hierarchical network architecture loosely represents the hierarchical organisation of the primate dorsal visual pathway. The forward connections between the retinal layer and the first layer of the network are derived from a topologically corresponding region of the preceding layer, using a Gaussian distribution of connection probabilities. These distributions are generated by a radius containing approximately 67% of the connections from the preceding layer. The connection probabilities for later layers are derived from a uniform distribution across all neurons from the preceding layer. This leads to an increase in the receptive field (RF) size of neurons through successive layers of the network hierarchy. The network dimensions used for this study are shown in. The natural visual images of size 128 × 128 pixels are pre-processed by an initial layer mimicking the response properties of V1 simple cells, wherein each (*x*, *y*)-location contains a bank of Gabor filter outputs corresponding to a hypercolumn generated by $$\begin{array}{r} {g(x,y;\lambda,\theta,\psi,\sigma,\gamma) = \exp\left( - \frac{x^{\prime 2} + \gamma^{2}y^{\prime 2}}{2\sigma^{2}} \right)\cos\left( 2\pi\frac{x^{\prime}}{\lambda} + \psi \right)} \\ \end{array}$$ $$\begin{array}{rcl} x^{\prime} & = & {x\cos\theta + y\sin\theta} \\ \end{array}$$ $$\begin{array}{rcl} y^{\prime} & = & {- x\sin\theta + y\cos\theta} \\ \end{array}$$ for all combinations of *λ* = 16, *γ* = 0.5, *σ* = 0.56*λ*, *θ* ∈ {0, *π*/4, *π*/2, 3*π*/4} and *ψ* ∈ {0, *π*, −*π*/2, *π*/2}. Subsequently the filtered images are fed into the first layer of the network architecture. The activation *h*_*i* of each neuron *i* in the network is defined as a linear sum of the inputs *y*_*j* from afferent neurons *j* in the previous layer weighted by the corresponding synaptic weights *w*_*ij*. That is, $$\begin{array}{r} {h_{i} = \sum\limits_{j}w_{ij}y_{j}} \\ \end{array}$$ where *y*_*j* denotes the firing rate of the presynaptic neuron *j*, and *w*_*ij* the synaptic strength from neuron *j* to neuron *i*. Within each layer competition is implemented in the following two stages and is graded rather than winner-take-all. First, to implement lateral inhibition the activations of neurons within a layer are convolved with a spatial filter *I*, where *δ* controls the contrast, *σ* controls the width, and *a* and *b* index the distance to the centre of the filter. $$\begin{array}{r} {I_{a,b} = \left\{ \begin{array}{ll} {- \delta e^{- \frac{a^{2} + b^{2}}{\sigma^{2}}}} & {\text{if} a \neq 0\text{or} b \neq 0,} \\ {1 - \sum\limits_{\begin{array}{l} {a \neq 0} \\ {b \neq 0} \\ \end{array}}I_{a,b}} & {\text{if} a = 0\text{and} b = 0.} \\ \end{array}\operatorname{} \right.} \\ \end{array}$$ Typical lateral inhibition parameters are given in. Next, contrast enhancement is applied by means of a sigmoid activation function $$\begin{array}{r} {y = f^{sigmoid}(r) = \frac{1}{1 + e^{- 2\beta(r - \alpha)}}} \\ \end{array}$$ where *r* is the activation after lateral inhibition, *y* the firing rate after contrast enhancement, and *α* and *β* the sigmoid’s threshold and slope respectively. The parameters *α* and *β* are constant within each layer, although *α* is adjusted to control the sparseness of the firing rates. For the simplified case of neurons with binarised firing rates ∈ {0, 1}, the sparseness of the firing within a layer is the proportion of neurons that are active. So, for example, in order to achieve a sparseness of 5%, the threshold *α* is set to the value of the 95th percentile point of the activations within the layer. Typical parameters for the sigmoid activation function are shown in. At the beginning of training, the synaptic weights throughout the network are initialised to random values. These synaptic weights are then updated each time an image of particular hand-object configuration in a specific retinal position is presented to the network. It is these learning updates to the synaptic weights that drive the development of hand-centred neuronal responses in the output layer. In previously published work, we used a trace learning rule to model the development of hand-centred receptive fields. However, in the new simulations with CT learning reported below, the synaptic connections between successive layers of neurons are updated using the Hebbian learning rule $$\begin{array}{r} {\Delta w_{ij} = \alpha y_{i}y_{j}} \\ \end{array}$$ where *y*_*j* is the firing rate of presynaptic neuron *j*, *y*_*i* is the firing rate of postsynaptic neuron *i*, and *α* is the learning rate which is set to 0.1. In order to limit the growth of each neuron’s weight vector ***w***_***i***, synaptic scaling is applied. To regulate the total synaptic input driving each neuron, while maintaining the relative strengths of individual synapses on the neuron’s dendrites, the length of each neuron’s synaptic weight vector is renormalized at the end of each timestep during training by setting $$\begin{array}{r} {\sqrt{\sum_{j}w_{ij}^{2}} = 1} \\ \end{array}$$ as usual in competitive learning. This computational step is in accordance with neuronal recordings by Turrigano et al. that showed synaptic learning to be accompanied by synaptic scaling as homeostatic regulation. In a computational model, Lazar et al. showed absence of weight vector renormalization caused neurons to encode redundant information by synchronizing their firing patterns. ### 2.1.1 Relation to HMAX Model VisNet has been often compared to HMAX, another computational model of the visual cortex that was established by Riesenhuber and Poggio and further improved by Serre. Like VisNet, HMAX aims for biological plausibility and utilises an unsupervised, hierarchical structure to find a suitable trade-off between selectivity and invariance in visual recognition. Typically, HMAX contains by a factor 100 more computational units than VisNet (65,536 for a 128 × 128 retina on all 4 layers). In HMAX, the final layer’s output is usually fed into a non-biologically plausible support vector machine or least squares algorithm for task classification, whereas VisNet uses Shannon information theory measures, based on dot product encoding, that are likely to occur in the brain in a similar fashion. While VisNet is usually trained with image sequences as they transform in the world (such that rotation, view, translation and illumination invariance can be learned), HMAX is dominantly trained with images from large databases that seek for object classification. In a detailed comparison, Rolls and Robinson, revealed that object classification works rather poor with both approaches (compared to backpropagation approaches) and that VisNet develops invariant representations that generalise to unseen transforms (not so for HMAX). Most importantly, VisNet did not respond to scrambled faces when trained on proper faces, suggesting the network to encode shape information, whereas HMAX retained stable firing rates for scrambled and unscrambled faces, suggesting the network to encode low-level features such as texture or particular facial spots. ## 2.2 Continuous transformation learning hypothesis Established by Stringer et al., CT learning is an unsupervised competitive learning mechanism relying on spatial similarity of successive transforms of a visual stimulus, for instance, as it smoothly shifts across the retina during a sequence of saccades. An idealized example of how CT learning guides the development of neurons with localized, hand-centred receptive fields in VisNet shows. The operation of CT learning is shown in a simplified network architecture with a single layer of bottom-up synaptic connections between an input layer and an output layer. The neurons in the input layer may be taken to provide an idealised visual representation of a particular spatial configuration of the hand and a nearby object. When such an input pattern is presented to the network, activation is evoked in a subset of neurons in the input layer. Then activation is transmitted through the initially random forward connections to the output layer, where one (or more) of the output neurons wins the competition and fires (shaded neuron). Next, the synaptic weights between active neurons in the input and output layers get strengthened according to the Hebbian learning rule (shown as solid lines between the layers). A gradual shift in the position of the visual stimulus (i.e. a hand-object configuration) across the retina will activate a sequence of spatially overlapping input patterns. Next, the activation of some input neurons with previously strengthened synaptic weights will drive the activation of the same output neuron and make it more likely to win the competition again. Consequently, the synaptic weight between this output neuron and the newly active input neuron(s) will get strengthened (dashed line). This procedure is repeated again and again as the visual stimulus shifts across the retina due to saccades during training. As a result, after training, the same output neuron will have learned to respond to the particular hand-object configuration across different retinal locations due to the high degree of spatial overlap between these images. In contrast to trace learning, it is important to emphasise that CT learning does not require that each spatial configuration of the hand and a target object is seen with temporal continuity as it shifts across the retina. In other words, CT learning can still bind together different retinal views of the same hand- object configuration even if different hand-object configurations are seen rapidly interleaved with each other through time. CT learning thus potentially offers a far more robust mechanism than trace learning for the development of hand-centred neurons, in that it can cope with a broader, more natural range of visual training sequences, in which the relative positions of the hand and nearby objects are sometimes changing rapidly as well as remaining stationary for a long period of time. Furthermore, although the explanation of CT learning given above and illustrated in would seem to imply that each hand-object configuration needs to be shifted continuously in small steps across the retina, it was shown by Stringer et al. that the learning mechanism is much more flexible than this. In fact, stimulus representations that are invariant to shifts in retinal location can still develop even if the same hand-object configuration is not shown in any two consecutively presented frames, like for example a moving object passing a static hand. In this case, output neurons may continue to develop (retinal) location invariant responses as long as the hand-object configuration is presented to the network across the entire space of overlapping retinal positions over the total training session. A related mechanism to CT learning, Slow Feature Analysis (SFA), has been shown to extract slowly varying features from a sequence of images to form translation or rotation invariant representations. As we suggest in this paper, CT learning drives the network to encode information along the directions of highest variance (i.e. to approximate a PCA), whereas SFA finds principal components with minimal eigenvalues (i.e. least variance). Importantly, SFA utilises the fact that the sensory signal of any photoreceptor varies quickly while the environment varies comparably slowly; thus it is the temporal structure of the signal determining the result of the algorithm. In contrast to that, CT learning allows for interleaving since it exploits purely the spatial overlap of the input signal, irrespective of any temporal component as shown in section 3.3. Therefore, as long as a feature has higher correlation within different views of the same hand-object configuration than in views of different hand-object configurations, then it is more likely to be represented by the same neuron. This allows to develop invariant representations not only if the related feature slowly varies over time but entirely independent of the temporal dimension. Accordingly, SFA seems to be more closely related to trace learning than to CT learning. However, unlike the latter two learning mechanisms, that are based on Hebbian learning, SFA is not regarded as biologically plausible because there is no evidence suggesting SFA to be computed by neurons in the brain. ## 2.3 Network performance analysis We developed techniques for analysing the response characteristics of neurons using two different methodologies. In the first approach, we tested whether neurons in the output layer had learned to encode a finite number of discrete hand-centred locations, with individual neurons responding to only one of these hand-centred locations regardless of the retinal position of the hand-object configuration. However, a population of hand-centred neurons in the brain needs to cover a continuous space of such hand-centred locations, with individual neurons having a finite sized receptive field covering some small region of the hand-centred space. Therefore, our second approach treated the hand-centred space as a spatial continuum, with the receptive fields of individual hand-centred neurons covering a small region of this space. We describe the mathematical details of these two analytical approaches next. ### 2.3.1 Analysis based on discrete hand-centred locations In our first approach to analysing the response characteristics of output neurons we test how well they have learned to represent a finite number *N* (up to ten) of discrete hand-centred locations across different retinal locations. Each ideal hand-centred neuron should respond to only one of the hand-centred locations, and should respond to that hand-centred location across all retinal positions of the hand-object configuration. In order to assess how well the output neurons represent a finite number *N* of discrete hand-centred locations, we utilize two information theory measures, single and multiple cell information. Both have been applied previously to analyse the performance of the VisNet model, and were introduced by Rolls and Millward. In this study, we define a stimulus as a particular hand-object configuration. Whenever an output neuron responds to a particular hand-object configuration across all retinal locations, but does not respond to any other hand-object configuration, then this neuron will convey maximal single cell information. The amount of single cell information carried by a specific neuron about a particular hand-object configuration is computed by $$\begin{array}{r} {I(s,R) = \sum\limits_{r \in R}P\left( r \middle| s \right)\log_{2}\frac{P\left( r \middle| s \right)}{P(r)}} \\ \end{array}$$ where the stimulus-specific information *I*(*s*, *R*) is the amount of information the set of responses *R* of a single cell has about a specific stimulus *s* (i.e. target location with respect to the hand), while the set of responses *R* corresponds to the firing rate *y* of a cell to each of the hand-object configurations presented in all retinal locations. The maximum single cell information possible is $$\begin{array}{r} {\text{Max.}\text{single}\text{cell}\text{information} = \text{log}_{2}(N),} \\ \end{array}$$ where *N* is the number of different spatial configurations of the hand and target object. For example, with *N* = 8 hand-object configurations, the maximum single cell information is 3 bits. In case a neuron does not respond to its preferred hand-object configuration across all retinal locations or in case it responds to another hand-object configuration, then the amount of single cell information carried by that neuron will not reach the theoretical limit. Thus, single cell information is a useful measure for assessing how hand-centred the response characteristics of a neuron are, with maximal single cell information implying that the neuron displays perfect hand-centred responses. Nevertheless, it does not address the question whether all *N* hand-object configurations are represented by the population of output neurons. For this reason, the multiple cell information computes the average amount of information about which hand-object configuration was presented on the basis of the responses of the most informative output cells. This procedure is used to verify whether, across the entire population of output cells, there is information about all of the different hand-object configurations that have been shown. Methods for calculating the multiple cell information measure have been previously described by Rolls and Millward. In brief, from a single presentation of a stimulus (i.e. hand-object configuration), we calculate the average amount of information obtained from the responses of all the cells regarding which stimulus is shown. This is achieved through a decoding procedure that estimates which stimulus *s*′ gives rise to the particular firing rate response vector on each trial. A probability table of the real stimuli s and the decoded stimuli *s*′ is then constructed. From this probability table, the mutual information is calculated as $$\begin{array}{r} {I\left( S,S^{\prime} \right) = \sum\limits_{s,s^{\prime}}P\left( s,s^{\prime} \right)\log_{2}\frac{P\left( s,s^{\prime} \right)}{P(s)P\left( s^{\prime} \right)}.} \\ \end{array}$$ Multiple cell information values are calculated for the subset of output cells which, according to the single cell information analysis, have the most information about which stimulus (hand-object configuration) is shown. In particular, the multiple cell information is calculated from 5 cells for each stimulus that had the most single cell information about that stimulus. Previous research has found this to be a sufficiently large subset of cells to inform the multiple cell information measure. All results for single and multiple cell information measures presented in the paper are averages across 10 simulations with different random seeds used to initialise the network synaptic connectivity and synaptic weights. ### 2.3.2 Analysis based on a continuum of hand-centred locations The single and multiple cell information described above suffer from the deficiency that they assume a finite number of discrete hand-centred locations as well as equidistant differences between all pairs of these stimuli. While a cell is computed as carrying maximal information if it responds selectively to one hand-centred location, responses for neighboring hand-centred locations are not taken into consideration. However, of course, neurons in the brain must represent a spatial continuum of target locations in hand-centred space. In this case, individual neurons respond to a small localised region of hand-centred space rather than a single point. Therefore, in our second approach to analysing the response characteristics of output neurons we verify its generalisation abilities by increasing the number of tested hand-object up to 30 highly overlapping hand-centred locations. Therefore, we introduce a method to quantify how well each cell has learned to represent a localised region of hand-centred space across different retinal locations. We classify a neuron in the output layer to be hand-centred using the following steps. First, we compute the firing rate responses of the neuron to each tested hand-object configuration presented in every retinal position. This gives a response matrix for the neuron, with one dimension being the hand-centred location of the target object and the other dimension being the retinal position of the whole hand-object configuration. Next, the response matrix for the cell is binarised by thresholding the firing rate responses. That is, if the firing rate response of the neuron is below the threshold *T* = 0.5 then it’s response is set to zero, otherwise it is set to one. Based on the resulting binary response matrix, also shown further below, a neuron is classified as hand- centred if it fulfils the following two criteria: The neuron displays a single localised region of response within its binary response matrix. The neuron displays an elongated response profile, whereby it responds to a relatively small region of the hand-centred space of target locations while responding to a much larger region of the retinal space for those preferred hand-object configurations. This is assessed using the following inequality: $$\begin{array}{r} {\sqrt{\frac{1}{n - 1}\sum\limits_{i = 1}^{n}\left( x_{i} - \overline{x} \right)^{2}} \geqslant \lambda \cdot \sqrt{\frac{1}{n - 1}\sum\limits_{i = 1}^{n}\left( y_{i} - \overline{y} \right)^{2}}} \\ \end{array}$$ where *n* is the total number of combinations of hand-centred target location *x*_*i* and retinotopic location of the whole hand- object configuration *y*_*i* that activate the cell, and *λ* is an adjustable parameter controlling the sensitivity of the criterion. Both, *x* and *y* dimensions are normalized in the range of \[0, 1\]. Examples of neurons with such elongated response fields are depicted further below in section 3.4. For all simulations *λ* was set to 4. Although these criteria can identify a population of hand-centred output neurons, this does not inform us about whether all local subregions of the hand- centred space of target locations are represented equally. Does the output population learn to represent the entire space of hand-centred target locations evenly, or does the network completely fail to represent some local subregions? In order to investigate this question, we carried out a further stage of analysis, which counts the amount of hand-centred neurons per training location, as follows. Each identified hand-centred neuron contributes to each location by the relative invariance of its response to that configuration over all retinal locations. For example, if a hand-centred neuron responded to all 10 retinal locations of one particular hand-object configuration and 5 retinal locations of a neighbouring configuration, then it is considered to be 67% and 33% hand-centred for each of those configurations, respectively. These values are computed for all identified hand-centred output neurons, and then used to assess the degree of coverage of the hand-centred space of target locations by the entire output population. Ideally, every location along the hand-centred space should be represented by approximately the same amount of hand-centred neurons. ## 2.4 Training procedure During training, VisNet was presented with greyscale images of 128 × 128 pixels consisting of a hand and a single, black circular target object (diameter = 36 pixels) at a specific position relative to the hand. Throughout all simulations, the position of the hand moves across the retina while the orientation of the hand is kept constant (i.e. simulations focus on learning translation rather than rotation invariance). In section 3.1, we started with 3 object positions arranged equidistant on a semicircle around the hand and gradually increased the number of object positions to 10 by lowering the intermediate distances. Each of these hand- object configurations was shown in 10 different retinal positions during training, each shifted horizontally by 2 pixels, in order to simulate rapid saccades around the visual scene. To clarify, we present each hand-object configuration in all 10 retinal locations before moving on to the next hand- object configuration. Exemplary training stimuli are shown in the bottom row of, while the upper row intends to clarify how the density of hand-centred locations was increased to effectively approximate a continuum. In section 3.2 we evaluate the generalisation abilities of the hand-centred neurons by training exactly like in section 3.1, but testing the neurons on 20 highly overlapping hand-object configurations. In section 3.3 we check whether the order of presentation of the stimuli during training affects the development of hand-centred output neurons. In particular, in order to confirm that the network is not relying on a form of temporal binding to develop hand-centred neurons, we present all of the hand-object configurations in each retinal location before moving to the next retinal location. This training order corresponds approximately to a target object moving quickly through space while the hand remains at a fixed position. Since in this setup, the relative position between the hand and the target object varies rapidly rather than being stable over a short period of time, these control simulations rule out temporal binding as the mechanism producing hand-centred neurons (like demonstrated by Galeazzi et al.). To be precise, they aim to confirm the relative flexibility of CT learning as the underlying generative mechanism. In section 3.4 we increase the number of hand-object configurations on which the network is trained to 15, 20 and 30 in order to investigate how well the output neurons learn to represent a near continuum of hand-centred target locations. When presenting a particular image during training, the activations of individual neurons and their firing rates are calculated within a layer, and then the afferent synaptic connection weights are updated, as described in section 2.1. The presentation of all hand-object configurations across all 10 retinal locations constitutes one epoch of training. The network is trained one layer at a time, from layer 1 upwards to layer 4. For all the simulations described here, the network was trained for 40 epochs per layer. # 3 Simulation results ## 3.1 Performance of network model when presented with discrete hand-object configurations The aim of the initial simulations presented in this section was to confirm that CT learning could drive the development of hand-centred neurons in the output layer of VisNet, where such neurons would respond selectively to one particular hand-object configuration over all retinal locations. We began by running simulations in which the target object was presented at 3 equidistant hand- centred locations, ‘left’, ‘top’ and ‘right’. Then we gradually increased the number of hand-centred object locations to 10 by reducing the distance between neighbouring object locations. In each simulation, the hand-object configurations are presented with temporal continuity during training in a similar manner to that performed in the trace learning simulations of Galeazzi et al.. That is, each hand-object configuration is shown shifting through all 10 retinal locations before moving on to the next hand-object configuration. shows the response profiles of 4 output neurons before and after training in a simulation with 7 hand-object configurations. Before training, the neurons responded randomly to the 7 hand-object configurations presented across different retinal positions. However, each of the same neurons learned to respond to a particular hand-object configuration across all retinal locations after training. For example, neuron (32, 27) learned to respond to the fourth hand-object configuration (i.e. the object is on top of the hand) and responded to this configuration across all 10 retinal locations. The 4 neurons shown in had thus developed perfect hand-centred firing properties after training. For a more global assessment of network performance, presents the single and multiple cell information analyses for 8 separate simulations with different numbers of hand-centred target locations ranging from 3 to 10. The results shown for each simulation are in fact averaged over 10 simulations with different random seeds, as described in section 2.3.1. The upper plot of shows the single cell information carried by all 1024 neurons in the output layer for each simulation before and after training. For each simulation, the corresponding plots show the amount of single cell information carried by each neuron in ranked order. It is evident that, for all 8 simulations, the entire population of output neurons carried relatively little single cell information before training. However, in each of the 8 simulations, after training more than 150 neurons developed maximum single cell information. Since the maximal information varies by log₂(Number of stimuli), the ordinate of is normalised to 1 for each simulation in order to facilitate comparison. Neurons with perfect hand-centred response characteristics, responded to just one hand-object configuration across all 10 trained retinal locations. We hypothesise that the higher number of hand-centred neurons for the case of 6 hand-centred target locations (531) compared with 3 target locations (212) results from the remaining representational capacity within our network. Furthermore, some drop off in performance could be seen in the simulations with the largest number (i.e. 9 or 10) of hand-centred target locations, with fewer neurons reaching the maximal level of single cell information in these simulations. In these cases, the hand-centred object locations are more tightly packed together and individual neurons may start to respond to two or more adjacent hand-centred object locations. However, this is entirely reasonable behaviour for a hand- centred neuron, since in the brain such neurons would in fact respond to small localised regions of hand-centred space. In other words, this last observation is a limitation of the current single cell information analysis, itself, which treats the hand-centred space as a finite number of discrete object locations rather than as a spatial continuum. It is for this reason that additional analytical methods were developed above in section 2.3.2 to treat the hand- centred space as a continuum, and which are applied to simulation data below in section 3.2. The lower plot of shows the multiple cell information for the same set of 8 simulations with different numbers of hand-centred object locations. For each simulation, the multiple cell information was computed across the 5 cells with maximal single cell information for each hand-object configuration. It can be seen that there was a large increase in the multiple cell information carried by output neurons after training. Moreover, in simulations with up to 8 target object locations, the multiple cell information asymptoted to the theoretical maximal value. This indicated that all of the different hand-object configurations were properly represented by the output neurons in these simulations. However, as seen with the single cell information analysis, there was an apparent drop in the asymptotic value of the multiple cell information for those simulations with the largest number (i.e. 9 or 10) object locations. Again, this is to be expected as the density of hand-centred object locations is increased, resulting in individual hand-centred neurons learning to respond to a small cluster of nearby hand-centred object locations. In summary, these simulation results confirm our hypothesis that CT learning, which relies on a purely Hebbian learning rule without a memory trace of recent neuronal activity as used in the trace learning model of Galeazzi et al., could drive the visually-guided development of hand-centred visual neurons. Such neurons respond selectively to a target object in a particular location with respect to the hand, and continue to respond to that configuration invariantly as it is presented across different retinal locations. The simulations described above demonstrate for the first time that these hand-centred cell response properties can develop through a biologically plausible, Hebbian learning mechanism. ## 3.2 Ability of network model to generalise to novel stimulus-object configurations In the simulations presented in the previous section, the network was trained and tested on the same set of hand-object configurations. However, the visual brain must be able to represent a spatial continuum of target object locations with respect to the hand. Can our model represent such a continuum of hand- centred locations after training on only a limited number of discrete hand- object configurations? In this section, we address this question by exploring the ability of the model to generalise its responses to a (near) continuum of mostly novel hand-centred target locations not encountered during training. In each of the simulations presented in this section, the network was trained on a discrete number of hand-object configurations, ranging from 3 to 10 configurations, as presented above in section 3.1. However, in contrast to the previous section, testing was done with an increased density of 20 highly overlapping hand-object configurations that were presented across different retinal locations. This meant that the model was being tested on a (near) continuum of hand-centred target locations, most of which had not been seen by the network during training. For example, presenting the object at 10 equidistant positions around the hand corresponded to an angle of 22° between neighbouring positions, while the angle is 10.6° respectively 6.9° for 20 and 30 locations. Accordingly, training on 10, but testing on 20 locations implies that the position of every second tested location to the closest known, training location is maximized; which clearly hampers task difficulty. Therefore, if hand-centred neurons are successfully shown to generalize to the object presented at the precise middle between two trained locations, then they will also generalize to all intermediate positions, which supersede the high computational effort of testing on a denser continuum of locations dispensable. The novel hand-centred target locations represented intermediate positions between the trained hand-centred target locations, like shown in. As the density of *tested* hand-centred target locations effectively approaches a continuum, we need to verify whether each output neuron learns to respond to a localised region or cluster of hand-centred locations, rather than responding to just a single hand-object configuration. Consequently, in this section we begin to employ the more flexible criterion described above in section 2.3.2 for assessing whether output neurons develop hand-centred responses. In this case, a neuron is classified as hand-centred if it responds to a small, localised cluster of hand-centred target locations across a large number of retinal locations. It is also important to check that the receptive fields of the entire population of hand-centred output neurons are distributed reasonably evenly across the 20 hand-centred target locations on which the network is tested. If we observe approximately the same number of hand-centred neurons selective to each of the 20 hand-centred target test locations, then this will imply that the trained model provides an adequate visual representation of the whole hand-centred space, and that the population of hand-centred neurons maintains a similar overall level of activity for objects shown in different locations with respect to the hand. Consequently, from the population of output neurons that were classified as hand-centred according to the criterion described in section 2.3.2, we calculated the number of hand-centred neurons that represented each of the (tested) target locations with respect to the hand. This was achieved in 2 successive steps as follows. In step 1, we recorded the number of times that each hand-centred output neuron *i* responded to each hand-object configuration *j* when that configuration was presented over all ten retinal positions, and this number of responses is denoted by $R_{j}^{i}$. In step 2, for each hand- centred output neuron *i* we added a contribution of $\frac{R_{j}^{i}}{\sum_{k = 1,N}R_{k}^{i}}$ to the *j*th value of the frequency distribution for each hand- object configuration *j*. Given this procedure, summing over all the final frequency values in the distribution gives the total number of hand-centred neurons. Based on this computational procedure, shows the number of hand-centred output neurons with receptive fields localised at different test positions across the hand-centred space. Results are shown for 6 separate simulations in which the network was trained on *N* = 3, 5, 7, 8, 9, 10 hand-centred target locations, but tested on 20 highly overlapping hand-centred target locations. The results shown are, in fact, averaged across 10 different simulations using altered random seeds to initialise the network connectivity. It is evident that for those simulations in which the network was trained with 7 or more hand-centred target locations, the network developed many (i.e. hundreds) of hand-centred output neurons as classified according to the criterion described in section 2.3.2. Furthermore, and most importantly, the receptive fields of the hand- centred output neurons were distributed reasonably evenly across the 20 hand- centred locations used to test the network. These results confirm that, for simulations in which the network is trained with only a finite number (i.e. 7, 8, 9 or 10) of discrete hand-centred target locations, the population of hand- centred output neurons is able to generalise its neuronal responses to, and thereby adequately represent, a spatial continuum of hand-centred locations. ## 3.3 Effect of temporal interleaving of different hand-object configurations during training A key predicted property of the proposed CT learning mechanism for the development of hand centred visual neurons is that CT learning should not require that each spatial configuration of the hand and a target object is seen with temporal continuity as it shifts across the retina. For example, CT learning should still be able to bind together images of the same hand-object configuration seen in different retinal locations even if different hand-object configurations are seen rapidly interleaved with each other through time. This important property, which sharply differentiates the CT learning mechanism proposed in this paper from the trace learning approach of Galeazzi et al., is now tested in this section. In both previous sections, each hand-object configuration was shown shifting through all 10 retinal locations before moving on to the next hand-object configuration. For the new simulations carried out in this section, we used the same set of training stimuli as used above in section 3.1. However, now we flipped the presentation order of stimuli during training. Specifically, in this next set of simulations all of the different hand-object configurations are shown at a particular retinal location before moving to the next retinal location. This led to a rapid temporal interleaving of different hand-object configurations during training such that not any two consecutively presented training stimuli revealed the same hand-object configuration. Given our supposition that CT learning should not be sensitive to the temporal ordering of stimulus presentation, we predicted that the new temporal ordering of stimulus presentation used in this section should give similar results to that obtained with the original stimulus presentation order used in section 3.1. compares simulation results obtained when the stimuli are presented in either the original presentation order used in section 3.1 (results shown in blue), in which each hand-object configuration is shown shifting through all 10 retinal locations before moving on to the next hand-object configuration, or presented using the new flipped order (shown in red). For each of these two different stimulus presentation orders, the figure shows the average number of perfectly hand-centred neurons in the output layer (carrying the maximum level of single cell information) for 8 simulations with different numbers of hand-centred configurations ranging from 3 to 10. It is evident from the figure that similar numbers of neurons developed perfect hand-centred response characteristics with the two alternative training conditions. We confirmed this observation by conducting *one-tailed t-tests* for the 8 separate sets of simulation results with 3 to 10 hand-centred object positions, none of which revealed a significant difference in the results obtained with the two training conditions. In order to get a single, global measure comparing the outcomes of all 8 sets of simulations, a *two-sided Wilcoxon signed-rank test* was used and returned a *p*-value of 0.511. The results described above confirm that the new CT learning mechanism proposed in this paper provides very robust performance with respect to variation in the presentation order of alternative hand-object configurations in different retinal positions during training. This robustness is a known property of CT learning, originally reported in the development of transformation invariant representations of objects in the ventral visual system. In particular, it is demonstrated that CT learning does not require that different retinal views of the same hand-object configuration are seen clustered together in time. Thus, CT learning enables the development of hand-centred neurons by binding together different retinal views of the same hand-object configuration on the basis of their spatial similarity or overlap, as described in section 2.2. ## 3.4 Performance of the model when trained with a continuum of hand-centred target locations In all the simulations described above, our model was trained with at most 10 different hand-object configurations. However, the relative configurations of the hand and nearby objects that humans see during everyday experience is much higher, approaching a continuum of object locations with respect to the hand. Consequently, in this section, we investigate whether our model is able to successfully develop hand-centred output neurons when the network is trained with a (near) continuum of hand-centred object locations. To this end, we conducted simulations with 15, 20 or 30 highly overlapping, equidistant hand- centred object locations arranged in a circular arc around the hand. In these cases, rather than learning to respond to only a single hand-object configuration, each output neuron should learn to respond to a localised region or cluster of hand-centred locations, like in section 3.2. Furthermore, the receptive fields of the population of hand-centred neurons should be evenly distributed throughout the space around the hand in order to provide an adequate visual representation of the whole space. Investigating the performance of our model on these more ecologically realistic training conditions, in which the hand-centred target locations form a (near) spatial continuum around the hand, denotes a crucial step in verifying the plausibility of CT learning as a mechanism that could drive the development of hand-centred neurons in the brain. As the density of hand-centred target locations effectively approaches a continuum, we verified whether each output neuron learns to respond to a localised region or cluster of hand-centred locations—like in section 3.2. shows the binarised firing responses of 4 example output neurons before and after the network was trained on 30 hand-centred target object locations presented across 10 retinal locations. Before training, the neurons initially responded randomly to the different hand- object configurations presented in different retinal positions. This can be seen from the scattered red fields within the binary response matrices shown in the top rows of the first two columns. Again, a red field indicates a response while a blue field indicates no response. The behaviour of the same neurons after training can be seen in the two right columns. It is evident that, after training, each of the 4 neurons developed sensitivity to a localised cluster of hand-centred target object locations, and responded to their preferred localised region of hand-centred space across different retinal locations. For example, neuron (32, 5) learned to respond when the object is presented at hand-centred positions 24 or 25 and responded to almost all shown retinal locations. As the response plots show, each of the 4 output neurons developed clearly localized, hand-centred receptive fields: responding to a short range of similar hand- object configurations almost completely irrespective of where this stimulus was presented on the retina. A theoretical explanation why training on a continuum of target locations still lead to the development of hand-centred neurons, rather than the undesired, alternative supposition of responding selectively to retinal locations, is presented in the next section. As discussed above, a key question is whether the receptive fields of the population of hand-centred neurons that develop during training are evenly distributed through the space around the hand, and whether these hand-centred neurons cover the entire hand-centred visual space and successfully represent all trained hand-object configurations. Following the procedure in Figs and shows the number of hand-centred neurons with receptive fields localised at different positions across the hand-centred space scaled to the range \[0, 1\]. Results are shown for 3 separate simulations in which the network was trained with 15, 20 or 30 hand-centred target locations. The plots for all three simulations show that the distribution of hand-centred neurons across the space around the hand is approximately uniform. As a further quantitative test, the standard deviations of the three plotted distributions were found to be in the range of only 5-12% of the mean values. These results confirm that the purely Hebbian CT learning mechanism produced approximately the same number of neurons representing each of the trained hand-object configurations, and hence an even visual representation of the space around the hand, for all 3 simulations. ## 3.5 A theoretical explanation for the development of hand-centred neurons The simulation results presented above in section 3.4 confirmed that when the network model is trained on a (near) continuum of hand-centred target locations, it continued to develop hand-centred output neurons with small localised receptive fields in hand-centred space. However, this leaves open the major theoretical question why the network developed such hand-centred output neurons rather than producing individual neurons that respond to many different spatially overlapping hand-object configurations at the same retinal location. The underlying problem is that when the network is presented with a continuum of highly overlapping hand-centred target locations, CT learning could potentially operate in alternative ways as follows. Firstly, CT learning could continue to drive output neurons to bind together small subsets of highly similar hand- object configurations presented across different retinal locations, and thereby produce hand-centred neurons with localised hand-centred receptive fields. Alternatively, CT learning could drive output neurons to bind together many different hand-centred target locations at a particular retinal location due to their spatial overlap, and hence produce output neurons that respond to many different hand-object configurations at that retinal position. This would correspond to horizontal rather than vertical red bars in. Consider, that both suppositions are in accordance with the CT learning hypothesis (2.2). Finally, CT learning could combine these two tendencies in an unconstrained and random manner, thus producing output neurons with highly erratic response characteristics. However, in practice, the simulations reported above in section 3.4 confirmed the continued development of perfect hand-centred neurons, even as the number of hand-centred target locations was increased to 100. In this section, we try to develop a deeper theoretical understanding of these important simulation findings. Specifically, we hypothesise that these simulation findings may be understood by considering a Principal Component Analysis (PCA) of the underlying sources of variance in the visual training stimuli consisting of alternative hand-object configurations presented across different locations on the retina. The VisNet model used in the simulations presented in this paper consists of a hierarchy of competitive neural layers. Competitive networks are frequently used as unsupervised pattern classification algorithms due to their ability to cluster together similar input patterns while separating dissimilar input patterns. In particular, they are able to produce far more compressed representations of the input patterns by removing redundancies in these representations. Within each layer, the excitatory neurons receive afferent synaptic connections from a topologically corresponding, localised region of neurons within the preceding layer. These bottom-up synaptic connections self- organise during visual training using the Hebbian learning rule followed by renormalisation of each neuron’s synaptic weight vector at the end of each timestep. This learning process bears some comparison with the operation of Oja’s learning rule, which may itself be derived from the operation of Hebbian learning with weight vector renormalisation and also bounds the length of the converged synaptic weight vector. Interestingly, Oja’s rule when applied to a single neuron has been shown to extract the first principal component of the input training data. In later work, Sanger chained together Oja’s neurons allowing for a full Principal Component Analysis (PCA) and called this technique the Generalized Hebbian Algorithm. We speculated that the neurons in the competitive output layer of VisNet may be carrying out a somewhat similar PCA operation. That is, individual output neurons may learn to represent input patterns with some correspondence to the largest sources of variation over the entire stimulus set used to train the network. The stimulus set used for each of our simulations consists of a number of different hand-object configurations shown across all retinal locations. We hypothesised that the greatest source of variation within the stimulus sets was due to changes in the position of target objects with respect to the hand rather than changes in the retinal location of hand-object configurations. In this case, applying PCA to such a stimulus set should lead to the highest principal components representing the hand-centred positions of target objects, with the retinal position of hand-object configurations represented by lower principal components. Most importantly, if the output neurons in our model are indeed learning to represent the directions of highest variance, then the lateral inhibition will drive them to be sensitive to the hand-centred locations of objects rather than the location of hand-object configurations on the retina. So, do the first eigenimages of those stimulus sets that were successfully used to develop hand-centred output neurons, represent the locations of target objects with respect to the hand? In order to investigate this, we applied PCA to the stimulus sets of 6 simulations. These simulations varied according to the number of hand-centred target locations, number of retinal locations of each hand-object configuration, and the size of the retinal shifts of hand-object configurations (in pixel), with these simulation parameters listed in. The eigenimages presented in resulted from applying PCA to the stimulus sets used for these simulations. This representation revealed the greatest sources of underlying variance in each stimulus set in descending order. The simulations represented in rows (a), (b), (c) and (e) developed hand-centred output neurons, while the simulations shown in rows (d) and (f) did not develop hand-centred neurons. From it can be seen that for those simulations that developed hand- centred output neurons, the first eigenimages represented the locations of targets with respect to the hand, rather than the location of the hand-object configuration on the retina. This effect is particularly evident for the simulation shown in row (a) of, for example. Here we can see that the first three eigenimages differentiate between different hand-centred positions of the target object. In row (a) we can see two light regions (signifying higher values) in the bottom left and bottom right of the eigenimage, where each of these light regions corresponds to a circular target object presented in a particular location with respect to the hand and translated horizontally across a number of retinal locations. Similarly, two dark regions (signifying lower values) can be seen in the centre left and centre right of the eigenimage, where each of these dark regions also corresponds to a circular target object presented in a particular hand-centred location and translated horizontally across a number of retinal locations. Importantly, though, there is no image of the hand present within the first eigenimage. Consequently, this eigenimage, which represents the greatest source of variance in the stimulus set, separates the upper from lower hand-centred object positions. In a similar manner, the second principal component in row (a) distinguishes between the leftmost and rightmost hand-centred object positions. Most importantly, this is independent of the retinal location of the hand-object configuration since the horizontal shifts of the hand are represented by the fourth principal component with a clear drop in the corresponding eigenvalue shown in the sixth column. Thus, the highest principal components for the simulation represented in row (a) clearly reflect the variance due to changes in target location with respect to the hand. Similar trends are also seen for the simulations represented in rows (b), (c) and (e). For example, the stimulus set for the simulation shown in row (b) contains a (near) continuum of 30 hand- centred target locations, and yet still develops hand-centred output neurons after training (see section 3.4. In row (b) we can see that the first four principal components differentiate between different hand-centred target locations, while the fifth principal component represents horizontal shifts of the hand. Thus, for those simulations that developed hand-centred output neurons, the PCA confirms that the greatest source of variance arises from changes in the hand-centred locations of target objects rather than changes in retinal locations of the hand or hand-object configurations. In contrast, the simulations that did not develop a single perfect hand-centred output neuron are represented in rows (d) and (f) of. For these simulations the hand-object configurations were presented shifting across a relatively large horizontal region of retinal space, where each hand-object configuration was presented across a grid of 50 pixels that was covered by 10 shifts of 5 pixels and 25 shifts of 2 pixels, respectively. With these settings, the variance in the stimulus set due to translation of hand-object configurations across the retina is highest. Consequently, it can be seen that the first eigenimages shown in rows (d) and (f) represents horizontal shifts of the hand. To further clarify our interpretation of these eigenimages, note that the first principal components in rows (d) and (f) each contain a pair of light and dark images of the hand in the central bottom region of the eigenimage, where the light and dark hand images occupy different retinal locations on the right and left, respectively. These eigenimages thus separate different retinal positions of the hand. Hence, for those simulations that did not develop hand-centred output neurons, the greatest source of variance arises from changes in retinal locations of the hand or hand-object configurations. The simulation represented in row (c), in which the hand-object configurations were shifted across a 10 × 10 grid rather than only horizontally as in other rows, still developed a large number of hand-centred output neurons. We hypothesize that this results from extending the underlying source of variance in the stimulus set by another orthogonal (vertical) dimension rather than increasing the variance in the already existing (horizontal) dimension. This explanation is suggested by the 4th and 5th eigenimages of row (c), which represent shifts of the hand in the two orthogonal directions. The PCA results shown in provide strong support for the hypothesis that it is the underlying sources of variance in the stimulus set that is driving self- organisation within the network when relying on the Hebbian learning rule with renormalisation of synaptic weight vectors at each timestep. In particular, the output neurons develop hand-centred responses when the greatest source of variance in the stimulus set, as may be revealed by PCA, is due to changes in the positions of target objects with respect to the hand. In conclusion, based on the above findings, we hypothesise that the output neurons in the network model are extracting and representing the dimensions of largest variance within the input data, which in these simulations is due to variation in the position of target objects with respect to the hand. Further, lateral inhibition between the output neurons, which introduces competition between the neurons, forces individual neurons to learn to respond to different parts along the dimensions of highest variance. Indeed, this corresponds to particular hand-object configurations. Moreover, CT learning, which binds together subsets of input patterns based on their spatial overlap or similarity, is encouraging output neurons to bind together input patterns along the dimension of least variance, that is, binding together particular hand-object configurations across different retinal locations. In this case, output neurons learn to respond selectively to specific hand-object configurations, or rather small localised regions of the space of hand-centred object locations when dealing with a spatial continuum, irrespective of retinal location. In other words, retinal locations are bound together for particular hand-object configurations rather than vice versa, inducing the development of output neurons that fire in a hand-centred frame of reference. # 4 Discussion In this paper we have presented the first approach to utilise an invariance learning mechanism known as Continuous Transformation learning to drive the development of hand-centred output neurons that respond invariantly over different retinal locations in a biologically plausible, hierarchical neural network model of the primate visual system. The CT learning mechanism operates by encouraging output neurons to bind together input patterns due to their spatial overlap or similarity, and in this way can train individual neurons to respond to particular hand-object configurations across the retina. CT learning as an explanation for the development of hand-centred visual neurons has a couple of advantages over a previous model developed by Galeazzi, which utilised trace learning rather than CT learning. Firstly, CT learning operates with a simpler, purely Hebbian form of learning rule which does not need to incorporate a memory trace of recent neuronal activity. Secondly, and more importantly, CT learning does not require that different retinal views of the same hand-object configuration are seen clustered together in time. Instead, CT learning is robust enough to cope with a much broader and more natural range of visual training sequences as simulated in section 3.3. This includes conditions in which the relative positions of the hand and nearby objects are changing rapidly or are static over a long period of time. In a nutshell, our simulation results provide the first evidence validating CT learning as a potential mechanism underlying the development of hand-centred neurons in the primate brain. Furthermore, the results specify and formalise the work of Haykin, who has claimed that an early benchmark model of the visual system developed by Linsker, a hierarchical network with topological bottom-up synaptic connectivity comparable to VisNet, learned to discriminate localised regions along the directions of highest information content (i.e. maximum variance) within the input data. In addition, both CT learning and PCA are independent of the temporal order in which the training stimuli are presented to the network, which accounts for the observation in section 3.3 that the temporal ordering of the training stimuli has no significant effect on the development of hand-centred output neurons in the network. In this paper, our goal was to visually tune neurons to fire in a specific frame of reference (which may be interpreted as a specific type of learning translation invariance). While translation invariance arises inherently in some supervised neural network architectures, such as convolutional neural networks (due to the shared weights), it needs to be pointed out that translation invariance is an inherently difficult task for self-organizing, competitive networks compared to e.g. learning rotation invariance. The latter case follows conventional principles of competitive learning: a cluster of overlapping input patterns is associated together by a weight vector pointing towards the centre of that cluster. This assumption of high correlation between different transforms cannot be made for translation invariance tasks where an object might occur at such broad positions in space that hardly any neurons become active for multiple object positions. This problem worsens when CT instead of trace learning is used, because any information carried along the temporal dimension is neglected. Hand-centred visual representations in the primate brain are thought to be used for accurately guiding hand movements towards target objects in the environment. Visually guided reaching, itself, is a highly complex process whereby visual information is transformed into motor actions with a feedback loop for online correction of movement. This paper does not attempt to model such full motor processes. Instead, we have focused on modelling how the hand-centred visual representations, themselves, might arise in a self-organizing fashion using a biologically plausible, unsupervised learning mechanism. The importance of hand- centred visual representations has recently been demonstrated in cognitive robotics by Juett and Kuiper. In their model, self-learned peri-hand representations arose naturally from random exploration and enabled a Baxter Robot to reach to target locations within its peri-personal space. However, Juett and Kuiper used visual information from multiple spatial angles and were not primarily concerned with the biological plausibility of their model architecture, while our model focuses on the development of hand-centred visual neurons within a biologically detailed, hierarchical neural network model of the primate visual cortex. Indeed, in future work we plan to supply the VisNet model with stereoscopic visual input from a simulated 3-dimensional visual training environment, in which the hand is seen with nearby objects. In theory, this could lead to the development of hand-centred neurons that represent 3-dimensional hand-object spatial configurations irrespective of the position of the configuration in 3D peri-personal space. A key issue investigated in this paper was the performance of the model when trained on a (near) continuum of hand-centred target object locations, as humans are exposed to in every day life. In the simulations reported in section 3.4 we found that the network continued to develop lots of perfect hand-centred output neurons. Moreover, we found that the receptive fields of this population of hand-centred neurons were uniformly distributed throughout the space around the hand, thus providing an even neuronal representation of this hand-centred space. These simulation results lend strong support to our hypothesised role for CT learning in the development of hand-centred visual neurons in the primate brain. However, we were still left with the fundamental question of why in practice the CT learning mechanism, which builds invariant neuronal responses by simply exploiting spatial overlap between similar input stimuli, encouraged individual output neurons to respond to particular hand-object configurations invariantly across the retinal space, rather than forcing neurons to respond invariantly across the continuum of hand-centred target locations at particular retinal locations. This question was addressed in section 3.5. Based on the close relationship between Oja’s learning rule and the Hebbian learning rule with synaptic weight vector normalisation used in VisNet, it was hypothesised that the output neurons in our model were carrying out an approximate form of Principal Component Analysis (PCA). That is, individual output neurons would learn to represent input patterns with correspondence to the largest sources of variation within the stimulus set used to train the network. In this case, the lateral inhibition between output neurons in our model will likewise drive them to be sensitive to the hand-centred locations of objects rather than the location of hand-object configurations on the retina. The plausibility of this explanation was investigated by performing PCA on the stimulus sets used in a number of different simulations. Consistent with our theoretical explanation for the development of hand-centred neurons when the model is trained on a (near) continuum of target locations around the hand, it was indeed found that the first eigenimages of the stimulus sets used in simulations that developed hand- centred neurons represented the locations of target objects with respect to the hand. presented two simulations, shown in rows (d) and (f), which failed to develop perfect hand-centred output neurons. In these simulations, the only difference was that the hand-object configurations were presented shifting across a larger horizontal region of retinal space. So what biological features might still be missing in our network model that might improve its ability to develop hand- centred neurons? In future work we intend to upgrade the biological realism of the model and investigate the effects of these modifications on model performance. These planned model improvements concern alterations to the bottom- up visual processing and the incorporation of top-down tactile and proprioceptive signals as follows. Firstly, let us consider the bottom-up visual processing within the existing model. We have not yet implemented a cortical magnification factor within the hierarchy of cortical layers within the network model, whereby the representation of a visual stimulus close to the fovea is enhanced compared to more peripheral vision. Neither have we attempted in this paper to model active fixation, that is, how a human subject actually shifts their eyes around visual scenes containing their hands with background objects. Although the feasibility of this approach has been demonstrated in an earlier study, in which we showed that hand-centred visual neurons still developed using trace learning when the model was fed with realistic visual input inferred from experimentally recorded human gaze changes based on an eyetracking system. Introducing these improvements to the biological accuracy of the model might perhaps reduce the region of retinal space over which the network learns translation invariant responses to hand-object configurations but also make the development of hand- centred neurons more robust. Secondly, motion-encoding neurons from the dorsal stream (e.g. MT), that can predict the future position of a target before it is actually reached, could pre-activate those hand-centred cells with the expected target position in their receptive field. Indeed, motion signals seem to be encoded in different reference frames, suggesting that visual decisions for action consider motion in a frame of reference best suited for the current task. Thus, early sub-threshold stimulation from speed-encoding neurons may help to overcome the current limitation of our model regarding the retinal space along which a particular hand-object configuration is shifted. Thirdly, the current model does not incorporate top-down tactile or proprioceptive signals. For example, neuronal areas preparing and guiding reaching movements receive afferent signals from the somatosensory cortex representing the touch of an object against part of the limb/hand, thus indicating the hand-centred location of a target object. Consistent with this, it has been found that some hand-centred or peri-hand neurons display bimodal visual/tactile receptive fields, responding to either the sight or touch of an object in a particular position with respect to the hand. Adding tactile signals to the model may allow us, for example, to account for the bimodal properties of some hand-centred receptive fields as shown previously by Magosso et al.. More recently, Pitti et al. integrated tactile information from an artificial skin with visual information in a recurrent spiking neural network model and even accounted for a successful replication of the rubber-hand illusion. However, these studies do not particularly address the problem of coordinate transforms in multisensory processing nor do they incorporate proprioceptive information in their model. In addition, reach related areas receive somatosensory signals carrying proprioceptive information about the position of the limb/hand in peri-personal space. Again, consistent with this, hand-centred neurons in posterior parietal cortex can maintain their firing behavior in absence of visual information, and can guide movement of the hand using proprioceptive information alone. Tactile and proprioceptive information could play an important role in guiding the development of hand-centred neurons in the brain. Introducing such signals into the model architecture may also lead to more robust development of hand-centred neurons. We gratefully acknowledge support from the Oxford Foundation of Theoretical Neuroscience and Artificial Intelligence. [^1]: The authors have declared that no competing interests exist. [^2]: **Conceptualization:** SMS JMG JB. **Data curation:** SMS JMG JB. **Formal analysis:** SMS JMG JB. **Funding acquisition:** SMS. **Investigation:** JB JMG. **Methodology:** SMS JMG JB. **Project administration:** SMS JMG JB. **Resources:** SMS JMG JB. **Software:** SMS JMG JB. **Supervision:** SMS JMG. **Validation:** SMS JMG JB. **Visualization:** JB. **Writing – original draft:** JB JMG SMS. **Writing – review & editing:** JB JMG SMS.

# Introduction Digestive system cancers are threaten human life and health. A recent study reported that the incidence rates of several digestive system cancers, including hepatocarcinoma, esophagus carcinoma, pancreatic cancer and intestinal malignancies, have a tendency to be elevated, although cancer-associated death has been continuously decreasing over the past two decades. Thus, the epidemiology of digestive system cancers (DSCs) remains grim. Several poor clinicopathological characteristics, such as advanced clinical stage, poor tumor differentiation and larger tumor size, were popularly recognized to be significantly associated with an unfavorable prognosis, but patients with similar clinicopathological characteristics often suffer different survival outcomes. Hence, identification of new reliable biomarkers that can more precisely predict the prognosis of patients with DSCs is very imperative. It has been reported that abnormal serum albumin is closely related to the progression of many diseases. Furthermore, previous studies have demonstrated that a decreased albumin level is correlated with an unfavorable prognosis of DSCs, including gastric, colorectal, and pancreatic cancers. In addition, serum globulin has also been demonstrated to have a close relationship with immunity and inflammation and acts as a valuable predictor of tumor progression. However, both serum albumin and globulin levels are easily influenced by non- cancer-related factors, including dehydration and edema, which can weaken their efficiency and accuracy in predicting the prognosis of cancer patients. To overcome this defect, numerous recent studies have tried to investigate the prognostic significance of the albumin/globulin ratio (AGR) in cancer patients, which combines two factors and thus might reduce the influence of confounding factors. For instance, a body of studies reported that a low pretreatment AGR is closely associated with worse prognosis in patients with digestive system cancers, such as gastric cancer, colorectal cancer, pancreatic cancer, hepatocellular carcinoma and so on. However, most studies published to date use a small sample size, which might affect the reliability of their conclusions. Therefore, it is imperative to perform a meta-analysis of studies investigating the prognostic value of the AGR in patients with DSCs to provide stronger evidence in favor of the prognostic value of the AGR. # Methods ## Literature search strategy A systematic literature search was performed in PubMed, EMBase and Web of science for eligible studies assessing the prognostic value of the AGR in digestive system cancers through September 8, 2017. The search strategy combined the following terms: (gastric or stomach or colon or rectal or colorectal or liver or hepatocellular or pancreatic or esophageal or esophagus or cholangio\* or gallbladder or bile duct) and (tumor or cancer or carcinoma or adenocarcinoma or malignan\*) and (“albumin to globulin ratio” or “albumin/globulin” or “albumin to globulin” or “AGR” or “albumin and globulin”), as well as (prognosis or prognostic or survival). ## Selection criteria The inclusion criteria for eligible studies included: (1) reported the association between the AGR and OS/DFS/CSS in digestive system cancers; (2) had full text and were published in English; and (5) had a hazard ratio (HR) with a 95% confidence interval (CI) or a survival curve. Furthermore, studies were excluded according to the following criteria: (1) case reports, reviews, letters and comments; (2) no sufficient data could be extracted to calculate the HR and 95% CI; (3) patients were not divided into two groups, including a low AGR group and high AGR group; and (4) studies were performed on animals. ## Data extraction and quality assessment The full texts of the included studies were carefully reviewed, and data were extracted by two independent researchers. A third investigator consulted to resolve inconsistencies. The following data were extracted: the first author’s name, country of research, year of publication, mean age of patients, cancer type, case number, study type, cut-off AGR, cut-off selection, treatment method, mean follow-up time, overall survival (OS), cancer specific survival (CSS), and diseases free survival (DFS). If the included studies provided both univariate and multivariate analysis results, only the multivariate results were extracted since they balanced many confounding factors. When HRs for OS, CSS and DFS could not be obtained directly, Engauge Digitizer version 4.1 (<http://digitizer.sourceforge.net/>, freely down-loaded software) was used to extract them from the Kaplan-Meier curves. The Newcastle-Ottawa Scale (NOS) was used to assess the study quality. Three items were evaluated, selection, comparability and outcome. The scores according to NOS varied from 0 to 9. A score of 6 or more was recognized as high quality. ## Statistical analysis Statistical analyses of this meta-analysis were fulfilled using Stata version 12.0 (Stata Corporation, College Station, TX, USA). Pooled HRs and 95% CIs were used to evaluate the quantitative aggregation of the survival results. The heterogeneity across studies was tested by Cochran’s Q and Higgins I² statistics. P \< 0.05 and I²\>50% were considered to be significant heterogeneity, while I² \< 25% and 25% \< I² \< 50% indicated no heterogeneity and moderate heterogeneity, respectively. A random effects model was used when statistical heterogeneity was detected (P \< 0.05, I² \> 50%); otherwise, the fixed effects model was applied. HR \> 1 (high AGR used as reference) means a higher risk of worse outcomes for low AGR, and the study was recognized as statistically significant if the 95% CI did not include 1 and P\<0.05. Sensitivity analysis was carried out by sequentially omitting individual studies at each step. If the results did not substantially alter when one study was excluded, this meant that the pooled results were stable. Publication bias was evaluated by Begg’s test, Egger’s test and funnel plot analysis, and P \< 0.05 with funnel plot asymmetry indicated that a statistically significant publication bias may exist. Duval’s nonparametric trim-and-fill method was used to evaluate the potential effect of publication bias, if significant publication bias exist. # Results ## Study search and study characteristics The search and selection strategy is shown in. A total of 133 studies were identified after searching PubMed, EMBase and Web of science. After 28 duplicates were removed, 105 studies were checked by two investigators who screened the titles and abstracts. Subsequently, 62 studies were excluded because they were reviews and comments (n = 11), covered irrelevant topics (n = 48) and were not published in English (n = 3), and 43 studies were left for full-text review. Finally, 15 studies were included in this meta-analysis after 16 excluding publications with no available data, 9 studies regarding non- digestive malignancies and 3 articles without full-length text. The sample sizes of these studies ranged from 66 to 5336. Among these studies, 5 studies were on colorectal cancer, 4 studies involved gastric cancer, 3 studies referred to esophageal cancer, and the rest of the included studies were on pancreatic cancer, liver cancer and cholangiocarcinoma. With respect to the study region, 10 studies were performed in China, 4 in Japan, and only one in the USA. The study period of the 15 studies ranged from 2013 to 2017. All of the included studies were designed retrospectively. According to the Newcastle–Ottawa scale (NOS), the study quality score varied from 5 to 7, which indicated that the study quality was moderate to high. The information above and other characteristics of the included studies are presented in. ## Prognostic value of the AGR in digestive system cancers ### A low AGR and overall survival (OS) in digestive system cancers Thirteen studies that included 9269 patients explored the association of the AGR with OS. A random-effect model was used to calculate the HR and 95% CI due to severe heterogeneity (I² = 66.4%, P\<0.0001). Pooled analysis showed that a low AGR was significantly connected with poor OS (HR = 1.94; 95% CI: 1.57–2.38; P \<0.001). Because a substantial heterogeneity existed, subgroup analyses, according to cancer type, region, sample size, cut-off value, cut-off selection, and treatment method, were carried out to explore the potential sources of heterogeneity. As the subgroup analysis results showed, a low AGR remained a predictor for worse OS in colorectal cancers (HR = 2.39; 95% CI: 1.45 to 3.94; P = 0.001), esophageal cancers (HR = 1.35; 95% CI: 1.08 to 1.68; P = 0.008), and gastric cancer (HR = 1.56; 95% CI: 1.19 to 2.05; P = 0.001). In addition, we still observed that a low AGR was related to unfavorable OS in China (HR = 1.75; 95% CI: 1.40 to 2.20; P\<0.001), Japan (HR = 2.13; 95% CI: 1.49 to 3.04; P\<0.001), and the USA (HR = 4.00; 95% CI: 2.06 to 7.77; P\<0.001). The correlation of a low AGR with OS was significant according to sample size (≤300 or \>300) (HR = 2.76; 95% CI: 1.75 to 4.36; P\<0.001 or HR = 1.53; 95% CI: 1.36 to 1.72; P\<0.001), cut-off value (≤1.30 or \>1.30) (HR = 2.83; 95% CI: 1.80 to 4.45; P\<0.001 or HR = 1.51; 95% CI: 1.35 to 1.70; P\<0.001), and treatment method (surgery alone or comprehensive treatment) (HR = 1.97; 95% CI: 1.40 to 2.77; P\<0.001 or HR = 1.95; 95% CI: 1.45 to 2.61; P\<0.001). Furthermore, from the results of the cut-off selection subgroup, we found that a low AGR was significantly associated with unfavorable OS in the ROC group (HR = 1.55; 95% CI: 1.31 to 1.84; P\<0.001), X-tile program group (HR = 2.17;95% CI: 1.00 to 4.71; P = 0.049), and for other selection methods (HR = 2.20; 95% CI: 1.25 to 3.28; P = 0.004), but no significance was observed in the cutoff finder group (HR = 2.55; 95% CI: 0.67 to 9.72; P = 0.171). A sensitivity analysis was conducted to evaluate the stability of the pooled HR for OS by omitting one study at each step. The result suggested that the results did not alter substantially, indicating the robustness of the results. Additionally, we assessed the publication bias according to the Begg’s funnel plot and Egger’s test. Nevertheless, the funnel plot was not symmetric, and a significant publication bias was found by Begg’s test (z = 2.14, P = 0.033) and Egger’s test (t \[bias\] = 2.91, P = 0.014). Then, the “trim and fill method” was applied to replace five missing studies, and the adjusted funnel plot was symmetric. Furthermore, after correction, the adjusted pooled HR was 1.521 (95% CI: 1.198–1.929, p = 0.001) based on the random-effect model, which indicated that the publication bias did not significantly influence the reliability of the association of a low AGR with poor OS. ### A low AGR and disease-free survival (DFS) The relationship of a low AGR with DFS was reported in 5 studies that included 6538 patients. Because of the obvious heterogeneity, pooled analysis was conducted with a fixed-effect model (I² = 47.2, P = 0.078) and showed that there was a significant association of a low AGR with worse DFS (HR = 1.49; 95% CI: 1.10 to 2.00; P \< 0.001). In addition, sensitivity analysis was conducted to evaluate the stability of the pooled HR for DFS by excluding a single study in turn. The result showed that the pooled HR was not altered substantially, suggesting that our result was robust. Publication bias was not applied when analyzing the correlation of a low AGR with DFS due to the limited number of studies that investigated the relationship of a low AGR with DFS. ### A low AGR and cancer-specific survival (CSS) in digestive system cancers Only 2 studies enrolling 479 patients explored the association of a low AGR with CSS. A fixed-effect model was applied to calculate the pooled HR with 95% CI due to no significant heterogeneity (I² = 3.3, P = 0.309). The result showed that a low AGR was significantly related to worse CSS (HR = 1.61; 95% CI: 1.13 to 2.28; P = 0.003). Publication bias and sensitivity analysis were not applicable when analyzing the correlation of a low AGR with CSS due to the limited number of studies investigating the relationship of a low AGR with CSS. # Discussion To the best of our knowledge, no meta-analyses have been conducted to date to assess the prognostic value of the AGR in patients with DSCs. In this meta- analysis, data from 15 studies investigating the association of the AGR with the prognosis of patients with DSCs was combined for statistical analysis. The results indicated that a low pretreatment AGR was significantly related to worse survival outcomes in digestive system cancers. With respect to the potential mechanisms, nutrition and inflammation might be responsible for the prognostic value of the AGR in tumors. In fact, there is a mutual promotion effect between cancer progression and inflammation. Malnutrition commonly occurs in patients with malignancies and slowly leads cancer patients into cachexia, which contributes to disease progression. Cancer- related inflammation refers not only to inflammatory factors produced by malignant cells but also to those generated when injured tissues are remodeled, rehabilitated and vascularized. Furthermore, it has been demonstrated that an abnormal elevation of serum or tumor-local inflammatory cytokines, including tumor necrosis factor; interleukin -1, -6, and -8; and vascular endothelial growth factor, could contribute to cancer progression by promoting tumor growth and metastasis. Thus, effective and practical biomarkers reflecting nutritional and inflammation conditions may be helpful to assess the prognosis of cancer patients. The AGR is generated from the combination of nutritional and inflammatory indices, so it may be a particularly helpful biomarker in this respect. Usually, the serum albumin level is considered to mirror the nutritional condition of the body. Furthermore, recent studies have also demonstrated that the serum albumin level is also able to reflect the body’s inflammatory status. Likewise, the serum globulin level is closely associated with the immune and inflammatory status of the body. It was demonstrated that elevated serum globulin levels caused by the accumulation of acute-phase proteins and immunoglobulins usually reflect a persistent inflammatory response. Serum albumin and globulin levels are easily affected by dehydration and fluid retention, which are relatively common in cancer patients. As such, they fail to reflect the authentic nutritional and inflammatory status of cancer patients. Therefore, the AGR takes the serum albumin and globulin levels into account concurrently, may more precisely mirror the body’s nutritional and inflammatory states and may be particularly helpful to predict the prognosis of cancer patients. To the best of our knowledge, our study is the first meta-analysis to evaluate the prognostic value of the AGR in patients with digestive system cancer. Although we found that a low pretreatment AGR was closely associated with a worse prognosis of patients with digestive system cancer, several limitations in our meta-analysis should be taken into consideration. First, most of the included studies were designed retrospectively, which unavoidably has a bias risk and thus affects the reliability of our pooled analysis. Second, the cutoff value ranged from 0.9 to 1.93, which may aggravate the heterogeneity among the included studies. Third, the included studies involved in patients’ population from different ethnicity groups, which might also lead to the heterogeneity and limit the generalization of the conclusion from our meta-analysis. Fourth, Egger’s and Begg’s tests indicated that there was a significant publication bias in our meta-analysis, for which the criteria that only published studies were included in our meta-analysis may be partly responsible. This may more or less influence the reliability of our pooled results, although the results of the “trim and fill” method to adjust for the publication bias suggested that the corrected pooled effect size for OS was still statistically significant. Finally, the number of studies included for the pooled estimates of RFS, DFS, CSS and PFS were rather few, and the small sample size may threaten the reliability of the pooled results in that regard. Therefore, the prognostic value of the AGR in patients with digestive system cancer requires further investigation. In conclusion, our meta-analysis demonstrated that a low pretreatment AGR was closely related to worse long-term outcomes in patients with digestive system cancer. Nevertheless, further large prospective and homogeneous studies should be conducted to validate the prognostic value of the AGR. # Supporting information [^1]: The authors have declared that no competing interests exist.

# Introduction Tail-anchored (TA) proteins are a unique class of integral membrane proteins that possess a cytosolic N-terminal functional domain, followed by a single transmembrane domain (TMD) near or at their C terminus, and a short C-terminal hydrophilic tail. TA proteins are also unique because, unlike the classical type II membrane protein family that possess the same topology (i.e., N_out-C_in), their C-terminal TMD emerges from the ribosome only after the termination of translation and, thus, their sorting from the cytosol to the proper organelle membrane occurs *a priori* in a post- translational manner. TA proteins are associated with all intracellular membranes and participate in a remarkably wide array of physiologically important processes. Consequently, a considerable amount of research has focused in the past few years on understanding their biogenesis, particularly the molecular mechanisms underlying their targeting to and insertion into specific membranes in yeasts and mammals. For instance, the targeting information in almost all TA proteins in these organisms has been demonstrated to be located within their C-terminal TMDs and flanking sequences. Furthermore, the functional nature of these C-terminal targeting signals with regards to their membrane selectivity have been shown to be conveyed, not by primary sequence motifs, but, rather, by distinct physico- chemical properties, such as their net charge and/or the overall hydrophobicity of the TMD. In terms of the machinery that mediate the targeting and/or insertion of TA proteins to their specific intracellular destinations, several of these have been recently characterized, again primarily in yeasts and mammals, and, with the exception of peroxisome-destined TA proteins, most TA proteins in these organisms utilize novel organelle import pathways that do not overlap with those used by their non-TA membrane protein counterparts. Mitochondrial TA proteins, for instance, bypass the translocase of the outer mitochondrial membrane (TOM complex) and utilize instead the mitochondrial sorting and assembly machinery (SAM), and/or the unique lipid composition of the mitochondrial outer membrane , in order to ensure their selective targeting. Likewise, the targeting and insertion of ER-destined TA proteins appears to be distinct from the classical signal recognition particle (SRP)/Sec61 co-translational/translocation pathway used by most other ER membrane proteins. ER-destined TA proteins rely instead on several alternative and possibly parallel pathways, including the GET complex, Hsp40/Hsc70, or, at least in mammals, other cytosolic chaperones and the unique lipid composition of the ER membrane. How other TA proteins in yeast and mammals or those in other evolutionarily diverse organisms are selectively targeted to their proper intracellular destinations, remain important unanswered questions. In plants, our understanding of TA protein biogenesis is rudimentary because only a few authentic plant TA proteins have been identified and characterized in terms of their targeting and/or membrane insertion. These include the peroxisomal isoform of ascorbate peroxidase, the ER, mitochondrial and/or plastidial isoforms of cytochrome *b*₅ (Cb5), certain members of the SNARE protein family, the 34-kDa receptor subunit of the translocon at the outer envelope of chloroplasts from pea (*Pisum sativum*; psToc34), and its homologs in *Arabidopsis thaliana*, Toc33 and Toc34. *Arabidopsis* Toc33 and Toc34 function as related, but distinct, substrate- specific GTPase receptors and/or regulators involved in plastid protein import. However, while Toc33 (the presumed ortholog of psToc34;) and Toc34 have been examined with respect to their targeting and membrane insertion, the majority of these studies, as well as those involving psToc34, have yielded conflicting results in regards to the nature of their targeting information. Likewise, it is unclear whether the insertion of Toc33 and Toc34 is mediated by distinct proteinaceous and/or membrane lipid factors, analogous to the unique insertion of ER- and mitochondrial-localized TA proteins, and whether these factors are utilized also by other TA and/or non-TA outer envelope proteins (OEPs),. Thus, elucidating the means by which other plastid TA proteins are targeted to and inserted into the outer envelope is important, not only in terms of their comparative sorting to Toc33 and Toc34, but also for our overall understanding of the sorting of OEPs in general. Towards this end, we describe here the results of a comprehensive series of *in vivo* and *in vitro* experiments aimed at characterizing and comparing the biogenesis of three *Arabidopsis* TA OEPs: Toc33, Toc34, and a novel 9-kDa putative TA protein (named here ‘OEP9’) that is of unknown function and was recently identified in bioinformatics-based screen for *Arabidopsis* TA proteins. Overall, we demonstrate that OEP9 is a *bona fide* TA plastid outer envelope protein and that, like other OEPs, including Toc33 and Toc34, it relies on the ankryin repeat-containing protein, AKR2A, as a chaperone/receptor for its initial targeting from the cytosol to plastids. OEP9 is distinct from Toc33 and Toc34, however, with regards to the nature of its molecular targeting signal and the membrane protein and lipid components involved in its insertion into the chloroplast outer envelope. The implications of these results in terms of the diversity of OEP sorting pathways, including those responsible for TA OEPs, and the membrane specificity of TA protein targeting in plant cells in general are discussed. # Results ## Protein sequence features of *Arabidopsis* OEP9, (co)expression profiling and evolutionary analysis of predicted OEP9 homologues in other plant species As illustrated in, OEP9 (At1g16000) is an 86 amino-acid-long putative TA protein, possessing of a single predicted TMD and a 32 amino-acid-long hydrophilic C-terminal sequence (CTS). According to the information provided for the OEP9 gene locus at GenBank and TAIR, the deduced protein is annotated to be of unknown function and possess no putative targeting signal motifs. Moreover, with the exception of its single, α-helix-forming TMD and two predicted intrinsically disordered (unstructured) segments located at its N and C termini, OEP9 is devoid of any obvious structural and/or functional domains. Analyses of various *Arabidopsis* tissue expression databases and co-expression mining algorithms, however, revealed that OEP9 expression is relatively high in roots (–c) and is co-regulated with several other genes, most of which encode cytosolic or plastid ribosomal proteins. These observations suggest OEP9 functions in plastid ribosome biosynthesis in root cells and provide a reasonable explanation for why OEP9 is absent in publicly available proteomic databases of *Arabidopsis* chloroplast envelope membranes (e.g., PPBD), since these studies were conducted with photosynthetic chloroplasts, which are distinct from root plastids in terms of their proteome composition. Web-based searches using BLASTp revealed that putative homologues with an overall high degree of amino acid sequence identity to OEP9 and, likewise, annotated to be of unknown function, exist in *Arabidopsis* (i.e., At1g80890) and several other plant species, including dicots (*Brassica*, tobacco and cotton), monocots (rice and maize), and moss (*Physcomitrella*). In addition, phylogentic analysis of these and other putative OEP9 homologues from diverse plant species revealed that they are all closely related and that, since no homologues of OEP9 appear to exist in non-plant organisms, such as yeasts, insects or mammals, they likely share a common ancestor that arose before the evolutionary split between vascular (dicots and monocots) and non-vascular (moss) land plants. ## OEP9, Toc33 and Toc34 are localized to the plastid outer envelope in BY-2 cells Although OEP9 was predicted by Kriechbaumer et al to be localized to mitochondria, this was not confirmed experimentally and it contradicted our results that this protein localizes specifically to plastids. As shown in, transient expression of N-terminal myc-epitope-tagged OEP9 (myc-OEP9) in tobacco BY-2 suspension cells, followed by indirect immunofluorescence confocal laser- scanning microscopy, revealed that the protein localized to numerous toroidal- shaped fluorescent structures that enclosed the punctate/spherical fluorescent structures containing the endogenous plastid stromal protein N-acetylglutamate kinase (NAGK). These results are consistent with OEP9 being localized to the outer envelope of plastids in these cells, i.e., undifferentiated heterotrophic plastids. Indeed, similar localization results were observed for N-terminal myc- tagged versions of Toc33 (myc-Toc33) and Toc34 (myc-Toc34). Control experiments including mock transformations with empty plasmid vector alone or omission of anti-myc IgGs during immunofluorescence staining of BY-2 cells transformed with myc-OEP9 yield no (epi)fluorescence, as expected. We showed also that either co-expression of myc-OEP9 and GFP-Toc33 (a chimera consisting of the green fluorescent protein \[GFP\] appended at its C terminus to Toc33) or co-expression of myc-OEP9, myc-Toc33 or myc-Toc34 and OEP7-GFP (consisting of the 7-kDa *Arabidopsis* non-TA OEP appended at its C terminus to GFP), resulted in all of these proteins co-localized in the plastid outer envelope in BY-2 cells. Similar results were observed for co-expressed myc-OEP9 and OEP7-GFP in transformed *Arabidopsis* suspension cells and co-expressed GFP- OEP9 (consisting of GFP appended at its C terminus to OEP9) and Tic40-RFP (consisting of the 40 kDa subunit of the translocon at the inner envelope of chloroplasts \[Tic40; Ref. 39\] fused to the red fluorescent protein \[RFP\]) in transformed *Arabidopsis* leaf epidermal cells. Moreover, we showed that co- expressed OEP9 lacking an N-terminal myc epitope tag and myc-Toc33 colocalized in BY-2 cells. OEP9 being immunodetected in these latter cells using polyclonal antibodies raised against a synthetic peptide corresponding to an amino acid sequence in the protein's CTS (refer to); however, due limited availability of this antibody reagent, mostly myc-tagged versions of OEP9 were employed in the remainder of the experiments described in this study. Nonetheless, these results with OEP9 and those presented also in and confirm that the sorting of (myc-)OEP9 in BY-2 cells, serving as a well-established *in vivo* targeting system for ectopically-expressed proteins, faithfully reflects its localization in *Arabidopsis* cells. Notably, OEP7-GFP in (co)transformed cells localized also to several, relatively smaller torus structures that were devoid of myc-tagged OEP9, Toc33 or Toc34 or did not delineate the punctate/spherical fluorescent structures attributable to endogenous stromal NAGK. These smaller OEP7-GFP-containing torus structures did, however, delineate the punctate/spherical structures attributable to endogenous E1β, a protein subunit of the pyruvate dehydrogenase complex located in the mitochondrial matrix. These results indicate that OEP7-GFP, a commonly used marker (fusion) protein for the chloroplast outer envelope in other studies involving transiently-transformed *Arabidopsis* leaf cell protoplasts, sorts also to the mitochondrial outer membrane in suspension cells, perhaps as a consequence of its ectopic (over)expression or a (cryptic) mitochondrial targeting signal that functions depending on the cell type and/or cell function. ## Membrane topology and insertion of OEP9, Toc33 and Toc34 To determine whether OEP9, Toc33 and Toc34 were actually oriented in the plastid outer envelope in BY-2 cells in a TA (N_out-C_in) manner, cells individually expressing these three N-terminal myc-tagged proteins were differentially permeabilized with either Triton X-100 or digitonin, and then examined by epi(immuno)fluorescence microscopy. As shown in, myc-tagged OEP9, Toc33 and Toc34 were all immunodetected in cells incubated either with Triton X-100, which permeabilizes all cellular membranes, or with digitonin, which permeablizes only the plasma membrane. In the corresponding same cells, however, endogenous stromal NAGK was only immunodetected in cells permeabilized with Triton X-100 and not in cells permeabilized with digitonin, whereas cytosolic α-tubulin was detected in both Triton X-100- and digitionin-permeabilized cells, as expected. Similarly, expressed, non-epitope-tagged OEP was only immunodetected (via antibodies raised against a synthetic peptide in the protein's CTS \[refer to \]) in cells permeabilized with Triton X-100 and not in cells permeabilized with digitonin. Taken together, these results confirm that, consistent with a TA topology, the myc epitope and thus the N terminus of OEP9 (and Toc33 and Toc34) is orientated towards the cytosol, while the C terminus of OEP9 faces the intermembrane space. Membrane insertion, in addition to topological orientation of myc-tagged OEP9, Toc33 and Toc34, was assessed also using an *in vitro* import system with isolated *Arabidopsis* chloroplasts. (lanes 2 and 3) shows that *in vitro* synthesized myc-tagged OEP9, Toc33 and, although to a lesser extent, Toc34, integrated stably into chloroplast membranes, as evidenced by their resistance to extraction with alkaline Na₂CO₃. The lack of molecular mass shift for these three membrane-integrated proteins (i.e., when compared to the size of their translation products alone \[lane 1\]) also confirmed that each protein is devoid of a cleavable transit peptide. Membrane-integrated, myc- tagged OEP9, Toc33 and Toc34 were also confirmed to be orientated in the proper TA (N_out-C_in) manner, since treatment of isolated chloroplasts containing these proteins with thermolysin yielded (smaller) protected protein fragments of the expected size, i.e., approx. 6-kDa, 4-kDa and 5-kDa fragments representing the predicted molecular mass of the C-terminal TMD and CTS of OEP9, Toc33 and Toc34, respectively (lanes 4 and 5). Similar results were observed for the membrane insertion and topology of non-epitope-tagged OEP9, reinforcing earlier conclusions, based on *in vivo* localization and topology experiments (see above), that the addition of the myc sequence to OEP9 (i.e., myc-OEP9) does not affect its normal targeting and insertion. Likewise, the results presented here for the membrane insertion and topology of myc-Toc33 and myc-Toc34 *in vitro* are consistent with those published previously for non- epitope-tagged Toc33 and Toc34. That the precursor form of the soluble small subunit of Rubisco (SSU) was efficiently imported into isolated chloroplasts and properly processed into its mature, thermolysin-protected form, as expected, confirmed the import competence of the chloroplasts used in our *in vitro* import assays. ## Characterization of the targeting information in OEP9, Toc33 and Toc34 To characterize the specific molecular targeting information required for sorting of OEP9, Toc33 and Toc34 to the plastid outer envelope, we conducted a comprehensive series of *in vivo* targeting experiments using chimeras consisting of either: i) different portions of each of these three TA proteins fused to GFP serving as a passenger protein; or ii) specific protein domains swapped between OEP9 and either Toc33 or the mitochondrial isoform of Cb5, one of the best-studied plant TA proteins in terms of its targeting and membrane insertion. We focused mostly on Toc33, rather than Toc34, in these mutational targeting experiments because these two proteins possess a relatively high degree (61%) of amino acid sequence identity and because Toc33 has been less studied in terms of its targeting information, and only using *in vitro*-based assays. As shown in, transiently-expressed myc-Toc33 sorted exclusively to endogenous NAGK-containing plastids, as expected (cf. cells expressing myc-Toc33 and co- stained for NAGK in). On the other hand, deletion of the so-called ‘NTC domain’ from Toc33, namely the C-terminal region of the protein consisting of the 20 amino acid residues immediately upstream (<u>N</u> terminal) of its TMD, the <u>T</u>MD, and <u>C</u>TS, resulted in the modified protein (myc-Toc33ΔNTC) being mislocalized entirely to the cytosol. When the NTC of Toc33 was appended to the C terminus of GFP, however, the resulting fusion protein (GFP-Toc33NTC) localized to numerous small punctate structures that are not plastids, as evidenced by the lack of colocalization of GFP-Toc33NTC and NAGK. Instead, as discussed below, these structures containing GFP-Toc33NTC are likely protein aggregates in the cytosol. Similar results were observed when the NTC domain of Toc34 was appended to GFP (GFP-Toc34NTC). That the NTCs of Toc33 and Toc34 were necessary, but not sufficient, for plastid targeting indicated that other important targeting information existed in the N-terminal regions of these proteins. To test this possibility, a chimera consisting of GFP appended to the Toc33 NTC plus an additional ∼100 upstream amino acid residues of Toc33 was constructed (i.e., GFP- Toc33_141–297). As shown in, GFP-Toc33_141–297, similar to GFP-Toc33NTC, localized to small punctate structures that were devoid of NAGK. On the other hand, GFP-Toc33_37–297, consisting of amino acid residues 37 to 297, including the protein's entire GTP-binding (G)-domain fused to GFP, sorted to plastids in a manner similar to full-length myc-Toc33, suggesting that almost the entire Toc33 protein, including its G-domain, is required for proper targeting to plastids. Similarly, a mutant form of myc-Toc33 in which the so- called ‘arginine finger’ residue of its G-domain (i.e., position 130) was replaced with alanine (myc-Toc33R₁₃₀ΔA) localized exclusively to plastids. While the results of numerous other studies have shown that this arginine mutation affects the ability of Toc33 to self-dimerize and, thus, function properly as an preprotein import receptor ( and references therein), the results obtained here indicate that the plastid targeting of Toc33 itself does not rely on arginine finger-dependent dimerization. We next characterized the targeting information in OEP9. As shown in, deletion of either the NTC (myc-OEP9ΔNTC) or CTS (myc-OEP9ΔCTS) of OPE9 resulted in mislocalization to the cytosol and to the cytosol and punctate structures, respectively, indicating that the CTS is minimally necessary for proper targeting of OEP9 to plastids. The punctate structures containing myc-OEP9ΔCTS did not co-localize with endogenous marker proteins for mitochondria, peroxisomes or Golgi, but they did colocalize, at least in some instances, in punctate structures containing the co-expressed fusion protein GFP-OEP7 (consisting of GFP fused to the N terminus to OEP7). GFP-OEP7 is known to form protein aggregates in the cytosol of transformed plant cells presumably due to the plastid targeting information near its N terminus being sterically disrupted by the (N-terminal) appended GFP moiety. Thus, partial co-localizations between myc-OEP9ΔCTS and GFP-OEP7 in several punctate structures, suggests that myc- OEP9ΔCTS (as well as various GFP-Toc33/34 fusion proteins that localize also to punctate structures \[see above\]) forms protein aggregates in the cytosol due to the disruption of its (C-terminal) plastid targeting information. shows also that the NTC of OEP9, unlike the NTC of Toc33 or Toc34, possesses sufficient plastid targeting information, since GFP-OEP9NTC (consisting of GFP fused at its C terminus of the OEP9 NTC;) localized exclusively to plastids. The OEP9 CTS alone (residues 54–86), however, was unable to target GFP to plastids, and instead this fusion protein (GFP-OEP9CTS) remained entirely in the cytosol. That the CTS of OEP9 is necessary, but on its own not sufficient (see above), for plastid targeting prompted us to test next whether this region contained the protein's key targeting information. Toward this end, we swapped the CTS of Toc33 in the context of GFP-Toc33NTC, which does not sort to plastids, with the CTS of OEP9, yielding a modified chimeric protein (GFP-Toc33NTCΔOEP9CTS) that localized exclusively to plastids. The CTS of OEP9 was sufficient also in sorting to plastids a modified version of the mitochondrial isoform of tung tree (*Aleurites fordii*) Cb5 (myc-Cb5ΔOEP9CTS) whereby the three amino-acid-long CTS of Cb5 (-RRK) was replaced with the CTS of OEP9. As shown in, while full-length myc-Cb5 sorted to E1β-containing mitochondria and the corresponding myc-Cb5 mutant lacking its CTS (myc-Cb5ΔCTS) mislocalized to the cytosol, as expected, myc-Cb5ΔOEP9CTS localized exclusively to plastids. On the other hand, a modified chimeric protein consisting of myc-OEP9 with its CTS replaced with the CTS of Toc33 (myc-OEP9ΔToc33CTS) did not localize to plastids, but instead, similar to myc-OEP9ΔCTS, mislocalized to small punctate structures and the cytosol. Overall, the data presented in indicate that OEP9, compared to Toc33 and Toc34, contains distinctly different plastid targeting information. The targeting signals in Toc33 and Toc34 being relatively long, consisting of almost the entire protein, including its C-terminal NTC and GTPase domain. By contrast, the targeting signal in OEP9 consisting of only its CTS and the adjacent TMD sequence. ## Detailed characterization of the plastid targeting signal in the CTS of OEP9 To gain further insight to the nature of the plastid targeting signal in the CTS of OEP9, we initially deleted the C-terminal half of this region of the protein. As shown in, myc-OEP9_1–70, which lacks the protein's C-terminal 16 amino acid residues (residues 71–86) did not localize to NAGK-containing plastids. Instead, similar to myc-OEP9ΔCTS , myc-OEP9_1–70 mislocalized to the cytosol and small punctate structures (presumably protein aggregates), indicating that this deleted portion and/or a combination of the both halves of the CTS are essential for targeting OEP9 to plastids. An examination of the OEP9 CTS sequence revealed it contains a number of positively- and negatively-charged residues, the majority of which were located between positions 65 to 77. Notably, this cluster of charged residues within the CTS was disrupted in the mutant myc-OEP9_1–70 and is conserved in all the putative homologues of OEP9. To assess, therefore, whether these charged residues in the CTS of OEP9 are important for its proper targeting, two mutants were constructed wherein several of either the positively-charged lysine and arginine residues or the negatively-charged aspartate residues were replaced with glycines. As shown in, the positively-charged mutant myc- OEP9K₆₉K₇₂R₇₄K₇₅ΔG mislocalized exclusively to E1β-containing mitochondria and the corresponding negatively- charged mutant, myc-OEP9D₆₈D₇₁ΔG, also mislocalized partially to mitochondria (i.e., myc-OEP9D₆₈D₇₁ΔG localized to both plastids and mitochondria). Interestingly, glycine substitutions of other (non-charged) amino acids within the same region of the OEP9 CTS also disrupted the protein's normal targeting to plastids, i.e., myc- OEP9Y₆₆M₆₇A₇₀ΔG mislocalized to punctate structures and the cytosol. Collectively, these data suggest that the net charge and/or charge distribution of the CTS, as well as the overall three-dimensional configuration of the CTS, mediates the plastid targeting specificity of OEP9. ## AKR2A interacts with OEP9 *in vivo* Recently, the *Arabidopsis* ankryin repeat-containing protein, AKR2A, was shown to function as an essential cytosolic mediator of OEP biogenesis, acting both as a chaperone to prevent the aggregation of nascent OEPs and as a receptor to facilitate their subsequent targeting from the cytosol to the chloroplast outer envelope. The evidence in support of this dual role for AKR2A provided in part by *in vitro* protein pull-down and/or nuclear mislocalization assays, which demonstrated that AKR2A interacts specifically with various OEPs. To investigate whether AKR2A interacts with OEP9 we also employed a nuclear mislocalization assay. Specifically, we constructed two chimeric proteins that consist of three tandem copies of the nuclear localization signal (NLS) from the SV-40 large T antigen fused to either the red fluorescent protein (RFP) alone (NLS-RFP) or to the RFP and AKR2A (NLS-RFP-AKR2A). For comparative purposes, a third chimera was constructed consisting of GFP fused to AKR2A alone (GFP-AKR2A) and that, unlike NLS-RFP-AKR2A, lacks an appended NLS. Consistent with the intracellular localizations reported previously for these three chimeric proteins in transiently-transformed leaf protoplasts, NLS-RFP and NLS-RFP-AKR2A both localized exclusively to the nucleus, while GFP-AKR2A localized to the cytosol in transformed BY-2 cells, indicating that the NLS was efficient in mislocalizing AKR2A (i.e., NLS-RFP-AKR2A) from the cytosol to the nucleus in these cells. Also consistent with previously published results, NLS-RFP-AKR2A was capable of mislocalizing co-expressed GFP-OEP7 to the nucleus in BY-2 cells. By contrast, GFP-OEP7 co-expressed with the NLS-RFP, similar to when GFP-OEP7 was expressed on its own, localized to numerous punctate structures that, as discussed above, are likely cytosolic aggregates of this fusion protein (cf. cells expressing GFP-OEP7 and co-expressing GFP-OEP7 and NLS-RFP in and, respectively). Notably, OEP7-GFP sorted to plastids and did not mislocalize to the nucleus when co- expressed with NLS-RFP-AKR2A, indicating that AKR2A does not bind efficiently to OEP7 when GFP is appended to its N terminus (GFP-OEP7), but does so when GFP is appended to its C terminus (OEP7-GFP); a conclusion that is consistent with previously published data on the functionality, or lack thereof, of the N-terminal plastid targeting signal in OEP7 and why GFP-OEP7 (and not OEP7-GFP) was employed here and elsewhere in nuclear mislocalization assays with AKR2A. In additional control experiments, Toc33-GFP (consisting of Toc33 fused at its C terminus to GFP;) co-expressed with NLS-RFP-AKR2A localized predominantly to the nucleus, confirming and extending previous results from *in vitro* pull-down assays showing that Toc33 interacts with AKR2A. On the other hand, when Toc33-GFP was either co-expressed with NLS-RFP or expressed alone it localized to the cytosol and not to plastids, presumably due to the disruption of the Toc33 plastid targeting information by the C-terminal-appended GFP moiety. shows that OEP9-GFP, consisting of OEP9 appended at its C terminus to GFP, localized to both plastids and to numerous punctate structures within the cytosol when expressed on its own. Analogous to the mislocalization of GFP-OEP7, these OEP9-GFP-containing punctate structures likely represent mislocalized aggregates of the fusion protein due to the partial disruption of the OEP9's plastid targeting information by the appended GFP moiety. OEP9-GFP also localized to both plastids and aggregates in the cytosol when co-expressed with NLS-RFP. However, when co-expressed with NLS-RFP-AKR2A, at least a portion of OEP9-GFP (mis)localized to the nucleus, i.e., in addition to being localized to plastids and the cytosolic aggregates, OEP9-GFP also accumulated in the nucleus when co-expressed with NLS-RFP-AKR2A (cf. cells co-expressing OEP9-GFP with NLS- RFP-AKR2A or NLS-RFP), indicating that OEP9 interacts with AKR2A. ## OEP9, compared to Toc33 and Toc34, requires different membrane-bound proteinaceous factors for integration and displays distinct differences in membrane lipid association Given our results indicating that OEP9, similar to other OEPs, relies on AKR2A as a mediator (i.e., chaperone/receptor) for its targeting from the cytosol to plastids, we examined next whether other protein(s), if any, are responsible for the subsequent insertion of OEP9 into the plastid outer envelope membrane. Toward this end, OEP9, Toc33 and Toc34 were compared initially for their ability to insert into isolated chloroplasts that were treated with the protease trypsin prior to the insertion reaction and, thus, removed surface-exposed outer membrane proteins including candidate receptor(s). As shown in, only a portion of *in vitro* synthesized, radiolabeled myc-tagged OEP9, Toc33 and Toc34 bound to and stably integrated into (as evidenced by their resistance to extraction with Na₂CO₃) trypsin-pretreated chloroplasts. That is, compared to the behavior of these three proteins in reactions containing untreated intact chloroplasts (refer to, lanes 2 and 3), their binding and integration into trypsin-pretreated chloroplasts was substantially reduced, although to a much lesser extent for myc-Toc34. The import and processing of SSU, however, was completely abolished by the pre-treatment of chloroplasts with trypsin. These latter results confirm that the protease had efficiently degraded proteins of the Toc complex, since the import of SSU is well known to be Toc complex-dependent. We demonstrated also that the integration, but not the binding, of myc-Toc33 and myc-Toc34 into chloroplast outer envelope membranes was significantly reduced when *in vitro* import reactions with either of these two proteins contained chloroplasts isolated from *ppi1* or *ppi3 Arabidopsis* mutant plants that lacked (via a T-DNA insertion) Toc33 or Toc34, respectively. These data indicate that Toc33 and Toc34 themselves serve as receptor proteins involved in their proper insertion. OEP9, however, does not appear to depend on the Toc33 or Toc34 receptors, since it bound and integrated, as well as orientated (based on thermolysin protection assays) in the proper (TA) manner, into both *ppi1* and *ppi3* chloroplasts in a manner similar to that for wild-type chloroplasts (cf.,). Instead, data presented in indicate that the OEP9 is dependent, at least in part, on some other surface-exposed proteinaceous factor(s). On the other hand, SSU was properly imported into *ppi3* chloroplasts, but was not into *ppi1* chloroplasts, consistent again with previous studies indicating that this protein, like other photosynthetic proteins, relies more so on Toc33 (than Toc34) for its import. The observation that a portion of OEP9, Toc33 and, to a greater extent, Toc34, inserted into protease-pretreated chloroplasts might be due to direct protein- lipid interactions and, thus, we tested whether these three TA proteins can bind to synthetic membrane lipids *in vitro*. Specifically, translation reactions containing myc-tagged OEP9, Toc33 or Toc34 were incubated with or without protein-free lipid membranes (liposomes) containing an average lipid composition similar to that of the chloroplast outer envelope membrane. All of the samples were subsequently subjected to sucrose gradient centrifugation followed by fractionation of the gradient into those containing either the liposomes and liposome-bound proteins (fractions 1 and 2), unbound proteins that remained in a specific portion of the sucrose gradient (fraction 3), the load fraction (fraction 4) or aggregated proteins that pelleted to the bottom of the gradient (fraction 5). As shown in portion of the myc-tagged OEP9, Toc33 and Toc34 added to the incubations was recovered in gradient fractions containing chloroplast-like liposomes (fractions 1 and 2, solid arrowheads), indicating that all three proteins were binding directly to this lipid bilayer. However, their binding efficiency to chloroplast-like liposomes varied considerably, i.e., while a substantial portion of Toc33 and Toc34 proteins were recovered in fractions with liposomes (fractions 1 and 2) compared to the gradient and load fractions (fractions 3 and 4), the majority of soluble OEP9 remained in the load and bottom fractions (fractions 4 and 5). Moreover, a portion of OEP9 was recovered in the top soluble fractions (fractions 1 and 2) of gradients without liposomes. Overall, these data indicate Toc33 and Toc34 bind much more efficiently to the chloroplast-like liposomes than OEP9. Shown also in, the majority of the soluble control protein SSU remained in the load and bottom fractions in gradients with or without liposomes, indicating that SSU, consistent with previous results, does not interact with chloroplast-like liposomes. We next assessed whether the targeting to chloroplast-like liposomes of Toc33, Toc34, and, although to a much lesser extent, OEP9 was specific for this lipid bilayer, since previous studies with psToc34 revealed its insertion into liposomes was dependent on the presence of lipids unique to the plastid outer envelope, namely the non-bilayer lipids mono/digalactosyldiacylglycerides (MGDG/DGDG), and on the concentration of anionic lipids, such as phosphatidylglycerol (PG). We therefore tested the ability of OEP9, Toc33 and Toc34 to bind liposomes that, unlike chloroplast-like liposomes, were devoid of MGDG, DGDG and PG, and possessed different amounts of other lipids that, overall, yielded a composition similar to mitochondrial membranes. Sucrose gradient flotation assays were employed for these experiments as described above, but, rather than SSU, an ER isoform of rat Cb5 (rCb5) served as a control protein, since this TA protein targets *in vitro* to any membrane, including synthetic liposomes. As shown in, neither myc-Toc33 nor myc-Toc34 bound to the mitochondrial-like liposomes, consistent with previous results for psToc34. By contrast, myc-OEP9 bound to the mitochondrial-like liposomes and did so in a manner similar to rCb5. Taken together, the data presented in suggest that OPE9, compared to Toc33 and Toc34, displays differences in its preference for binding membrane lipids, and that this behavior may serve as an important determinant in the targeting specificity of these three TA proteins. # Discussion ## The sorting of TA proteins to plastids involves at least two distinct pathways Plastids participate in a wide array of essential metabolic processes, all which rely on the acquisition of distinct nuclear-encoded protein components from the cytosol, and as such, the protein composition of the organelle is influenced by both nuclear gene expression and the activity of intracellular targeting pathways specific for plastid biogenesis. In fact it is now well appreciated that multiple import pathways serve in the uptake of both soluble and membrane- bound proteins into plastids. However, compared to other proteins our understanding of plastid TA protein biogenesis is lacking. Given the importance of TA proteins in other critical aspects of cell metabolism and physiology, we undertook a comparative analysis of the targeting and insertion mechanisms of three plastid TA proteins, including Toc33 and Toc34, both of which function as Toc complex-receptor GTPases, and OEP9, a newly-identified TA protein of unknown function. Overall, our results provide evidence in support of at least two pathways for plastid TA biogenesis that are distinguished by the nature of their molecular targeting signals and the membrane protein and lipid components involved. These findings should now not only facilitate a more detailed analysis of these membrane components, some of which may be shared in terms of their underlying biochemical mechanisms, but also complement the growing body of evidence for the complex diversity of plastid protein sorting pathways, as well as the diversity of sorting pathways for TA proteins localized to other organelles (e.g., mitochondria and ER). ## OEP9 is integrated in the plastid outer envelope in a TA manner and may be involved in ribosome biosynthesis in roots OEP9 was one of over 500 *Arabidopsis* candidate TA proteins identified in a recent bioinformatics screen based primarily on the three main structural characteristics that have traditionally defined the TA protein family, including: 1) the presence of a single putative TMD within the C-terminal ∼50 amino acid residues; 2) the absence of any other TMDs; and 3) the lack of an N-terminal hydrophobic secretory signal sequence. Consistent with each of these characteristics, we showed here using differential detergent permeabilization and protease protection assays that OEP9 is stably integrated in the chloroplast outer envelope in a TA manner (i.e., N_out-C_in). This interpretation of OEP9's TA topology was reinforced by results from a parallel series of assays with Toc33 and Toc34, all of which were in full agreement with the previously reported TA topology of these two proteins. Similar to many other OEPs, the function of *Arabidopsis* OEP9 is unknown. Nevertheless, the existence of conserved OEP9 homologues in other diverse plant species and the absence of homologues in non-plant organisms, suggests that its function(s) is plant specific. Indeed, some indirect evidence obtained from various web-based *Arabidopsis* (co)expression datasets supports the possibility that OEP9 functions in the plastids of root cells and in plastid ribosome biosynthesis. Whether OEP9 is actually involved in plastid ribosome biogenesis in roots, however, remains to be tested experimentally; a task that will likely require inducible RNAi mutants of OEP9, since knock-out (T-DNA) mutants of this gene or its paralogue (At1g80890) are not available, suggesting also that OEP9 is essential for plant growth and development. ## Properties of the OEP9 targeting signal Since plant cells possess an additional organelle (the plastid) that is absent in most other eukaryotic cells (e.g., yeast and mammals), there is an added level of complexity in the intracellular trafficking system for plant TA proteins that warrants a close examination of the targeting signals involved. For almost all TA proteins, regardless of their organelle destination, the initial targeting event is mediated by *cis*-acting sequences within the C-terminal region of the protein. Consistent with this paradigm, the OEP9 CTS and TMD together are both necessary and sufficient for targeting the protein from the cytosol to plastids. However, these sequences within OEP9 also appear to play distinct roles: the CTS contains the protein's key plastid targeting information and the TMD, which in addition to being required for thermodynamic association and integration into membranes, possesses general physico-chemical properties, such as overall hydrophobicity, length, and/or propensity to form an α helical structure, that act together to convey the proper context for the CTS to function as a targeting signal. Perhaps the best support of this conclusion is that the CTS of OEP9 on its own is not sufficient for targeting GFP to plastids, but is sufficient in re-targeting the mitochondrial isoform of Cb5 or the NTC domain of Toc33 fused to GFP to plastids. By contrast, the Toc33 NTC domain on its own was not sufficient for targeting GFP to plastids. As discussed below, these latter data indicate that Toc33 (and Toc34) possesses a different type of targeting signal than OEP9 since it relies on additional targeting information present within the N-terminal GTPase domain of the proteins. Inspection of the CTS sequences of OEP9 and putative OEP9 homologues in other plant species revealed several conserved features that possibly represent distinct targeting signal motifs. For instance, all of these proteins possess a cluster of conserved positively- and negatively-charged amino acid residues (residues 65–77;) that are similar to the charged residues known to be important for the proper sorting of other OEPs, namely *Arabidopsis* OEP7 and OEP64. Interestingly, both OEP7 and OEP64 possess a single TMD, but unlike OEP9, it is located near the protein's N terminus and yields an N_{intermembrane space}-C_cytosol orientation in the outer envelope membrane. Moreover, the clusters of charged residues in OEP7 and OEP64 have been implicated in preventing interaction with SRP and thus entry into the Sec61 co- translational pathway of the secretory system. We also found that the charged residues in OEP9 (CTS) are critical for its proper targeting to plastids, suggesting that OEP9 shares the same targeting information and, hence, as discussed below, utilizes the same plastid biogenetic pathway as OEP7/64. One important difference between OEP9 and OEP7/64, however, is that mutations to certain charged residues in the CTS of OEP9 resulted in the modified proteins being mistargeted to mitochondria, rather than to the secretory system. This difference in (mis)targeting is most likely due to both OEP7 and OEP64 possessing an N-terminal TMD that, when their charged residues are mutated, engages the SRP/Sec61 co-translational pathway, whereas OEP9 (wild-type or mutant) possesses a C-terminal TMD that emerges from the ribosome only after the termination of translation and, thus, targets strictly in a SRP-independent post-translational manner. The mitochondrial mislocalization of OEP9 mutants with alterations to certain charged residues within the CTS also suggests that the TA targeting pathways for chloroplasts and mitochondria are independent, but competing, and that the specific sorting of OEP9, as well as other TA proteins, to either of these two organelles (or to other organelles) is not based strictly on the overall net positive charge of the CTS. While the actual distribution of the charges in the CTSs of these proteins may be an important aspect in mediating targeting specificity, the basic mechanism(s) that underlies the proper sorting of TA proteins in plant cells does not appear to match that in mammals, wherein a net positive charge in the CTS conveys sorting to mitochondria and a net negative or null charge conveys sorting to the ER. It seems instead that plant TA protein targeting signals have acquired additional information that ensures higher fidelity association with the correct organelle. Consistent with this premise, mutational analyses of the OEP9 CTS revealed that, in addition to the charged- related characteristics, the overall secondary and/or three-dimensional configuration of this region appears to play an important role in plastid targeting specificity. This is potentially an important featureof the OEP9 CTS, since at least some protein structure prediction programs indicated that this region, as well as the N terminus of the protein, has the propensity to be intrinsically disordered and disordered segments in other proteins can serve as specific binding/recognition elements and/or flexible linkers involved in macromolecular assembly. Spectroscopic and prediction-based structural modeling of the CTS of wild-type and mutant OEP9 proteins, as well as large-scale and systematic mutational analyses of the putative targeting signals in the CTSs of other (predicted) plastid TA proteins are now being planned in order to determine whether the putative unstructured domains and/or physico-chemical and sequence-specific features in OEP9 are functionally conserved. ## Role of the GTPase domain in the targeting and membrane insertion of Toc33 and Toc34 Compared to OEP9 and most other TA proteins examined to date, Toc33 and Toc34 appear to be unique in that targeting is not mediated only by sequences within their C-terminal TA sequence. That is, while the NTC sequences of Toc33 or Toc34 are necessary for their targeting to plastids, they are not sufficient in redirecting GFP from the cytosol to plastids. Only when the entire G-domain of Toc33, along with the NTC region, was fused to GFP was targeting to plastids observed. Taken together, these data are consistent with previous *in vitro* studies indicating that the G-domain of Toc34, along with the TMD and CTS sequences, is important to varying degrees for insertion into isolated chloroplasts. Interestingly, we observed also that a G-domain-mutant version of Toc33 (myc-Toc33R₁₃₀ΔA), which exists primarily as a monomer *in vitro*, targets to plastids *in vivo* in a manner similar to its wild-type counterpart. Thus, while it appears that the G-domains of Toc33 and Toc34 are, at a minimum, critical structural determinants important for maintaining the overall targeting- and/or insertion-competent conformation of these receptor proteins, the so-called ‘arginine fingers’ within these G-domains and, hence, the self-dimerization process that they presumably mediate, is not a prerequisite for proper targeting. However, since Toc33/34 at the chloroplast surface are important for mediating their own insertion (; see below), resident Toc33/34 homologs in BY-2 cells, which presumably contain a corresponding intact arginine finger, may account, in part, for the successful plastid targeting of myc-Toc33R₁₃₀ΔA. It is also possible that the apparent differences in the role(s) of the ‘arginine fingers’ in targeting and/or insertion/assembly versus homodimerization of Toc33/34 reflects the complex nature of (TA) OEP membrane biogenesis in general and the different approaches (*in vivo* versus *in vitro*) employed to study this multi-step process. ## Role for ‘kinetic factors’ in the organelle-specific targeting and membrane insertion of OEP9, Toc33 and Toc34 In recent years, considerable progress has been made towards understanding the biogenetic pathways responsible for the intracellular localizations of TA proteins. Based almost entirely on studies carried out with yeast and mammalian model systems, and with TA proteins that localize to mitochondria, peroxisomes or ER, the current working model for TA protein biogenesis involves two main steps: (i) delivery of the nascent protein from its sites of syntheses in the cytosol to the surface of the appropriate organelle, a process that must also ensure the avoidance of interaction with inappropriate organelles; and (ii) the subsequent insertion of the TA protein into its proper membrane bilayer. Both of these steps rely on, depending on the TA protein, one or more so-called ‘kinetic factors’ (e.g., *cis*-acting targeting and insertion sequences, cytosolic proteins, membrane proteins and/or lipids, etc.) that ultimately serve to accelerate the integration and, thus, the retention of the TA protein into its proper organelle membrane destination. In the case of plastid TA proteins, our results and those presented elsewhere indicate that the first step in their biogenetic pathway is mediated, at least in part, by the cytosolic chaperone/receptor AKR2A. That is, our data from nuclear mislocalization assays suggests that AKR2A controls the intracellular distribution *in vivo* of both OEP9 and Toc33, as well as the non-TA (control) protein OEP7. While this conclusion for AKR2A and OEP9 likely requires addition experimental support, it is reasonable to presume that ARK2A does not appear to function as a general mediator of other (non-plastid) membrane proteins, including TA proteins, since AKR2A did not interact *in vivo* with the mitochondrial isoform of Cb5 or *in vitro* with the 22 kDa peroxisomal membrane protein or mitochondrial TOM20. On the other hand, AKR2A interaction specifically with OEPs appears to be mediated by the plastid targeting sequences since AKR2A does not bind *in vitro* to OEP7 or OEP64 that are devoid of their targeting signals or *in vivo* either to OEP7 when the protein's targeting signal is sterically blocked by an N-terminal appended GFP moiety or to OEP9 lacking its CTS. The mechanism by which AKR2A recognizes TA and non-TA OEPs and how AKR2A functions as a chaperone to maintain nascent OEPs in a targeting- and insertion-competent state are open questions. We showed also that membrane-bound protein factors play an important role in the insertion of OEP9, Toc33 and Toc34 into the plastid outer envelope. However, the specific membrane proteins involved and, thus, the underlying mechanisms that mediate the insertion of these three TA proteins appears to be different for OEP9 compared to that for Toc33 and Toc34. For instance, while binding to and insertion into the membrane was sensitive to trypsin pretreatment of chloroplasts for all three TA proteins *in vitro*, only Toc33 and Toc34 did not insert into chloroplasts isolated from mutant plants lacking Toc33 (*ppi1*) or Toc34 (*ppi3*), albeit less so for Toc34. These data suggest that Toc33 and Toc34 themselves are essential for their insertion. Moreover, that Toc33 and Toc34 still bound, but did not integrate into, *ppi1* or *ppi3* chloroplasts and that the targeting efficiency of Toc34 to trypsin-treated chloroplasts *in vitro* was greater than that of Toc33 is consistent with previous conclusions that the biogenesis of these two receptor proteins relies on additional, perhaps different, membrane proteins. While the identity of the membrane protein factor(s) involved in the binding and/or insertion of OEP9 into the plastid outer envelope also remain to be determined, both Toc33 and Toc34 are not likely candidates in this regard since OEP9 inserts efficiently and in the correct (TA) topology into *ppi1* and *ppi3* chloroplasts, supporting further the notion that the mechanism of insertion of OEP9 is different than that of Toc33 and/or Toc34. Indeed, since OEP9 appears to share the same targeting information as OEP7 (see above), it may utilize the same insertion machinery, i.e., Toc75, the protein-conducting channel of the Toc complex that serves, in addition to its role in Toc complex-mediated preprotein translocation, in the membrane insertion of the OEP7 homolog from pea (OEP14). By contrast, Toc75 does not appear to participate in the insertion of psToc34 into the chloroplast outer envelope. In addition, OEP9, Toc33 and Toc34 appear to rely on membrane lipids, but yet they do so in different ways. For instance, while all three TA proteins bound to protein-free liposomes with a composition that resembled that of the chloroplast outer envelope, OEP9 did so much less efficiently. By contrast, OEP9, but not Toc33 and Toc34, bound to mitochondrial-like liposomes. These results, combined with those published previously for the specific insertion of psToc34 into protein-free chloroplast-like liposomes, but not into isolated mitochondria, and the proposed role of lipids in the targeting specificity of TA proteins in general, suggests that the unique lipid composition of the chloroplast outer envelope allows Toc33 and Toc34 to discriminate between the surface of chloroplasts and that of other incorrect organelles. An interaction that may also help explain, in part, the evolution of a targeting process for these two (receptor) proteins that is dependent on themselves. In contrast to Toc33 and Toc34, membrane lipids of the chloroplast outer envelope membrane appear to serve primarily to mediate normal thermodynamic association and integration of OEP9, and, therefore, protein factors (e.g., AKR2A and possibly Toc75) likely determine its plastid-specific targeting and integration. This premise is similar to the model developed for the ER-specific isoform of rCb5, which inserts into all membranes in a cell free system, but targets exclusively to the ER *in vivo*, presumably by the action of (cytosolic) protein factors that prevent its nonspecific insertion into other (incorrect) organelle membranes. While this proposed thermodynamic role for membrane lipids in OEP9 biogenesis remains to be confirmed experimentally, it is tantalizing to speculate that the targeting of OEP9 to mitochondrial-like liposomes compared to chloroplast-like liposomes *in vitro* also reflects an underlying affinity of this protein for certain membrane lipids that may be present at specific sites or domains in the chloroplast outer envelope. For instance, if one considers that the lateral distribution of lipids in the chloroplast outer envelope is likely not uniform, it is possible that specific lipid domains exist within this membrane and that these, in combination with certain protein factors, help mediate the proper association and integration of OEP9 into the plastid outer envelope membrane. # Materials and Methods ## Recombinant DNA procedures and reagents Standard recombinant DNA procedures were preformed as described by Sambrook et al. Molecular biology reagents were purchased from New England Biolabs Ltd. (Pickering, Canada) and Invitrogen Canada Inc. (Burlington, Canada) and plasmid DNA was isolated using commercially available kits either from Qiagen (Mississauga, Canada), Invitrogen, or Bio-Basic Inc. (Markham, Canada), all in accordance with the manufacturer's instructions. All DNA constructs were verified using dye terminated cycle sequencing preformed at either Arizona State University DNA Laboratory (Tempe, AZ) or the University of Guelph Genomics Facility (Guelph, Canada). Plasmid DNA mutagenesis reactions were carried using the QuikChange site-directed mutagenesis kit (Stratagene, La Jolla, CA). Synthetic oligonucleotides were synthesized by either Sigma-Genosys Canada (Oakville, Canada) or University of Guelph Laboratory Services (Guelph, Canada). ## Construction of Plasmids A complete description of all plasmids used in this study and a list of the sequences of oligonucleotide primers used in plasmid constructions are provided in and, respectively. ## Tobacco BY-2 cell cultures and microprojectile bombardment Tobacco (*Nicotiana tabacum* cv BY-2) and *Arabidopsis thaliana* (var Lansberg erecta) suspension cell cultures were maintained and prepared for biolistic bombardment as described previously. Transient transformations, including those involving *Arabidopsis* leaf epidermal cells, were performed using 10 µg of plasmid DNA (or, with one exception \[see below\], 5 µg of each plasmid for co- transformations) with a biolistic particle delivery system (Bio-Rad Laboratories Ltd., Mississauga, Canada). For nuclear mislocalization assays, 500 ng of plasmid DNA encoding the GFP fusion protein(s) was used for co-transformations. Following bombardment, cells were incubated for 6–20 h to allow for expression and sorting of the introduced gene product(s) and then processed for immunofluorescence microscopy. ## Immunofluorescence microscopy Biolistically bombarded tobacco Bright Yellow-2 (BY-2) or *Arabidopsis thaliana* (var Landsberg erecta) suspension-cultured cells were processed for immunofluorescence microscopy as described by Lingard et al. Briefly, both cells were fixed in 4% (w/v) formaldehyde, and then incubated for 2 h with either (for BY-2 cells) 0.01% (w/v) pectolyase Y-23 (Kyowa Chemical Products, Osaka, Japan) or (for Arabidopsis cells) 0.03% (w/v) cellulysin (Calbiochem) and 0.1% (w/v) pectinase (Sigma-Aldrich Ltd., Oakville, Canada). Thereafter, cells were permeabilized with either 0.3% (v/v) Triton X-100 or 25 ug/mL digitonin (Sigma- Aldrich Ltd.) for 30 min. Primary antibodies and sources were as follows: custom rabbit anti-OEP9 antibodies were raised against a keyhole limpet hemocyanin- conugated synthetic peptide corresponding to the OEP9 amino acid sequence DKADKARKARLSSSSSANK (residues 68 to 86 \[refer to \]) (Cedarlane Laboratories Ltd., Hornby, Canada); mouse anti-myc antibodies in hybridoma medium (clone 9E10; Princeton University Monoclonal Antibody Facility, Princeton, NJ); rabbit anti-*Arabidopsis* N-acetyl glutamate kinase (NAGK) ; rabbit anti-pea E1β ; rabbit anti-cottonseed catalase ; rabbit anti-pea reversibly glycosylated polypeptide ; and mouse anti-α-tubulin (Sigma-Aldrich Ltd). Fluorescent dye- conjugated secondary antibodies sources were as follows: goat anti-mouse and goat anti-rabbit Alexa 488 and goat anti-rabbit Cy5 (Invitrogen); goat anti- mouse and goat anti-rabbit rhodamine red-x (Jackson ImmunoResearch Laboratories, Inc., West Grove, PA). Epifluorescent images of suspension cells were acquired using a Zeiss Axioscope 2 MOT epifluorescence microscope (Carl Zeiss Inc., Thornwood, USA) with a Zeiss 63X Plan Apochromat oil-immersion objective. Image capture was performed using a Retiga 1300 charge coupled device camera (Qimaging, Surrey, Canada) and Openlab 5.0 software (Improvision, Waltham, MA). CLSM images were acquired using a Leica DM RBE (Leica Microsystems Inc., Richmond Hill, Canada) microscope with a Leica 63x Plan Apochromat oil-immersion objective a Leica TCS SP2. Fluorophore emissions were collected sequentially in double-labelling experiments; single- labelling experiments exhibited no detectable crossover at the settings used for data collections. Confocal images were acquired as single optical sections and saved as 512×512 pixel digital images. Note also that epifluorescence images All fluorescence images of cells shown in the figures are representative of \>50 independent (transient) transformations from at least two independent transformation experiments. Figure compositions were generated using Northern Eclipse (v. 5.0) software (Empix Imaging Inc., Mississauga, Canada) and Adobe Photoshop CS (Adobe Systems Canada, Etobicoke, Canada). ## Arabidopsis growth conditions All wild-type and mutant (*ppi1* and *ppi3*) *Arabidopsis* plants were of Columbia-0 ecotype. *ppi1* and *ppi3* seeds were provided by J. Froehlich (Michigan State University). Seeds were surface-sterilized and sown on Petri plates containing 4.3 g/L Murashige and Skoog salt and vitamin mix with buffer (Bioshop Canada Inc., Burlington, Canada), 10 g/L sucrose and 0.8% (w/v) agar as previously described. Seeds were then chilled at 4°C and grown under a long-day cycle (16 h light, 8 h dark) at 20–25°C until being harvested ∼14 days after germination for chloroplast isolations (see below). ## Targeting to chloroplasts and liposomes *in vitro* *Arabidopsis* chloroplasts were isolated as described by Wang et al. Phospholipid vesicles (liposomes) with various lipid content were prepared by extrusion in 10 mM Tris-HCl (pH 7.5) as described previously. Phosopholipid vesicles of chloroplast-like composition (based on the outer envelope of chloroplasts from spinach with the exception that 6% sulfoquinovosyl diacylglyerol was omitted) contained (as moles percent) 30∶20∶32∶10∶6∶2 digalactosyldiacylglyceride (DGDG)/monogalactosyldiacylglyceride (MGDG)/phosphatidylchloine (PC)/phosphatidylglycerol (PG)/phosphatidylinositol (PI)/phosphatidylethanolamine (PE). Phosopholipid vesicles of mitochondria-like composition (based on Henderson et al for *Xenopus* mitochondria) contained (as moles percent): 48∶10∶28∶10∶4 PC/PI/PE/phosphatidylserine (PS)/cardiolipin. MGDG and DGDG were purchased from Larodan Fine Chemicals (Malmo, Sweden) and all other phospholipids were purchased from Avanti Polar Lipids Inc. (Alabaster, AL). With the exception of Cb5, all *in vitro* synthesized proteins (OEP9, Toc33, Toc34 and SSU) were generated using the appropriate plasmid DNAs along with a T7-coupled transcription-translation system containing wheat germ extract and \[³⁵S\]-Methionine (Perkin-Elmer NEN Radiochemicals, Waltham, MA) according to the manufacturer's instructions (Promega, Nepean, Canada). Cb5 was synthesized *in vitro* using pSP/CytoB5 plasmid DNA, SP6 polymerase (MBI Fermentas, Burlington, Canada), and RNAs translated using rabbit reticulocyte lysate in the presence of \[³⁵S\]-Methionine as previously described. Targeting of *in vitro*-translated proteins to chloroplasts was carried out as described by Smith et al. Briefly, translated proteins were incubated with 50 µg of chloroplasts in HEPES-sorbitol buffer (20 mM HEPES \[pH 7.5\], 300 mM sorbitol), import master mix (consisting of: 50 mM HEPES-KOH, 330 mM sorbitol, 5 mM magnesium acetate, 25 mM potassium acetate), 1 mM dithiothreitol, 5 mM ATP, 1 mM GTP, 10 mM methionine and incubated at 26°C for 30 min. Following targeting, chloroplasts were reisolated by centrifugation at 2,000 *g* for 5 min at room temperature and then resuspended in either SDS-PAGE sample buffer or 100 mM Na₂CO₃ (pH 11.5). Chloroplasts resuspended in Na₂CO₃ were incubated on ice for 10 min and then centrifuged at 40,000 *g* for 30 min at 4°C using an Optima Max ultracentrifuge (Beckman Coulter Canada, Inc., Mississauga, Canada). Following centrifugation, the supernatant was isolated and subjected to trichloroacetic acid precipitation and the resulting pellet was resuspended in SDS-PAGE sample buffer. Radiolabeled proteins were analyzed by SDS-PAGE and phosphorimaging using a Bio-Rad Personal Molecular Imager FX (Bio-Rad Laboratories Ltd). Thermolysin digestion of chloroplasts was carried out as previously described. Import reactions were incubated with either 10 µg/mL thermolysin (Sigma-Aldrich Ltd.) (for import reactions with OEP9 and SSU) or 100 µg/mL thermolysin (for import reactions with Toc33 and Toc34). After a 30 min incubation on ice, EDTA was added to a final concentration of 10 mM to inactivate the protease. Thermolysin-treated chloroplasts were then repurified through a 35% (w/v) Percoll cushion and washed. Pretreatment of chloroplast membranes with trypsin was carried out by incubating isolated chloroplasts with 80 µg/mL trypsin (Sigma-Aldrich Ltd.) at 25°C in the dark for 1 h as previously described. The protease was then inactivated by the addition of PMSF to a final concentration of 2 mM. Trypsin-pretreated chloroplasts were then repurified, used in targeting assays, and radiolabeled proteins analyzed by SDS-PAGE/phosphoimaging as described above. Liposome-binding assays were carried out as described previously. Radiolabeled proteins were incubated with one equivalent of liposomes (40 µg) for 1 h at 24°C. Thereafter, sucrose was added to a final concentration of 1.6 M. Samples were then transferred to centrifuge tubes and sucrose gradient buffers (0.8 M and 0.25 M sucrose steps) were sequentially layered on top of the sample. After centrifugation for 18 h at 100,000 *g*, gradients were fractionated from the top into five fractions of equal volume (with the solubilized pellet as the bottom fraction) and analyzed by SDS-PAGE, using a Tris-Tricine buffer system and phosphorimaging using a Storm 840 phosphorimager and allied software (GE Healthcare Life Sciences, Piscataway, NJ). All data shown from experiments with isolated chloroplasts or liposomes are representative of at least two independent experiments. ## Bioinformatics Analyses Putative intrinsically disordered segments in OEP9 were identified using SMART (<http://smart.embl-heidelberg.de/>) and I-TASSER (<http://zhang.bioinformatics.ku.edu/I-TASSER>) protein-structure prediction programs are indicated with stippled lines. Predicted OEP9 homologues were identified by performing a WU-BLASTn (2.0) search of the Institute for Genomic Research (TIGR) plant transcript (EST) assemblies database (<http://blast.jcvi.org/euk-blast/plantta_blast.cgi>). Deduced amino acid sequences were then obtained from TIGR and/or GenBank (<http://www.ncbi.nlm.nih.gov/>) and aligned using the ClustalW algorithm (<http://npsa-pbil.ibcp.fr/>). The maximum likelihood phylogenetic tree represents results from neighbor-joining analysis of amino acid sequences obtained using the ClustalW2 program (<http://www.ebi.ac.uk/Tools/clustalw2/index.html>). Sequences used for analysis were obtained from GenBank, the *Arabidopsis* Information Resource (TAIR) ([www.arabidopsis.org](http://www.arabidopsis.org)) and TIGR. # Supporting Information We thank J. Miernyk, G. Moorhead and K. Dhugga for providing antibodies, J. Froehlich for *ppi1* and *ppi3* seeds, and P. Jarvis for pPZP221-Toc33/Toc34. We also thank D. Goring and R. Fieldhouse for assistance with electronic expression profiling and protein-structure prediction programs, respectively, and Y.T. Hwang, A. Howard and other members of our group for their assistance, advice and encouragement on various aspects of this research. Special thanks are extended to B. Abell for providing experimental results prior to publication, T. Mullen for assistance with preparation of the manuscript, and J. Dyer for insightful discussions. K. Gibson maintained the plant cell cultures, E. Bothwell assisted in the construction of pRTL2/Tic40-RFP, and G. Smith constructed pRTL2/NLS-RFP and pRTL2/NLS-RFPΔTAA. [^1]: Conceived and designed the experiments: PKD MDS DWA RTM. Performed the experiments: PKD LGR SKG MPAH. Analyzed the data: PKD LGR MDS DWA RTM. Contributed reagents/materials/analysis tools: MDS DWA RTM. Wrote the paper: PKD RTM. [^2]: The authors have declared that no competing interests exist.

# Introduction The high morbidity and mortality of ischemic strokes is a global healthcare concern. Intracranial atherosclerotic stenosis leads to a sharp decline in blood flow to brain tissue and is one of the important causes of ischemic strokes. The intracranial stenosis incidence is quite different due to the race, ethnic and region, and it is of importance in certain parts of the world such as some countries and regions including China. Past research implicates middle cerebral artery (MCA) stenosis as the most common site of symptomatic intracranial atherosclerosis. The degree of narrowing at the site of stenosis is associated with new-onset ipsilateral vascular events. A hemodynamic mechanism is one of the main factors contributing to recurrent ischemic strokes in these patients. In recent years, intravascular stent placement for treatment of carotid stenosis has been gradually improving. New techniques have emerged which permit intracranial vascular stenting. In patients who are not responding to antithrombotic therapy or have recurrent symptomatic MCA stenoses, intravascular stenting has been the preferred preventative treatment and can significantly improve low levels of local perfusion due to MCA stenosis, thereby preventing ischemic strokes and significantly reducing the incidence of strokes. However, intracranial stenting is feasible but that it is not substantiated by any randomized clinical trial, and, methods for intracranial detection of stent placement are still limited. While DSA is the gold standard for the diagnosis of intracranial arterial stenosis, it is nonetheless expensive, invasive, and cannot present hemodynamic changes. As a result, repeated DSA tests within a short time span are not preferred and therefore cannot be used for follow-up testing after stent placement. Transcranial color-coded sonography (TCCS) provides real-time, dynamic morphological information through color or power Doppler flow imaging. Doppler wave spectrum and the measurement of blood flow parameters enable evaluation of hemodynamic changes and cerebral vasospasms. In recent two decades, it is proved and accepted as an effective tool in intracranial vascular disease. However, there is seldom study of TCCS in intracranial stenting because the stenting procedure is not substantiated by any randomized clinical trial. Morreira et al. analyzed DSA/MRA and ultrasound imaging data of patients with neck and intracranial stents. Based on the results of the 2-year follow-up, a reliable determination of stent restenosis can be made based on blood flow velocity measurements established through ultrasound. With the development of ultrasound imaging, some novel techniques such as like vascular enhancement technology (VET) and three-dimensional (3D) imaging are approved to be valuable in clinical diagnosis. 3D imaging can achieve more space information of hollow viscera organs than traditional two-dimensional imaging. The VET technique can provide clear images of the vessel lumen by “reverse enhancement” of micro-vascular display capabilities. The report of 3D and VET mode imaging on intracranial stenting is rare. We designed this study to evaluate the clinical application of ultrasound in the management of stent placement for MCA stenosis. We used traditional TCCS as a follow-up method to assess hemodynamic changes at various time points before and after the MCA stent placement, and, applied VET and 3D imaging to part of the patients to test whether could we get additional information. # Materials and Methods ## Research subjects Between July 2008 and July 2012, we enrolled 43 patients with cerebral artery stenosis at our institution consecutively. Thirty-two patients were male and 11 were female ranging in age between 32–69 years (mean age of 54.5±10.9 years). Patients with the clinical symptoms of cerebral artery stenosis included the following: sudden cerebral ischemia or limb weakness, paralysis, sensory disturbances, difficulties in producing language or hemianopia. By TCCS examination, only patients with a sufficient bone window and with acceleration of the MCA blood flow velocity were included into the study. In patients included in the study, ultrasound detected abnormally high rates of blood flow velocity at the point of the stenosis. The location of the stenosis was at the M1 segment of the MCA. All patients were examined and confirmed through digital subtraction angiography (DSA). We used the Gateway-Wingspan stent (Boston, United States), which is 2.5–3.5 mm in diameter and 15–20 mm in length. ### Ethical approval of the study protocol All participants (or their legal guardians) gave written informed consent according to the Declaration of Helsinki. The study was approved by the Human Subjects Review Committee of the Fourth Military Medicine University (Xi'an, China). ## Instruments and methods The Siemens Antares and Acuson Sequoia512 color Doppler ultrasound system were used. The probe model was 4P1 or 4Vc1 with a frequency of 1.8– 2.0 MHz. The MCA probe depth was 120– 160 mm, and transcranial Doppler imaging settings were used. For preoperative examinations, subjects were required to be sober and calm but were not required to undergo special preparations. During temporal window examinations, patients were positioned in the lateral position and the probe was placed above the zygomatic arch between the lateral orbital margin and the earflap. Then, the probe was moved to the front, middle or rear windows depending on the image display. In 2D ultrasound, the clear “heart-shaped” hypoechoic structure originating from the midbrain was used as a positioning marker. Color Doppler flow imaging (CDFI) or power Doppler functions of the instrument can be used to display the major blood vessels of the circle of Willis at the base of the brain. Individual blood vessels can be located using blood flow direction and vascular anatomy information obtained through CDFI. Adjustments were made to the color scale and gain to yield clear images. Red indicates blood flow toward the probe and blue indicates blood flow away from the probe. Through the temporal window, it is possible to observe the ipsilateral MCA, which is shown in red. Further assessments can be made on the even distribution of the blood flow stream and the presence of stenotic lesions. Spectral Doppler was then enabled and line sampling and color flow beam were obtained at an angle \<60°. Samples were taken at vessel locations once every 0.3–0.5 cm. The spectral shape of blood flow as well as changes of the audio signal at each sampled location was monitored. The following blood flow parameters were measured: systolic peak flow rate (Vmax), end-diastolic flow rate (Vmin), time-averaged maximum flow rate (Vmean), and resistance index (RI). 2D ultrasound was used to determine stent location and morphology before, during (during recovery from anesthesia), and after stent placement. The CDFI function was used to display major arteries and blood flow across the stent. Spectral Doppler measurements were made to reveal hemodynamic parameters. Vascular enhancement technology (VET) and 3D-TCCS functions were used for image post- processing and to determine stent location, shape, and diameter. All the image taking and spectral Doppler measurements were performed directly online by one experienced sonographer (Yu Wang). The VET image analysis and 3D-TCCS image postprocessing were finished offline by another sonographer (Jian Mei Chen). Both of the interpreters were blinded to the DSA results. The time point was chosen at pre-stent to access the degree of stenosis, post- stent immediate (in the observation room waiting period right after operation) and post-stent one-week to check for patency of the stent, six-month and two years follow-up to detect restenosis. All the imaging procedures were identical to those described above. ## Statistical Analysis Each set of data was continuously measured at least three times. The highest flow rates at the site of stenosis were used for data analysis. The SPSS11.5 statistical package was used for data analysis. Mean±standard were used to report measurements. Variance was analyzed based on the preoperative, perioperative, and postoperative values. *P*\<0.05 was qualified as statistical significance. # Results ## Transcranial color Doppler imaging 2D imaging did not display clear images of the stenosis for all 43 patients pre- angioplasty and stenting. Color Doppler flow imaging (CDFI) revealed bright and mosaic-like color changes in blood flow of the stenotic lesions before the procedure. Spectral Doppler showed a broadened spectrum, suggesting a high impedance spectrum with a significantly increased velocity of blood flow. Audio signals at points of narrowing were low and blunt and showed partial wheeze-like vascular murmurs mixed with high-profile vascular murmur. All the diagnosis was confirmed by DSA before or during the procedure, which demonstrate local diagnosis of M1 segment of MCA. Ultrasound was performed in all 43 patients immediately after stent placement (during recovery from anesthesia). Hyperechoic images of the stents, on which stent length can be measured, can be seen in 41 patients out of 43 after the procedure. The stent echo cannot be clearly determined in the remaining 2 patients. CDFI demonstrated the disappearance of the original multicolored mosaic-like blood flow patterns while revealing a bright red bundle of blood flow through the hyperechoic stent. When comparing preoperative levels, the Vmax, Vmin, and Vmean at perioperative and postoperative examinations were significantly lower (*P*\<0.05) in 42 cases. However, perioperative blood flow velocity was a little bit increased in one patient (preoperative flow rate is 223 cm/s and perioperative flow rate is 231 cm/s). Upon CDFI reexamination one week after the procedure, normal blood flow velocity was restored and blood flow was significantly decreased (*P*\<0.05) compared to preoperative and perioperative values. In certain patients, no significant changes in blood flow velocity were found at six-month and two-year follow-up when compared with values at one week after stent placement (*P*\>0.05). Doppler measurements before and after stent placement for all patients are shown in. ## Ultrasound vascular enhancement technology (VET) VET imaging was applied in 32 cases. During preoperative exploration, major intracranial vascular lumen could be clearly visualized. Visible lumen narrowing were seen at the lesion site in stenotic blood vessels, and presented as “corset-like” lesions. In parts of the stenosis, post-narrowing expansions can be seen. Postoperative stent exploration could clearly reveal the lumen and the intraluminal stent position, length, shape, and diameter. ## 3-D Transcranial color-coded sonography (3D-TCCS) We performed 3D-TCCS in 5 patients. Through 3D reconstruction, stent location and vessel wall can be visually displayed. The net-like structure of the stent can be viewed from different angles through reconstruction of 0.2 mm cross- sections. During follow-up examination, 1 patient had increased local blood flow velocity at 178 cm/s. TCCS and VET imaging showed no more detail image on this case. 3D-TCCS imaging revealed localized stenosis characterized by hyperechoic signals at the intimal surface, indicating intimal hyperplasia. The patient was subsequently diagnosed with mild stent restenosis (\<50%) with DSA. # Discussion In recent years, intracranial vascular stenting is reported as improved therapy for restoring intracranial blood supply and widely recognized. But similar to other endovascular procedures, an effective and convenient method for assessing treatment efficacy and follow-up has not been developed. Insufficient research is available for detection of hemodynamic changes after stent placement; therefore, our group has previously investigated trancranial Doppler ultrasound for diagnosis of intracranial artery stenosis. We have found that, TCCS leads to higher sensitivity, specificity, and accuracy in diagnosing intracranial artery stenosis. Therefore, this technique may be used for the diagnosis and preoperative assessment of cerebrovascular stenosis. In the current study, we enrolled 43 symptomatic patients with MCA stenosis. Through continuous ultrasound imaging, we assessed its clinical value for intracranial artery stenting. Using traditional ultrasound imaging for the hemodynamics assessment of the stenosis, the degree of stenosis was determined mainly based on the velocity of blood flow. Prior to endovascular stenting, DSA was used for diagnostic confirmation in all patients. The assessment of postoperative management is dependent on the parameters recorded through Doppler blood flow velocity measurements. Hemodynamic changes were evaluated perioperatively, for one week after the procedure, and at six-month and two-year follow-up. Based on our results, intravascular stenting led to immediate relief in stenosis accompanied by a dramatic decrease in flow. After one week, due to the reshaping of the stent and vascular remodeling, further improvements in hemodynamics were observed and flow rate continued to decline toward normal levels. At six-month and two-year follow-up, some patients exhibited no significant increases in blood flow velocity, indicating the superior efficacy of stenting. No occurrences of in-stent restenosis were found. In a large population study evaluating MCA implantations, the incidence of in-stent restenosis determined through DSA was 5%, indicating that these procedures are safe and efficacious. The follow-up population of this study is relatively small. At six-month and two-year follow-up we did not find any cases of increased blood flow velocity or in-stent restenosis in 20 patients. Due to the metal composition of the stent, TCCS can clearly display the stent, thereby determining the location and shape of the stent. Color Doppler can be used to assess postoperative blood flow velocity and has indicated that stenting can achieve adequate patency. 2D imaging has shown that TCCS could display in- stent stenosis in 41 patients while lesions in 2 patients were unclear. Factors that may affect this include age, gender, and skull attenuation. However, this study did not show that these patients were female and older in age. Additionally, the location of the stent may also be a factor, where the acoustic window may place further limitations. In this study, the instantaneous blood flow velocity in one female patient increased after the stenting procedure, which may be affected by her degree of anesthesia and the subsequent recovery. During physical examination, the patient was found to be irritable and hypertensive. The patient's blood pressure was correlated with the cerebrovascular hemodynamics. Hypertension can lead to changes in vascular function and a transient increase in blood pressure may indirectly lead to an elevation in blood flow velocity. Additionally, hyperperfusion syndrome may also be a potential cause, where hemodynamic measurements may indicate increased blood flow velocity in the stented cerebral hemisphere and acute-onset vasospasms may lead to transient abnormal hemodynamics. At one-week follow-up with TCCS, this patient's blood flow velocity had significantly decreased when compared with perioperative values. At six-month follow-up, the maximum blood flow velocity was slightly higher (but within normal limits) than at one-week follow-up with indicating superior efficacy. The VET technique builds a silhouette image based on blood vessels carrying blood. This technique can provide “reverse enhancement” of microvascular display capabilities and vascular sensitivity, leading to clear images of the vessel lumen and wall structures and significantly reduced volume artifacts. In contrast to 2D imaging, VET can be used to visualize small arteries and intracranial vascular lumina. In the detection of intracranial vascular lesions, VET has advantages over color Doppler imaging on displaying the lumen of vessels because of less frequency reduction. In the study, when compared with (they come from the same patient), the clarity of TCCS stent lumen images was enhanced with the VET technique, enabling a more objective evaluation of the stent morphology. Compared with conventional 2D imaging, 3D imaging can provide further detailed information (e.g. multi-angle view of the lesion) such as the threading of stents. For patients with in-stent restenosis, conventional TCCD can detect blood flow acceleration whereas 3D-TCCS can display details of intimal hyperplasia at the stent surface. However, only a few cases in our study utilized 3D imaging, thereby limiting further analysis of its efficacy. Nonetheless, TCCS can be used for the preoperative diagnosis and postoperative follow-up in the treatment of intracranial artery stenosis. This study is a single-center research. The sample case is limited to accomplish more correlation study to clinical outcomes and comparison study to postoperative DSA or other non-invasive techniques such as CTA and MRA. This is the main limitations of the study. With the accumulation of cases, assessment on reproducibility of the methods would be performed in the future work. In conclusion, TCCD can be considered a quick and effective clinical detection method to evaluate the stenting treatment for MCA stenosis. Based on the change in hemodynamics, an objective assessment of the clinical efficacy can be performed. New imaging technologies 3D and VET may achieve additional image informations with increased clarity and reliability. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: YYD XL. Performed the experiments: YW JMC. Analyzed the data: JW XL. Contributed reagents/materials/analysis tools: LHL JPD. Wrote the paper: XL YYD.

# Introduction Sexual conflict in mating systems, due to differences in investment and direct mating costs, can lead to intersexual arms races, as well as being a potential engine of speciation. Rapid evolution of male adaptations and female counter- adaptations, explicitly predicts that changes in male and female characters should be correlated as coincidental transformations on internal branches of a phylogeny. However, despite claims of the widespread occurence of sexual conflict fuelling arms races, credible empirical examples from a phylogenetic perspective, are largely lacking and the major question today is whether sexual conflict really generates sexually antagonistic evolution. *Drosophila* provide a model system on a microevolutionary scale, and in particular, artificial selection experiments have convincingly demonstrated sexual antagonistic coevolution, involving accessory gland substances \[some 80 different Acps\] in the male ejaculate that in various ways affect the female. The exponentially increasing literature on sexual conflict involving various organisms includes population crosses, mating experiments, meta-analyses of groups with different mating systems, comparisons of evolutionary rates in sexual versus asexual characters, theoretical models, experimental manipulations of sexual traits and studies of the economics and costs of matings and all contribute to the growing body of evidence. Indeed a full-scale Kuhnian paradigm shift is starting to be acknowledged where traditional sexual selection in terms of male-male competition and female choice, rather than being the golden standard for interpreting mating interactions and outcomes, merely becomes “*an interesting emergent property of a more ancient and far-reaching evolutionary dynamic: sexual conflict*”. But the paradigm shift is not friction free and controversies remain, in particular centred around what the alternatives predict and thus what data is really needed to corroborate sexual conflict as oppose to traditional sexual selection – something perhaps not as straight forward as previously believed. However the claim that the substantial direct costs of multiple matings to females can be compensated for by their offspring acquiring indirect genetic benefits or “good genes” has been questioned by recent studies. On a macroevolutionary level examples are scarce, but water striders (Gerridae) and hermaphroditic snails provide the only putative example of correlated characters involved in a sexual arms race. Since macroevolutionary tests of coevolutionary scenarios require a phylogenetic context, incomplete taxon sampling and uncertainty in phylogenetic reconstruction often undermine the credibility of empirical examples. Recently developed Bayesian applications to phylogenetics offer new possibilities to test evolutionary scenarios while accounting for the uncertainty in phylogenetic reconstruction. In addition, thorough taxonomic sampling in phylogenetic studies have proved to be of fundamental importance for accurate reconstruction as well as answering various questions on evolution and speciation. Here we provide an example of an antagonistic intersexual arms race, where confounding effects of incomplete taxon sampling is minimized and phylogenetic uncertainty statistically accounted for. Diving beetles (family Dytiscidae) are predatory aquatic insects. Sexual conflict in diving beetles has been suggested to explain various dorsal modifications in females. The mating system is typical for sexual conflict scenarios that involve substantial direct costs ; with no courtship, males simply attack bypassing females, females resist behaviourally by attempting to dislodge the male in rapid and erratic swimming moves \[27, \], and mating interactions are characterized by a very long postcopulatory guarding phase. To grab the female, males have their front feet modified into a palette equipped with suction cups that are attached to the female dorsal surface at initial contact. A possible unique mating cost to females relates to the fact that these secondarily aquatic insects are dependent on atmospheric oxygen, which is carried under the elytra and replenished with frequent trips to the surface (normally once every 8–15 minutes). Only the male however, has access to air during the mating. Indeed, precopulatory violent shaking by the male seems to increase the female's need for oxygen which is followed by copulation when the female is exhausted and non-resisting. During the approximately six hours of postcopulatory guarding in the diving beetle *Dytiscus alaskanus*, the male periodically tilts the female upward allowing her to replenish the air supply. The evolution of this behaviour further emphasizes the importance of this cost to females, although it has yet to be experimentally quantified. An alternative explanation for the resistance behaviour of females is the classical female choice, where, although the behaviour may be costly, this is compensated for through good genes passed on to her offspring since only fit males able to withstand her resistance will fertilise her eggs. Until unambiguously tested with breeding experiments and fitness measures in offspring such a scenario cannot be excluded, but more and more evidence suggest that this is unlikely to compensate for the substantial direct costs. To trace the evolutionary history of sex-specific characters in diving beetles, we used the recently revised genus *Acilius* which contains 13 extant species distributed over the Northern Hemisphere. The genus is characterized by having dense macropunctures on the dorsal surfaces, females with prominent, setose furrows on the elytra, and males with broadly expanded protarsi equipped ventrally with three large and many minute suction cups. These structures even attracted the attention of Charles Darwin, who regarded the setose female furrows in *Acilius* as an example of an aid for males to better grip females during mating. However, it is clear from basic physical laws and simple experiments, that the mechanically working male suction cups function best on smooth surfaces where complete contact around their circumference enables attachment. # Results and Discussion We analyzed one mitochondrial gene (Cytochrome oxidase I), two nuclear genes (Histone III and Wingless) and morphological characters to infer the evolutionary relationship of the 13 *Acilius* species. Bayesian and parsimony analysis on the concatenated data converged on the same, fully resolved phylogeny shown in, with very high clade support values (posterior probabilities: 0.94–1.0). Although missing DNA data from the possibly extinct Chinese species *A. sinensis*, the morphological character matrix strongly inferred the placement of this species in the phylogeny. Palearctic and Nearctic species groups are clearly supported as monophyletic with the exception of the Mediterranean species *A. duvergeri*, which is sister to all other *Acilius*. Bayesian posterior probability of this topology is 0.85 with the next best topology 0.05. Moreover, only three additional topologies from the Bayesian analysis together make up a 95% credible set of trees (cumulative posterior probability\>0.95). This is interpretable as the true tree being among these four topologies with a probability of 0.95, assuming the model is correct. Accomodating the phylogenetic uncertainty, ancestral character reconstructions were carried out on all four topologies and inferred coincidental character transformations discussed below are unaffected by their topological variation. Dorsal macropunctures and female setose furrows show transformations at the ancestor of *Acilius*, at the ancestor of all taxa except *A. duvergeri*, at the ancestor of the sister taxa *A. mediatus* and *A. fraternus* and at the terminal branch, *A. kishii*. Reconstruction of size and numbers of male suction cups show major coincidental character transformations on the same ancestral nodes except the branch subtending *A. kishii*. Thus the male of the ancestral lineage that developed the dorsal macropunctures, radically evolved from having a set of few medium sized suction cups to having three large and multiple minute suction cups. On the next internode the evolution of female setose elytral furrows coincides with a second major male transformation where the suction cups further differentiated into one basal very large suction cup, two medium sized and five times the number of minute cups. Finally, the loss in females of setose furrows in the ancestor to *A. fraternus*/*A. mediatus* coincides with a reversal in male characteristics resembling the initial condition before females changed ; differentiation among the three larger cups is nearly completely reversed, whereas the number of minute cups decreases only moderately in the ancestor, but further in *A. mediatus*. The pattern mimics results from a comparative population-level study where male suction cups increase in differentiation as a response to increasing frequency of a granular female morph of the diving beetle *Graphoderus zonatus*. However, coincidental character transformations at the same ancestral nodes cannot indicate which character evolved first and which was the response. An independent line of evidence regarding the order of coevolution exists in the recently diverged species *A. kishii.* This species is confined to a single mountain lake and population in Fukui prefecture, Japan. Females of this species have secondarily lost the setose elytral furrows, although slight impressions of furrows are still apparent. Males of *A. kishii* are very similar to the sibling species *A. japonicus* but ratio comparisons between the three largest cups show that S1 is relatively larger compared to S2 and S3 in *A. kishii* (MANOVA, P\<0.01, n = 7, Tukey's HSD test). However, the basic pattern of one very large cup, two medium sized and a large number of minute cups, characterising all other species with setose females, also characterises males of *A. kishii*. This provides the only direct evidence of the order of evolution in the *Acilius*-arms race; suction cup differentiation in males is the response to previous changes in female dorsal surfaces. The advantage of differentiating the cups into smaller and larger ones as the surface evolves sculpture can tentatively be explained by the equations defining the suction force and attachment time. Both are influenced most by the radius of the cup and the radius of the leaking channels; small cups can increase the suction force and attachment time by reducing the leaking channels, while larger suction cups achieve the same effect by increasing the volume and area of the cups. In addition, the *A. japonicus*/*kishii* species pair may be an example of speciation at an early stage driven by sexual conflict. At least this singular example fulfils the prediction that characters involved in the arms race change faster than others (e.g. male genitalia are identical while being diagnostically different between all other *Acilius* species). DNA similarity of CO1 between *kishii* and *japonicus* (99.5–99.6%), lies in the range of the within species variation for *A. japonicus* (99.5–99.8% for three sequenced individuals). Branch length estimates from the Bayesian analysis also illustrate the recent history of the split. At present the ranges of these two taxa are completely allopatric, but this disjunction certainly dates from the withdrawal northwards of the last glacial advance during which northern biota expanded further south in Japan. Thus, it is expected that the isolation dates from the Holocene and speciation is likely to have occurred in a few thousand generations, whereas theory on sexual conflict suggests speciation can occur in even less time. Quantifying the cost to females of multiple matings is a challenge for future studies and we tentatively predict that in the natural history of diving beetles, aquatic insects dependent on atmospheric oxygen, lies the key to the extraordinary female counter-adaptations in *Acilius* and other diving beetles. While the actual cost for females of multiple matings is unknown, observations on *Acilius* from the field reveal a very high male harassment rate. Likewise, while the biomechanical effect of a structured surface for suction cups can be predicted from physical laws or simple measurements on dead animals, only experimental mating trials will reveal the net effect when compensatory behaviour can interact with the equation. # Materials and Methods ## Taxon selection and molecular methods Outgroup exemplars include representatives from the genus *Graphoderus*, the sister group to *Acilius*. *Acilius* exemplars include all 13 extant species for morphology and all except *A. sinensis* for sequence data. Genomic DNA was extracted from legs and thoracic muscle tissue from ethanol-preserved material collected between 2000 and 2003 using standard protocols. DNAs and controls were amplified using PCR and a DNA Engine DYAD Peltier Thermal Cycler using the primers *Haf* and *Har* (for *histone III*), *LepWg1* and *LepWg2a* (for *wingless*), and *C1-J-2183* and *TL2-N-3014* (for *cytochrome oxidase I*). Product yield, specificity of amplification and contamination were investigated using agarose gel electorphoresis. PCR products were purified and cycle sequenced using ABI Prism Big Dye (version 3). Sequencing reactions were column purified and fractionated with an ABI 3730xl DNA analyser (DNA Sequencing Center, BYU). Fragments were sequenced from complementary strands and these were examined and edited using the program Sequencher. Genebank accession numbers are given in. A morphological matrix of 46 discrete characters was compiled for all taxa. ## Alignment and phylogenetic analyses The three genes used were not length variable and the unambiguous alignment was based on conservation of amino acid reading frame. The concatenated alignment consisted of 1647 base pairs. The genes and the morphological data matrix were combined and analysed simultaneously in a Bayesian framework using MrBayes ver. 3.0b4 with partition specific model settings; a separate HKY85+Γ+I model was assigned to each of four partitions of the molecular data defined as; 12posCO1, 3posCO1, 12posH3/Wingless, 3posH3/Wingless. The partitions and assigned model is by necessity a balance between reality and avoiding assigning more parameters than limited data can possibly estimate. For the morphological matrix a Markov k model+Γ was used, accounting for the fact that only parsimony informative characters were scored. It was recently shown that model-based phylogenetic inference forcing differently evolved data to a uniform model is hazardous. In particular, estimating the branch length as an average from the total data may yield erroneous likelihood calculations for each partition. Consequently branch lengths were unlinked and independently estimated for all five partitions, avoiding the potential pitfall. Four independent Markov Chain Monte Carlo runs with 10 million generations each were sampled every 1000 generation. The first 2 million generations were discarded in each run as burn-in and the last 8000 sampled trees were pooled from the four runs and summarized, identifying the topology with highest posterior probability. Model parameter estimations are given in. Proper mixing of chains, acceptance probabilities and convergence of likelihood values, tree topologies, branch lengths and model substitution parameters were checked in the four runs before the samples were pooled. Clade support values and posterior probability of topologies were calculated as the frequency of each clade/topology among the sampled trees. We also ran a parsimony analysis on the total evidence data with all characters equally weighted and unordered. Implicit enumeration in TNT ensured finding the shortest tree with 15 taxa in reasonable time. Of the 1693 characters, 290 were parsimony informative (CO1: 152, H3: 39, Wingless: 53, Morphology: 46). ## Character reconstruction Suction cup diameter S1–S4 was measured for five individuals of each species where fresh material was available. For S4 an average of 10 cups was calculated. Measurements and counting of S4 cups were carried out by digital photo (Kodak Megaplus camera Model 1.6i) through an Olympus microscope at 64× and imported to the image analysis software Optimas 6.5 (Media Cybernetics, 1987–1999). To accommodate uncertainty in the phylogeny reconstruction, character optimizations were done on all tree topologies that together cumulated a posterior probability of \>95%, ordered in decreasing posterior probability from the Bayesian analysis. Discrete morphological characters were mapped using Fitch optimization, while continuous characters were optimized as additive using Farris optimization as implemented in TNT. # Supporting Information We thank T. Bergsten for help in exploring the physical properties of suction cups and M.F. Whiting, A.N. Nilsson, S.L. Cameron, L. Rowe, G. Arnqvist, F. Johansson and P. Ingvarsson for comments on an earlier draft of this manuscript. [^1]: Conceived and designed the experiments: JB KM. Performed the experiments: JB KM. Analyzed the data: JB. Contributed reagents/materials/analysis tools: KM. Wrote the paper: JB KM. [^2]: The authors have declared that no competing interests exist.

# Introduction Enterococci are part of the normal human microbial flora. Historically, the majority of invasive enterococcal infections were caused by *Enterococcus faecalis*, followed by *Enterococcus faecium*. In recent decades, however, the epidemiology of invasive enterococcal infections appears to be changing worldwide, and a number of trends have been recognized, notably, the global emergence of enterococci as important nosocomial pathogens and the emergence of resistance to commonly used antimicrobial agents, including penicillins, aminoglycosides and glycopeptides. An increase in the number of *E. faecium* strains in hospitals in different countries has been documented during the last decade. These isolates had in common not only the antibiotic resistance traits (to ampicillin, quinolones and to glycopeptides in some cases) but also several virulence factors that might have contributed to the success of *E. faecium* as a leading nosocomial pathogen. Although these strains were initially classified within a single clonal complex 17, it appears that the genetic diversity of this CC allows the classification of all isolates in three main lineages (17, 18, and 78), which is a more accurate representation of the recent evolution of these isolates. Management of severe infections due to resistant enterococcal strains, especially *E. faecium*, has therefore become a therapeutical challenge. However, most of the reported experiences regarding enterococcal infections concern the general, non-immunocompromised population, and they mainly involve vancomycin-resistant strains. Additionally, the majority of the studies published to date have been carried out in the United States, where the epidemiological situation is very different from that occurring in Europe. Furthermore, information regarding bloodstream infection (BSI) due to *E. faecium* in immunosuppressed patients with cancer is particularly scarce. Given the above, the aim of the present study was to describe the incidence and risk factors for vancomycin-susceptible *E. faecium* BSI in a large prospective cohort of cancer patients. We also aimed to ascertain the clinical features, antimicrobial susceptibility, genotypes and outcome of BSI due to *E. faecium* in this population. # Materials and Methods ## Setting, patients and study design We conducted a prospective observational study in a 200-bed cancer referral centre for adults in Barcelona, Spain. From 1 January 2006 to 30 September 2012 all hospitalized cancer patients and haematopoietic stem cell transplant recipients with at least one episode of BSI were included in the study. Information on baseline characteristics, clinical features, empirical antibiotic therapy and outcome was carefully recorded in a specific database. All episodes of BSI due to vancomycin-susceptible *E. faecium* were compared with those caused by vancomycin-susceptible *Enterococcus faecalis* in order to identify the risk factors for ampicillin resistance acquisition and to assess differences in clinical features and outcome. We also compared patients who died with those who survived in order to identify risk factors for mortality. All BSI episodes at our hospital are reported and followed up by an infectious disease physician. Changes in antimicrobial treatment and general management were advised when necessary. ## Ethics statement This observational study was approved by the Institutional Review Board Comité Ético de Investigación Clínica del Hospital Universitari de Bellvitge (Ethics Committee of Clinical Research-Hospital Universitari de Bellvitge), with the following reference number PR 232/10. To protect personal privacy, identifying information of each patient in the electronic database was encrypted. Informed consent was waived by the Clinical Research Ethics Committee because no intervention was involved and no patient identifying information was included. ## Definitions Neutropenia was defined as an absolute neutrophil count \<500/mm³. Current corticosteroid therapy was recorded when a patient was receiving corticosteroids at the time of the BSI episode or in the previous month. Prior antibiotic therapy was defined as the receipt of any systemic antibiotic for \>48 hours during the previous month. BSI was considered to be from an endogenous source in those patients with neutropenia in whom no other BSI sites were identified. In those patients without neutropenia, an unknown source was considered if an evident origin of the infection was not identified. Shock was defined as a systolic pressure \<90 mmHg that was unresponsive to fluid treatment or which required vasoactive drug therapy. Empirical antibiotic therapy was considered inadequate if the treatment regimen did not include at least one antibiotic active *in vitro* against the infecting microorganism. Early case-fatality rate was defined as death within 48 hours of the BSI episode. Overall case-fatality rate was defined as death by any cause within the first 30 days of the onset of BSI. ## Microbiological studies Blood cultures were performed by standard methods. Two sets of two blood samples were drawn from patients with suspected bloodstream infection. Blood samples were processed by the BACTEC 9240 system (Becton Dickinson Microbiology Systems) with an incubation period of five days. Positive blood samples were sub-cultured onto chocolate agar. Identification and antibiotic susceptibility were performed using commercially available plates (MicroScan, Siemens), following the manufacturer’s instructions. The antimicrobial susceptibility of isolates was interpreted according to current Clinical Laboratory Standard Institute criteria. Thirty *E. faecium* strains isolated between 2006 and 2012 from single bacteraemic patients were available for genotyping. Pulsed field gel electrophoresis (PFGE) was performed in all strains after *Sma*I restriction of chromosomal DNA, as previously described. PFGE patterns were interpreted both by visual inspection, using the criteria of van Belkum et al. and by analysis with the FINGERPRINTING TM II software, version 3.0 (BioRad Laboratories, Inc., Madrid, Spain). Dendrograms were constructed using Dice coefficients, with optimization and band position tolerance being set to 0.5% and 1% respectively. A similarity coefficient of 80% was selected to define the patterns. Multilocus sequence typing (MLST) was conducted on 17 representative strains of each *Sma*I-PFGE type, as described by Homan et al.. Sequence types (STs) were assigned according to the *E. faecium* MLST database ([<u>http://efaecium.mlst.net</u>](http://efaecium.mlst.net)). ## Statistical analysis Continuous variables were compared by means of the Mann-Whitney U test and *t*-test. Qualitative variables were compared using the chi-square test, and odds ratios and 95% confidence intervals were calculated. Multivariate conditional logistic regression analysis of factors potentially associated with *E. faecium* acquisition and mortality included all statistically significant variables in the univariate analysis, sex and age, and all clinically important variables regardless of whether they were statistically significant or not. This analysis was performed with the stepwise logistic regression model of the SPSS software package (SPSS v. 17). # Results During the study period 1287 consecutive episodes of BSI were recorded. Of the 550 (42.5%) episodes caused by Gram-positive bacteria, 105 were due to enterococci (19%). Thirteen episodes of enterococcal BSI were not included in the study because they were caused by species other than *E. faecium* or *E. faecalis* (*E. gallinarum* 6, *E. casseliflavus* 2, *E. avium* 2, *E. hirae* 1, *E. durans* and *E. raffinosus* 1). Thus, 54 episodes of BSI caused by *E. faecium* and 38 by *E. faecalis* were finally included in the study. Four patients with two episodes of enterococcal BSI were included since they were considered to present different episodes, separated by at least four weeks. The incidence of *E. faecium* BSI increased significantly over time (22 episodes/126610 admissions from 2006 to 2009 vs 32 episodes/80586 admissions from 2010 to September 2012; *p*=0.002). By contrast, the incidence of *E. faecalis* BSI remained stable over time (p=0.215). shows the baseline and clinical characteristics of patients with enterococcal BSI compared by groups. Patients with BSI due to *E. faecium* were more likely to have received previous antibiotics (mainly carbapenems), previous antifungal prophylaxis and previous blood transfusion. Likewise, they had more prolonged neutropenia than did patients with *E. faecalis* BSI, and were more likely to have a concomitant infection. In addition, there was a trend in the former group of patients towards haematological malignancy as the most frequent underlying disease, and for there to be a venous catheter in place. The most frequent origin of BSI was an endogenous source in 38% of cases (41% in the *E. faecium* group vs 34% in the *E. faecalis* group), followed by catheter infection in 15% of cases (15% vs 16%, respectively) and cholangitis in 13% (15% vs 10.5%, respectively). BSI originating in the urinary tract tended to be more frequent in the group of *E. faecalis*, whereas neutropenic enterocolitis tended to be more frequent in the *E. faecium* group. After applying a logistic regression model the only variable found to be an independent risk factors for *E. faecium* acquisition was previous use of carbapenems (OR 10.24; 95% CI, 1.35-77.66). ## Microbiological studies All *E. faecium* isolates were vancomycin and teicoplanin susceptible. Only two strains (2.9%) of the 54 isolates were ampicillin susceptible with MICs ≤1 µg/mL. During the study period 97.1% of strains showed high-level resistance to ampicillin, and the rates of high-level resistance to streptomycin and gentamycin were 74.6% and 32.8%, respectively. Quinolone resistance accounted for 88.8% of the isolates, with MICs ≥4 µg/mL. Among 30 ampicillin-resistant *E. faecium* strains available for genotyping, 13 PFGE patterns were obtained corresponding to 7 STs. ST117 was dominant, accounting for 16 isolates (53.3%) belonging to PFGE type D (n=10; 62.5%) and PFGE type A (n=6; 37.5%). Five isolates (16.7%) belonged to ST17 of PFGE type B (n=4) and H (n=1), three isolates (10%) belonged to ST 78, two (6.7%), to ST18, two (6.7%) to ST203 and one to ST192. A single isolate belonged to a new ST844. Considering the new classification recently proposed by Willems and co-workers, 5 isolates belonged to lineage 17 (ST17), two isolates belonged to lineage 18 (ST18) and 23 isolates belonged to lineage 78 (ST78, ST117, ST192, ST203 and ST844). From 2006 to 2009, one isolate out of eight available for genotyping belonged to ST117 (year 2009). In contrast, from 2010 to 2012, 15 out of 22 isolates belonged to ST117, with PFGE patterns D (n=10) and A (n=5). The allelic profiles for all different STs are detailed in. Antibiotic treatment and patient outcomes are detailed in. The large majority of patients received empirical antibiotic therapy (91%). The most frequently antibiotic used was a glycopeptide (42% of cases), followed by third- and fourth-generation cephalosporins (29%), β-lactam plus β-lactam inhibitors (26%) and carbapenems (25%). More patients with *E. faecalis* BSI received a combination of a β-lactam plus β-lactam inhibitor empirically, whereas more patients in the *E. faecium* group were given a glycopeptide. Patients with *E. faecium* BSI were more likely to have received inadequate initial empirical antibiotic therapy than were patients with *E. faecalis* BSI, and time to adequate empirical antibiotic therapy was also longer in the former group. No significant differences were found between the two groups regarding other outcomes such as early and overall mortality rates. Risk factors associated with overall case-fatality in the cohort of patients with enterococcal BSI are shown in. The use of current corticosteroids, shock at presentation, intensive care unit (ICU) admission, mechanical ventilation and unknown source of BSI were the variables most frequently found in the group of patients who died. However, after applying a logistic regression model the only variables found to be independent risk factors for overall case-fatality were the current use of corticosteroids (OR 4.18; 95% CI, 1.34-13.01) and ICU admission (OR 9.97; 95% CI, 1.96-50.63). BSI due to *E. faecium* was not identified as a risk factor for overall mortality. # Discussion We observed a dramatic increase in the incidence of ampicillin-resistant, vancomycin-susceptible *E. faecium* BSI in patients with cancer over the study period. Historically, *E. faecalis* was the responsible for the large majority of all clinical enterococcal infections, with *E. faecium* being less frequently isolated. However, in the late 1990s the ratio of *E. faecalis* to *E. faecium* infections in the United States shifted in favour of *E. faecium*, while in Europe the first reports of increased numbers of infections due to *E. faecium* were published in the mid-1990s. Despite this, information regarding BSI caused by vancomycin-susceptible *E. faecium* in immunosuppressed cancer patients is scarce in the literature, and all the articles published to date are retrospective studies. Thus, our study is the first prospective analysis of a cohort of cancer patients with vancomycin-susceptible *E. faecium* BSI to be conducted at a time when enterococcal infections are gaining importance worldwide, both in terms of dissemination and antimicrobial resistance. It is also the first study to describe how vancomycin-susceptible *E. faecium* outnumbers *E. faecalis* in this high-risk population of patients with BSI in our geographical area. Although several studies have focused on risk factors for vancomycin-resistant enterococcal infection or colonization, little is known about risk factors for BSI due to vancomycin-susceptible *E. faecium*. In our study we identified previous use of carbapenems as the only independent risk factor associated with *E. faecium* BSI. Since the early research of Boyce et al., which related ampicillin-resistant enterococci to the use of imipenem, several authors have demonstrated the association of ampicillin-resistant enterococci with β-lactams. Antibiotics may facilitate colonization and infection by depleting the gastrointestinal tract of its normal anaerobic flora and by selecting enterococci due to limited bactericidal activity against these organisms. The use of broad spectrum β-lactams (including carbapenems) in patients with cancer and frequently-associated febrile neutropenia is very common in clinical practice. Furthermore, the emergence of multidrug-resistant organisms (especially extended-spectrum β-lactamase-producing *Enterobacteriaceae*) among cancer patients in our centre often forces us to use carbapenems as the treatment of choice. A judicious use of antibiotics is therefore needed in order to avoid the development and dissemination of bacterial resistance. Our study shows that all the enterococci isolates remained susceptible to glycopeptides. Although vancomycin resistance has become an emerging health problem worldwide, it is less important in Europe than in the United States, and as yet it does not seem to be a problem in our geographical area. However, the emergence of *E. faecium* is of potential concern as it is more commonly associated with vancomycin resistance than are the other enterococci. The large majority of *E. faecium* isolates showed a high level of resistance to ampicillin, and only two strains were ampicillin susceptible. Notably, the two patients carrying the two susceptible strains had not received carbapenems previously. PFGE and MLST analysis of 30 available *E. faecium* isolates showed that all isolates belonged to the three major recently described hospital-associated *E. faecium* lineages (17, 18 and 78). Of note, the ST117 (lineage 78) was especially frequent in the last three years, which may explain, at least partially, the emergence of *E. faecium* in our centre. This finding is in line with recent molecular epidemiological studies that have identified these three lineages (formerly CC17) as being responsible for the worldwide emergence of ampicillin-resistant *E. faecium*. These three lineages have adapted extremely well to the hospital environment, including the acquisition of ampicillin resistance and the *esp* virulence gene, which is associated with biofilm formation. Therefore, these lineages have become the leading cause of hospital- acquired *E. faecium* infections and outbreaks. The partial replacement of ampicillin-susceptible *E. faecalis* by hospital-associated lineages of ampicillin-resistant *E. faecium* is worrying, since it may set the stage for the emergence of vancomycin-resistant *E. faecium*. There is controversy in the literature regarding the association between *E. faecium* infection and mortality. Some authors have reported increased mortality in those patients with BSI due to ampicillin-resistant *E. faecium*. However, it is still unclear if the increase in mortality actually depends on infection or, rather, whether infection behaves as a marker of the severity of underlying diseases. In our study, and in line with some previous reports, we found no association between *E. faecium* BSI and increased mortality. Interestingly, some in vitro studies have suggested that enterococcal virulence determinants are more frequently found in *E. faecalis* isolates than in *E. faecium* isolates. On the other hand, some studies have reported that *E. faecium* is more often resistant to phagocytosis than is *E. faecalis*. Whether there is a clinically relevant difference in virulence between vancomycin-resistant enterococci and vancomycin-susceptible enterococci, or between different enterococcal species is unknown. The only variables found to be independent risk factors for mortality in our study were ICU admission and current corticosteroid therapy. ICU admission is associated with severe sepsis and shock, which are known to be risk factors for mortality in patients with BSI. Patients receiving corticosteroid therapy mainly corresponded to debilitated patients with severe uncontrolled underlying disease, who are known to be a risk group for poor outcome. Patients with *E. faecium* BSI in our study were more likely to receive inadequate empirical antibiotic therapy. Inadequate empirical antibiotic therapy has previously been reported to be associated with mortality, especially in patients with Gram-negative BSI and in those with vancomycin-resistant enterococcal infections. However, this association was not observed in our study. A retrospective study by DiazGranados et al. identified vancomycin- resistance as a risk factor for mortality in neutropenic cancer patients. However, it was not associated with inadequate empirical antibiotic therapy, but rather was attributed to prolonged duration of BSI. Factors influencing mortality among cancer patients are often difficult to asses. The emergence of *E. faecium* in immunosuppressed patients with cancer is a concern, since there are limited therapeutic options for these organisms. Although new antimicrobials, such as linezolid, daptomycin and quinupristin/dalfopristin, have recently been developed to treat serious enterococcal infections, resistance to these agents has already emerged. This study has a number of strengths. It describes the incidence of *E. faecium* BSI in a large cohort of BSI episodes prospectively collected in a specific immunosuppressed high-risk population. Additionally, it provides information regarding the *E. faecium* clonal complexes identified during the study period. However, it also has certain limitations. The small number of patients in the two groups may have prevented us from identifying significant differences between them. Also, as this was a single-centre study the results may have been influenced by local epidemiological variables, thereby limiting their applicability to other settings. In conclusion, we found a significant increase in vancomycin-susceptible *E. faecium* BSI among cancer patients, especially those treated previously with carbapenems. Clonal complex 17 was responsible for the large majority of *E. faecium* infections, particularly in recent years. The emergence of *E. faecium* among immunosuppressed cancer patients is worrying since there are limited treatment options and it may presage the emergence of vancomycin-resistant enterococci. Addressing this trend for enterococci requires a rational approach that combines infection control with antimicrobial stewardship. # Supporting Information [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: CG JC. Performed the experiments: CG JA MC MAD CA. Analyzed the data: CG CGV MO. Contributed reagents/materials/analysis tools: MB M. Arnan M. Antonio. Wrote the manuscript: CG MAD FA JC.

# Introduction Selection of favorable alleles through domestication and breeding has led to dramatic changes in seed and fruit attributes, plant habit, and productivity. For tomato (*Solanum lycopersicum* L), breeding has involved the competing forces of narrowed genetic variation due to best by best crosses followed by selection, and the expansion of genetic variation due to the introgression of genes for biotic stress resistance from wild species. The long history of crossing to wild relatives has broadened the genetic diversity in contemporary germplasm relative to vintage and landrace germplasm. In addition, breeding for distinct market classes and production systems has led to genetic differentiation in contemporary germplasm. For example, processing and field- grown fresh market tomatoes are now distinct sub-populations. Novel sequencing technologies have uncovered sufficient variation to investigate the effect of human selection across an entire plant genome. Single nucleotide polymorphisms (SNPs) are a predominant form of sequence variation among individuals representing as much as 90% of the genetic variation in any species. SNPs are distributed throughout a genome, they provide stable markers for genetic analysis, and their detection is amenable to automation. It is increasingly cost and time efficient to genotype large populations in a high- throughput manner. Because of these advantages, SNPs have become a marker system of choice for genetic analysis in plant species. In tomato, SNPs have been discovered using several methods: *in silico* mining of expressed sequence tag (EST) databases, intron and amplicon sequencing of conserved orthologous set (COS) genes, and hybridization to oligonucleotide arrays. Recently, 62,576 non- redundant SNPs were identified based on transcritpome sequences for six tomato accessions. High-throughput SNP discovery has been accompanied by the development of array- based genotyping platforms that permit rapid scoring of several thousand markers in parallel. Such SNP genotyping methods have facilitated high-density genetic map construction and genome-wide association analysis. For example, the use of a maize array with 49,585 SNPs produced two genetic linkage maps with 20,913 and 14,524 markers, respectively. In tomato, we used the “SolCAP” array with 7,720 SNPs to generate high-density genetic maps using two F₂ interspecific populations: 3,503 markers in the *S. lycopersicum* LA0925 x *S. pennellii* LA0714 (EXPEN 2000) population and 4,491 markers in the Moneymaker x *S. pimpinellifolium* LA0121 (EXPIM 2012) population. An array with 44,100 SNPs was used to genotype 413 diverse accessions of rice and analyzed for association with 34 quantitative traits. SNP data from such arrays can also be a resource for germplasm management in breeding programs and has a role in genomic selection strategies for crop improvement. The analysis of variation in tomato populations has often focused on differences between cultivated and wild species. In contrast, relatively little is known about which genes or genomic regions distinguish market classes within cultivated genepools. In the early 1900s, significant effort was placed on the evaluation of wild germplasm as a source of new resistance genes. Wide crosses were used to introduce new sources of resistance, with several varieties in commerce carrying introgressed genes by the late 1930s and early 1940s. These efforts often had a regional focus, and it remains unclear how introgressed regions are dispersed within breeding programs. In addition, there was a concerted effort to breed for distinct plant habits and fruit characteristics for the processing and fresh market tomato industries. These efforts were initiated in the 1940s and resulted in the first cultivars suitable for machine harvest by the early 1960s. At least three sources of *S. pimpinellifolium* were incorporated into processing breeding in order to develop compact plants amenable to “once over” (i.e. destructive) harvest. In order to investigate genetic variation on the tomato genome due to contemporary breeding, we subjected a collection of 426 accessions to high- throughput genotyping using the SolCAP SNP array. The tomato accessions represented different market classes of cultivated tomato and closely related wild species. Knowledge of the genetic and physical organization of the SNPs allowed us to conduct population level analysis based on the germplasm panel. SNP genotypes from the array were analyzed to identify sub-populations, and to assess genetic differentiation and diversity between and within sub-populations. We also investigated patterns of linkage disequilibrium (LD) within each chromosome in three sub-populations of large-fruited cultivated accessions (processing, fresh market, and vintage). The population level analysis revealed regions of the genome with high genetic variation between sub-populations, suggesting that historical breeding practices have led to different patterns of genetic variation in cultivated tomato germplasm. # Results ## Array-based SNP Genotyping We used an array consisting of 7,720 SNPs distributed throughout the genome, to genotype 426 accessions (410 inbreds and 16 hybrids; referred to as the SolCAP germplasm). The hybrids were chosen to maximize heterozygosity and develop a cluster file for the GenomeStudio software (Illumina Inc., San Diego, CA, USA). In order to establish an accurate and automatic genotype calling procedure for the GenomeStudio software that is usable across the entire genepool of cultivated tomato, clustering based on the SolCAP germplasm was cross-validated with a cluster file based on 92 hybrids developed independently by TraitGenetics. The resulting high-quality cluster file is available through the eXtension website. In order to assess the quality of SNP calls, we duplicated 34 accessions that were randomly selected using independent DNA preparations and genotyping facilities. The average proportion of consistent calls across all accessions was 98.7%. Cluster analysis was performed, and for all 34 accessions, the nearest neighbor was the duplicate accession. The proportion of consistent calls increased to 99.3% by excluding three accessions, OH981136 (processing, 91.8%), Purple Clabash (vintage, 90.0%), and Principle Borghese (vintage, 94.1%). One of the duplicate samples for OH981136 showed higher rates for no call (8.8%) and heterozygote call (8.4%) relative to the other sample (0.7% for no call and 0.2% for heterozygote call). The data quality between two samples of Purple Clabash was also different from each other (0.7% vs. 5.7% for no call). Heterozygote call rates were similar for the two samples. The duplicate samples of Principe Borghese differed by 5.9%, but showed similar rates for no call (0.8% vs. 1.0%) and heterozygote call (0.2% vs. 0.2%). These results suggest that overall reproducibility is high, and that differential calls may result from DNA quality that affects the percentage of “no call”, residual heterozygosity, and variation within accessions. The data were analyzed to determine the polymorphism rate based on the inbred accessions. A total of 7,500 SNPs (97.2%) were polymorphic and 61 SNPs (0.8%) were monomorphic. Among the 7,500 polymorphic SNPs, there were 7,375 SNPs with \<10% missing data, 84 SNPs with 10–20% missing data, and 41 SNPs with \>20% missing data. The SNPs with a high frequency of missing data were randomly distributed across accessions. Polymorphism of 123 SNPs could not be unequivocally determined because of a large amount (≥10.0%) of missing data (34 SNPs) or because polymorphism detection was due to a single homozygous allele and only heterozygote calls for the alternate allele (89 SNPs). It is possible that the heterozygotes actually represent duplicated genes, or the alternate allele was simply not present as a homozygote. These 89 SNPs were also randomly distributed across the genome. In addition, 36 SNPs failed to produce a genotype call because of poor signals. The proportion of heterozygous SNPs was assessed within cultivated tomato germplasm and *S. pimpinellifolium* accessions. The heterozygosity with all scorable markers (7,684 SNPs) was 0.02 for processing, 0.01 for large-fruited fresh market, 0.01 for large-fruited vintage, and 0.04 for *S. pimpinellifolium* accessions. Using only markers that were polymorphic within each sub-population (range 3,700– 6,022 polymorphic markers), the processing, fresh market, and vintage accessions showed heterozygosity levels of 0.02, 0.01, and 0.02, respectively. The heterozygosity for *S. pimpinellifolium* was 0.05. ## Genetic Differentiation between Sub-populations The 410 inbred accessions were first divided into five sub-populations based on *a priori* knowledge of germplasm pedigree, age, market class, and origin: 141 processing, 122 fresh market, 88 vintage, 43 wild cherry, and 16 *S. pimpinellifolium*. Principal component analysis (PCA) and linkage disequilibrium (LD) analysis supported separation of the fresh market sub-population into a group of 110 large-fruited accessions and 12 cherry accessions (cultivated cherry). Similarly, the vintage sub-population was divided into 61 large-fruited accessions, 15 cherry accessions which clustered with the cultivated cherry sub- population, and 12 landrace accessions. This iterative analysis led to the definition of seven sub-populations for further analysis: 141 processing, 110 large-fruited fresh market (hereafter referred to as fresh market), 61 large- fruited vintage (hereafter referred to as vintage), 27 cultivated cherry, 12 landrace, 43 wild cherry, and 16 *S. pimpinellifolium*. In the PCA analysis, the first two principal components (PC1 and PC2) explained 22% and 16% of the total variance, respectively. PC1 separated the *S. pimpinellifolium* accessions from all other sub-populations (*P*\<0.001), while PC2 separated processing, fresh market, and vintage germplasm (*P*\<0.01;). A subsequent PCA was conducted using only the processing, fresh market, and vintage sub-populations. The second, more focused analysis validated the significant separation of the sub-populations. In addition, PCA based on only the cultivated accessions suggested further divisions within both processing and fresh market sub-populations. Although most of the accessions were clustered into a corresponding sub-population, there were a few accessions that appear to be transitional and/or misclassified accessions. For example, Peto 460 (PI 600920; coordinates −22.1, −4.8 in) and Heinz 1370 (PI 341134; coordinates 4.6, −3.4) were grouped with the vintage accessions due to their date of release, but were developed as processing tomatoes. Similarly, Rio Grande (coordinates −12.5, 9.1) was classified as a fresh market accession, but clustered with the processing accession. This cultivar is an early “Roma” type tomato and was originally developed as a processing tomato. Clustering supports the processing origin of these three accessions. Pairwise analysis of *F*_st was used to test the significance of genetic differentiation between sub-populations. In order to test effects of marker choice for this analysis, we estimated pairwise *F*_st using two independent subsets of SNP data derived from the array. Analysis was performed on all 410 inbred accessions for each marker set. In addition a re- sampling analysis was performed by randomly selecting n = 40 for processing, fresh market, and vintage sub-populations. Repetition of this analysis suggests that our conclusions are supported with balanced populations and across multiple marker subsets. Cultivated germplasm, including processing, fresh market, and vintage accessions were all significantly diverged (*F*_st  = 0.29 to 0.41, *P*\<0.005). The cultivated cherry, wild cherry, and landrace accessions were not differentiated from each other but were distinct from all other sub- populations (*F*_st  = 0.13 to 0.64, *P*\<0.005). The *S. pimpinellifolium* accessions were separated from all other sub-populations with estimates of pairwise *F*_st ranging between 0.57–0.81 (*P*\<0.005;). The *F*_st analysis verified genetic differentiation between sub- populations defined by PCA. Further divisions within both processing and fresh market sub-populations detected in PCA were also tested by estimating pairwise *F*_st. The processing accessions can be divided into two groups consisting of 82 and 52 accessions, respectively. Seven accessions were not grouped because they were outliers, these tended to be CULBPT accessions which derive from recent crosses to fresh-market material. The two main processing groups were significantly differentiated from each other (*F*_st  = 0.27 *P*\<0.005) and distinct from the other germplasm. Two groups of the fresh market tomatoes including 61 and 49 accessions each showed significant differentiation between them and from the other sub-populations (*F*_st  = 0.10 to 0.76, *P*\<0.005;). ## Levels of Polymorphism within Sub-populations The level of genetic diversity within each sub-population was measured using allelic richness (*A*), expected heterozygosity (*He*), and polymorphic information content (PIC). Within cultivated germplasm, the highest estimates of *A*, *He*, and PIC were found in the cultivated cherry accessions. Among the remaining cultivated sub-populations, the fresh market accessions showed higher *A*, while the landraces had the highest *He* and PIC values. The sub-population of vintage accessions contained the lowest variation for all three descriptors. The further division of accessions within both processing and fresh market sub- populations showed higher estimates of these descriptive statistics relative to the vintage sub-population. The highest number of polymorphic markers (6,234 SNPs) was identified in the cultivated cherry sub-population. There were 5,909 and 5,650 polymorphic markers in the wild cherry and *S. pimpinellifolium* sub-populations. For the contemporary accessions, 4,648 and 6,022 markers were polymorphic in the processing and fresh market sub-populations, respectively. Fewer polymorphic markers were found in the vintage (3,700 SNPs) and landrace (3,159 SNPs) sub- populations. Distribution patterns of polymorphic SNP markers across 12 chromosomes were different between sub-populations. For example, chromosome 8 showed proportionally lower SNP numbers for the processing and cultivated cherry accessions, as did chromosome 12 for the large-fruited fresh market and wild cherry tomatoes. Chromosome 10 had lower polymorphism rates for the vintage and landrace groups compared to contemporary processing and large-fruited fresh market sub-populations. The number of polymorphic markers in sub-populations increases with the size of the population. This fact led to concerns that our estimates of diversity might be skewed by population size differences despite the fact that *A* and *He* are adjusted for population size. To address the concern, we performed rarefaction analysis to estimate polymorphic marker accumulation curves. For all seven sub- populations, the curves reached an asymptote within the population size sampled. The *S. pimpinellifolium*, wild cherry, and cultivated cherry sub-populations were the most diverse groups based on the curves. Within the large-fruited cultivated sub-populations, fresh-market accessions were more diverse than vintage accessions. The accumulation curves for the processing and vintage accessions merged at n = 60, despite the fact that they were well separated between n = 10 to 25. This result suggests that differences in number of polymorphisms detected between these sub-populations may be population size dependent. ## Minor Allele Frequency of SNP Markers Minor alleles for 7,310 SNP markers with physical map positions were determined based on genotypic data of 410 inbred accessions. Minor allele frequency (MAF) was then estimated within each sub-population. We graphed MAF in order to visualize genetic variation between three representative sub-populations of cultivated tomatoes (processing, fresh market, and vintage) along with *S. pimpinellifolium* accessions. These plots revealed different MAF patterns over the entire genome with chromosomes 2, 4, 5, 6, and 11 being particularly variable between the cultivated sub-populations. The processing and fresh market sub-populations showed unique MAF patterns on these chromosomes relative to the vintage sub-population. MAF patterns on chromosome 5 distinguished the processing sub-population from the fresh market sub-population. Common alleles in the *S. pimpinellifolium* accessions tended to be minor alleles in the cultivated sub-populations. Since the processing and fresh market sub- populations were further divided into two sub-groups, respectively, MAF was estimated within each sub-group and graphed. The processing sub-groups were distinguished by unique MAF patterns on chromosomes 5 and 11. The most variable MAF patterns between the large-fruited fresh market sub-groups were found on chromosome 11. ## Loci Explaining Variation within Germplasm PCA loadings measure the correlation between PC and SNP, and provide an estimate of how much each SNP contributes to variance. We extracted loadings from the PCA of only cultivated germplasm and displayed these values relative to physical position in order to visualize SNPs that contribute most to variance in the germplasm panel. There were 778 SNPs with absolute values of \>0.02 for PC1. Most of these SNPs were found on chromosomes 4 (24.0%), 5 (32.6%), and 11 (31.1%). For PC2, 709 SNPs with absolute values of \>0.02 were found on chromosomes 5 (12.3%) and 11 (52.9%). We also investigated loci that may be under positive selection between the three cultivated sub-populations using an *F*_st outlier method based the expected distribution of *F*_st and *He*. We identified 339 candidates for loci under positive selection between processing and fresh market as falling outside of the 95% confidence interval. A high portion of these loci (61.4%) were derived from chromosome 5, while 0.6–10.0% of the SNPs were distributed across the other 11 chromosomes. Comparison between processing and vintage germplasm detected 128 candidates for loci under positive selection (, and). Among these, 57 loci (44.5%) were located on chromosome 4 and 35 loci (27.3%) on chromosome 5. For comparison between fresh market and vintage sub-germplasm, 208 loci were outliers based on a 95% confidence interval. Most of the loci were derived from chromosome 4 (42.6%) and chromosome 11 (43.5%). For all three pairwise comparisons, the candidate loci under positive selection were not randomly distributed within a chromosome. To visualize the position of candidate genes that may have been selected for during breeding, we superimposed the position of 28 loci onto the physical map of MAF pattern, PCA loading, and *F*_st outlier detection. Candidate loci included genes that affect fruit size, shape, and color, disease resistance, and plant morphology. Three genes for fruit shape and size, *fw2.2*, *OVATE*, and *lc* are located on chromosome 2. Of these genes, only *lc* appears to be a candidate for a locus under selection based on MAF patterns, PCA loadings, and *F*_st outlier detection of linked SNPs. The *OVATE* locus is polymorphic within all cultivated sub-populations, and therefore SNPs lack power to discriminate populations. The large fruited allele of *fw2.2* is fixed in the cultivated sub-populations. The *SUN* mutation which controls elongated fruit shape is found on chromosome 7 and is present at a low frequency in both processing and vintage accessions. The genomic region for the *SUN* mutation contains some SNPs with high absolute values for PCA loadings, though none of these were detected in the *F*_st outlier analysis. Another fruit shape gene, *FAS* is found in a region of chromosome 11 with high loadings for PC2, and LD between *FAS* and SNPs in this region may be responsible for detection of *F*_st outliers between processing (which lack *FAS*) and vintage (which are polymorphic for *FAS*) germplasm. Phenotypic data are available in flat file format through the SolCAP website and in searchable form through the Sol Genome Network (SGN) ontology database using advanced search options with the stock number prefix, SCT; stock type, accession; stock editors, SolCAP project; and the organism as either *Solanum lycopersicum* or *S. pimpinellifolium*. A detailed analysis of phenotypic data is provided in a separate publication. Fruit color genes affecting skin and flesh color are distributed on chromosomes 1 (*y*), 3 (*Psy1*), 6 (*B*), 8 (*gf*), and 10 (*u*). Although *y*, *Psy1* and *gf* are found in regions of the genome with SNPs that have moderately high absolute values associated with PCA loadings and somewhat variable MAF between sub-populations, these regions do not appear to be highly discriminatory. Regions on chromosome 1 and 3 were also not coincident with loci identified by the *F*_st outlier approach. The *old gold crimson* (*og^c*) allele of the *B* gene encodes a mutation in the fruit specific lycopene beta- cyclase and is segregating in all three cultivated populations. Thus minor variation in MAF patterns, PCA loadings, and the detection of a SNP under selection between processing and vintage are more likely a reflection of the closely linked *SP* allele. In contrast, the region of chromosome 8 containing *gf* contains several SNPs detected as outliers based on *F*_st between processing and vintage germplasm, which is consistent with the absence of *gf* mutations in processing populations and the presence of the mutant alleles in several vintage accessions. The allele of *u* gene for uniform ripening appears to fall in a region that is fixed in the processing germplasm. This allele is present at a high frequency in fresh market germplasm and at a low frequency in vintage germplasm. The region contains several SNPs with high absolute values for PCA loadings and also SNPs detected as *F*_st outliers. The signals of genetic differentiation on chromosomes 5, 6, and 11 might be due to allelic variation in disease resistance genes, and reflect different introgression histories. Two resistance genes to bacterial disease, *Pto* and *Rx3* on chromosome 5, appears to be candidates for loci under selection based on MAF patterns, PCA loading, and *F*_st outlier detection. The region of chromosome 6 containing *Cf-2* and *Mi1.2* genes, shows signals of selection distinguishing processing accessions from fresh market and vintage accessions. In general, *Cf* genes are deployed more frequently in fresh market material while *Mi* has been widely used in processing germplasm in California, but rarely used in processing accessions in the Midwestern U.S. Chromosome 11 contains one of the oldest introgressions, *I* for *Fusarium* resistance. This gene falls in a region of the genome that distinguishes contemporary germplasm from vintage germplasm based on MAF patterns. The region also contains some SNPs with high PC loadings. SNPs were detected as outliers at the 95% confidence level between processing and vintage germplasm. The region of chromosome 11 containing *Rx-4/Xv3* and *I-2*, appears to be discriminatory between the cultivated sub-populations based on MAF patterns, PC2 loadings, and outlier SNPs. Other regions with signals for loci under selection include chromosomes 7 (*I-3*), chromosome 9 (*Ve1*, *Tm2*^∧*2*, *Sw5*, and *Ph-3*), and chromosome 10 (*Ph-2*). The genomic regions for *Ph-2* and *Ph-3* show MAF patterns that distinguish processing from fresh market and vintage sub- populations. SNPs were also detected in the regions by PCA loadings and *F*_st outlier analysis. These resistance genes have been deployed in several processing tomatoes. The region of the genome containing *Tm2*^∧*2* distinguishes fresh-market from vintage and processing accessions based on the presence of SNPs with a MAF of 10–15%, consistent with deployment of this resistance in fresh market accessions. However, this region was not detected based on PC loadings or *F*_st outlier detection, possibly due to low allele frequencies for markers in linkage disequilibrium with *Tm2*^∧*2*. Genes affecting plant morphology are distributed across several chromosomes. A region on chromosome 2 that appears to differentiate sub-populations contains the compound inflorescence gene (*s*). Chromosomes 5 and 6 contain genes (*SP5* and *SP*, respectively) that control plant habit and may be key selection points in differentiating vintage from contemporary sub-germplasm and processing from fresh market. The *SP5* gene is in a region of the genome containing SNPs with high loadings for PC2, but does not appear to have been detected by *F*_st outlier analysis. The *jointless* gene (*j*) on chromosome 11 falls in a region containing SNPs with high absolute values for PCA loadings and outlier SNPs. This region distinguishes processing from fresh market and vintage accessions, reflecting the near fixation of jointless accessions in this germplasm category. The genomic region for the *j-2* gene on chromosome 12 contains SNPs with high absolute values for PCA loadings. ## Linkage Disequilibrium (LD) Analysis The extent of LD across each chromosome was analyzed for the processing, large- fruited fresh market, and large-fruited vintage sub-populations. Pairwise *r²* was calculated using 1,572 polymorphic SNP markers with MAF of ≥0.1 for processing, 1,504 for fresh market, and 700 for vintage accessions. The *r²* values were plotted against the genetic distance, and curves of LD decay were fitted using locally weighted scatterplot smoothing (LOESS) and non-linear regression (NLR). The LOESS and NLR methods estimated similar LD decay on 9 chromosomes for processing, 8 chromosomes for fresh market, and 11 chromosomes for vintage germplasm. In these chromosomes, the observed difference of the LD decay between the methods ranged from 0 to 4.9 cM. Over 10 cM difference for LD decay estimated by LOESS and NLR was found on chromosome 6 (28.1 cM for processing and 14.1 cM for fresh market); chromosome 8 (20.2 cM for processing); and chromosome 11 (9.6 cM for fresh market and \>40 cM for vintage). The use of either a fixed *r²* value of 0.2 or a value estimated using the 95^th percentile method resulted in similar values for LD decay over chromosomes in the three sub-populations. Different patterns of LD decay were observed between chromosomes and sub- populations. Baseline *r²* values estimated using the 95^th percentile method ranged from 0.11 to 0.23. In the processing sub-population, the baseline *r²* value based on the 95^th percentile method was 0.23 and led to estimates of LD decay that ranged from 0.8 cM on chromosome 11 (both LOESS and NLR) to 7.7 cM on chromosome 8 (LOESS) and 35.2 cM on chromosome 6 (NLR). LD decay estimated from the 95^th percentile baseline *r²* value of 0.12 in the fresh market sub-population ranged from 3.8 cM on chromosome 4 to 47.6 cM on chromosome 11 for LOESS, and ranged from 4.1 cM on chromosome 12 to 38 cM on chromosome 11 using NLR. With the 95^th percentile baseline *r²* value of 0.11 in the vintage sub-population, LD decay occurred over the shortest distance on chromosome 12 (1.9 cM for LOESS and 1.1 cM for NLR), while the greatest distance for LD decay was found on chromosome 1 (9.9 cM for LOESS) and chromosome 3 (11.8 cM for NLR). Estimates of LD decay vary more on a chromosome to chromosome basis than those based on the method used to establish the LD cut-off or decay curve. # Discussion A high density array with 7,720 SNP markers was used to genotype the SolCAP germplasm panel consisting of 410 inbred accessions. This diverse germplasm enabled the development of a cluster file for accurate SNP calling with the GenomeStudio software (Illumina Inc. San Diego, CA, USA). Over 98% of SNP markers generated consistent calls between duplicate samples across 34 accessions, with differences due mostly to DNA quality. Also, 7,375 SNP markers (96%) were polymorphic with \<10% missing data in the entire germplasm. We found that the level of heterozygosity (proportion of heterozygotes) was low, but variable between market classes of germplasm. The SNPs provide excellent genome coverage, but their distribution is not reflective of chromosome size. For example, chromosome 1 is cytologically one of the largest, yet is underrepresented by SNPs relative to chromosome 11 which is cytologically small. It is possible that this distribution represents a distortion of the true measure of polymorphism, heterozygosity, or genetic diversity due to the sampling of markers. However, the germplasm presented in this study were well represented in the sequencing and SNP discovery pipeline with five of the seven sub-populations contributing to sequenced germplasm. We found more variation in the level of polymorphism between chromosomes within a sub-population relative to the variation between sub-populations. These results suggest that there is little ascertainment bias within the red-fruited species, and that the chromosome to chromosome variation reflects breeding history rather than polymorphism discovery. The possibility of ascertainment bias when applied to more distant germplasm exists, and will be the subject of a future study. Cultivated germplasm was divided into distinct sub-populations. The genetic differentiation between processing and fresh market germplasm reflects human selection for distinct ideotypes tailored to the needs of specific production systems. The contemporary processing and fresh-market germplasm were also distinct from vintage germplasm, which is consistent with previous findings. The further division within processing germplasm confirms an earlier analysis based on the Bayesian model implemented in STRUCTURE which demonstrated sub-structure consistent with breeding history and environmental adaptation. We have not seen evidence for sub-structure within fresh market germplasm previously. Upon close inspection, the cluster demarked by PC1 coordinates (-10, 10) and PC2 coordinates (−30, −15) contained 87% of the fresh market accessions from Florida and 65% of the accessions from North Carolina. In contrast, the cluster from PC1 (10, 30) and PC2 (-15, 10) contained 100% of the fresh market accessions from Oregon and California. The Oregon and California accessions were not part of our previous collection, and the sub-structure identified in this analysis suggests that diversifying forces may play a role in shaping fresh market as well as processing tomato germplasm. Genetic differentiation within market classes may reflect founder effects in breeding programs, selection for specific traits, environmental adaptation, or a combination of these factors. Cultivated cherry, wild cherry and landrace accessions were not well separated. These results support previous studies showing that Latin American accessions and feral accessions often share alleles. The lack of differentiation between cultivated cherry and wild cherry or landrace accessions was somewhat surprising, but is consistent with a diverse breeding base for the cherry market class. Genetic diversity for each sub-population was measured using allelic richness, expected heterozygosity, and polymorphic information content. The descriptive statistics in the contemporary sub-populations exceeded levels found in the vintage sub-population, but were lower relative to the cherry sub-populations and *S. pimpinellifolium*. Although differences between processing and vintage sub-populations may be population size dependent, rarefaction analysis provides strong support for increased variation in fresh-market sub-population relative to vintage sub-population. Tomato has undergone genetic bottlenecks during domestication and through selection after the introduction of the crop into Europe. Since the early 1900s, wild relatives have been used to introgress new alleles into cultivated tomato. This practice is expected to increase allelic diversity in contemporary processing and large-fruited fresh-market germplasm. The analysis of minor allele frequency (MAF) across the genome demonstrates genetic differentiation between sub-populations in specific chromosome regions. The different MAF patterns between three cultivated sub-populations (processing, fresh market, and vintage) were particularly evident on chromosomes 2, 4, 5, 6, and 11. The same regions of the genome were highlighted by SNPs contributing high absolute values to PCA loadings and by SNPs detected as *F*_st outliers based on a deviation from the expected distribution of *F*_st and *He*. Thus, these chromosomes appear to be under diversifying selection relative to other regions of the genome. In order to investigate potential regions of the genome under selection, we superimposed MFA patterns, PCA loadings, and SNPs identified from *F*_st outlier detection with the position of genes affecting fruit shape, size and color, disease resistance, and plant morphology ( and). Taken together, several regions of the genome containing candidate genes which may be under selection were detected based on coincident MAF patterns, PC loadings, or *F*_st outlier analysis. However, the resolution of this analysis does not provide unequivocal evidence that selection for these candidates explains variation between sub-populations. Given our marker resolution and observed LD decay, candidate gene analysis was not highly informative. In addition, our ability to detect outliers may be influenced by allele frequencies and distribution across the sub-populations. In general, several regions under selection are compatible with the introgression of genes for fruit size and shape (e.g. *lc* and *FAS*), fruit color (e.g. *gf* and *u*), disease resistance (e.g.*Cf-2* and *I-2*), and plant morphology (e.g. *s* and *SP5*). Our results also suggest several chromosomes or regions of chromosomes with no obvious candidate genes to explain the observed differentiation. Chromosome 4 was identified as highly important for genetic differentiation between the sub- populations of cultivated germplasm. The role of this chromosome is less clear, though multiple loci affecting plant habit (e.g. *dmt, Epi, glo,* and *si*) have been mapped relative to morphological markers on this chromosome (<http://tgrc.ucdavis.edu>). Other areas under selection that are poorly explained by candidate genes include the chromosome 5 centromere which differentiates processing germplasm from the other sub-populations. Common alleles at many loci in the *S. pimpinellifolium* accessions are minor alleles in the other sub-populations, including cherry tomato. A recent study with 144 cherry accessions demonstrated a separation into two groups: one close to the cultivated tomato and one showing admixture of cultivated tomato and *S. pimpinellifolium*. Genetic diversity was higher in the *S. pimpinellifolium* relative to the wild cherry and cultivated tomato accessions, a result that is consistent with our findings. We expect accessions of *S. pimpinellifolium* to be relatively diverse due to their range of mating systems (many accessions exhibit facultative outcrossing) and wide geographic distribution in the native region. The wild cherry accessions may be a mixture of feral cultivated accessions and wild progenitors of cultivated tomato. The decay of LD over genetic distance is important to determine the density of markers appropriate for genetic analysis and selection strategies. LD levels vary both within and between species. Previous estimates of LD decay in tomato were based on the entire genome with an average of 6–14 cM in processing accessions, 3–16 cM in fresh market accessions, and 15–20 cM in commercial European greenhouse accessions. Given the marker density across each chromosome, we estimated LD decay on a chromosome by chromosome basis for processing, fresh market, and vintage accessions. LD decay was variable between chromosomes and sub-populations, suggesting that historical recombination is not uniform across the genome. For example, chromosome 11 showed decay over 0.8 cM for the processing sub-population, and decay over 19.7 cM for the fresh market sub- population. Although the similar estimates of LD decay was found between the LOESS and NLR methods on most of chromosomes, high levels of difference were found on chromosomes 6, 8, or 11 depending on the sub-populations. These chromosomes also showed high levels of non-homogenous distributions for pairwise *r²* values relative to the other chromosomes. This variation may reflect structure within sub-populations due to selection. When the *a priori* vintage germplasm sub-population was based on age of variety, regardless of fruit size or geographical origin, the LD decay pattern for chromosome 4 displayed extensive LD and what appeared to be parallel patterns of decay. When the vintage accessions were separated into large fruited vintage and cultivated cherry/landrace groups, the dual pattern of LD decay disappeared and LD was reduced. The patterns of LD decay we observed appear to be a consequence of introgression and directional selection through breeding. The range of LD decay also suggests that recombination remains limiting in cultivated tomato because of the inbreeding mating system and an emphasis on backcrossing and pedigree selection in tomato breeding programs. With LD decay exceeding 1cM, the SolCAP tomato array will be useful for most selection strategies and association studies as the average marker intervals range from 0.8 to 1.6 cM in the tomato genetic maps. High-throughput SNP genotyping has provided a means to both visualize and quantify the effect of human selection on the tomato genome. Selection has reduced genetic diversity in vintage cultivated forms relative to wild and feral forms, and has changed the frequency of predominant alleles. In contrast to domestication, contemporary breeding has increased allelic diversity and heterozygosity in populations relative to vintage tomatoes. Selection has also led to sub-populations which are characterized by distinct haplotype blocks, patterns of allelic diversity, and recombination history. This analysis has highlighted specific regions of the genome that appear to be under selection, with SNPs on chromosomes 2, 4, 5, 6, and 11 clearly distinguishing fresh market and processing lineages. Our findings are not surprising given the history of breeding activities. However, incorporating this information into future breeding strategies offers both challenges and potential for creativity. Going forward, we will want to balance a desire to promote recombination and generate new combinations of alleles, with the need to preserve desirable combinations of genes. These analyses also highlight a role for several chromosomes in the differentiation of cultivated tomatoes. A role for chromosome 4 has not been highlighted by previous studies yet this chromosome appears to have regions that have been selected during breeding activities. # Materials and Methods ## Plant Material A collection of 426 tomato accessions, referred to as the Solanaceae Coordinated Agricultural Project (SolCAP) germplasm panel, was assembled from the National Plant Germplasm System (NPGS), the C. M. Rick Tomato Genetics Resource Center (TGRC), and from public plant breeding programs. Accessions from eight universities in the United States and Canada were represented including Cornell University (USA), North Carolina State University (USA), Ohio State University (USA), Oregon State University (USA), Pennsylvania State University (USA), University of Florida (USA), University of California-Davis (USA), and Ridgetown College, University of Guelph (Canada). The germplasm panel represented only the red-fruited species *S. lycopersicum* and *S. pimpinellifolium*, and included 141 processing, 110 fresh market, 61 vintage, 27 cultivated cherry, 12 landrace, 43 wild cherry, 16 *S. pimpinellifolium*, and 16 hybrid accessions. Processing and fresh market germplasm represented contemporary accessions, while vintage accessions (sometime referred to as heirlooms) represented early tomato selections that in some cases predate the application of Mendelian principles to crop improvement. Cherry tomatoes, often referred to as *S. lycopersicum ‘cerasiforme’*, in our germplasm panel were separated into two sub-populations: cultivated and wild accessions. Landraces included Latin American cultivars and represent early domesticates from regions near the centers of origin and domestication. The collection also contained germplasm that was considered commercially relevant, with several inbred lines that are parents of commercial hybrids. The collection also included the parents of several important recombinant inbred and inbred backcross populations,, segmental substitution lines, and a mutation library. Finally, two accessions that have been the subject of public sequencing efforts, Heinz 1706 and LA1589, were also included. ## SNP Genotyping The SolCAP germplasm panel was genotyped using a tomato array with 7,720 SNPs as implemented in the Infinium assay (Illumina Inc., San Diego, CA, USA). Details of the SolCAP SNP discovery pipeline are described previously, as are details of the array. In addition, all SNPs on the array have been incorporated into the SGN database, the SNP annotation file is available through the SolCAP website, and sequences are available through the National Center for Biotechnology Information Sequence Read Archive (accession number SRP007969). For each accession of the SolCAP germplasm, genomic DNA was isolated from fresh young leaf tissue according to a modified CTAB method. Double-stranded DNA concentrations were quantified using the PicoGreen assay (Life Technologies Corp., Grand Island, NY, USA) and normalized to 50 ng/ul with 10 mM Tris-HCl pH 8.0, 1 mM EDTA. Genotyping was conducted with 250 ng of DNA per accession following the manufacturer’s protocol for the Infinium assay. For SNP calls, the resulting intensity data were loaded in GenomeStudio version 1.7.4 (Illumina Inc., San Diego, CA, USA). In order to determine SNP genotype, we first used the automated cluster algorithm to generate initial calls. Clustering for every SNP was assessed by visual inspection and modified when the default clustering was not clearly defined. Particular attention was paid to a clear definition of the boundaries for heterozygote calls, which were reduced manually for a number of SNPs in order to reduce the number of ambiguous calls. As a result, the rate of alleles with no call was increased slightly. ## Data Analysis Physical positions of 7,666 SNPs were previously determined relative to the tomato genome sequence. The SNPs with physical positions were filtered based on polymorphism and missing data. The 7,323 polymorphic SNPs with \<10% missing data included 13 markers with inconsistent chromosome assignments between physical and genetic map positions. We removed these SNPs and used the 7,310 SNPs for minor allele frequency (MAF) and rarefaction analyses. For principal component analysis (PCA), pairwise *F*_st, descriptive statistics, and *F*_st outlier detection, we used 4,393 polymorphic SNPs (excluding the 13 markers) with \<10% missing data that were genetically mapped in the Moneymaker x LA0121 F₂ population of 184 plants. A second set of 3,473 markers was selected based on genetic position in the EXPEN 2000 genetic map and used for pairwise estimation of *F*_st. Polymorphism of these markers was determined based on the observation of at least one alternative allele in the 410 inbred accessions. For linkage disequilibrium (LD) analysis, the 4,393 SNP markers were filtered for a MAF of ≥10% within each sub- population. ### Principal component analysis The genetic relationship between sub-populations was analyzed using PCA as implemented in R. GenomeStudio SNP data were converted to proportional scoring where 2 is equal to homozygous for the common allele; 1 is equal to heterozygote; and 0 is homozygous for the rare allele. Missing data were imputed using the R package pcaMethods in which missing data calls are based on the SVDimpute algorithm. PCA was conducted for the entire data set, and subsequently for a data set consisting of only the three major sub-populations of cultivated tomato (processing, fresh market, and vintage). The relationship between accessions was visualized by plotting scores for PCs. Marker contributions to the loadings of each PC were extracted and displayed relative to chromosome position in order to visualize regions of the genome containing markers that maximize variation within the germplasm collection and assuming that these represent regions of the genome with maximum diversity. Significant differences between sub-populations were tested via analysis of variance (ANOVA) for the eigenvalues of the first two principal components. ### Genetic differentiation and diversity As a measure of population differentiation, pairwise *F*_st was estimated using the Microsatellite analyzer v4.05. This analysis was conducted using two sets of markers (3,473 and 4,393 SNPs) for all 410 inbred accessions. We also estimated pairwise *F*_st with a reduction of sample size (n = 40) for processing, fresh market, vintage sub-populations in order to estimate *F*_st without bias due to different population sizes. The analysis was iterated for three separate sets of accessions with n = 40 for each market class. The *P*-value for the pairwise *F*_st was obtained from 10,000 permutations of genotypes and a Bonferroni correction was applied. Genetic diversity within each sub-population was assessed based on allelic richness (*A*), expected heterozygosity (*He*) and polymorphic information content (PIC) using the 4,393 SNPs for the 410 accessions. *A* and *He* were estimated using the MSA software which corrects for sample size, while PIC was calculated using the equation:where n is the number of allele and *p_i* is the frequency of the *i*^th allele. ### Rarefaction analysis This analysis was used to estimate how the number of polymorphic markers increases relative to sample size. Analysis and graphing was conducted using the species accumulation curve function “specaccum” in the R package, vegan. The method “random” was applied to estimate means and standard deviations from 100 sub-samples without replacement. This method provides a data summary and permits boxplot methods to be used to graph the curves. The boxplot function was used to superimpose standard deviations. ### Minor allele frequency The minor allele was determined based on the allele calls for 410 inbred accessions. Minor allele frequency (MAF) was then calculated across all chromosomes for each of sub-populations and sub-groups within the processing and fresh market sub-populations. This approach permitted the comparison of allele frequencies between sub-populations. MAF was graphed to visualize genetic variation based on physical map position across 12 chromosomes using the R package ggplot2. We also positioned genes for disease resistance (13 genes), fruit shape and size (5 genes), fruit color (5 genes), and plant morphology (5 genes) on the MAF plot. Physical positions of the genes relative to the tomato genome sequences were determined using coding sequences or flanking marker sequences. These sequences were aligned to the Tomato WGS chromosome v SL2.40, using BLAST. ### Loci under positive selection In order to identify loci under positive selection between processing, fresh market, and vintage sub-populations, we used an *F*_st- outlier detection method as implemented in the LOSITAN workbench. Outliers were detected based on a distribution of *F*_st and expected heterozygosity (*He*). Five simulations for each pairwise comparison were run with 10,000 iterations. Simulations were conducted using the neutral mean, forced mean, and a mutation model with infinite alleles. Under these options, LOSITAN estimated a 95% confidence interval, and SNPs that fell outside of this range were identified as under positive selection (higher than expected *F*_st for an estimated *He*). ### Linkage disequilibrium (LD) analysis The extent of LD across each chromosome was measured using the SNP genotypes in three representative groups of cultivated tomatoes (processing, fresh market, and vintage). Pairwise *r²* between markers within each chromosome was then calculated using TASSEL v 2.1 and GGT v 2. In order to determine the decay of LD, these *r²* values were plotted against the genetic distance (cM) between markers. Curves of LD decay were fitted using locally weighted scatterplot smoothing (LOESS) and non-linear regression (NLR) using R. For LOESS, smoothing parameters between 0.1 and 0.5 were tested, and a final parameter of 0.3 was chosen based on curve fits. For NLR, the predicted *r²* values between adjacent markers were calculated using the model of Hill and Weir. Two methods were chosen to determine baseline *r²* values: a fixed value of 0.2 and the parametric 95^th percentile of the distribution of the unlinked markers. # Supporting Information We would like to thank Esther Van der Knaap at The Ohio State University, OARDC; the C. M. Rick Tomato Genetics Resource Center (TGRC); and the National Plant Germplasm System (NPGS) for providing tomato seeds. We thank Carolyn Male for discussion and suggestions of representative heirloom accessions and Joanne Labate for helpful discussion. We also thank internal reviewers at The Ohio State University, OARDC for comments and helpful suggestions on the manuscript. [^1]: The author, M. W. Ganal has competing interests as a member of TraitGenetics, which is a commercial company that performs molecular marker analysis with the tomato SNP array. TraitGenetics has also a commercial interest in the data generated with the array since it increases the value of their services to their customers. This does not alter the authors’ adherence to all the PLOS ONE policies on sharing data and materials. There are no further products in development or marketed products or patents to declare. [^2]: Conceived and designed the experiments: SCS AVD RTC SFH JWS RGG DRP MM JRM DMF. Performed the experiments: SCS AVD KS DSD DZ DMF. Analyzed the data: SCS MWG DMF. Contributed reagents/materials/analysis tools: MWG RTC SFH JWS RGG DRP MM JRM DMF. Wrote the paper: SCS MWG DMF.

# Introduction *Brucella* spp. are Gram-negative bacteria that belong to the α-2-proteobacteria group. They are responsible for one of the world’s most widespread zoonotic infections, brucellosis, causing miscarriage and infertility in animals and a febrile disease in humans known as Malta or undulant fever. *Brucella* is a facultative intracellular pathogen that can infect professional and non- professional phagocytes. After internalization, bacteria persist and replicate inside a modified vacuole called the *Brucella*-containing vacuole (BCV), a compartment with membranes derived from the endoplasmic reticulum. During intracellular trafficking and active replication inside the BCV, *Brucella* cells encounter low oxygen tension, low pH, reactive oxygen and nitrogen species, and nutrient deprivation. The primary basis for the tenacity of *Brucella* infections lies in their capacity to persist for prolonged periods in the phagosomal compartment of these phagocytes. In order to survive under these extreme conditions, *Brucella* is able to sense and respond to environmental factors by appropriately modifying RNA and protein expression. *Brucella*, like many other bacteria, can sense and respond to environmental changes through two-component systems (TCS), which typically consist of a sensor histidine kinase (HK) and its cognate response regulator (RR). Usually, sensor HKs are proteins consisting of a sensor domain at the N-terminal end, linked to a C-terminal HK domain. Activation of the sensor domain by a chemical or physical signal leads to an increase in autophosphorylation of a His residue in the HK domain. Subsequently, the phosphate is transferred to a conserved Asp residue in the receiver (REC) domain of the RR protein. Some RRs only have a REC domain, and others have a C-terminal effector domain which could be a DNA or protein-binding domain, or could have enzymatic activity. This effector domain elicits an output response, which finally modulates gene expression and cellular physiology. *Brucella* spp. genomes encode a LOV-domain-containing protein that is associated to a PAS (Per, Arnt and Sim) domain plus a HK domain (LOVHK), which belongs to the HWE family. LOV domains are blue-light sensory modules which bind a flavin molecule as cofactor. Activation of LOVHK by blue-light increases the autophosphorylation activity of the HK domain. LOVHK plays a role in the intracellular survival of *Brucella*, since deletion of this gene in *B*. *abortus* shows an attenuated phenotype in cell-culture infection assays. Light is the main source of energy of the biosphere but it also has damaging effects on many biological molecules. Many bacteria have evolved photoreceptor proteins that sense light and in particular blue-light, to elicit adaptive responses to this stress factor. LOV-domain-containing proteins from other bacteria have been associated with stress responses. For example, light perception through LOV proteins has been associated with activation of stress responses in the Gram-positive bacteria *Bacillus subtilis* and *Listeria monocytogenes*. Recently, a relationship between LOV proteins and the RR, PhyR, has been shown in two members of the α-proteobacteria group: *Caulobacter crescentus* CB15 and *Erythrobacter litoralis* HTCC2594. While in *C*. *crescentus* the interaction between a LOV protein and PhyR was demonstrated using a genetic approach, in *E*. *litoralis* this relationship was shown by phosphotransfer assays *in vitro*. LOV proteins of both species not only interact with PhyR but they are also able to interact with single-domain response regulators. PhyR is the main regulator of the General Stress Response (GSR) system that protects bacteria against a wide variety of stress conditions. This system is conserved within and restricted to all α-proteobacteria, and has been described in *Methylobacterium extorquens* AM1, *Bradyrhizobium japonicum* USDA 110, *C*. *crescentus* CB15, *Sphingomonas* sp. Fr1, *Sinorhizobium meliloti*, *Bartonella quintana*, *Methylosinus* sp. B4S, *Rhizobium etli*, *Rhodopseudomonas palustris* TIE-1, and *Brucella abortus* 2308. The GSR system consists of an alternative extracytoplasmic function (ECF) sigma factor known as RpoE that, in the absence of stress or a particular stimulus, binds the anti-sigma factor NepR and maintains the system in an inactive state. In the presence of a stimulus such as peroxides, increase of osmotic pressure, UV light, or desiccation, the anti-sigma factor is sequestered by PhyR allowing the ECF sigma factor to associate with the core RNA polymerase, and direct the expression of specific genes in response to stress. PhyR is a RR that has an N-terminal sigma factor- like domain which binds NepR, and a C-terminal REC domain with a conserved Asp residue. *In vitro*, phosphorylation of this residue using acetyl-phosphate or beryllium trifluoride is required for PhyR/NepR interaction in many α-proteobacteria. In *B*. *abortus*, unphosphorylated PhyR and RpoE1 have similar interaction affinities with NepR. However, phosphorylation of PhyR increases its affinity for NepR, thus stabilizing PhyR-NepR complexes. Interestingly, *in vivo*, in the absence of stress PhyR protein is degraded by the ClpXP protease. In the presence of stress, PhyR protein level is stabilized by inactivation of ClpXP protease and/or PhyR phosphorylation. Although this system responds to different stress factors, up to now a HK able to phosphorylate PhyR has not been described in *Brucella* spp. In the present work, using a combination of two-hybrid assays and phosphotransfer experiments we identify two response regulators as partners of LOVHK and provide further information on the LOVHK-associated regulatory network. The results presented here increase the understanding of the role of this signaling pathway in *Brucella* virulence. # Materials and Methods ## Bacterial strains and culture conditions Genetic manipulations were carried out in *Escherichia coli* DH5α in Luria- Bertani medium (LB, Difco) at 37°C on a shaker at 250 rpm with the appropriate antibiotics added. *E*. *coli* BL21(DE3)pLysS was used for protein expression and was grown in autoinducer medium ZYM-5052, MD-5052 medium or in LB. *E*. *coli* S17-1 was used for conjugation with *B*. *abortus*. When used, antibiotics were added to the following final concentrations: 100 μg ml^-1 ampicillin, 25 μg ml^-1 kanamycin, 12.5 μg ml^-1 tetracycline, 25 μg ml^-1 chloramphenicol. *Brucella abortus* 2308 was grown at 37°C on a shaker set at 250 r.p.m. in rich medium Tryptic Soy Broth (TSB, Difco) from an initial OD_600nm of 0.05, or in Tryptic Soy Agar (TSA, Difco). When used, antibiotics were added to the following final concentrations: 50 μg ml^-1 ampicillin, 25 μg ml^-1 kanamycin, 10 μg ml^-1 nalidixic acid, and 20 μg ml^-1 chloramphenicol for agar plates and 5 μg ml^-1 chloramphenicol for liquid cultures. All *Brucella* strains used in this study were manipulated in a biosafety level 3 laboratory available at Leloir Institute following national regulations. All *E*. *coli* and *Brucella* strains used in this study are listed in. ## DNA manipulations DNA manipulations were performed according to standard techniques. PCR reactions were performed using *B*. *abortus* 2308 genomic DNA as template and Pfx polymerase (Invitrogen) according to the manufacturer’s instructions. The amplified products were digested with the indicated restriction enzymes and cloned into the same restriction sites of the corresponding plasmids. All plasmids and primers used are listed in and , respectively. ### Cloning of the *lovhk* and *pdhS* genes for the two hybrid assays The *lovhk* (BAB2_0652) gene was amplified using primers Full_LOV_forward and Full_LOV_reverse. The 1.4 Kb PCR product obtained was cloned into pBT plasmid (BacterioMatch II Two-Hybrid System Kit, Stratagene) using NotI and XhoI. The resulting pBT construct expresses the *lovhk* gene in frame with the λcl gene. The *pdhS* gene was also cloned into pBT plasmid to be used as a positive- interaction control using primers PdhS_c-ter_forward and PdhS_c-ter_reverse. ### Cloning of RR genes for the two hybrid assays The *B*. *abortus* 2308 genome has 24 genes encoding proteins containing REC domains. All REC domains were cloned except for the gene that corresponds to a hybrid HK (BAB1_0346) and a pseudogene (ΨBAB1_1059). Primers listed in were used to amplify each gene. All forward oligonucleotides had the NotI restriction site, except for the BAB2_0630_forward that had an EcoRI restriction site. All reverse oligonucleotides had an XhoI restriction site. The PCR products digested with the corresponding restriction enzymes were cloned into the pTRG plasmid (BacterioMatch II Two-Hybrid System Kit, Stratagene) to obtain constructs in- frame with the RNAPα gene (pTRG). A minilibrary of REC-containing genes was created by mixing equal amounts of pTRG constructs. ### Cloning of *Brucella* genes for recombinant protein expression *lovhk* gene (BAB2_0652) was amplified with primers LOVHK_Full_NheI_Fw and LOVHK_Full_XhoI_Rev, containing the NheI and XhoI restriction sites respectively. The *hk* domain of *lovhk* was amplified with primers BaLOV_5Dhpt_NheI and BaLOV_3CAL_SalI. The *lovR* gene (BAB1_0099) was amplified with primers LovR_FWD_NheI and LovR_REV_XhoI. The *phyR* gene (BAB1_1671) was amplified with primers 1671_NdeI_FWD and 1671_His_Stop_REV. The *lovhk* and *hk* domain were cloned into pET24a (Novagen) expression plasmid, using the corresponding restriction enzymes. *phyR* gene was first cloned into pGEM-T Easy (Promega) generating the plasmid pGEM-PhyR, and subsequently subcloned into pET24a (Novagen) expression plasmid, using NdeI and NotI restriction enzymes. *lovR* was cloned into the pTrcHisB (Invitrogen) expression plasmid using the corresponding restriction enzymes. All primers are listed in. All the constructions were sequenced to confirm the absence of mutations. Plasmids expressing the different recombinant proteins were transformed into BL21 (DE3) pLysS. ## Bacterial two-hybrid assays Two-hybrid assays were carried out using the BacterioMatchII two-hybrid system (Stratagene) following the manufacturers’ instructions. This system uses an *E*. *coli* reporter strain (BacterioMatch II Two-Hybrid System Reporter Cells), and detects protein-protein interactions based on transcriptional activation of the His3 gene. The reporter cells were co-transformed with the corresponding plasmids and incubated at 28°C from 24 to 48 h until colonies were visible. Interactions were determined to be positive by growth on selective screening medium containing minimal M9 medium plus 5 mM 3-amino-1,2,4-triazole (3-AT), 25 μg ml^-1 chloramphenicol and 12.5 μg ml^-1 tetracycline, and validated by growth in dual selective screening minimal M9 medium plus 5 mM 3-AT and 12.5 μg ml^-1 streptomycin, 25 μg ml^-1 chloramphenicol, and 12.5 μg ml^-1 tetracycline (double positive clones). LOVHK was used as bait and a library with the 23 RR as prey. For the second two-hybrid assay, LOVHK was used as bait and a library containing 22 RRs (LovR was not included) as prey. In order to identify to which RR the double positive colonies corresponded to, the inserts were amplified by PCR and sequenced. Protein interactions in the two-hybrid assays were quantified using the FW102 O_L2-62 *E*. *coli* reporter strain as indicated in Carrica *et al*., 2012, except that protein expression was induced for 3 h. Interactions with the corresponding empty vectors were used as negative controls, and the interaction between strains co-transformed with pBT-GF2 and pTRG-Gal11 plasmids was used as a positive control. Both experiments were conducted under normal light conditions, and β-galactosidase activity was assayed using a standard Miller protocol. ## Expression and purification of recombinant proteins For LOVHK protein purification, culture was grown in minimal autoinducer medium (MD-50529) and only in this case all culture manipulations, protein purification and incubations were carried out under dark or dim red light conditions. *E*. *coli* cells expressing HK or PhyR recombinant proteins were grown in rich autoinducer medium (ZYM-5052) at 37°C for 3 h, and then incubation was continued at 18°C for 16 h. Cells were harvested by centrifugation at 8,000 x g for 20 min and resuspended in lysis buffer (20 mM Tris-HCl pH 8.0, 500 mM NaCl, 20 mM imidazole, 0.05% Triton X-100 and 1 mM phenylmethylsulphonylfluoride—PMSF), and disrupted by sonication with a probe tip sonicator (QSonica, LLC, Misonix XL-2000 series). The total cell lysates were centrifuged at 35,000 x g for 15 min and supernatants were incubated with nickel nitrilotriacetic acid-agarose resin (Ni-NTA agarose, Quiagen) with gently agitation at 4°C for 1 h. The resin was packed in a column, washed twice with washing buffer (20 mM Tris-HCl pH 8.0, 500 mM NaCl, 20 mM imidazole), and proteins were eluted with 5 ml of elution buffer (20 mM Tris-HCl pH 8.0, 500 mM NaCl, 200 mM imidazole). The fractions containing the corresponding proteins were dialyzed against dialysis buffer (20 mM Tris-HCl pH 8.0, 100 mM NaCl, 1 mM EDTA, 1 mM PMSF) and stored at 4°C until use. LovR was purified under denaturing conditions. Bacteria were grown in LB medium at 37°C for 3 h. Protein expression was induced with 1 mM isopropyl-thio- β-D-galactopyranoside (IPTG) and incubation was continued for 16 h at 18°C. Cells were harvested by centrifugation at 8,000 x g for 20 min, resuspended in lysis buffer with 8 M urea added, and disrupted by sonication. The purification protocol was continued as above with 8 M urea added to all buffers. Purified LovR was refolded by dialysis against dialysis buffer at 4°C, and then stored at 4°C until use. The purification procedures of all proteins were evaluated in 12–15% SDS-PAGE and stained with Coomassie Brilliant Blue. The PAS-HK-NtrY recombinant protein was expressed and purified as previously described and the recombinant REC domain of NtrX was a kind gift of Ignacio Fernández. For anti- PhyR antibody production, PhyR recombinant protein was further purified in a Superdex-75 gel filtration column (GE healthcare) equilibrated with buffer 20 mM Tris-HCl pH 8.0, 300 mM NaCl. The samples containing PhyR were pooled and concentrated by centrifugation using a centrifuge filter (Millipore). ## Phosphotransfer assays Purified LOVHK protein at a concentration of 2.5 μM was irradiated with 10 flashes of white light (a fluence of 4000 umol m-2 sec-1 of a xenon flashlamp) in phosphorylation buffer (20 mM Tris-HCl pH 8, 50 mM NaCl, 5 mM MgCl₂, 100 μM ATP) containing 10 μCi of \[γ-³²P\] ATP (111TBq/mmol, PerkinElmer Life Sciences). After 15 min at 37°C, purified RRs were added to the mixture to a final concentration of 2.5 μM each (final volume: 100ul). At the indicated times, aliquots of 10 μl were drawn and the reaction was stopped with an equal volume of 2X Laemmli sample buffer. Incubations were carried out under regular white light room illumination. Samples were separated by 15% SDS-PAGE, dried, exposed to a Storage Phosphor Screen (GE Healthcare), and scanned by a Storm 840 Molecular Imager (GE Healthcare) or exposed to a radiographic film. The intensity of each band was estimated using the ImageQuant 5.2 program (Molecular Dynamics) or Quantity One 4.6.3 Bio-Rad program. The relative intensity of each band to total intensity at time 20 seconds is reported in figures. ## Mutant-strain construction and complementation The *B*. *abortus* 2308 *lovhk*::*km* mutant strain was previously obtained by a kanamycin 00CAssette insertion. *∆lovR* and *∆phyR* mutant strains were obtained by clean deletion of those genes in *B*. *abortus* 2308 wt. Briefly, the 5´ and 3´ flanking regions of each gene were amplified with primers indicated in, and both fragments (containing complementary regions) were ligated by overlapping PCR using the flanking oligonucleotides. The resulting fragment was cloned into plasmid pk18mobsacB obtaining plasmids pk18mobsacB\_*∆lovR* and pk18mobsacB\_*∆phyR* respectively, which do not replicate in *Brucella*. Plasmids were transformed into *E*. *coli* S17-1 and transferred to *B*. *abortus* 2308 by conjugation. Single recombinants were selected by resistance to kanamycin and sensitivity to 10% sucrose in TSA plates. Double recombination events were selected by sensitivity to kanamycin and resistance to 10% sucrose in TSA plates. The excision of the plasmid and the generation of the mutant strains (named *∆lovR* and *∆phyR*, respectively) by allelic exchange were confirmed by colony PCR and sequencing of PCR products. For complementation assays, the *lovhk*::*km* mutant strain was conjugated with pMR10*cat*\_*lovhk* vector, which expresses *lovhk* under the control of its own promoter, obtaining the complemented strain *lovhk*::*km*/pMR\_*lovhk*. The pMR10*cat*\_*lovhk* vector was constructed by restriction free cloning method. Briefly, the *lovhk* gene with a fragment of 581 bp upstream of the start codon was amplified with primers pMR10_pLOVHK_F and pMR10_pLOVHK_R, and the resulting fragment was used as a mega-primer for a PCR using pMR10*cat* vector as template, obtaining the pMR10*cat*\_*lovhk* vector. This plasmid was transformed into *E*. *coli* S17-1 and transferred to *lovhk*::*km* mutant strain by conjugation. Bacteria bearing the pMR10*cat*\_*lovhk* plasmid were selected by resistance to nalidixic acid, kanamycin and chloramphenicol in TSA plates. The presence of the plasmid was confirmed by PCR. The complemented strain showed a slower growth in TSB compared to the wt and *lovhk*::*km* mutant strains. This effect is probably due to the expression of *lovhk* gene from a multicopy plasmid, which could be toxic as it has been previously reported for other HKs. ## Total RNA isolation from *Brucella* *Brucella abortus* 2308 wt and the mutant and complemented strains were grown in TSB medium at 37°C. An aliquot of bacterial culture in logarithmic phase was mixed with 1/10 volumes of stop solution (ethanol:phenol, 19:1) and left at room temperature for 2 minutes. Cells were harvested by centrifugation at 10,000 x g for 2 min, the supernatant was removed, and total RNA isolation was carried out with MasterPure RNA Purification Kit (Epicentre, Illumina) following the manufacturer’s instructions, with some modifications. The pellets were resuspended in 500 μl of 1X T&C lysis solution (containing 1 μl Proteinase K at 50 μg μl^-1 per 300 μl of 1X T&C lysis buffer), incubated at 65°C for 30 min with vortexing every 5 min, and then placed on ice for 5 min. Then, 0.25 ml of MPC solution was added to each tube and centrifuged at 14,000 rpm at 4°C for 10 min. The supernatants were mixed with 0.75 ml of isopropanol, and stored at -20°C for 16 h to 3 days. The samples were then centrifuged at 14,000 rpm at 4°C for 10 min. Pellets were air-dried and resuspended in 195 μl of DNAse buffer with 5 μl of DNaseI RNAse-Free, and incubated at 37°C for 40 min. Then, 0.2 ml of 2X T&C lysis solution was added to each tube and vortexed. 0.2 ml of MPC solution was added, vortexed again and placed on ice for 5 min. Samples were then centrifuged at 14,000 rpm at 4°C for 10 min. Then 0.5 ml of isopropanol was added to the supernatant and centrifuged at 14,000 rpm at 4°C for 10 min. The pellets were washed twice with 70% ethanol. Pellets were air-dried and then resuspended in 30 μl of RNase-free water. RNA was quantified using a Nano-Drop spectrophotometer (ND-1000, Thermo Fisher Scientific). DNA was subsequently removed by digestion with RQ1 RNase-free DNase (Promega) according to the manufacturer’s instructions. ## Real Time quantitative RT-PCR assays Reverse transcription was performed with SuperScript III First Strand Synthesis System (Invitrogen) following the manufacturer’s instructions using random decamer primers (Invitrogen) and RNasin ribonuclease inhibitor (Promega). The cDNA samples obtained were used as templates in qRT-PCRs. Primers used in the reactions were designed with Primer Express v3.0.1 3 (Applied Technologies) and adjusted by visual inspection. qRT-PCR reactions were performed with SYBR Green in 96-well plates in an Mx3005P instrument (Stratagene) or in a StepOnePlus instrument (Applied Technologies), and analyzed with MxPro3005P or StepOne programs respectively. The results of each target were normalized to *B*. *abortus* Translation Initiation Factor-1 (*if-1*, BAB1_0282) mRNA, and are presented as log₂ of fold induction relative to wt. ## Antibody production Specific antiserum against PhyR was raised in mice. Five mice were immunized on days 0, 15, 30 and 45 with 100 μg of PhyR recombinant protein mixed with Incomplete Freund’s Adjuvant (Sigma-Aldrich). After the fourth dose, mice were sacrificed and total blood was extracted. Anti-PhyR titer was determined by ELISA, and serum was stored at -20°C until use. All research involving animals has been conducted according to the National Institutes of Health Guide for the Care and Use of Laboratory Animals, and all experimental protocols have been approved by the Institutional Animal Care and Use Committee (IACUC) of Leloir Institute (Protocol number: 2009/08/41/FG). ## Western Blot analysis *B*. *abortus* cells were grown in TSB up to logarithmic phase, and equal amounts of bacteria of each strain were harvested by centrifugation for 2 min at 10,000 x g, resuspended in Laemmli sample buffer and inactivated by heating at 100°C for 20 min. These samples were then separated on a 15% SDS-PAGE and transferred to a nitrocellulose filter (Millipore). RibH1, a lumazine synthase (LS) isoenzyme, was used as loading control as it exhibits constitutive expression in *B*. *abortus*. Membranes were probed with primary polyclonal mouse antiserum anti-PhyR (1:5,000), or polyclonal rabbit antiserum anti-RibH1 (1:2,000) in PBS-Tween 0.05% and 1% milk, with gently agitation at 4°C for 16 h. Membranes were then incubated with secondary antibody HRP-conjugated goat anti- mouse (1:2,000) (Sigma A4416) or HRP-conjugated goat anti-rabbit (1:5,000) (Sigma A6154), with gently agitation at room temperature for 1 h. Blots were developed using Pierce ECL Plus Western Blotting Substrate (Thermo Scientific), following the manufacturer’s instructions. Signal intensity was measured using a Storm 840 Molecular Imager (GE Healthcare), and then quantified using the ImageQuant 5.2 program. ## Stress induced by starvation *Brucella abortus* 2308 wt and mutant strains were grown in TSB medium at 37°C, up to logarithmic phase (OD₆₀₀ ≈ 1–1.5). An aliquot of each bacterial culture was taken (time 0 h), and the rest of the culture was washed once with a modified MM1 minimal medium (57.3 mM K₂HPO₄, 35.9 mM KH₂PO₄, 0.1% w/v yeast extract, pH 7.0), and resuspended in the same medium. The cultures were incubated in a shaker at 37°C, and aliquots were drawn at 30 minutes, 1 h and 2 h. Total RNA was isolated from each sample and retro-transcribed as previously described. *phyR* mRNA expression was analyzed by qRT-PCR as previously described. ## PhyR stability This experiment is based on an experiment conducted by Kim *et al*., 2013, with some modifications as follows. *Brucella abortus* 2308 wt, *lovhk*::*km* and *∆lovR* mutant strains were grown in TSB medium at 37°C up to logarithmic phase (OD₆₀₀ ≈ 1). An aliquot of each bacterial culture was taken (time 0 h) and then, in order to inhibit general protein synthesis, chloramphenicol was added to a final concentration of 500 μg ml^-1. Cultures were incubated in a shaker at 37°C, and aliquots were drawn at 3 h and 5 h. PhyR protein level was analyzed by Western Blot. Relative protein level to time 0 h is shown for each strain. ## *VirB* promoter activity The *virB* promoter was amplified using primers pvirbup and pvirbdown, and cloned into pBBR-*lacZ* vector (a pBBR1MCS-4_Amp^R vector containing the *lacZ* promoter-probe cassette) generating the plasmid pBBR- prom-*virB*-*lacZ*, where the *virB* promoter is fused to the *lacZ* reporter gene. This replicative plasmid was electroporated in *B*. *abortus* wt, *lovhk*::*km*, *ΔlovR* and *ΔphyR*. For complementation assays, the pBBR- prom-*virB*-*lacZ* plasmid was introduced by conjugation into the *lovhk*::*km/*pMR\_*lovhk* complemented strain. The pMR10*cat* and pBBR1 based vectors are mobilized by RP4 and compatible with each other. Empty pBBR-*lacZ* was introduced in the wt 2308 strain and used as negative control. For promoter activity determination, two colonies from each strain were analyzed. Bacteria were grown in TSB from logarithmic to stationary phase, and two aliquots from each culture were drawn at each time. β-galactosidase activity was determined in each sample as previously described, with the following modification: the reaction mixture was centrifuged before A₄₂₀ determination. β-galactosidase activity is expressed in Miller Units \[*A*₄₂₀/(time x volume x OD₆₀₀) x 1000\]. This experiment was repeated three times. ## Statistical analysis Data are presented as mean ± standard error or standard deviation of the mean. Significance is reported using one-way ANOVA followed by the appropriate post- hoc test as indicated in the legend of each figure. All analysis was performed using the R statistical package. # Results ## Identification of two cognate RRs for LOVHK Two-component systems usually consist of two proteins: a sensor domain linked to a HK domain and a RR containing the receiver and output domains. Many HK and RR proteins are encoded in the same operon, evidence suggesting their functional interaction. In other cases, the HK and the RR are actually encoded as a single protein and are named hybrid HKs \[e.g. the LOV-HK-RR protein in *Pseudomonas syringae*, \]. In *Brucella*, LOVHK is encoded as an orphan histidine kinase with no obvious RR immediately downstream or upstream. In order to determine the identity of the RR functionally interacting with LOVHK, we carried out a bioinformatic analysis. Searches in *Brucella* genomic databases allowed us to identify 24 proteins and a pseudogene (ΨBAB1_1059) containing REC domains, the receiver-domain signature for RRs. One of these REC-containing proteins is a hybrid HK (BAB1_0346), three only contain a REC domain (BAB1_0099, BAB2_0628 and BAB2_0042), 18 are linked to a DNA binding domain and the remaining two are linked to a GGDEEF (BAB2_0630) and a sigma-like (BAB1_1671) domain. Based on the fact that the REC domain determines the specificity of interaction with a HK, we cloned all REC domains from the containing genes from *B*. *abortus* (excluding the hybrid HK) into a bacterial two-hybrid vector, and screened this collection using full-length *Brucella* LOVHK as bait. A positive control was included to test the functionality of this two-hybrid experiment. The PdhS HK gene which is known to interact with DivK was cloned into the same bait vector and used to screen the REC minilibrary. Both baits, LOVHK and PdhS, gave a similar number of positive clones, while with the empty vector very few colonies were obtained. Sequence analysis of five PdhS-double positive clones showed the expected DivK and 20 of the LOVHK-double positive clones showed that all corresponded to the BAB1_0099 gene from *B*. *abortus*. This gene encodes a small RR protein with only a REC domain. BAB1_0099 was named LovR as it is a single-domain RR that interacts with a HK protein containing a LOV domain, as in the case of LovR and EL_LovR from *C*. *crescentus* and *E*. *litoralis* respectively. However, there is low sequence similarity among them. In a second round, the BAB1_0099-containing plasmid was removed from the library and the screening using LOVHK as bait was repeated. The total number of positive clones was lower than the first round but higher than control. Ten colonies from the double positive clones were analyzed by PCR and sequencing: seven of them coded for the BAB1_1671 gene, two for BAB1_1538 and one for BAB2_0041. Previous results indicate that BAB1_1538 and BAB2_0041 interact with other HK proteins. The RR encoded by BAB1_1671 corresponds to PhyR, and it has been previously identified as part of the General Stress Response (GSR) system in *B*. *abortus*. We centered our study in the most relevant interactions observed: LOVHK with LovR, and LOVHK with PhyR. In order to test for the observed interactions between both RRs and LOVHK, equal amounts of each clone were co-transformed with the LOVHK bait vector into the FW 102 O_L2-62 *E*. *coli* reporter strain, which allows quantification of the interaction between the bait and prey by induction of β-galactosidase protein expression. Both RRs showed a significantly higher interaction with LOVHK than with the empty vectors used as negative controls. These results suggest that LOVHK interacts with two RRs: LovR and PhyR. ## Phosphotransfer from LOVHK to LovR and PhyR RRs are activated by the transfer of a phosphoryl group from a histidine residue in the HK domain to a conserved aspartic acid residue in the REC domain of the RR. Both LovR and PhyR contain the conserved aspartic acid that is the putative phosphor-acceptor residue (highlighted in yellow). In order to determine whether PhyR and LovR could be phosphorylated *in vitro* by LOVHK, we first conducted a time course for LOVHK autophosphorylation. After a brief white-light pulse, LOVHK phosphorylation reached saturation after 15–30 min in subsequent darkness. The unirradiated control showed a similar time course although the saturation level was lower than that of the illuminated sample. There followed a slow loss of phosphate for the illuminated sample over the next several hours. Then, we conducted phosphotransfer assays, by adding recombinant purified PhyR or LovR to a previously autophosphorylated LOVHK. Firstly, recombinant LOVHK was preloaded with phosphate by briefly irradiating it with white light in the presence of radioactive \[γ-³²P\]-ATP and incubating for 15 min prior to the addition of each RR. The 15 min wait insured strong LOVHK labeling at the time of RR addition. Both RRs were phosphorylated by LOVHK, although they presented different kinetics. PhyR phosphorylation becomes evident after 40 seconds of incubation and reaches a plateau after 20 minutes, coinciding with a drop in the phosphorylation level of LOVHK. LovR shows a fast increase in phosphorylation level reaching a maximum after 2 minutes and then becomes dephosphorylated in 10 minutes. When LOVHK is incubated with PhyR, the total amount of phosphate of the system remains constant suggesting that the phosphorylation level of this RR is stable. However, the addition of LovR strongly decreases the phosphorylation level of LOVHK and consequently the total amount of phosphate present in the system. This result suggests that the phosphate transferred to LovR is rapidly hydrolyzed. In order to test whether LOVHK is able to phosphotransfer to PhyR in the presence of LovR, we repeated the autophosphorylation labeling of LOVHK and added stoichiometric amounts of both RRs at the same time. The phosphorylation kinetics of LovR was similar to the pattern observed in the previous experiment where the RR was incubated with LOVHK. The PhyR protein was also phosphorylated although the kinetics was slightly slow as compared when this RR was alone with LOVHK, which led to a minor level of PhyR\~P. However, the presence of LovR did not completely inhibit the phosphotransfer to PhyR indicating that LOVHK is able to interact with both RRs simultaneously. Under similar conditions, the phosphorylated HK domain of LOVHK did not phosphotransfer to an unrelated *Brucella* RR, NtrX, and a similar result was obtained when PhyR was incubated with an unrelated *Brucella* HK, PAS-HK-NtrY. Taking together, these results strongly suggest that the interaction between LOVHK with both RRs is specific and that LOVHK is able to phosphotransfer to the central regulator of the GSR system in *Brucella*. ## LOVHK modulates the GSR response PhyR is one of the main components of the GSR system in α-proteobacteria, which has recently been described in *B*. *abortus* 2308 and other microorganisms belonging to the same group \[–, –\]. The GSR system positively regulates its own transcription, so that activation of the GSR system by a stress factor leads to an increase in the expression of genes belonging to the GSR system including *phyR*, *nepR* and *rpoE*. Stationary phase is generally associated with nutrient starvation, which is a stress factor and leads to the activation of the GSR system in many Gram-positive and Gram-negative bacteria. Therefore, bacteria in stationary phase are more resistant to changes in the environmental conditions compared to exponential phase. As a consequence, all experiments were conducted with bacteria under exponential phase, in order to avoid the stress- induction effect in stationary phase. In order to test whether LOVHK can modulate the GSR system in the absence of any stress, we estimated the basal activation of the GSR system by following *phyR* mRNA expression. Wild type and the isogenic *lovhk*::*km* and *∆lovR* mutant strains were grown in rich medium up to logarithmic phase and the amount of *phyR* expression was measured by qRT-PCR. We found that *phyR* expression was significantly decreased in the *lovhk*::*km* mutant compared to wt, while the *∆lovR* mutant did not show any difference. In order to verify if this effect was due to the absence of *lovhk*, we introduced a plasmid expressing this gene under the control of its own promoter (pMR10*cat*\_*lovhk*) in the *lovhk*::*km* mutant strain. The *phyR* expression of this complemented strain (named *lovhk*::*km*/pMR\_*lovhk*) was similar to wt and significantly different from the *lovhk*::*km*. In addition, we analyzed PhyR protein levels under the same conditions by western blot to assess whether the protein level was also modified in the mutant strains. As expected, we found that the *∆phyR* mutant strain did not express PhyR and that PhyR protein level was decreased in the *lovhk*::*km* mutant as compared to wt. However, it did not show a significant change in the *∆lovR* mutant strain. In the case of the *lovhk* complemented strain, it showed a PhyR protein level similar to the wt strain. Taking together, these results suggest that LOVHK is required for maintaining the basal expression level of PhyR in rich media. Furthermore, as LOVHK is a light sensor, we tested whether *phyR* expression varied between light and dark conditions in the wt strain. However, we did not find a difference in RNA and protein expression of PhyR (data not shown), even though we tried different periods of time of exposure to light and culture media. Kim *et al*., 2013, identified a series of genes that are regulated by the GSR system in *B*. *abortus*, including *rpoH1* (RNA polymerase factor sigma-32 RpoH1), *dps* (DNA starvation/stationary phase protection protein Dps) and *lovR*. Thus, we decided to investigate further whether the expression of these genes is also modified in the absence of LOVHK. The *B*. *abortus* 2308 wt and the isogenic *lovhk*::*km*, *∆lovR*, *∆phyR* and *lovhk*::*km*/pMR\_*lovhk* strains were grown in rich medium up to logarithmic phase. After RNA extraction, the expression of *rpoH1*, *dps* and *lovR* genes was measured by qRT-PCR. We found that expression of these three genes was significantly down-regulated in both the *lovhk*::*km* and *∆phyR* mutant strains compared to wt, and fully restored in the *lovhk*::*km*/pMR\_*lovhk* complemented strain. By contrast, *rpoH1* and *dps* gene expression was not modified in the *∆lovR* mutant. Moreover, *lovhk* basal expression was not modified in the *∆lovR* and *∆phyR* mutant strains compared to wt. These results suggest that LOVHK contributes to regulate the basal transcription level of genes regulated by the GSR system in *B*. *abortus in vivo*. In many α-proteobacteria, the GSR system responds to different stress factors including exposure to hydrogen peroxide, UV light, high osmolarity, nutrient starvation, or desiccation. We tested whether starvation could induce the GSR system. *Brucella* wt cells were grown up to logarithmic phase, transferred to a modified MM1 minimal medium and *phyR* expression was determined. After 2 h in this medium, bacterial viability was not affected and the GSR system was induced indicating that starvation condition can trigger the stress response. Thus, in order to test if the GSR system is activated by starvation in the mutants, *Brucella* cells were grown in rich medium up to logarithmic phase and transferred to a modified MM1 minimal medium, and *phyR* epression was analyzed at different time points. During the time of the experiment, bacterial viability was not affected. The *lovhk*::*km* mutant strain showed diminished expression at time 0 h as compared with wt strain. Starvation condition was able to induce the expression level of *phyR* in the three strains evaluated. However, the induction profile of *lovhk*::*km* mutant was delayed with respect to wt control. After 1 h of incubation the wt reached the fully induction state while in the *lovhk*::*km* mutant the induction level was still reduced. At 2 h, the *phyR* expression level was similar in the three strains. The *∆lovR* mutant strain showed a *phyR* induction profile not significantly different from wt, except at 0.5 h that showed a slight decrease. In conclusion, the absence of LOVHK does not prevent the GSR from responding to starvation, but LOVHK is needed to obtain the fully induction state in a short period of time under starvation stress. Kim *et al*., 2013, reported that PhyR protein levels decrease in the absence of stress by the action of the ClpXP protease. However, in the presence of stress PhyR protein levels are stabilized by phosphorylation and/or inactivation of ClpXP. Therefore, we decided to analyze whether LOVHK and LovR affect PhyR protein stability *in vivo*. *B*. *abortus* wt, *lovhk*::*km* and *∆lovR* mutant strains were grown in rich medium up to logarithmic phase and then the protein synthesis was inhibited by adding chloramphenicol. In the wt and *lovhk*::*km* mutant strains, the PhyR protein level decreased during the experiment with a similar profile. By contrast, in the *∆lovR* mutant strain PhyR protein level did not significantly changed during the 5 hours of the experiment. Although the effect seems to be modest, this result suggests that PhyR is more stabilized in the absence of LovR. ## Deletion of LOVHK affects the *virB* operon expression Swartz *et al*., 2007, found that the *lovhk*::*km* mutant is less infective than the wt strain in the J774A.1 macrophage cell line. However, neither *∆lovR* nor *∆phyR B*. *abortus* mutants showed significant differences compared to wt in cell infection assays (data not shown). Kim *et al*., 2013, also reported a similar result for the *B*. *abortus ∆phyR* and *∆rpoE1* strains in primary murine macrophage infection and initial colonization of BALB/c mouse spleens. However, they found that the GSR system is important for maintenance of a chronic infection in mice (more than 1 month p.i.). In addition, *B*. *melitensis* 16M *ΔrpoE1* mutant strain becomes attenuated in BALB/c mice at 4 weeks p.i.. While the GSR system in *Brucella* is important for persistence in mice and not at short times p.i., LOVHK is important during the early infection phases. Therefore, apart from the GSR system, LOVHK may be signaling to other intracellular components. Although there have been described many virulence factors in *Brucella*, the most characterized is the *virB* operon that encodes a Type IV Secretion System (T4SS). This secretion system translocates protein effectors from the bacterium to the cytoplasm of the eukaryotic cell, modifying the response of the host. Interruption of *virB* genes leads to complete loss of virulence both in cell cultures and in mice. Thus, v*irB* expression is crucial for macrophage infection, survival and establishment of the replicative niche inside host cells. When bacteria are grown in rich medium, *virB* is transcriptionally activated at the beginning of the stationary phase. Hence, we tested whether LOVHK could modify *virB* expression. For this purpose, we measured the activity of the *virB* promoter by a transcriptional fusion to the *lacZ* reporter gene in *B*. *abortus* 2308 wt and in *lovhk*::*km*, *∆lovR* and *∆phyR* mutant strains. Bacteria were grown in rich medium and β-galactosidase activity was assayed at different times from exponential to stationary phases. The activity of the *virB* promoter is induced when *Brucella* enters in the stationary phase in all strains. However, the induction of *virB* promoter showed a decrease of about 50% in the *lovhk*::*km* mutant as compared to the wt. While in *∆lovR* and *∆phyR* strains the activity of the *virB* promoter was very similar to wt. Complementation of *lovhk*::*km* with the plasmid expressing LOVHK protein completely restores the *virB* induction. This result suggests that deletion of *lovhk* decreases induction of the *virB* promoter by a GSR independent mechanism. # Discussion In a previous study, LOVHK was described as a blue-light sensor *in vitro*, as exposure to light increases its autophosphorylation activity. It was also found to be a virulence factor, as a *lovhk*::*km* mutant strain presents an attenuated phenotype in cell culture infections. Herein, we have characterized a novel and functional signaling pathway in *Brucella* where three components have been elucidated, LOVHK and two RRs: LovR and PhyR. The results presented in this work allow us to propose that *Brucella* LOVHK is signaling to the GSR system and affects the expression of the *virB* operon, an important factor in bacterial virulence. In *B*. *abortus* both *lovhk* (BAB2_0652) and *lovR* (BAB1_0099) are coded in different chromosomes, with no apparent partner in their vicinity. In *C*. *crescentus*, *lovK* and *lovR* are encoded in a single locus. The genome of *E*. *litoralis* encodes three *lov-hk* genes. *EL362_lovhk* and *lovR* are encoded in an operon, while *EL368_lovhk* and *EL346_lovhk* are orphan HKs, with no predictable cognate RR in the proximate region. Taking into account the genomic localization of these LOV-HKs and the single-domain RRs that interact with them, this information suggests that LOV-HKs cognate partners are not easily predictable, and that their identification needs to be studied carefully. Using bacterial two-hybrid assays and phosphotransfer experiments (, Figs), we identified two RRs as partners of LOVHK. Activation of LOVHK leads to an increase in autophosphorylation activity and phosphotransfer to its two RRs partners. Phosphotransfer from LOVHK to LovR presents a faster kinetics than transfer of phosphoryl moiety to PhyR. However, the phosphoryl group attached to PhyR\~P is more stable than the LovR\~P, which loses its phosphoryl group in a few minutes and leads to a rapid loss of LOVHK phosphate. Based on the *in vitro* phosphotransfer assays, we propose that LovR could be functioning as a phosphate-sink resulting in LOVHK inactivation *in vitro*. The LovR dephosphorylation could be due to self-catalyzed dephosphorylation or by a possible phosphatase activity of LOVHK. The mechanism of LovR dephosphorylation will be characterized in a further work. Previous works described analogous signaling systems also composed of a LOV- domain-containing HK belonging to the HWE family, a single-domain RR and PhyR in *C*. *crescentus* and *E*. *litoralis*. In *C*. *crescentus*, the relationship between LovK and both RRs, LovR and PhyR, was demonstrated by a genetic approach, while in *E*. *litoralis* the interactions among these proteins was studied by *in vitro* phosphotransfer assays. In *E*. *litoralis* the three LOV- HK proteins, named EL368_LOVHK, EL346_LOVHK and EL362_LOVHK differ in their phosphotransfer profiles. While EL368_LOVHK and EL346_LOVHK phosphorylate both RRs *in vitro*, EL362_LOVHK only transfers phosphate to LovR. The phosphorylation kinetics from *Brucella* LOVHK to PhyR is very similar to the kinetics from EL368_LOVHK to PhyR. In both cases, PhyR\~P signal intensity stabilizes after 20–30 minutes of incubation. However, the dephosphorylation kinetics of LovR\~P is faster in *Brucella* than in *E*. *litoralis*. We have not observed an increase in the dephosphorylation rate of PhyR\~P when LovR is present in the reaction such as was found in *E*. *litoralis* PhyR. Two possible scenarios could explain this observation: A) PhyR is dephosphorylated by LovR and LOVHK re-phosphorylates again PhyR protein maintaining the level of PhyR label; B) LovR only acts as a phosphate-sink for LOVHK and the phosphorylated PhyR is stable. Further work is needed to elucidate between these two alternative hypotheses. In *B*. *abortus*, in the absence of stress, LOVHK increases basal transcription of *phyR* and other genes regulated by the GSR system (Figs). Upon stress by starvation, the absence of LOVHK does not impair the final GSR response, but LOVHK is needed to obtain the fully induction state in a shorter period of time thus, confirming that LOVHK contributes to the activation of the GSR system. On the other hand, even though LovR decreases the phosphorylation state of LOVHK *in vitro*, contributing to return LOVHK to the inactive state, it does not have a significant impact on transcription of the GSR regulon *in vivo* (Figs,). The effect of mutating the *lovR* gene could be compensated by other not yet characterized mechanisms. However, the absence of LovR increases PhyR stability *in vivo*. This effect could be the consequence of a decreased level of LOVHK\~P when LovR is present, a condition that finally leads to lower phosphorylation levels of both RRs. Further work is needed to characterize in more detail the *in vivo* role of LovR. The expression of *lovhk* is not regulated by the GSR system in *B*. *abortus* under normal growing conditions, while *lovR* transcription depends on GSR activation. These results are in agreement with a previous microarray analysis showing that *lovhk* expression is not modified in the *B*. *abortus rpoE1* mutant as compared with the wt strain under oxidative stress conditions, while *lovR* expression is markedly decreased in the same conditions. Additionally, neither the *lovhk* nor the *lovR* promoters have the DNA sequence motif recognized by RpoE1, suggesting that *lovR* is only indirectly regulated by the GSR system. On the contrary, in *C*. *crescentus* the σ^T (the orthologue of RpoE1) recognition motif is present in the promoter of *lovK* and *lovR* genes, and their transcription is regulated by the GSR system showing that both genes are directly regulated by the sigma factor σ^T. The GSR system is responsive to many environmental conditions suggesting that PhyR integrates the input signals of several sensory HKs. In α-proteobacteria, the GSR system is characterized for a conserved genomic context, which involves the genes coding for the ECF sigma factor *rpoE*, *nepR* and *phyR*, and in many cases also includes one or two sensor HKs. Three of this kind of sensor HKs have been characterized as possible regulators of PhyR phosphorylation level: PhyK from *C*. *crescentus*, PhyP from *Sphingomonas* sp. Fr1, and RsiC from *S*. *meliloti*. In *Brucella*, the GSR genomic locus is also associated with two genes coding for putative HKs: BAB1_1669 and BAB1_1673. At the time of writing, both HKs have been analyzed and none of them phosphorylate PhyR. However, our results suggest that in *Brucella* other HKs could be activating PhyR, as the absence of LOVHK does not impair the GSR system to respond to stress under starvation conditions. *Brucella* has many virulence factors, although the most characterized is the T4SS *virB* system, which is essential for *Brucella* infection. The regulation of the *virB* operon is very complex, involving at least five regulatory factors. Herein, we show that the absence of LOVHK markedly decreases the *virB* operon promoter activity by a GSR independent manner. This result suggests that, in addition to modulation of the GSR system, LOVHK may be participating in other signaling pathways. Evidence in favor of this hypothesis is that *virB* expression was not modified in a microarray analysis conducted between *B*. *abortus* wt and the *∆rpoE1* mutant strain under stress conditions induced by hydrogen peroxide. Moreover, the consensus RpoE1 recognition motif was not found in the *virB* promoter from *B*. *abortus* and *B*. *melitensis*. These results suggest that the *virB* operon is not directly regulated by the GSR system. However, a decreased amount of VirB8 protein expression was reported in a *B*. *melitensis ∆rpoE1* mutant strain grown under normal conditions. Altogether, these results suggest that LOVHK may be controlling gene expression by at least two signaling pathways, and the mechanism by which *virB* expression is altered in the absence of *lovhk* remains unknown. At the time of writing, Gourley *et al*., 2014, demonstrated by a microarray analysis that genes related to the GSR system and *virB* operon are differently regulated in a *lovhk* mutant strain of *B*. *melitensis* as compared with wt. Here we further demonstrated a biochemical and physiological regulation of these genes in *B*. *abortus*, characterized the LOVHK signaling pathway and identified new components of the GSR system. In addition to other stress factors, light is also considered a stress signal. Even though it has been demonstrated that *Brucella* LOVHK is a blue-light sensor protein *in vitro* and that light modulates bacterial virulence in macrophage cell line infections, we could not demonstrate light as a modulator of the GSR system in bacterial cultures. However, the presence or absence of LOVHK modulates the GSR response. A similar result was obtained for *C*. *crescentus* LovK, and the authors suggested that LovK could be sensing the cytoplasmic redox potential through the LOV domain. Further experiments will test whether *Brucella* LOVHK is sensing another signal besides blue-light. Recent studies of Kim *et al*., 2014 suggest that the PAS domain of *Brucella* LOVHK is involved in the response to oxidative stress. *Brucella* can be acquired by aerosols, direct contact with mucosa or skin wounds with body fluids from infected animals, or by consumption of contaminated dairy products. Light sensed by bacteria present in aborted placenta may prepare the bacteria for infection of the following host. Kim *et al*., 2013, demonstrated that *B*. *abortus* PhyR\~P/NepR is a long-lived complex, so that when PhyR is activated by phosphorylation, bacteria would be prepared not only for the present stress factor, but also for future stresses, as has also been proposed for *M*. *extorquens*, and *B*. *japonicum*. In accordance with this model, light or some other signal sensed by LOVHK prior to infection may prepare *Brucella* for future stresses encountered in the host. However, in the absence of PhyR or RpoE1, alternative mechanisms may compensate in order to respond to stresses encountered in the host. In conclusion, this report provides important insights into the LOVHK signaling cascade. Our work identifies a novel TCS pathway in *Brucella*, composed of LOVHK, LovR and PhyR, and defines a functional role for the light sensor LOVHK. We also identified the first HK that interacts with PhyR in this bacterium, and we establish a connection between LOVHK, the GSR system and *virB* expression. Combining the results presented here together with the information from previous reports, we propose a model for LOVHK intracellular signaling pathway. This model is in concordance with what has already been described for the GSR system in *B*. *abortus*. Furthermore, this is the first mammalian pathogen for which a physiological role and intracellular interactions for a LOV-domain-containing protein have been described. # Supporting Information We thank Dr. Rodrigo Sieira for his assessment and suggestions; Dr. Gary Splitter for his helpful comments; Ignacio Fernández for sharing NtrY/NtrX recombinant proteins; and Dr. Xavier De Bolle for pMR10*cat* vector. F.A.G and G.P. are researchers of CONICET. G.S. is recipient of a fellowship of CONICET. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: GS GP. Performed the experiments: GS MCC TST GP. Analyzed the data: GS MCC TST GP RAB FAG. Wrote the paper: GS GP WRB.

# Introduction The presence of a physician before hospital arrival is believed to lead to effective cardiopulmonary resuscitation (CPR) for patients with out-of-hospital cardiac arrest (OHCA). Several studies have described ambulance crews staffed with a physician including cardiac ambulance crews, helicopter ambulance teams and physician-manned ambulance (PMA) teams comprised of two paramedics and an anesthesiologist. However, findings concerning whether the presence of a physician before hospital arrival leads to improved patient outcome during CPR are mixed. A randomized, controlled trial in the 1980s showed that the mortality rate in a group of patients for whom a physician was present was 35% lower than predicted, while the mortality in the non-physician group was as predicted. It has been pointed out, however, that the study design might have carried inherent bias. A study performed in Nottingham, England reported a trend towards improved survival in patients treated by PMA teams. However, the study might have been influenced by selection bias since the subjects were not randomized. In addition, it is unclear how the results of this 1970s study might apply to the current emergency medical service (EMS) system. A study performed in Taiwan reported improved survival in patients treated by non-PMA teams. However, the overall survival rate was only 1.4%, and the validity of the study has been questioned. According to a systematic review of non-randomized studies between 1990 and 2008, four of five studies in patients with out-of-hospital cardiac arrest (OHCA) revealed a significantly higher survival rate in the PMA group than in the non-PMA group. However, these studies were limited by several methodological issues, such as the use of samples from a single center and type II errors due to the inclusion of fewer than 100 patients. In these previous studies, the findings are inconsistent and the effectiveness of a physician's presence during pre-hospital CPR has yet to be established. To verify the effectiveness of the presence of a physician during CPR of patients with OHCA before hospital arrival, the influences of other factors, such as patients, bystanders, and CPR, need to be controlled. Although a randomized, controlled study should be performed, such a study is challenging because of ethical reasons. Therefore, using a propensity analysis, we analyzed OHCA cases using national data in Japan from 2005 to 2010 and determined whether the presence of a physician in the ambulance was associated with immediate and 1-month survival in patients with OHCA. # Methods ## Study design This was a prospective observational study using national registry data. The subjects were patients who were 18 years of age or older, had OHCA before the arrival of EMS personnel, were treated by EMS personnel and were then transported to medical institutions in Japan between 1 January 2005 and 31 December 2010. Using the national registry data in different time periods, we have reported the effects of epinephrine use and lactated Ringer solution use upon resuscitation outcome in patients with OHCA. The results of the present study derived from the included patients are new findings. The study was approved by the Ethics Committee at Kyushu University Graduate School of Medicine. The requirement for written informed consent was waived. ## Data collection In Japan, 47 municipal governments provide EMS through 807 fire stations with dispatch centers. All patients with OHCA who were treated by EMS personnel were transported to hospitals, excluding those with decapitation, incineration, decomposition, rigor mortis or dependent cyanosis because the Japanese guidelines do not allow EMS providers to terminate resuscitation in the field. Based on the standardized Utstein style template, the Fire and Disaster Management Agency (FDMA) registered all OHCA cases in Japan in a prospective, nationwide and population-based manner. In particular, data concerning bystander cardiopulmonary resuscitation (CPR), automated external defibrillator use and the components of CPR used by EMS personnel (*i.e.*, initial rhythm, defibrillation, intubation and epinephrine use) were collected using EMS records. The EMS person responsible for each patient with OHCA met the physician who treated the patient at the hospital and collected 1-month follow-up data. Then, the data from the 807 fire stations with dispatch centers in the 47 prefectures were electronically integrated into the national registry system on the FDMA database server. ## Ambulance crew An ambulance crew consisted of three emergency providers, including at least one emergency life-saving technician, but no physician. The certifying paramedic curriculum in Japan generally includes 180 h of lectures and practice in school and experience in 30 successful cases in the operating room under the instruction of an anesthesiologist. Emergency life-saving technicians are permitted to insert adjunct airways and to use semi-automated external defibrillators. With approval from an online emergency physician, specially trained emergency life-saving technicians have been permitted to insert intravenous lines since July 2004, and certified emergency life-saving technicians have been permitted to administer intravenous epinephrine since April 2006. A physician who happens to be with a patient when the patient collapses outside a hospital, or who happens to be in an ambulance for the training of the ambulance crew, might be engaged in pre-hospital cardiopulmonary resuscitation (CPR) until the patient's arrival at the hospital. In this study, the criterion for the presence of a physician in the resuscitation team was whether a physician rode in an ambulance from the scene of a patient's collapse to their arrival at the hospital. The presence of a physician was completely unplanned, and his/her role was not clear. If a physician accompanies the patient in an ambulance, the physician can independently perform advanced life support (ALS) (*i.e.*, perform tracheal intubation, insert an intravenous line and/or use epinephrine). It is possible that a physician might administer ALS in addition to basic life support, such as chest compression and rescue breathing, ECG analysis or team management in an ambulance. In this study, ALS was performed in 15.08% of all cases in which a physician rode with the patient in an ambulance. ## Variables The collected data included information on OHCA cases, CPR initiated by a bystander, life support provided by EMS personnel and patient outcome. Patients who survived cardiac arrest were followed for up to 1 month after the event, and data on the survival and neurological and physical status were collected. Neurological outcomes 1 month after successful resuscitation were evaluated using the cerebral performance category (CPC) scale, with five categories (1, good cerebral performance; 2, moderate cerebral disability; 3, severe cerebral disability; 4, coma or vegetative state; 5, death), and the overall performance category (OPC) scale, also with five categories (1, no or mild neurological disability; 2, moderate neurological disability; 3, severe neurological disability; 4, coma or vegetative state; 5, death). At 1 month after the cardiac event, the EMS person responsible for the patient with OHCA contacted the physician in charge of that patient and collected CPC and OPC data. These data were entered into the same national database. The variables used in the study are listed in. In particular, the etiology of cardiac arrest (*i.e.*, cardiac or non-cardiac) was determined clinically by the physician in charge, with the aid of EMS personnel. Because an automated external defibrillator (AED) analyzed the patient's rhythm automatically and delivered a shock only when it detected ventricular fibrillation (VF), the patient's first recorded rhythm was regarded as VF when laypersons delivered shocks with the use of a public access AED. Additionally, the category of VF included ventricular tachycardia (VT). The endpoints consisted of four types. These endpoints were return of spontaneous circulation (ROSC) before hospital arrival, survival at 1-month after cardiac arrest, 1-month survival with CPC category 1 or 2, and 1-month survival with OPC category 1 or 2. Of these, survival at 1-month after cardiac arrest was regarded as a major outcome measure. ## Statistical analysis The data that met the criteria concerning the patient age, time course and the presence of a physician in the ambulance were analyzed (*n* = 619,928). Using data for all 619,928 patients, three types of unconditional logistic regression models were fitted using the endpoints listed in as the dependent variables. Several variables have been shown to be predictors of the resuscitation outcome in patients with OHCA, including age, sex, bystander eyewitness, relationship of bystander to patient, bystander chest compression, bystander rescue breathing, use of public-access AED by the bystander, first documented rhythm and time from call to arrival at the scene. The effect of the presence of a physician was examined using three types of analysis models that differed in the degree of controlling for the effects of the covariates. Specifically, the first model did not control for the effects of the covariates, the second model controlled for the effects of the predictor variables, and the third model controlled for all covariates. Variability existed with respect to the quality of emergency medical services; thus, 46 dummy variables were introduced to 47 prefectures in Japan, and the third model controlled for the effect of the areas. In each analysis model, three types of subjects were used. The first model type used the total number of subjects. To exclude the effect of the number of ambulance crew members (*i.e.*, three vs. four), the second model type used cases with three- member ambulance crews. To evaluate the effect of the presence of a physician in the ambulance, the third model type used cases that excluded subjects in which the physician performed CPR but did not ride in the ambulance. Of the four primary outcome variables, 1-month survival was used for the sample size calculation. With an actual 1-month survival rate of 11.15% in the group accompanied by a physician in the ambulance and 4.98% in the group without the presence of a physician, 619,928 samples per group provided a power level of 100.00% with a type I error of 1%. The presence of a physician in the ambulance was not assigned randomly in the patient population; therefore, we developed a propensity score for the presence of a physician in the ambulance and controlled for potential confounding and selection biases. Without regard to patient outcome, the propensity score for the physician presence in the ambulance was determined using multivariate logistic regression analysis. Specifically, a full non-parsimonious logistic regression model was fitted with the physician presence in the ambulance as the dependent variable; the independent variables included all study variables except for endpoint variables plus dummy variables for the 47 prefectures in Japan (*i.e.*, 46 variables). A propensity score for the presence of a physician in an ambulance was calculated from the logistic regression equation for each patient. This propensity score represented the probability of the physician presence in the ambulance. Using the propensity score in the SAS Macro program by Parsons et al., cases in which a physician rode in the ambulance were matched to unique control cases in which a physician was not present. Using data on the propensity-matched subjects, five types of conditional logistic regression models were fitted, with each of the endpoint variables listed in as the dependent variable. To precisely evaluate the effect of the presence of a physician on a patient's outcome, the effect of every covariate needed to be controlled. In the propensity-matched sample, there were significant differences between groups with and without a physician with respect to several variables. In addition, several variables have been reported previously to be predictors of the resuscitation outcome in patients with OHCA. Thus, the effect of the presence of a physician was examined using five types of analysis model that differed in the degree of controlling for the effects of the covariates. Specifically, the first model does not control for the effects of covariates; the second model controls for the effects of propensity; the third model controls for the effects of propensity and significant variables in propensity- matched samples; the fourth model controls for the effects of propensity, significant variables in propensity-matched samples and variables reported to be predictors of the resuscitation outcomes ; and the fifth model controls for the effects of propensity and all covariates. Of the four primary outcome variables, 1-month survival was used for sample size calculation. With an actual 1-month survival rate of 15.61% in the group accompanied by a physician and 12.66% in the group not accompanied by a physician, 9,231 samples for each group provided a power level of 100.00% with a type I error of 1%. The significance level for all tests was *p*\<0.05 (two-sided). All analyses were performed using the SAS software (ver. 8.2; SAS Institute, Cary, NC, USA). # Results Between 1 January 2005 and 31 December 2010, 668,481 cardiac arrests occurred. Of these cases, 619,928 patients with OHCA met the inclusion criteria. The types of missing values in 37,171 cases are indicated in. With respect to the ‘total cardiac arrest cases in Japan between 1/1/2005 and 12/31/2010’ (*n* = 668,481), ‘assessed for eligibility 18–110 years old’ (*n* = 657,099) and ‘cases used for analyses’ (*n* = 619,928), the mean ages of patients with OHCA were 71.75±18.42, 72.34±16.58 and 72.91±16.32 years, respectively. The numbers (percentages) of female patients with OHCA among the three groups were 277,580 (41.4%), 277, 580 (42.2%) and 273,111 (44.1%), respectively. The numbers (percentages) of eyewitness cases were 272,331 (40.7%), 272,331 (41.4%), and 268,513 (43.3%), respectively. The numbers (percentages) of cases of cardiac origin were 369,131 (55.2%), 369,131 (56.2%) and 365, 608 (59.0%), respectively. shows the baseline characteristics of patients with OHCA according to the presence of a physician in the ambulance. There was no significant difference in the prevalence of OHCA by origin for the two groups (*p* = 0.35). With respect to the other variables listed in, there was a significant difference between those who rode in an ambulance car with a physician and those who did not. A significant difference was observed between those who rode in an ambulance with a physician and those who did not with respect to the prevalence of variables that are known predictors of the resuscitation outcome in patients with OHCA (*i.e.*, age, sex, bystander eyewitness, relationship of bystander to patient, bystander chest compression, bystander rescue breathing, use of public-access AED by a bystander, first documented rhythm and time from call to arrival at the scene). The time from the call to arrival at the scene was 7.38 and 7.31 min, respectively, among those who rode in an ambulance with a physician and those who did not. ## Physician's ambulance car ride and patient survival summarizes survival outcome based on a physician's presence during an ambulance car ride among three types of subjects. Among all subjects, with respect to the four types of outcome variables, there was a significant and positive association between a physician's presence and the four outcome measures in the unadjusted, adjusted for selected variables, and adjusted for all covariates models (all *p*\<0.001), except for 1-month survival in the adjusted for all covariates model, which showed a significant and negative association. Among the three ambulance crew groups, there was a significant and positive association between a physician's presence during an ambulance ride and ROSC before hospital arrival in the unadjusted, adjusted for selected variables, and adjusted for all covariates models (all *p*\<0.001). Among subjects, excluding cases for which a physician did advanced life support (ALS) but did not ride in the ambulance, there was a significant and positive association between a physician's presence during the ambulance ride and the four outcome measures in the unadjusted, adjusted for selected variables, and adjusted for all covariates models (all *p*\<0.01), except for 1-month survival, CPC (1 or 2) and OPC (1 or 2) in the adjusted for all covariates model. ## Physician's ambulance car ride and survival in propensity-matched patients To calculate the propensity score, a full non-parsimonious logistic regression model was fitted. This model yielded a *c* statistic of 0.85, which indicated a good ability to differentiate between cases in which a physician rode in the ambulance car and cases without a physician present. The propensity scores ranged from 0.003 to 1.000, which indicated that the probability of a physician's presence was between 0.003 and 1. In the study, 9,231 patients who were with a physician in the ambulance were matched to 9,231 patients who were without a physician in the ambulance. With respect to every predictor variable, except for the relationship of bystander to patient, the presence of an emergency life-saving technician in the ambulance car, ALS by a medical doctor (MD), and time from call to scene arrival, no significant differences were detected between groups with and without a physician. summarizes survival outcomes based on a physician's presence in the ambulance among propensity-matched patients. Of the four types of endpoint variables, a positive association was detected between a physician's presence during the ambulance ride and ROSC in all models (all *p*\<0.001). In terms of an association between a physician's presence during the ambulance ride and 1-month survival, a positive association was detected in all models, with the exception of the unadjusted models (*p*\<0.05 for the last model and *p*\<0.001 for the remaining models). In terms of an association between a physician's presence during the ambulance ride and CPC (1 or 2) or OPC (1 or 2), a positive association was detected in the models that adjusted for propensity and significant variables in the propensity-matched samples in, which adjusted for propensity and significant variables in the propensity-matched samples in and selected variables, and which adjusted for propensity and all covariates (*p*\<0.05 for the last model and *p*\<0.001 for the remaining models). # Discussion Based on a valid propensity analysis that controls for the effects of selection bias and confounding factors, we are the first to reveal that a physician's presence during the ambulance car ride is independently associated with short- and long-term outcome. Since previous findings were inconsistent, we believe that our findings are important both theoretically and practically. As for the possible positive effect of a physician's presence during an ambulance car ride upon resuscitation outcome, several possible reasons have been suggested. First, the presence of a physician improves outcomes of cardiac arrest because advanced procedures, such as airway management and epinephrine use, can be done by the physician. However, this possibility contradicts previous studies that found both intubation and epinephrine use to be independent predictors of poor outcome in patients with OHCA. Second, physicians are more likely to comply with treatment guidelines and possess up-to-date knowledge than other ambulance personnel. According to our analysis of CPR by physician's presence status, the following was revealed (see the Supporting Information): (1) of procedures for which effectiveness was demonstrated (*i.e.*, chest compression and defibrillation), multiple procedures were more frequently performed in the physician group than in the non-physician group; (2) a set of procedures that should be performed simultaneously (*i.e.*, initial shockable rhythm and defibrillation) were more frequently performed in the physician group than in the non-physician group; and (3) no use of ineffective procedures (*i.e.*, epinephrine use and ALS devices), along with the use of effective procedures, was more frequently used in the physician group than in the non-physician group. Thus, the present findings might be due to the second reason. In addition, physicians are reportedly more efficient in managing procedures such as ECG analysis and team management. The presence of a defined team leader, with experience and knowledge to provide oversight during resuscitation, could explain the increased focus on quality of care. Although this possibility might be applicable to the present study, we could not verify this point in the study. In summary, the better outcome in the group of patients accompanied by a physician in the ambulance might be related to the quality of medical care due to the physician's up-to-date knowledge and better compliance with treatment guidelines. Since analysis based on all subjects was influenced by the effects of selection bias and confounding factors, such results were not consistent with those based on propensity-matched subjects, except for the association between the presence of a physician and ROSC. However, since the results based on different analysis models agreed, the association between the presence of a physician and ROSC found among all subjects can be trusted. It is possible that the improved quality of CPR might not be due to the specific presence of a physician, as it might be due to having four persons on the PMA versus three persons on the first responding non-PMA. Among subjects whose crew member number was three, there was a significant and positive association between a physician's presence and ROSC before hospital arrival. These results suggest that the improved quality of CPR was, at least in part, due to the presence of a physician specifically. Of the 602,742 cases without a physician during the ambulance ride, there were 81,015 cases for which a physician did ALS. Among the remaining subjects, excluding cases for which a physician did ALS but did not ride in the ambulance car, there was a significant and positive association between a physician's presence during the ambulance ride and ROSC before hospital arrival. Thus, a physician's presence might have a positive effect on resuscitation outcome in patients with ROSC. Regarding the analytical method used in this study, because the proportions of ROSC before hospital arrival differed between groups with and without a physician in an ambulance, it might be speculative to suggest an additional effect on the other three types of endpoints \[*i.e.*, 1-month survival, CPC (1, 2) and OPC (1, 2)\]. However, long-term survival cannot be achieved without first restoring circulation. In addition, ROSC is used widely as a measure of short-term survival. Thus, in this study, ROSC was entered into the analysis model as an independent variable when evaluating the association between the presence of the physician in an ambulance car and long-term survival. Several limitations and caveats to our study must be acknowledged. First, we performed propensity analysis and made a rigorous adjustment for selection bias and confounding factors, which would be expected with a standard multivariate analysis. Nevertheless, since a physician's presence during an ambulance car ride was not assigned by random allocation, we need to acknowledge that we can only partially control and adjust for factors actually measured. Second, data on in-hospital CPR after hospital arrival were not included in the analysis. It is possible that our findings might have been influenced by differences in in- hospital resuscitation, such as hypothermia and mechanical chest compression devices, between those who were with a physician during the ambulance car ride and those who were not. Although the quality of in-hospital resuscitation might influence 1-month survival, we could not control for the effects of such factors. Third, the specialty(ies) of the physicians who rode in ambulance cars was unknown, and it is probable that a physician who was adept at CPR of OHCA patients rode in the car. If this was the case, then the association between the physician's presence and resuscitation outcome might have been over-estimated. However, we could not control for the effects of such factors. Fourth, to evaluate the effect of the presence of a physician in an ambulance on the resuscitation outcome of patients with OHCA, data are required regarding the survival rate after hospital discharge and 6 months later. However, we could not evaluate the effect of physician presence in an ambulance on these outcome variables due to the lack of relevant data. In summary, despite the limitations of this study, the associations between a physician's presence during an ambulance car ride and increased short- and long- term outcomes were consistent. Additional analysis also indicated that the presence of a physician was beneficial for CPR of patients with OHCA. Our findings should be verified by studies that include in-hospital resuscitation data. # Supporting Information [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: AH TN. Analyzed the data: AH TA YN. Contributed reagents/materials/analysis tools: MH. Wrote the paper: AH TA.

# Introduction Understanding the role of selection in the resistance against cancer should benefit from the study of animal populations. Domestic populations that develop spontaneous neoplasms similar to those encountered in humans have been increasingly investigated to unravel the complex genetic determinism of some cancers. Studies that examine cancer in natural populations are, however, quite rare, despite their potential for improving our knowledge of cancer resistance mechanisms in an ecological context. Observations made in this field are limited to a small number of unique cancer evolution cases, as those described in the naked-mole rat, the elephant, and the Tasmanian devil. In particular, horizontally transmitted cancers remain rarely observed events in nature, and so far they have been described in dogs, the Tasmanian devil and several species of bivalves. A well-known case of transmissible cancer is the one affecting the Tasmanian devil (*Sarcophilus harrisii*), the largest existing marsupial carnivore. This cancer, known as the Devil Facial Tumor Disease (DFTD), is characterized by a recent emergence, a high propensity to metastasize, and a mortality rate close to 100% within 12 months after infection. The disease was first detected in north-eastern Tasmania in 1996 and has since spread across 95% of the species’ range through biting injuries during social contact. DFTD has induced a rapid decline and a very strong selective pressure on the Tasmanian devil–with local population losses over 90%–giving rise to serious concerns regarding the species survival on the short term. DFTD has become a very well documented case study over the past decade. This amount of data is a valuable resource to better understand cancer biology and is expected to provide insight into the mechanisms underlying immunosurveillance of cancer initiation and metastatic spreading. A recent analysis of Tasmanian devil genomic time-series led to the identification of two regions exhibiting signatures of selection in response to DFTD which contain genes related to cancer risk and immune function in humans. Here, we extend this research by using a customized maximum-likelihood method that has been shown efficient for investigating rapid selection in experimental populations when genomic time- series data are available. In comparison to the approach used in, our method has in particular the advantage of efficiently disentangling the effects of strong selection from those of strong drift in a population-specific manner. In total, 97 genomic regions harboring 148 protein coding genes with human orthologues were identified. The functional analysis revealed that all the functionally characterized orthologues have a link with cancer and around 15% are involved in the functioning of the Central Nervous System (CNS), with about fifteen genes associated with behavioral disorders. Our data support the view that the evolutionary response to DFTD consists in several strategies that rely on a larger range of genetic variants than previously thought. These findings should contribute to a better understanding of the ecology and evolution of both the Tasmanian devil and cancer. In particular, this should help to pinpoint some genes that may provide host populations with a “resistance”–either by improving cancer survival or by reducing the risk of infection–to a transmissible cancer. # Results ## Signatures of selection in the Tasmanian devil genome Our analysis resulted in the identification of 97 signatures of selection dispersed throughout the chromosomes, accounting for about 0.3% of the genome. Most signatures of selection were found to be population-specific, and only one was common to the three investigated populations. The West Pencil Pine (WP), Freycinet (FN) and Narawntapu (NP) populations displayed 54, 38 and 37 signatures of selection, respectively. The ability to detect signatures of selection from the dataset was influenced by the experimental sampling protocol and SNP detection, which was specific to each population. In the WP population, characterized by an adequate time-series sampling but with a low SNP density, a large number of small signatures of selection were detected. In the NP population, despite a high SNP density, a low proportion of candidate SNP (0.4%) was detected. This can be explained by the non-optimal sampling time-series, with the latest data point too close to the date of arrival of the disease, which left little time for selection to produce visible effects. The FN population presented a much better SNP density and sampling time-series, compared to the other populations, which allowed detecting the highest number of candidate SNP (210) and larger signatures of selection. As a consequence, a larger proportion (68%) of candidate regions were located less than 100 kb from a protein coding gene in the FN population. In total, for the three populations, 60 signatures of selection were detected in the vicinity of protein coding genes with a human orthologue (see and for the details) allowing us to identify 148 candidate genes according to Ensembl genome browser 90. ## The functional annotation of candidate genes reveals a strong link with cancer We used the IPA knowledge base analysis tool (Ingenuity Systems®, [www.ingenuity.com](http://www.ingenuity.com/)) to identify biological functions and disease-related categories associated with our candidate genes. The top molecular and cellular functions included several “cell-related functions” such as cell cycle, morphology and organization. Other functions were related to nucleic acids, primary metabolism and the immune system, as shown in. Among the top 100 disease-related categories, 73 were associated with “cancer”. The “solid tumor” category was the most overrepresented (p-value = 1.16×10⁻¹⁰_, FDR = 4.39×10⁻⁷) with 138 genes out of 147 associated with this term. According to IPA, all the 60 signatures of selection were located in the vicinity of coding sequence host genes potentially related to cancer. Some candidate genes are key regulators of signaling pathways mediating cancer progression involved in cell cycle, apoptosis and genome instability. Among these candidates are the DTWD1, NEK6 and NSD3 genes that are regulators of the G2/M transition, and BAG4 and TRIM66, which prevent apoptosis. The CEP131 and PINX1 genes are involved in chromosome stability. A great number of candidate genes are associated with signaling pathways that usually mediate embryonic development but may also have a role in cancer progression. Alterations of the FGFR1 gene, a tyrosine-kinase receptor (RTK), have been described in many tumors. Other candidate genes are known to be involved in oncogenic RTK pathways, such as SHC4 and ST5/DENND2B. The CEP131 gene is involved in cell proliferation and migration through the activation of the phosphoinositide 3-kinase (PI3K/Akt) signaling and the SMAD3 gene is a key mediator of the transforming growth factor-β (TGF-β) signaling pathway involved in cancer progression. The candidate FOXN3 is a key gene of the Wnt/β-Catenin pathway that is altered in cancers. Other candidates, such as the SPTBN1, NSD3 and CDK14 genes, are able to influence cancer progression through Wnt signaling. Key mediators of other pathways related to both development and cancer, such as Notch and Hippo, were among our candidates. The NOTCH2 gene is involved in the direct cell-cell interactions of the Notch pathway that can inhibit cancer progression. The TEAD4 gene belongs to the TEAD (transcriptional enhancer factor domain) transcription factors that are required for activating the proliferation genes targeted by Hippo signaling in cancer. ## An important subset of candidate genes are mediators of metastasis It is noteworthy that a great number of candidate genes identified in the present study have the potential to influence invasiveness and metastasis of cancer cells. In particular, we identified seven metastasis-related genes that encode adhesion molecules (ADGRA2, ADGRD2, CDH8, MCAM, THY1, TSPAN9, and TSPAN11). Other candidates, such as PRRX2 and CST3, confer metastatic properties to cancer cells through the TGF-β pathway. Similarly, FOXN3, HMGCS2, LSM1, PHGDH, RIN2, TRPM8, are known regulators of metastatic processes. In total, over 30 selection candidates identified in this study are known to participate in metastasis-related mechanisms. ## Behavior-related genes among the identified candidate genes The second major functional category associated with our candidate genes refers to the architecture of complex behaviors. Functions associated with development and the nervous system were revealed by IPA analysis. In particular, 16 candidates were classed within the annotation term “development of neurons”. Several of them may contribute to the development and homeostasis of important cellular compartments in the central nervous system (CNS). For example, SYNDIG1 is essential for the formation of excitatory synapses in the hippocampus. NKX2-2 is a key regulator of serotonergic neuron development. SDK2 is an adhesion molecule that regulates synaptic connections, thereby influencing the arrangement of neural circuits in the CNS. KIF13B, a member of the kinesin motor protein superfamily, has a key role in the regulation of axon development. Other candidates have putative roles at the synapse level. SLITRK5 and SLC38A10 play important roles in neurotransmission. CDC42EP4 and HERC1 are involved in synapse homeostasis. In addition, the candidate list harbors several genes encoding subunits of ion channels (e.g., CACNB2, KCNA4, KCNIP3, or KCTD3) that are involved in signal transmission in the CNS. Some human homologues of our candidate genes are also associated with behavioral disorders. For example, the large signature of selection displayed on scaffold GL834709 (nucleotides 2,702,597 to 2,946,330) corresponds to human locus 8p11.23, which is related to cancer, but has also been proposed as a “neurodevelopmental hub” associated with autism spectrum disorders (ASD) according to a recent meta-analysis. The NDUFAF5 candidate is involved in the assembly of the first complex of the respiratory chain, which is frequently impaired in CNS disorders like ASD. Several other candidates, such as CACNB2, CDH8, HERC1, KCTD3, KIF13B, SERINC2, and SLC39A11 have been associated with ASD. Some candidates have been linked to intellectual disability (ID), such as CRBN, GPKOW, HERC1 and KCNA4, or to other behavioral disorders, such as CDC42EP4 and SLITRK5. ## A few candidates may contribute to immunity Immunity, a function that may be related to cancer progression, is also represented through a few genes. In particular, six candidates (TRIM10, TRIM15, TRIM26, TRIM39-RPP21, TRIM62, and TRIM66) are members of the large tripartite motif (TRIM) family that consists of ubiquitin ligases involved in both innate immunity and cancer progression. With the exception of TRIM39-RPP21, all the TRIM genes identified here are known to be related to cancer according to IPA. TRIM10, TRIM15, TRIM26, and TRIM39-RPP21 belong to the human leukocyte antigen (HLA) region, which gathers many immunity regulators. TRIM62 and TRIM66 have also been suggested to have a role in immunity. Candidates from the TSPAN (TSPAN9, TSPAN11) and the aGPCR (ADGRA2, ADGRD2) families, which were cited above for their putative roles in metastasis, may also be implicated in immune response. The versatile SMAD3 candidate has been shown to influence the immunosurveillance of cancer. # Discussion The Tasmanian devil dataset investigated here was initially presented and analyzed by Epstein et al., who reported two putatively selected regions harboring seven candidate genes. These two regions were also detected in our analysis and correspond to signatures of selection located in scaffolds GL841593 (nucleotides 4,501,785 to 4,979,756) and GL849657 (nucleotides 283,671 to 283,701). Epstein et al. performed their analysis with a composite test statistic that took into account temporal changes in both allele frequency and integrated extended haplotype homozygosity of an individual SNP site (iES). Such an approach is based on the idea that an SNP undergoing strong selection would rapidly rise in frequency, while the haplotypic diversity of the surrounding region would suddenly decrease. To avoid false-positives, Epstein et al. restricted their analysis to the candidate regions shared by the three populations. The limit of this approach is that the total number of candidate genes detectable is highly dependent on the SNP density, especially since the West Pencil Pine (WP) population presented a low genotyping density with only \~ 5000 available SNP. We chose a different approach than that proposed in. First, the three devil populations were analyzed separately since there is substantial genotyping heterogeneity among populations. In addition, the genetic structuration of the Tasmanian devil implies that some observed polymorphisms are expected to be population-specific. In particular, the existence of population-specific genotypes has already been emphasized to account for some phenotypic differences in response to DFTD among devil populations. Second, we used a customized method that looks for selection in each population independently. Our method is based on a Wright-Fisher model coupled to maximum-likelihood computations, which makes it possible to determine whether the observed temporal variation in SNP frequency is more likely to be caused by selection and drift than by drift alone in each population. Importantly, our method implements a test statistic that takes into account the amount of drift through the effective population size (*N*_*e*), in each population. Despite the very low *N*_*e* in the Tasmanian devil (*N*_*e* \~ 30, see), our method allowed the identification of 97 signatures of selection. This confirms that the selection imposed to the Tasmanian devil during the emergence of DFTD was extremely intense and that the evolutionary response to a transmissible cancer may involve the contribution of many genes, which is in line with the complex multistep processes of cancer. Our analysis identified a majority of population-specific candidate regions for DFTD-linked selection. The low number of candidate region overlapping among populations may be explained by different factors. The first relates to the large sampling heterogeneity in genotyping and in SNP densities among populations, as demonstrated by the fact that only 5% of the SNP were common among populations. The second reason involves the complex host-cancer interaction, which strongly depends on the genetic variation available in the host population. We can expect selection to act on different genes in different populations because between-population genetic variation exists. This does not necessarily imply that the adaptive mechanisms selected to resist cancer are dramatically different among devil populations; rather, this refers to the redundancy of gene functions, with different genes or different pathways being able to act in a similar and functional manner. The Tasmanian devil-DFTD interaction is a nice example of host-pathogen reciprocal evolutionary process. DFTD-driven selection acts to increase host resistance and reduce the negative effects of the tumor on individual fitness. In an evolutionary antagonistic process, DFTD evolves to counter-adapt in response to the host adaptive changes. Our study, as in the previous analysis by Epstein et al., focuses on the genetic changes arising in the host, rather than the tumor itself. DFTD-driven selection has targeted genes in the host cells of the tumor microenvironment, where non-cancerous cells such as fibroblasts, adipocytes, inflammatory cells, etc., contribute to the malignant progression. Genes selected in the host may therefore limit normal cell recruitment and activation by the cancerous cells, and regulate metastasis-related processes. Through our analysis, we identified more than 30 candidate genes that may contribute to metastasis. In particular, some of them belong to different families of cell adhesion molecules (e.g., tetraspanins, cadherins, adhesion G protein-coupled receptors, immunoglobulins), which are essential in processes that lead to metastasis. For example, both MCAM and THY1, which encode cell adhesion molecules of the immunoglobulin superfamily (IgSF-CAMs), are frequently overexpressed in metastatic tumor tissues. Moreover, tetraspanins and adhesion G protein-coupled receptors also influence metastasis as well as the cadherin CDH8. These results must be considered in the light of the high prevalence of DFTD metastases in infected devils and the importance of cell adhesion processes in shaping the tumor microenvironment. All this suggests that the control of metastasis, in particular through the maintenance of the host tissue integrity, could be an essential component of the host evolutionary response to DFTD. In addition to the strong association with metastasis mentioned above, the functional annotation of our candidate genes allowed identifying putative key mediators of cellular processes frequently deregulated in cancers, such as cell cycle control and apoptosis. Well-known signaling pathways related to both cancer and development, such as MAP-kinase, TGF-β and Wnt pathways, were also targeted by selection. In a marsupial such as the Tasmanian devil, the development of neonates represents a critical window for selection. As a consequence, we must consider the possibility that selection has targeted some genes for their developmental role or even for their specific tumor-suppressive action during development. Overall, our analysis suggests that natural selection may have targeted multiple cellular circuits to limit the acquisition of cancerous features in devil tumor tissues, which should ultimately prevent or at least delay cancer growth and metastatic processes. In the context of a transmissible cancer, such genetic changes are expected to provide the host with increased fitness, since host individuals with a slow DFTD progression will be able to reproduce despite infection. Among the target genes identified in this study, there are several human orthologues associated with behavioral disorders affecting synaptic connections and characterized by deficits in communication and social interaction, such as ASD and ID. This observation is very intriguing, especially in the light of recent results indicating that more aggressive individuals were at greater risk of developing DFTD. DFTD-driven selection could therefore have favored less aggressive individuals, which are less likely to be infected due to reduced physical interactions with other Tasmanian devils. This empirically supports the hypothesis recently discussed in that an evolutionary response to cancer could rely not only on cellular pathways involved in cancer progression, but also on adjustments of life-history traits and behavior. Even if evolutionary costs driven by sexual selection may be opposed–because socially dominant individuals engage more often in mating–, Roche et al. suggest that “the avoidance of contagious cancers could be a selective force for specific behavior”, which agrees with our results. The Tasmanian devil-DFTD interaction provides a good natural model for further investigating this hypothesis through additional field and resequencing studies. Overall, our results suggest that DFTD has extensively shaped the genome of the Tasmanian devil and that several functions have been targeted in the host by DFTD-driven selection. This supports the evidence that adaptation occurs rapidly even in situations of limited genetic variation. Our study also shows that genomic time-series data are particularly useful for detecting signatures of selection associated with complex phenotypes even when the number of generations of selection investigated (\~ 5 generations) and the effective population size (*N*_*e* \~ 30) are very small. In this study, we only had the opportunity to explore the host perspective of the host-tumor interaction. Further studies incorporating genomic time-series data from both the host and the transmissible tumor will be required to better understand the host-tumor coevolution. # Methods ## Tasmanian devil dataset The data analyzed in the present article were initially reported in and made publicly available as a Dryad data package. This data package consists of SNP genotyping data produced by Stacks following RAD-seq (Restriction-site Associated DNA sequencing) assays from tissue samples collected in 360 Tasmanian devils across Tasmania. Samples were collected at different time points between 1999 and 2014, allowing the analysis of genomic time-series that take into account the impact of DFTD through time on Tasmanian devil populations. We restricted our analysis to the same samples as in, from the localities of Freycinet (FN), Narawntapu (NP) and West Pencil Pine (WP). We reproduced the SNP filtering strategy reported in. In brief, we performed filtering according to (i) MAF computed over the whole dataset (SNP with MAF less than 0.01 were discarded), (ii) observed heterozygosity computed over the whole dataset (SNP with heterozygosity over 0.5 were discarded), (iii) the proportion of missing genotypes (SNP with less than one-third of genotypes either in the whole dataset or in a sample were discarded, except for the two smallest samples in which SNP with less than half genotypes were removed), (iv) the linkage disequilibrium between neighboring SNP (using PLINK, we removed one SNP from pairs of SNP harboring *R*^*2*\> 0.99 over 20 successive SNP and 50 kb of distance in any sample). We obtained filtered datasets of 16978, 27173 and 5401 SNP for the FN, NP and WP populations, respectively. Relevant information about the dataset for the present work is summarized in. Further information about sample collection, genotyping and data processing can be found in. ## Signatures of selection We submitted the genomic time-series of the three investigated Tasmanian devil populations to a customized method for detecting footprints of selection (available at <https://github.com/hubert-pop/signasel>). This method was initially described and successfully implemented to detect genomic regions targeted by short-term selection in experimental wheat populations. Briefly, this method compares two Wright-Fisher models, one including drift and the other including drift plus selection, in a maximum-likelihood framework. The model that best fits the SNP frequency variation observed over time is identified through a Likelihood Ratio Test (LRT). Each SNP is individually tested and associated with a p-value that quantifies to what extent the temporal variation in SNP frequency may be due to selection, under a null hypothesis postulating an effect of drift only. A strength of this method is to model and disentangle the effects of drift plus selection as opposed to those of drift alone in a population-specific manner. This has proven to be efficient for detecting SNP under selection from genomic temporal samples separated by a few generations in small populations undergoing intense selection, as in the Tasmanian devil. For example, we performed simulations mimicking a strong selection (applying a constant selection coefficient of 0.5) in the FN population (considering a constant effective population size, *N*_*e*, of 34, and a sampling interval of 6 generations). In such a case, the power to detect a truly selected SNP was 61%, while the false-positive rate was about 1%, suggesting that our method would provide enough power to find signatures of selection from the Tasmanian devil dataset. In the real data analysis, we corrected for multiple testing by applying the Benjamini-Hochberg false-discovery rate (FDR) procedure. Therefore, we considered as relevant signatures of selection the genomic regions that included at least one SNP with a p-value \< 0.0001 (which corresponds to a FDR of \~ 8%) or at least two neighboring SNP with p-values \< 0.01 (which corresponds to a FDR of \~ 13.5%). We looked for candidates for selection by applying our method to two temporal samples in the FN (the sample from 1999 and the combination of those from 2012 and 2013) and WP (the sample from 2006 and the combination of those from 2013 and 2014) populations, and to three temporal samples (samples from 1999, 2004 and 2009) in the NP population. Given the sampling times and an assumed generation time of 2 years in the Tasmanian devil, we considered that the time-series covered a period of 6, 5, and 3 complete generations in the FN, NP, and WP populations, respectively. As our method relies on Wright-Fisher models, the effective size (*N*_*e*) of each investigated population must be provided. Simulations indicated that a bias in the estimation of *N*_*e* may under some circumstances affect the power to detect selection but has almost no impact on the false-positive rate (data not shown). We used the values of *N*_*e* suggested in, that is, 34, 37 and 26 for the FN, NP and WP populations, respectively. These estimates come from the Jorde and Ryman method for inferring unbiased contemporary *N*_*e* from genetic data. ## Candidate gene identification We identified as candidate genes all the protein coding genes standing at less than 100 kb from the detected signatures of selection and having a human orthologue according to the Ensembl genome browser 90 ([https://www.ensembl.org](https://www.ensembl.org/)). Ensembl stable ID of candidate genes and corresponding orthologues were retrieved using the Ensembl Genes 90 database and the Devil_ref v7.0 Tasmanian devil reference assembly obtained from BioMart. ## Ingenuity pathway analysis We used the manually curated database IPA® (Ingenuity Pathway Analysis, QIAGEN Inc., <https://www.qiagenbioinformatics.com/products/ingenuity-pathway- analysis>) to identify the diseases and developmental disorder as well as the molecular and cellular functions associated with our candidate genes. We submitted the list of 147 human orthologues of our candidate genes to a “Core Analysis” with the “Ingenuity Knowledge Base” reference set to find overrepresented functions and diseases in our gene set. This is achieved by IPA by applying a right-tailed Fisher Exact test to estimate the likelihood that the overlap between the set of genes and a given function or disease is due to random chance. # Supporting information The authors wish to thank Wendy Brand-Williams for linguistic revision of the manuscript and Andrea Rau for useful suggestions. [^1]: The authors have declared that no competing interests exist.

# Introduction The home range is defined as the area used by an animal during its normal activities. Establishment of spatially confined home ranges, which may also define an actively defended territory, is a widely observed pattern in nature. The exploration, extension and stability of home ranges have fundamental ecological and evolutionary consequences, for example, by determining where predator-prey or intra-specific agonistic interactions occur. The mechanistic idea behind the home range concept is that an animal moves following random stimuli (i.e., diffusion movement) but with an added tendency to remain around a specific point, which constrains the fraction of the available potentially suitable habitat to one that is actually used. Among the different mechanistic movement models that have been proposed for describing home range behavior in animals, biased random walks are probably the most widespread. Describing the drift that constrains the animal around the center of the home range by a bivariate Ornstein—Uhlenbeck (OU) process dates back at least to 1997, and this specific implementation has been repeatedly used since then providing mechanistic descriptors of home range behaviour for a range of wild-living animals. Understanding the exact mechanisms driving the establishment of home ranges is not only relevant from a fundamental perspective of behavioral ecology, but can also inform the effectiveness of spatial management actions, such as the design of protected areas. However, the methods usually used to obtain positional data in aquatic systems via telemetry suffer from substantial observational error. Addressing this methodological issue is crucial to generate reliable inferences about the drivers of the home range establishment in nature. Global positioning systems do not work in aquatic environments. Hence, alternative biotelemetry methods have been proposed for positioning fish and other aquatic animals, such as satellite tracking. However, the positioning error caused by geolocation in satellite-based biotelemetry applications is usually large (up to several km), which reduces the usefulness of this technology for the fine-scale mechanistic studies of home range behavior in coastal or freshwater fishes thriving in smaller lakes or river sections. Alternative acoustic telemetry systems have been developed for the study of the behavior of marine coastal fishes. In such applications, an acoustic transmitter is implanted in the fish that emits a periodic series of ultrasonic pulses that are eventually detected by one or more submerged receivers. The standard data that one obtains is a time-series of detections (including of course many missing data) from an array of spatially spaced receivers. Generating precise positions is only possible by examination of multivariate time-series resulting from arrays where hydrophones are located reasonably close to each other. Improved acoustic telemetry systems have recently been designed for high resolution fine-scale behavioral studies based on automated time synchronization of acoustic signals received by various hydrophones (e.g., VPS system from Vemco^® or the MAP system by Lotek^®,). However, the most commonly used system for learning about movement behavior in marine fish currently consists of arrays of fixed automatic omnidirectional receivers without a fine time synchronization. A detection event occurs when a tagged fish with a transmitter is sufficiently close to the receiver. The fish’s position is still frequently interpolated from the position of the receivers that have detected the fish over a predetermined period of time or time-step. However, this procedure may result in biased positioning and may induce incorrect conclusions regarding the characterization of the home range behavior or, in general, any other characteristics of fish’ movement. Moreover, there is substantial evidence that the probability of detection is not only function of the distance between a fish and a hydrophone but that it can also be highly influenced by several environmental factors that affect sound propagation in water, such as water currents, tidal phase or environmental noise. All these factors—either isolated or combined—may introduce bias when positioning an animal and they can in turn affect the inference of the mechanisms of home range behavior. Therefore, it has been recommended to set acoustics transmitters at known positions within the array (which are known as beacon tags or control tags) for calibrating the environmental effects on the probability of detection. To make use of information from these control tags, it is important to develop novel statistical methods that are able to incorporate observational error when the aim is to infer precise positional data from acoustic tracking. Only then can the movement mechanisms that generate a given home range pattern be accurately estimated and the between-individual variability in home range behaviour be properly described. In this context, state-space models (SSM) have emerged as one of the most promising tools to study animal movement in the wild because they nicely combine a process model (i.e., the movement model that predicts fish position at any time) with an observation model (i.e., the model that properly infers the fish position from the data generated by the tracking system). In SSM, the process model predicts the future fish position given its current position and the mechanistic properties of the movement model, while the observation model provides the probability of obtaining a particular observation (i.e., the number of detections events per receiver per time unit) conditional on the true (and unobserved) fish position. The environmental-related changes in the probability of detection can be monitored through control tags located at known positions within the array, and they are included in the modelling process through the observation model. Pedersen and Weng developed an innovative SSM approach that combines a bivariate OU movement model with an appropriate observational model for acoustic tracking data, and demonstrated its robustness and usefulness for estimating the parameters characterizing the home range movement in a coral reef fish. Here, based on the same conceptual SSM proposed by Pedersen and Weng, we present an alternative Bayesian fitting strategy for estimating the parameters of an OU movement model (exploration rate, location and size of the home range). Pedersen and Weng’s SSM solution is based on frequentist statistics, while ours is based on a Bayesian framework. Our methods are not meant to revive the frequentist- Bayesian debates. Instead, our approach should be considered as a convenient alternative for those used to the technicalities and the Bayesian way of reasoning. The Bayesian SSM developed here is highly flexible and it properly deals with the data-sets produced by acoustic tracking arrays. Moreover, it is easily customizable by any other end-user because an R-code is provided associated with our paper. We developed the model framework and tested its robustness via extensive computer simulation before fitting it to a real data- set of a small-bodied coastal fish. The case study presented here is deliberately simple, but it is representative for many small-bodied sedentary coastal species (e.g., those inhabiting reefs or other temperate coastal habitats) in the sense that many of the coastal species have relatively small and stable home ranges. # Materials and Methods ## Ethics Statement The real data-set is composed of a collection of acoustic detections of wild free-ranging pearly razorfish, *Xyrichtys novacula* tagged with acoustics tags in 2011. The capture and tagging of the individuals were authorized by those responsible for marine natural resources and the Marine Protected Area (MPA) of Palma Bay (Mallorca Island), the Fisheries Department of the Balearic Islands, through a permit to the CONFLICT Project (ref: CGL2008-00958) and to the REC2 Project (ref: CTM2011-23835), both of them funded by the Spanish Ministry of Science and Competiveness. Our study did not involve endangered or protected species, and no animals were sacrificed. Acoustic tags were attached to fish after anesthetization with MS-222, and all efforts were made to minimize fish handling and harm. ## Theoretical assumptions The SSM developed here assumes that actual fish positions constitute a hidden (unobserved) Markovian state variable that must be estimated from the pattern of detection events on each of the acoustic receivers while following a predetermined mechanistic movement model. Receiver detections of a sound signal emitted by a fish are assumed to constitute stochastic events that depend not only on the distance between the fish and the receiver but also on environmental variables affecting sound propagation. Therefore, our approach combines two different modules: *(i)* the fish movement model, and *(ii)* the observational model. ## Fish movement module: a process model to describe the mechanistic pattern leading to the establishment of a home range The most widely used model for describing animal movement are random walks (RW). Many different forms of RW have been used to describe the different types of movements encountered in different scenarios and species. The RW case is uncorrelated, i.e., the direction of movement at a given time step is independent of the previous directions of motion, which means that the location at a specific time step depends on the location at the previous time step plus a random term. Moreover, RW assume no bias, i.e., there is no preferred direction of movement. Movement under such circumstances is Brownian, and the pattern produced at the population level is standard diffusion. Simple RWs are, therefore, not a reasonable choice for describing the movement of the increasing number of fish species for which relatively small and temporally stable home ranges have been reported. In these cases, animals do not move freely within large patches of suitable habitat. Instead, there is a need for an additional, possibly memory-driven behavioral rule, according to which each individual will tend to be attached to a specific site. Such a movement within relatively small home range can be described by an OU process. Accordingly, fish move within a harmonic potential field, the strength of which describes the extent of their home range. The rationale behind this model is that fish still move within a homogeneous environment following random stimuli (e.g., food patches or predatory threat), but this rule is combined with a tendency to remain around a specific point, designated as the center of the home range. Specifically, we consider that the trajectory of a fish, *r(t)* = *(x(t)*, *y(t)*), is described by the stochastic Langevin equation: $$\overset{\rightarrow}{r}(t) = {\overset{\rightarrow}{r}}_{H}(t) + \overset{\rightarrow}{\Delta}(t)$$ where ${\overset{\rightarrow}{r}}_{H}$ denotes the position of the center of the home range at time *t*. In general, the center of the home range can be dependent on *t*, expressing for example that during day-time the fish wanders around a particular feeding place, while it is constrained to a different spatial area during night-time. The displacement $\overset{\rightarrow}{\Delta}(t)$ from the instantaneous center of the home range at any time is given by the OU process $$\frac{d\overset{\rightarrow}{\Delta}(t)}{dt} = - k(t)\overset{\rightarrow}{\Delta}(t) + \overset{\rightarrow}{\xi}(t)$$ which represents a fish that is attracted toward the center of its home range by a central harmonic force of instantaneous strength *k(t)*, while it is also subjected to an external random force. The random force is described by the Langevin term $\overset{\rightarrow}{\xi}(t)$, which is a bi-dimensional, white Gaussian process of zero mean, variance (*ε*) in each spatial coordinate and no correlation among them (, but see for an alternative definition that may translate in elliptical home range). Again, the time dependence of *k* and *ε* expresses that the fish behavior may change across *t*, for instance some species exhibit two different states (e.g., foraging or resting type of movement). The general solution of is: $$\overset{\rightarrow}{r}(t) = {\overset{\rightarrow}{r}}_{H}(t) + e^{- Q(t)}\left\lbrack {{\overset{\rightarrow}{\Delta}}_{0} + {\int\limits_{0}^{t}{\xi(t\prime)}}e^{Q(t\prime)}dt\prime} \right\rbrack$$ where *Q(t)* is given by: $$Q(t) = {\int\limits_{0}^{t}{k(t\prime)dt\prime}}$$ A suitable discretization (*t* = *nΔt*) of the fish trajectory described by Eqs and is given by: $${\overset{\rightarrow}{r}}_{n + 1} = {\overset{\rightarrow}{r}}_{n + 1}^{H} + e^{- (Q_{n + 1} - Q_{n})}\left( {{\overset{\rightarrow}{r}}_{n} - {\overset{\rightarrow}{r}}_{n}^{H}} \right) + {\overset{\rightarrow}{R}}_{n}$$ where ${\overset{\rightarrow}{r}}_{n}^{H}$ denotes the position of the center of the home range at time *t* = *nΔt*, $$Q_{n + 1} - Q_{n} = {\int\limits_{n\Delta t}^{(n + 1)\Delta t}{k(t\prime)dt\prime}}$$ and ${\overset{\rightarrow}{R}}_{n}$ is a stochastic, normally distributed, variable with zero mean and standard deviation (*σ*): $$\sigma_{n} = \sqrt{\frac{\varepsilon_{n}(1 - e^{- 2k_{n}\Delta t})}{2k_{n}}}$$ Eqs to apply to the general case in which ${\overset{\rightarrow}{r}}^{H}$, *k* and *ε* may be time-dependent and define different behavioral states. However, here we develop the simplest case applied to species with diurnal active life-styles. For example, our case species, the pearly razorfish remains inactive and buried in the soft bottom during night-time (see below for more details of the biological model selected for the real data-set). When the parameters of the movement model are constant (e.g., during day-time in the pearly razorfish), Eqs – simplify, and the movement of the fish can be described by $${\overset{\rightarrow}{r}}_{n + 1} = {\overset{\rightarrow}{r}}_{}^{H} + e^{- k\Delta t}\left( {{\overset{\rightarrow}{r}}_{n} - {\overset{\rightarrow}{r}}_{}^{H}} \right) + {\overset{\rightarrow}{R}}_{n}$$ where ${\overset{\rightarrow}{R}}_{n}$ is a stochastic, normally distributed, variable with zero mean and standard deviation (*σ*): $$\sigma_{} = \sqrt{\frac{\varepsilon_{}(1 - e^{- 2k\Delta t})}{2k}}$$ The biologically relevant effect is that the movement of the fish is stochastic within a given spatial area surrounding the center of the home range. The “radius” of the circular home range (*radius*, the radius of the area within which a fish has a 95% probability of being found when a large period of time is considered) depends on *k* and *ε*: $$radius_{} = \sqrt{\frac{- \varepsilon_{}\text{~ln}\left( {1 - 0.95} \right)}{k_{}}}$$ Palmer et al. developed the biological interpretation of this specific version of a random walk for marine coastal fishes. While the size of the circular home range (*radius*) depends on the ratio *ε*/*k* and determines the potential size of the space use of the individual in meters, the parameter *k* of the model is the rate of exploration (in min^-1), which determines the slope of the curve describing the cumulative space used in function of time or how quickly the individual explores the whole home range. Thus, *k* represents the speed by which an individual moves through its home range. ## Observational module: modelling the probability of detection using control tags The second module of our SSM deals with the observational model. As commented above, the true fish positions $\overset{\rightarrow}{r}(t)$ are unobserved. Instead, the only information obtained by an array of acoustic receivers consists of a detection pattern (i.e., how many detections are registered during the *n* time-steps by each of the acoustic receivers of the listening array, or *ND*_*n*,*j*). The probability of detecting a signal (*PD*_*n*,*j*) is described by a logistic function of the distance $d_{n,j} = \sqrt{{(x_{n} - x_{RECj})}^{2} + {(y_{n} - y_{RECj})}^{2}}$ between the true fish position at *n*, (*x*_*n*, *y*_*n*), and receiver *j* (*j* in R receivers), located at (*x*_*RECj*, *y*_*RECj*), $$Log\left( \frac{PD_{j,n}}{1 - PD_{j,n}} \right) = \alpha + \beta\left\lbrack \sqrt{{(x_{n} - x_{RECj})}^{2} + {(y_{n} - y_{RECj})}^{2}} \right\rbrack$$ ## Parameter estimation Given the input data (a matrix consisting in the number of detections, *ND*_*n*,*j*, at each one of the *j* receivers during *n* time-steps), the goal is to estimate both the value and uncertainty of the movement parameters (${\overset{\rightarrow}{r}}^{H}$, *k* and *radius*). In the fish movement module (upper level), we used the movement model defined by Eqs and, and for the observation module (lower level), we used. Concerning such an observation module, it is well known that a detection event mainly depends on the distance between a fish and a receiver (*α* and *β*) but it is also influenced by environmental conditions. These dependencies are explicitly modelled and estimated from the input data, and *α* and *β* were estimated in our application using a control tag moored at known distances from each of the receivers. The temporal scale at which *α* and *β* should be estimated is case- specific. In our case, with no tidal variations, a daily scale was chosen for simplicity (see below). This means that the values *α*_*day* and *β*_*day* were considered constant at the within-day scale. Again, for simplicity, *α*_*day* and *β*_*day* were estimated in preliminary and independent statistical analyses, and were considered fixed and supplied as data to the Bayesian model detailed below. The movement parameters (and uncertainty) of the SSM were estimated using a Bayesian fitting strategy. The model was implemented and run using the R2jags library of the R package (<http://www.r-project.org/>), which opens JAGS (<http://mathstat.helsinki.fi/openbugs/>). Three Markov Chains Monte Carlo simulations (MCMC) were run, and minimally informative prior knowledge was assumed. Specifically, *k* was assumed to follow a uniform distribution bounded between zero and 1 min^-1. The *radius*, and the latitude and the longitude of the center of the home range were assumed to follow a normal distribution. In all four cases, the parameters of the prior distributions ensured a nearly flat prior distribution. In addition, for demonstrative purposes, we compared the posterior distributions resulting when minimally informative priors were used with those obtained when biological information is available for setting the priors, and when alternative prior distributions were imposed. The first 10,000 iterations for all of the parameters were discarded (burn in period), and a thinning strategy was adopted to ensure the temporal independence of successive values within the chain (only one out of every 10 consecutive values was kept). The convergence of the MCMC chains of all parameters was assessed by visual inspection of the plots of the iterations and tested using the Gelman-Rubin Statistic, with values \< 1.1 indicating convergence. Convergence was reached after a variable number of iterations. Depending on the simulation or the real case of tagged fish (see below), between 3,000 and 9,000 valid iterations were retained after burning and thinning for describing posterior distributions. A fully customizable R-code (corresponding to one simulation experiment; see below) is provided in. ## Precision and accuracy of the analytical approach: simulation experiments Before applying the Bayesian SSM described here to a real data-set, the accuracy and precision of the estimations and the effect of the prior distribution in the posteriors were checked via computer simulation. The simulation experiments were aimed at disentangling the effects of two issues when estimating the movement parameters: *(i)* different combinations of movement parameters in which the exploration rate (*k*) and the *radius* of the home range varied mirroring the between-fish variability observed in the real study-case (sim 1 to 4), and *(ii)* the effect of the observational time-step, defined as the fraction of time where the detections are pooled (5, 10, 15, 30, 60 and 90 min). Note that in all the simulation experiments, the transmitter emitted one acoustic signal per minute, which is the actual emission period in the case of the pearly razorfish (PT-2, Sonotronics, Inc., Tucson, Arizona, USA;). Therefore, the simulated fish was moved every minute and the detection (or not) by each of the receivers in the array was checked with the same periodicity. However, depending on the specific simulation experiment, the detections were pooled in different time-steps as mentioned above and would be typical in real applications. Two series of simulations were conducted. In the first, we generated a total of 24 fish trajectories using Eqs and considering four realistic combinations of movement parameters (*k* and *radius*;), which were analysed after pooling the number of detections at the 6 different observational time-steps. A simulated squared array of 25 evenly spaced (300 m) omnidirectional receivers and a sequence of *α*_*day* and *β*_*day*, estimated using control tags, both inspired by the settings of the acoustic tracking study described by Alós et al., were used to generate a matrix of acoustic detections per time-step (ND_*n*,*j*) for each one of the 24 simulated trajectories. The tracking period was 12 days. However, to mirror diel behavior of the razor fish, the fish was moved only during 14 light-hours per day, which means that a fish path lasted for 10,080 positions (or 12×14×60 min). Therefore, the data input for the Bayesian analyses of each simulation experiment consisted of a multivariate temporal series of number of detections per time-step, which was a matrix of 25 columns (receivers) by 2016, 1008, 672, 336 or 168 rows (for a time-step of, respectively, 5, 10, 15, 30, 60 and 90 minutes). Concerning the detection probability, we generated a time sequence of *α*_day and *β*_day based on the between-day variability currently observed, which has been assessed by fitting the number of detections actually obtained by control tags moored in a known position in the array of acoustic receivers. In our case, the choice of a daily scale for the detection probability is justified by the low temporal variability observed in the real case study; this scale should be modified to fit the case specificities of future acoustic tracking studies. In summary, for each of the 24 simulations, fish were moved within the array and a new position was defined after one minute (Eqs). The probability of detection by each one of 25 omnidirectional receivers was then calculated as a function of *(i)* the distance between the fish and the receivers and *(ii)* the day- specific values of *α*_*day* and *β*_*day*. These predicted probability values were compared with random values extracted from a uniform distribution between zero and one to simulate detection (or not). Finally, the number of detections by each receiver was cumulated according to the specific time-step. We then fitted the Bayesian SSM to the input data produced by these 24 simulation experiments and the estimated movement parameters (posterior median and Bayesian Credibility Intervals, BCI, 2.5% and 97.5% for *k*, *radius* and the position of the center of the home range) were compared with the true (known) values. The second series of simulations aimed assessing the accuracy and the precision of the Bayesian model, in particular the effect of the priors in the posterior distributions. These second series of simulations were focused in the most extreme scenarios in terms of movement parameters (sim 1 and sim 4). The goal was to obtain precise estimates of accuracy and precision considering a time- step of 15 min and 30 min according to the first block of simulations (see below for relevant bias starting at time-steps of 30 min). Three simulations scenarios were then considered: *(i)* sim 1 with a 30 minutes time-step, *(ii)* sim 4 with a 30 minutes time-step and *(iii)* sim 4 with a 15 minutes time-step in line with the results of the first set of simulations. In these three cases, instead of a single fish, the simulation experiment was repeated for 50 fish to obtain 50 replicates of each simulation. The percentages of simulations where the known parameters were properly estimated (where the estimated BCI of the parameter included the true value) were quantified. Finally, the outcomes of imposing different priors were evaluated using a single simulation by comparing the posterior distribution and a set of five different prior distributions. We focused only on the case of the *radius* of the home range because data were more easily available for this parameter. In the case of the pearly razorfish, the 95% of the kernel utilization distribution occurred within an averaged (between-fish) accumulated area and s.d. of 0.32 ± 0.13 km² which is equivalent to a radius of 314 ± 67 m. By providing a fish with a true radius of 245 m (sim 1 above), the BCI were compared after setting five different priors. ## Case study—the pearly razorfish, *Xyrichtys novacula* The Bayesian SSM was applied to a collection of acoustic detections from an acoustic tracking experiment done in 2011 where the movement of several pearly razorfish was monitored for a short period of time (\~20 d: the length of tracking period is limited by the battery life span, which in turn is limited by the fish size) using an array of 21 omnidirectional acoustic receivers (model SUR-1, Sonotronics, Inc., Tucson, Arizona, USA) in the waters of Mallorca Island, NW Mediterranean (see the details of the receivers array in). The pearly razorfish is a small protogynous monandric hermaphrodite with marked sexual dimorphism and prefers habitats characterized by sandy soft bottoms; the species is highly targeted by the recreational fisheries in temperate areas in the Mediterranean. We selected six tracked individuals in 2011 that generated sufficient data following the decision-tree criteria to discard potential mortalities described in March et al.. Moreover, it is well known that after the implementation of the acoustic tag *X*. *novacula* show a short period of abnormal behavior during which the fish remain buried in the soft. Accordingly, we used Continuous Wavelet Transformations (CWT) using the *sowas* library in R-package to detect the normal behavior to set the initial day of the time-series of acoustic detections following Alós et al.. We only considered the day-time detections as the pearly razorfish remains inactive and buried in the soft bottom during the night-time, which prevents detections. The resulting time series of detections for each tagged fish are presented in and are representative of the typical data generated in this type of acoustic tracking study based on arrays of omnidirectional receivers. We fitted the Bayesian SSM to these 6 tagged fish, and the posterior distribution of the movement parameters (latitude and longitude of the center of the home range (${\overset{\rightarrow}{r}}^{H}$), *radius* and *k*) and their uncertainty (BCI) were summarized for each individual. Following the results of the simulation exercise, we decided to fit the Bayesian SSM considering a time-step of 15 min. # Results ## Simulation experiments The Bayesian SSM retrieved the movement parameters of the simulated fish trajectories with acceptable precision and accuracy in most of the combinations of movement parameters (sim 1, 2, 3 and 4) and for the six time-steps that were considered. The Bayesian Credibility Intervals (BCI, 2.5% and 97.5%) were usually tight, and in most cases (84.4%) the true value was within the BCI, which indicated that the results obtained with the Bayesian approach were accurate in most of the cases. Regarding the estimation of the latitude and longitude of center of the home range (${\overset{\rightarrow}{r}}^{H}$ in meters), the model fit yielded BCIs that included the real value in almost all cases. However, the uncertainties associated with estimating ${\overset{\rightarrow}{r}}^{H}$ were larger in sim 1 (*k* = 0.001 min^-1 and *radius* = 245 m) and sim 2 (*k* = 0.001 min^-1 and *radius* = 387 m) than in sim 3 (*k* = 0.01 min^-1 and *radius* = 245 m) and sim 4 (*k* = 0.01 min^-1 and *radius* = 387 m). Regarding the exploration rate (*k* in min^-1), the Bayesian SSM retrieved BCIs that included the true value except for the two largest time-steps (i.e., 30, 60 and 90 min), where *k* was overestimated. Moreover, precision of *k* was smaller in sim 3 and 4, suggesting positive relationship between uncertainty and how fast the individual explores the home range. The estimations regarding the *radius* were similar to the results obtained for *k*: most of the estimated BCIs included the true value, with the exception of the two largest times-steps (60 and 90 min) where the *radius* was overestimated. In the case of the *radius*, the uncertainties associated with the estimation of the parameter were similar for all combinations. Overall this first series of 24 simulation experiments suggested a good performance and accuracy of the Bayesian SSM unless the time-step is large (30, 60 or 90 min), especially for the simulations involving large exploration per unit of time (i.e., high *k* in simulation runs 3 and 4). Moreover, the Bayesian SMM also seems to properly retrieve the original trajectory of the simulated fish, which supported a good performance for positioning the fish. The results obtained from the second series of simulation experiments (50 replicates) confirmed the general picture depicted in the first series of simulation experiments for a single fish. When *k* and *radius* were small (sim 1), the largest time-step considered (30 min) had no or small effect, and the estimates of the movement parameters were accurate, which suggested a good performance of the Bayesian SSM. However, when *k* and radius were larger (sim 4), the largest time-step produced slightly biased (overestimation) estimates for *k*, while the other parameters remained unbiased. The overestimation observed in *k* was notably reduced when smaller time-steps (15 min) were used, and the percentage of BCI containing the true value raised from 8% to 78% suggesting a better performance of a 15 min time-step (or smaller) than 30 min (or larger) when *k* and radius were large. The outcomes of imposing different priors were evaluated using a single simulation. For a uniform distribution bounded between 0 and 10,000 m, the estimated BCI of the *radius* (real value 245 m) was 185 and 405 m. For a normal distribution with zero mean and a large variance (tolerance = 10⁻⁸), the BCI was 178 and 417 m, and for a normal distribution with mean = 314 m and the observed between-fish variance, the BCI was 198 and 357 m. Finally, for a normal distribution with mean = 314 m and a variance ten times larger than the observed between-fish variance, BCI was 193 and 357 m. Therefore, when minimally informative or reasonable priors were assumed, posteriors were largely narrower than priors, and BCIs were similar and included the true value. Conversely, for a normal distribution with mean = 100 m and a narrow variance (tolerance = 0.01), BCI was 133 and 157 m. Therefore, as expected, when very informative, but biased priors were assumed, posterior distributions did not include the true value. ## Case study of pearly razorfish The results of the simulation experiments above suggested that a time-step of 15 minutes or less provided the most accurate and precise estimates of the movement parameters in most of the simulation cases that were evaluated. As computation time exponentially increases with the size of the input data, we considered a time-step of 15 min for fitting the real data set (\~ 3 h per individual). The Gelban-Rubin statistic, which assessed the convergence of the model parameters, was below of 1.1 in all cases. shows the BCIs of the estimated movement parameters for the 6 individuals of pearly razorfish that were analysed. The BCIs did not overlap in many cases, suggesting the existence of individual differences in the movement parameters. displays the estimated trajectories of the individuals tracked showing different patterns of home range behavior. # Discussion A particular movement behavior that constraints the animal within a small area or home range has several ecological, evolutionary and managerial consequences for many exploited fish species. With the recent miniaturization of acoustic tracking devices, fisheries scientists and ecologists have now a suitable tool for disentangling the mechanisms behind the behavior that constraints fish within a home range or, more generally, behind any movement behavior, even for small-bodied fish such as the pearly razorfish. However, given that the data collected by arrays of omnidirectional acoustic receivers (i.e., number of detections per time unit) are only indirectly related to the fish’s position, the estimation of the movement parameters is often imprecise. This may in turn limit the biological interpretation of acoustic tracking data. In this paper, we presented a Bayesian approach for fitting a SSM, which joins a mechanistic home range movement model (in our case, a random walk weighted by an OU process), with an appropriate observational model. This observational model predicts detection probabilities not only from the distance between a fish and the receivers, but it also takes into account that the distance-related detectability can be affected by environmental factors. Related work combining movement and observational models based on a frequentist solution for estimating the movement parameters in a SSM have recently been published. The Bayesian solutions we propose follows this research and constitute a suitable alternative for those used to the Bayesian way of reasoning. With both a frequentist and a Bayesian SSM approach available, the methodological ground is developed for improving the mechanistic understanding of home range behavior and its ecological and evolutionary consequences across a wide range of animals. SSMs are among the most promising analytical tools for analyzing movement data generated from tracking systems, including acoustic applications. All SSM are based on combining two models: *(i)* the process and *(ii)* the observational model. Concerning the process model, the specific movement model we choose for describing the movement behavior of our study species is able to unravel the behavioral mechanisms behind the emergence of a spatially confined home range, not only in terms of its size (*radius*), but also the location of the center of the home range, and to study how quickly a fish explores its home range (parameter *k*,). These parameters can be interpreted as individual traits that might be under selection by natural or anthropogenic forces. Moreover, the SSM modelling approach is flexible and it can easily accommodate other types of movement behaviors if they fit the specificities of the tracked species (e.g., correlated random walks following an environmental driver,). One of the key improvements of SSM is the incorporation of an observational error module. In the same way that is demonstrated in, our approach allows for the explicit incorporation of the effects of environmental variability over the general pattern of a distance-dependent detection probability. Specifically, the detection probability of an acoustic pulse transmitted by a fish was described by a distance-dependent logistic model, shown in and elsewhere. The parameters of the logistic curve (*α* and *β*) can be estimated at the desired temporal scale using a control tag (as in our case), thereby providing a solution for addressing the variability in the probability of detection related to environmental factors. One of the main novelties of the approach proposed here is technical rather than conceptual, in the sense that the model parameters are estimated using the Bayesian machinery. Bayesian inference has been widely proposed as an efficient way to estimate the parameters of complex movement models with large number of parameters, as is the case in the SSM proposed here. Another potential advantage is that the Bayesian inference method allows combining existing knowledge (through prior probabilities) with additional knowledge derived from new data (through likelihood) to obtain the posterior distribution of the model parameters (in our case the movement parameters and movement path). The posterior distribution of the MCMC summarizes the degree of belief of the parameter estimates, given the data. However, rather than promoting frequentist- Bayesian debates, which is not intended by our work, the results we obtained strongly support that posterior distributions and its biological interpretation are virtually the same when imposing either minimally informative priors or reasonable informative priors. Conversely, only when very informative yet possibly biased priors are imposed, the analysis may result in biased posteriors as demonstrated in our simulation analysis. However, our simulation experiments have also demonstrated that the accuracy and precision in the estimation of the movement parameters as well as in the positional data by our model were reasonable. In all cases, the (known) movement parameters in the four realistic simulation settings were properly retrieved. Only when the observational time- step considered were 30, 60 and 90 min the parameters were not well estimated, but only for certain movement trajectories characterized by large distance travelled per unit of time (*k*). This finding suggests that mobile fishes may experience relevant changes in both position and detection probability within the same time-step which may induce a bias in their movement path when large observational time-steps are considered. Therefore, a trade-off between how fast an individual moves and the duration of the time-step should be considered, both in the modelling process and in designing the array of receivers (i.e., between- array distance) as well as the periodicity at which acoustic pulses should be optimally emitted. To that end, pilot studies aiming at generating preliminary information about the movement pattern of a given species is highly advisable for optimizing the design of an acoustic array and the transmitter settings. The approach proposed by Pedersen et al. or the R-code provided in the should help for conducting simulation experiments for fulfilling this task. The application of the Bayesian SSM to case study of the pearly razorfish revealed that the species’ activity is constrained within a very small circular home range with a *radius* of 145.04 ± 49.5 m (between-fish mean and s.d.) and that the exploration rate (*k*) of the home range was 0.005 ± 0.002 min^-1. That is, despite that the pearly razorfish is abundant within a large area of connected sandy and soft bottom habitats, a given individual fish only uses a very small fraction of such large area of suitable habitat, at least at the temporal scale considered here (up to effective 15 tracking days). This pattern has been observed in other sequential hermaphrodites living in coastal areas of the Mediterranean. For example, the radius of the circular home range estimated for the Mediterranean rainbow wrasse, *Coris julis*, was only 227.6 m. These findings confirmed that even relative small marine protected areas may provide a significant protection to the adult stock of the pearly razorfish. Moreover, our methodological approach provides a tool to quantify the large variability of behavior between individual fishes fish. This is especially relevant when considering the growing evidence for the ecological and evolutionary role of between-fish variability in behavioral traits. In fact, Alós et al. demonstrated how among-individual variability in the home range behavior can generate selection gradients through harvesting individuals characterized by larger home range radius and large exploration rate (*k*) that, when combined, implies that fishes with faster swimming speeds are selectively harvested. SSM, therefore, provides a novel tool for future studies aimed at investigating the ecological and evolutionary consequences of the home range behavior in exploited fishes, with especial emphasis on the relationship among home range behavior and harvesting-induced selection. Although useful as discussed, the Bayesian SSM proposed here has limitations too. First, it is a method based on a MCMC-algorithm, thus it is computation- intensive and reaching convergence for all the parameter may require long computation times compared with frequentist solutions. In our case, the real data-sets obtained for the pearly razorfish was however analysed in a reasonable computing time (\~3 hours). Nevertheless, computation time may be a severe constraint our Bayesian approach when long time series (months or years) are available. The recent approach suggested by Albersten et al. implemented in the novel R-package Template Model Builder (TMB) for fitting SSM models to movement data provides a promising tool to alleviate MCMC computational costs in the future when applying the methods presented in this paper. Second, our approach is currently underexploiting some of the potential benefits of SSM when applied to movement data. SSM is a specific case of a family of Hidden Markov Models (HMM) where data (in our case positional data) are observed with uncertainty. HMM are becoming popular for understanding fine-scale animal behavior in reality mining applications because of their ability to differentiate different behavioral states from movement data. One interesting application has been proposed by Patterson et al. who demonstrated the usefulness of the method for discriminating if a given fish is in a resident or migratory state using electronic tagging data in the southern bluefin tuna (*Thunnus maccoyii*). Therefore, as it was suggested in the theoretical framing of this manuscript, the Bayesian SSM can be readily expanded for discriminating different behavioral states such as foraging, hunting or spawning. Third, random effects can also be readily incorporated into hierarchical Bayesian models to facilitate the convergence of the model parameters where the individual movement parameters are estimated from the individual-level data but they are assumed to be distributed around a population mean. All three limitations maybe solved in the future to improve the performance of the application of Bayesian SSM to acoustic tracking data. To conclude, the use of SSMs opens new possibilities for analysing movement data of any animal, including of course sedentary marine fishes, which can provide novel insights for improving our understanding of home range behavior. Fish path and movement parameters can also be used for deriving biologically relevant information with direct application for promoting the sustainability of exploited fish and for providing better understanding of relevant spatial ecological processes. The approach demonstrated here is flexible to include different movement processes and can be easily adapted to acoustic tracking study with other receiver configuration arrays. Similarly, a wide range of behavioral models can be analysed, increasing the potentials of biotelemetry as recently suggested by Krause et al., Donaldson et al. and Hussey et al.. # Supporting Information We thank Antoni Grau and the staff of the *Direcció General de Pesca (Govern de les Illes Balears)* for supporting this study. We also especially thank the *Club Nàutic Portitxol* and the researchers involved in the field for their help, especially M. Cerdà, M. Cabanellas, C. Díaz and R. Rosselló. We also thank RECUMARE and the Unitat Asociada LIMIA-IMEDEA. We thank the comments made by two reviewers in early versions of the manuscript. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: JA MP. Performed the experiments: JA MP SB. Analyzed the data: JA MP SB RA. Contributed reagents/materials/analysis tools: JA MP SB RA. Wrote the paper: JA MP SB RA.

# Introduction In the Age of the Internet, network theory has been undergoing exponential growth. One of its important problems is to algorithmically construct networks (or *graphs*) with predefined parameters, or to uniformly sample networks with these parameters. For general background, the interested reader can turn to the now-classic book of Newman, Barabási and Watts () or to the more recent book of Newman (). One of the earliest and still most important problems in graph theory is uniformly sampling all possible graph realizations of given degree sequence. (For the definitions see Section “”.) One possible method for this is a simple MCMC approach (proposed by Kannan, Tetali and Vempala); take an arbitrary realization of the degree sequence, then perform a series of randomly chosen local transformations (called *swap* or *switch*). They conjectured that the process is *rapidly mixing*, i.e., a random realization is achieved after polynomial many steps. The first result with a flawless proof in connection with this conjecture is due to Cooper, Dyer and Greenhill (2007,) for the special case when the degree sequence is regular. Greenhill proved in 2011 the analogous result for (in- and out-)regular directed graphs (). In 2013 Miklós, Erdős and Soukup proved the conjecture for *half-regular* bipartite graphs (). Here the degree sequence on one class is regular, while there is no constraint on the other class. (A comprehensive survey on the topic is or is.) In modern network applications sampling the solutions is just one requirement. Sometimes the actual number of all solutions (or at least a good approximation of it) is also important. It is well known (Jerrum, Valiant and Vazirani 1986,) that for *self-reducible counting problems* a rapidly mixing sampling method also provides a quick estimation of that number (with small relative error and with very high probability). Unfortunately none of the sampling problems listed above belongs to this class. The main purpose of this paper is to remedy this imperfection. For that end we introduce a slightly more general degree sequence problem, which has all the good characteristics of the sampling procedures above (including their rapidly mixing nature), furthermore, which belongs to the class of self-reducible counting problems. This new problem is a common generalization of the regular directed graph and of the half-regular bipartite graph cases. Therefore, showing the rapidly mixing nature of the corresponding MCMC procedure provides new proofs for both problems. We prove only the existence of a polynomial upper bound on the mixing time, but do not prove the tight upper bound in Greenhill’s theorem in. In Section “Degree sequences” we recall the known definitions and facts on degree sequences problems in simple graphs. Then we introduce and study in full generality our proposed new *restricted degree sequence* (or **ReDeSe** for short) problem, where we deal with forbidden edges. Next, we study a specific instance of the general ReDeSe problem: bipartite degree sequences with a forbidden (but not necessarily perfect) 1-factor and a forbidden (but maybe empty) star. In Section “Sampling” we first discuss some known results to sample degree sequence realizations. Then, we formulate our main result (Theorem 10): the proposed MCMC process on half-regular bipartite degree sequences with a well- defined small forbidden edge set is rapidly mixing. Our proof is based on Sinclair’s *multicommodity flow method* (), and follows closely the proof in. We discuss the similarity between the two proofs in this section, while in Sections “Milestones” and “The analysis” we study the details of our new Markov chain approach which require different treatment. In Section “Counting” we show that the studied sampling problem leads to a self- reducible counting problem. Therefore, our almost uniform sampling method provides a good approximation on the size of the set of all realizations, strengthening also Greenhill’s result on regular directed graphs (), and Miklós, Erdős and Soukup’s result on half-regular bipartite graphs (). # Degree Sequences In this paper, every graph is assumed to be *simple*; there are no loops or multiple edges. ## Degree sequences and realizations Let *V* be a labeled set of *n* elements. The *degree sequence* **d**(*G*) of a graph *G* = (*V*, *E*) is the sequence **d**(*G*)_*i* = *d*(*v*_*i*) of its vertex degrees. A non-negative integer sequence **d** = (*d*₁, …, *d*_*n*) is *graphical* iff **d**(*G*) = **d** for some simple graph *G*, and then *G* is a *graphical realization* of **d**. The first successful approach to decide whether a degree sequence is graphical is due to Havel (), his simple but surprising observation provides immediately a greedy algorithm to build such a realization. His work was rediscovered independently later by Hakimi (). Their method is based on the so-called *swap* operation. (The expressions *switch* or *rewiring* are also widely used. In this paper the word *switch* will be used for a similar, but slightly more general, operation.) The *swap* operation is defined as follows: Let *G* be a simple graph and assume that *a*, *b*, *c* and *d* are different vertices. Furthermore, assume that (*a*, *c*), (*b*, *d*) ∈ *E*(*G*) while (*b*, *c*), (*a*, *d*) ∉ *E*(*G*). Then $$E\left( G^{\prime} \right) = E(G) \smallsetminus \left\{ (a,c),(b,d) \right\} \cup \left\{ (b,c),(a,d) \right\}$$ is another realization of the same degree sequence. We denote this operation by *ac*, *bd* ⇒ *bc*, *ad*. Havel’s nice observation is the following: **Lemma 1** (Havel,). *Assume that in graph G vertex v is adjacent to vertex x but not to vertex y. Assume furthermore that d*(*x*) ≤ *d*(*y*). *Then one can find a swap operation which produces a new graphical realization G′ of the degree sequence* **d**(*G*) *with the property:* Γ′(*v*) = Γ(*v*) ∖ {*x*} ∪ {*y*}. *(These are the corresponding neighborhoods of vertex v.)* The analogous notions for bipartite graphs are the following: if *B* is a simple bipartite graph then its vertex classes will be denoted by *U*(*B*) = {*u*₁, …, *u*_*k*} and *W*(*B*) = {*w*₁, …, *w*_ℓ}, and we keep the notation *V*(*B*) = *U*(*B*) ∪ *W*(*B*). The *bipartite degree sequence* of *B*, **D**(*B*) is defined as follows: $$\begin{array}{r} {\mathbf{D}(B) = \left( \left( d\left( u_{1} \right),\ldots,d\left( u_{k} \right) \right),\left( d\left( w_{1} \right),\ldots,d\left( w_{\ell} \right) \right) \right).} \\ \end{array}$$ We can define the swap operation for bipartite realizations similarly to but we must take some care: it is not enough to assume that (*b*, *c*), (*a*, *d*) ∉ *E*(*G*) but we have to know that *a* and *b* are in one vertex class, and *c* and *d* are in the other. To make clear whether a vertex pair is not forbidden to be an edge we will call a vertex pair a **chord** if it can hold an actual edge in a realization. Those pairs that cannot accommodate an edge are **non-chords**. (For example, pairs from the same vertex class of a bipartite graph are non-chords.) It can also be found in \[, Theorem 6\]. Denote $\overset{\rightarrow}{G}$ a directed graph (no parallel edges, no loops, but oppositely directed edges between two vertices are allowed) with vertex set $X(\overset{\rightarrow}{G}) = \{ x_{1},x_{2},\ldots,x_{n}\}$ and edge set $E(\overset{\rightarrow}{G})$. For every vertex *v* we associate two numbers: the *in-degree* and the *out-degree* of *v*. Instead of introducing the matching definitions, we will apply the following *representation* of the directed graph $\overset{\rightarrow}{G}:$ let $B(\overset{\rightarrow}{G}) = (U,W;E)$ be a bipartite graph where each class consists of one copy of every vertex of $\overset{\rightarrow}{G}$. The edges adjacent to a vertex *u*_*x* in class *U* represent the out-edges from *x*, while the edges adjacent to a vertex *w*_*x* in class *W* represent the in-edges to *x* (so a directed edge *xy* corresponds the edge *u*_*x**w*_*y*). If a vertex has zero in- (respectively out-) degree in the directed version, then we delete the corresponding vertex from $B(\overset{\rightarrow}{G}).$ (Actually, this representation is an old trick used already by Gale.) There is no loop in our directed graph, therefore there is no (*u*_*x*, *v*_*x*) type edge in its bipartite realization—these vertex pairs are non-chords. Consider two different realizations, *G* and *H*, of the same degree sequence (either simple or bipartite one). It is a well-known fact that the first can be transformed to the second one (and vice versa) with consecutive swap operations. Formally, there exists a series of realizations *G* = *G*₀, …, *G*_*i*−1, *G*_*i* = *H*, such that for each *j* = 0, …, *i*−1 there exists a swap operation which transforms *G*_*j* into *G*_*j*+1. For simple graphs this was proved already in 1891 by Petersen. It can be shown that lemma 1 also provides a solution via the so-called *canonical realizations*. The analogous result for bipartite graphs (with possible multiple edges but no loops) is due to Ryser (). For simple bipartite graphs this is common-knowledge. For directed graphs an analogous result is known. It was discovered by Kleitman and Wang and later rediscovered in. It is important however to recognize that in case of directed graphs the “classical” Havel-type swap operation is not always adequate. To see this, it is enough to consider an example with three vertices: each vertex incident to one in-edge and one out-edge. There are exactly two possible realizations of this directed degree sequence, therefore a swap operation with six chords is necessary: we exchange three edges with three non- edges in one step. In papers it was shown that this extra operation is always sufficient. ## Restricted degree sequences In this paper we study the following common generalization of all previously mentioned degree sequence problems: The **restricted degree sequence** (i.e. ReDeSe) problem **d**^ℱ consists of a degree sequence **d** and a set $\mathcal{F} \subset \left( \frac{V}{2} \right)$ of **forbidden** edges. The problem is to decide whether there is a simple graph *G* on *V* with the given degree sequence and with *E*(*G*) ∩ ℱ = ∅. It is clear that this problem is essentially identical with Tutte’s f-factor problem: the f-factor to be found is our degree sequence while the graph what the f-factor is searched for is the complement of the forbidden edges. Therefore Tutte’s theorem and the famous blossom algorithm of Edmonds apply nicely for the ReDeSe problem. However the focus of our approach is quite different from the f-factor problem: at first we are interested several (or all) solutions of the ReDeSe problem instead of finding one solution, and often enough we want to find “typical” solutions. At second: this sampling problem seems to be hopeless in general. In our studies we restrict ourself for carefully chosen small instances. The **bipartite restricted degree sequence** problem **D**^ℱ consists of a bipartite degree sequence **D** on (*U*, *W*), and a set ℱ ⊂ \[*U*, *W*\] of **forbidden** edges. The problem is to decide whether there is a simple bipartite graph *G* on *V* with the given degree sequence and with *E*(*G*) ∩ ℱ = ∅. Clearly, a bipartite restricted degree sequence problem **D**^ℱ on (*U*, *W*) is the restricted degree sequence problem **d**^ℱ′ on *U* ∪ *W*, where ℱ′ = ℱ ∪ \[*U*\]² ∪ \[*W*\]². Furthermore, we already studied one instance of the bipartite restricted degree sequence problem, namely the bipartite representation of directed degree sequences: here ℱ is one 1-factor, which corresponds to the forbidden loops. It is important to add that while the fundamental result of Jerrum, Sinclair and Vigoda on sampling perfect matchings in graphs () provides a uniform sampling approach for the possible realizations, their method is not useful in practice. That is the reason that so much effort has been made on this topic. We return to this issue at the end of Section “Counting”. In the remaining of this subsection we study the general ReDeSe problem. The next subsection will be devoted to a particular bipartite restricted degree sequence problem which will play a central role later in the paper. **Definition 2.** Let **d**^ℱ be a restricted degree sequence problem and let *G* be a realization of it. The sequence of vertices 𝓒 = (*x*₁, *x*₂, …, *x*_2*i*) is a **chord-circuit** if: (D1) all pairs *x*₁*x*₂, *x*₂*x*₃, …, *x*_2*i*−1*x*_2*i*, *x*_2*i**x*₁ are chords; (D2) each of these chords is different. A chord-circuit is **elementary** if (D3) no vertex occurs more than twice; (D4) when two copies of the same vertex exist, then their distance along the circuit is odd. **Definition 3.** The chord-circuit 𝓒 is said to **alternate** in *G*, if the chords along 𝓒 are in turn edges and non-edges in *G*. (For example *x*_2*j*−1*x*_2*j* are edges for 1 ≤ *j* ≤ *i*, while the other chords are non-edges in *G*.) Deleting the actual edges along 𝓒 from *G* and adding the other chords as edges constructs a new graph *G*′ which is again a realization of **d**^ℱ. This is a **𝓒-swap** and this operation is known in general as a **circular *C*_2*i*-swap**. Finally, two *different* vertices *x*, *y* of the alternating chord-circuit 𝓒 form a **PV-pair** if the distance of the vertices along the circuit (the number of chords between them) is odd and greater than 1. If all PV-pairs are non- chords (so they belong to ℱ), then this circular 𝓒-swap is called a **ℱ-compatible swap** or **ℱ-swap** for short. The ℱ-swap is one of the central notions of this paper. When *i* = 2 then the circular *C*₄-swap coincides with the classical Havel type swap. When *i* = 3 then we get back the notion of the **triangular *C*₆-swap**, which occurs in connection with directed degree sequences. We define the **weight** of the ℱ-compatible circular *C*_2*i*-swap as *w*(*C*_2*i*) = *i*−1. This definition sets the weight of the classical Havel type swaps to 1 and the weight of a *C*₆-swap to 2, which agree with the definitions used in paper. Furthermore it is well known (see for example again) that (*i*−1) Havel type swaps are needed to alternate the edges along *C*_2*i* in case of simple graphs with no forbidden edges. As we will see next the same applies for any elementary circular *C*_2*i*-swap: **Lemma 4.** *Let G be a realization of* **d**^ℱ *and let the elementary chord-circuit 𝓒 of length* 2*i be alternating. Then the circular 𝓒-swap operation can be carried out by a sequence of ℱ-swaps of total weight i*−1. In other words there exists a sequence *G* = *G*₀, *G*₁, …, *G*_ℓ of realizations such that for each *j* = 0, …, ℓ−1 there exists an ℱ-compatible swap operation from *G*_*j* to *G*_*j*+1. The difference between *G* and *G*_ℓ is exactly the alternating circuit 𝓒. Finally, the total weights of those ℱ-swap operations is *i*−1. We will say that this swap sequence does **process** the prescribed circular swap operation. *Proof*. We apply mathematical induction for the length of the chord-circuit: when *i* = 2 then the statement is trivial. Assume now that this is true for all circuits of length at most 2*i*−2. Then take an alternating elementary chord- circuit 𝓒 of length 2*i* in a realization of **d**^ℱ. If each PV-pair in 𝓒 is a non-chord, then the circular *C*_2*i*-swap itself is a ℱ-swap of weight *i*−1. So we may assume that there is a PV-pair *uv* in 𝓒 which is a chord. This chord together with the two “half-circuits” of 𝓒 form chord-circuits 𝓒₁ and 𝓒₂ using the chords of the original circuit 𝓒 and twice the chord *uv*. One of them, say 𝓒₁, is alternating. The length of 𝓒₁ is 2*j* \< 2*i* therefore there exists a ℱ-compatible swap sequence of total weight *j*−1 to process it. After the procedure the *status* of *uv* (the property of the chord whether it is an edge or a non-edge) will alter into the other status. With this new status of the chord the circuit 𝓒₂ becomes an alternating one with length 2*i*+2−2*j*, so it can be processed with $\frac{2i + 2 - 2j}{2} - 1$ total weight—and after this procedure the chord *uv* is switched back to its original status. We found a swap sequence of total weight *i*−1 which finishes the proof. **The space of all realizations of d**^ℱ: Consider now the set of all possible realizations of a restricted graphical degree sequence **d**^ℱ. Let *G* and *H* be two different realizations. The natural question, similar to the case of classical degree sequence problems, is whether *G* can be transformed into *H* using ℱ-swaps? The answer is affirmative: **Theorem 5.** *The space* 𝔾 = (𝕍, 𝔼) *of all realizations of the restricted degree sequences problem* **d**^ℱ *is connected*. *Proof*. What we have to prove is the following: let *G* and *H* be two realizations of **d**^ℱ. Then we have to find a series of realizations *G* = *G*₀, …, *G*_*i*−1, *G*_*i* = *H*, such that for each *j* = 0, …, *i*−1 there exists an ℱ-swap from *G*_*j* to *G*_*j*+1. Consider the symmetric difference of the edge sets of the two realizations: Δ = *E*(*G*)△*E*(*H*). This set is two-colored by the original hosts of the edges: there are *G*-edges and *H*-edges. It is clear that for each vertex *v* in the graph 𝓖 = (*V*, Δ) the numbers of *G*-edges and *H*-edges incident to *v* are the same: *d*_*G*(*v*) = *d*_*H*(*v*). It is well known that this can be decomposed into alternating circuits 𝓒₁, …, 𝓒_ℓ. We will use the notions of **circuit** and **cycle** in a simple graph *G* as usual: therefore a circuit is a chord-circuit where all chords are edges. A cycle is a circuit without repeated vertices. A circuit is **alternating** in Δ if the edges come in turns from *E*(*G*) and *E*(*H*). When this is the case then the corresponding chord-circuit in realization *G* (as well as in *H*) is also alternating. We can find a decomposition, such that no circuit contains a vertex *v* twice and their distance *δ* (the number of edges between the copies is even. Indeed, if *δ* is even, then *δ* is at least four, consequently the vertex *v* splits the original circuit into two smaller, but still alternating circuits. Furthermore, if a circuit contains a vertex *v* at least three times, then there are at least two of them with even distance. It is clear that any alternating circuit decomposition can be transformed into a decomposition where each (chord)-circuit is elementary with successive transformations. It is also clear that if, by chance, we start with a circuit decomposition of maximal number of circuits, then all circuits in this decomposition are automatically elementary. (Of course, finding such a decomposition may be very hard.) The application of Lemma 4 proves that each circuit 𝓒 can be processed with \|𝓒\|/2−1 total weight. This finishes the proof. It seems to be interesting that using a result from paper one can determine the minimum weight of an ℱ-compatible swap sequence which transforms *G* into *H*, however we do not discuss this question here. ## Bipartite 1-Factor + 1 Star Restricted Degree Sequences In the previous subsection we studied the restricted degree sequence problem in its full generality. However, our real interest lays in a quite simple case: **d**^𝕱 is called a **1-Factor + 1 Star Restricted Degree Sequence** problem (or **1F1S** problem for short), if (Ψ) the set 𝕱 of forbidden edges is a bipartite graph where the edges are the union of an 1-factor and a star with center *s*. Similarly, if **D** is a bipartite degree sequence, and (Ψ) holds for 𝕱, then **D**^𝕱 is called a **Bipartite 1F1S** problem. Everything discussed in this subsection applies to all 1F1S degree sequence problems in simple graphs. However, we are particularly interested in the bipartite case, therefore we will discuss these observations for the bipartite case only. We fix the underlying vertex set *V* = (*U*, *W*). Then **D**^𝕱 is a bipartite 1F1S problem where the center *s* of the forbidden star belongs to *U*. **Lemma 6.** 1. *The space of all realizations of* **D**^𝕱 *is closed under* 𝕱-*compatible swap operations*. 2. *The* 𝕱-*compatible swap operations are circular C*₄- and *C*₆-*swaps*. *Proof*. (i) As we saw already that any bipartite 1F1S can be understood as an 1F1S on simple graphs, therefore considering ℱ = 𝕱 ∪ \[*U*\]² ∪ \[*W*\]² and applying Theorem 5 for the problem **d**^ℱ proves (i). \(ii\) Let us consider any alternating elementary circuit *C* in the symmetric difference △ of two different realizations. There is a vertex *u* ∈ *C* ∩ *U* which is ≠ *s*. There is at most one forbidden chord in 𝕱 which is adjacent to *u*. If *C* has more than 6 vertices, then *C* has at least 4 vertices in *W* therefore there exists a vertex *w* ∈ *C* ∩ *W*, such that *uw* is a chord and *uw* is not in *C*. Therefore the corresponding *C*-swap is not compatible with 𝕱. As we already mentioned, Tutte’s *f*-factor theorem can always be utilized to find actual graphical realizations of the bipartite 1F1S problem. However, in this special case we can prove a Havel type result (similar to Lemma 1) and can construct a greedy algorithm to produce such realizations. Consider the bipartite 1F1S degree sequence problem **D**^𝕱. If the forbidden star is not empty, then let *u* ≔ *s*. Otherwise let *u* ∈ *U* be any given vertex and denote *N*(*u*) ⊂ *W* the set of those vertices which form chords together with *u*. (It is clear that if *u* ≠ *s* then \|*W*\|−1 ≤ \|*N*(*u*)\|.) **Observation 7.** *For any y* ∈ *N*(*u*) *there is at most one vertex, denoted by y*^𝕱, *such that yy*^𝕱 *is a non-chord, so it belongs to* 𝕱. *Furthermore if y, z* ∈ *N*(*u*) *and y*^𝕱 = *z*^𝕱 *then y* = *z*. Now a linear order ≺_*u* on *N*(*u*) is called **good** if it satisfies the following properties: for *y*, *z* ∈ *N*(*u*) and *y* ≺_*u* *z* we have *d*(*y*) ≥ *d*(*z*) and in case of *d*(*y*) = *d*(*z*) we also have *d*(*y*^𝕱) ≥ *d*(*z*^𝕱). It is obvious that there always exists a good order on *N*(*u*). Furthermore whenever *d*(*y*) = *d*(*z*) and *d*(*y*^𝕱) = *d*(*z*^𝕱), then there are more than one good order. **Lemma 8.** *Let G be a graphical realization of the 1F1S sequence* **D**^𝕱, *let u* ≔ *s if the forbidden star is not empty and take any u* ∈ *U otherwise. Let y, z* ∈ *N*(*u*) *where y* ≺_u *z with uz* ∈ *E while uy* ∉ *E. Then there exists an alternating chord-cycle C of length at most 6 in G with y, u, z* ∈ *C. Processing C with* 𝕱-*compatible swap operations, we have* Γ_*G*′(*u*) = Γ_*G*(*u*) ∖ {*z*} ∪ {*y*} *in the acquired new realization*. *Proof*. We have *uz* ∈ *E* but *uy* ∉ *E*. At first assume that there exists a vertex *μ* ∈ *U* ∖ {*u*}, such that *μy* ∈ *E*, and *μz* ∉ *E* but *μ* ≠ *z*^𝕱. When such vertex exists then 𝓒 = (*u*, *z*, *μ*, *y*) is a suitable alternating chord-cycle. When *d*(*y*) \> *d*(*z*) then there are two vertices *μ* and *μ*′ ∈ *U* such that *yμ* ∈ *E* and *zμ* ∉ *E*, and *yμ*′ ∈ *E* and *zμ*′ ∉ *E*. Now either *zμ* or *zμ*′ is a chord. However, if *d*(*y*) = *d*(*z*) then it can happen that *z*^𝕱 *y* ∈ *E* and $$\left. \text{for}\text{all} x \in U \smallsetminus \left\{ u,y^{\mathfrak{F}},z^{\mathfrak{F}} \right\}\text{we}\text{have} xy \in E\Leftrightarrow xz \in E. \right.$$ It is important to observe that in this case *y*^𝕱 *z* ∉ *E*, otherwise some *x* would not satisfy (in order to keep *d*(*y*) = *d*(*z*)). So the only case when we do not find automatically an appropriate circular *C*₄-swap with *u*, *y* and *z* is when *d*(*y*) = *d*(*z*), *yz*^𝕱 is an edge and *zy*^𝕱 is a chord but not an edge. In this case, we can find a *μ* ∈ *W* ∖ {*y*, *z*} such that *y*^𝕱*μ* ∈ *E* but *z*^𝕱*μ* ∉ *E* since *d*(*y*^𝕱) ≥ *d*(*z*^𝕱). Observe that *z*^𝕱*μ* is a chord because *μ*^𝕱 ≠ *z*^𝕱. Now 𝓒 = (*y*, *u*, *z*, *y*^𝕱, *μ*, *z*^𝕱, *y*) is the required alternating chord circle. When *uμ* is a chord, then the circular *C*₆-swap is not 𝕱-compatible, but we can process the cycle properly (as it was shown in the proof of Lemma 4). When it is a non-chord, then the circular *C*₆ is an 𝕱-compatible operation. Lemma 8 provides the following easy Havel type greedy algorithm to decide whether our bipartite 1F1S restricted degree sequence is graphical. (The algorithm is essentially the same as the original Havel procedure.) **A greedy algorithm** to decide whether a bipartite 1F1S degree sequence is graphical: the degree sequence is **D** while the set of forbidden edges is 𝕱. (H₁) Let *u* ≔ *s* if the forbidden star is not empty, otherwise take an arbitrary *u* ∈ *U*. Consider a good order ≺_*u* on *N*(*u*). Connect *u* the first *d*(*u*) vertices from (with respect to ≺_*u*) of *N*(*u*). If *d*(*u*) \> \|*N*(*u*)\| then the algorithm `FAILS`, our sequence **D**^𝕱 is not graphical. Delete *u* from *U*, and update the degree sequence and the set *W* accordingly. Finally delete the edges adjacent to *u* from 𝕱. (H₂) Repeat the previous step while *U* is not empty. **Theorem 9** (Generalized Havel theorem for the bipartite 1F1S ReDeSe problem). *The 1F1S restricted degree sequence* **D**^𝕱 *is graphical if and only if the previous greedy algorithm provides a realization*. *Proof*. The proof is exactly the same as in the original case, described by Havel; if the sequence is graphical, then consider a realization. Fix vertex *u* ∈ *U* as described in Lemma 8. Recursive applications of the lemma provide a realization where *u* is connected to the first *d*(*u*) vertices in *N*(*u*) with respect to ≺_*u*. The repeated application of the previous reasoning finishes the proof. # Sampling Degree Sequence Realizations with Sinclair’s Multicommodity Flow Method There are several available methods to sample uniformly the space of all realizations of a given degree sequence. One of these approaches is a Markov Chain Monte Carlo method, proposed by Kannan, Tetali and Vempala (1999,). They consider a local transformation (the swap operation) on the realizations, which in turn defines an irreducible, reversible and aperiodic finite Markov chain on these realizations; at any given realization they choose two independent edges from some probability distribution and perform the corresponding swap operation if it is feasible. They conjecture that the resulted MCMC is *rapidly mixing*. They were studying the particular case when the degree sequence is regular bipartite using Sinclair’s multicommodity flow method. Their conjecture was proved for regular graphs by Cooper, Dyer and Greenhill (2007,). (Their result does not apply to bipartite graphs, since their version does not allow forbidden edges.) An analogous theorem was proved by Greenhill on regular directed graphs (). Here she proved at first that for regular directed degree sequences circular *C*₄-swaps alone make the space of the realizations connected, then she gave a strong upper bound on the mixing time. (However, as we saw it earlier, the space of directed realizations are not always connected when using only *C*₄-swaps.) Finally, in 2013 Miklós, Erdős and Soukup proved () that the corresponding Markov process is rapidly mixing on each bipartite **half-regular** degree sequence, superseding the original study of Kannan, Tetali and Vempala (). In this paper we study the realizations of half-regular bipartite 1F1S restricted degree sequences **D**^𝕱. The vertex set is (*U*, *W*) where the center of the forbidden star *s* is ∈ *U* and where all vertex in *U* (except possible *s*) have the same degree. The degrees in *W* are not constrained. The state space of our **Markov chain** is the graph 𝔾 = (*V*(𝔾), *E*(𝔾)) where *V*(𝔾) consists of all possible realizations of our problem, while the edges represent the possible swap operations: two realizations (which will be indicated by upper case letters like *X* or *Y*) are connected if there is a valid 𝕱-swap operation which transforms one realization into the other one (and the inverse swap transforms the second one into the first one as well). Recall that there are two kinds of 𝕱-compatible swap operations: the circular *C*₄-swaps and certain *C*₆-swaps (in the latter case opposite vertex pair in the *C*₆ must be non-chord), Furthermore, these two kinds of operations make the state space connected (see Theorem 5). The *transition (probability) matrix* *P* of the Markov chain is defined as follows: let the current realization be *G*. Then (a) with probability 1/2 we stay in the current state (that is, our Markov chain is **lazy**); (b) with probability 1/4 we choose uniformly two-two vertices *u*₁, *u*₂;*v*₁, *v*₂ from classes *U* and *W* respectively and perform the swap if it is possible; (c) finally with probability 1/4 choose three—three vertices from *U* and *W* and check whether they form three pairs of forbidden chords. If this is the case, then we perform a circular *C*₆-swap if it is possible. The swaps moving from *G* to its image *G*′ is unique, therefore the probability of this transformation (the *jumping probability* from *G* to *G*′ ≠ *G*) is: $$\text{Prob}\left( G\rightarrow{}_{b}G^{\prime} \right) ≔ P\left( G^{\prime} \middle| G \right) = \frac{1}{4} \cdot \frac{1}{\binom{|U|}{2}\binom{|W|}{2}},$$ and $$\text{Prob}\left( G\rightarrow{}_{c}G^{\prime} \right) ≔ P\left( G^{\prime} \middle| G \right) = \frac{1}{4} \cdot \frac{1}{\binom{|U|}{3}\binom{|W|}{3}}.$$ (These probabilities reflect the fact, that *G*′ should be derived from *G* by a regular swap or by a *C*₆-swap.) The probability of transforming *G* to *G*′ (or vice versa) is time-independent and *symmetric*. Therefore *P* is a symmetric matrix, where the entries in the main diagonal are non-zero, but (probably) distinct values. Our Markov chain is *irreducible* (the state space is connected), and it is clearly aperiodic, since it is lazy. Therefore, as it is well known, the Markov process is reversible with the uniform distribution as the globally stable stationary distribution. Our main result is the following: **Theorem 10.** *The Markov process defined above is rapidly mixing on each bipartite half-regular 1F1S restricted degree sequence.* **Remark 11.** When we apply this setup for directed graphs then the out-degrees are regular (except, perhaps, the out-degree of the vertex *s*), while we have no constrains on the in-degrees. However, it is important to see, that while this result provides a rapidly mixing sampling procedure on regular directed graphs as well, the applied Markov chain is not the same as the one in Greenhill’s model. Hence, this result does not supersede Greenhill’s result. The proof of Theorem 10 follows closely the proof developed in paper. (More precisely we need to slightly generalize it. The required minor technical issue will be discussed in the Section “”.) Consider two realizations *X* ∈ 𝔾 and *Y* ∈ 𝔾 of the problem **D**^𝕱, and take the symmetric difference Δ = *E*(*X*)Δ*E*(*Y*). As we saw already in the proof of Theorem 5 for each vertex *v* in the bipartite graph (*U*, *W*; Δ) the number of adjacent *X*-edges (= *E*(*X*) ∖ *E*(*Y*)) and the number of the adjacent *Y*-edges are equal. Therefore Δ can be decomposed into alternating circuits and later into alternating cycles. The way the decomposition is executed is described in details in Section 5 of the paper. Here we just summarize the high points: At first we decompose the symmetric difference Δ into alternating circuits on all possible ways. In each cases we get an ordered sequence *W*₁, *W*₂, …, *W*_*κ* of circuits. (Usually there are a huge number of different circuit decompositions.) Each circuit is endorsed with a fixed cyclic order. Now we fix one circuit decomposition. Each circuit *W*_*i* from the ordered decomposition determines one unique alternating cycles decomposition: $W_{i} = C_{1}^{i},C_{2}^{i},\ldots,C_{k_{i}}^{i}.$ (This unique decomposition is one of the most delicate points of the entire proof in. The main problem is that a circuit can be “long”—linear in the number of vertices—therefore, it can happen that it is decomposed into a linear number of cycles. Keeping track of all possible changes along the circuit is necessary, and without clever data handling it may require an unacceptable big data set. Section 5.2 in paper found a way around this problem.) The ordered circuit decomposition of Δ together with the ordered cycle decompositions of all circuits provide a well defined ordered cycle decomposition *C*₁, …, *C*_ℓ of Δ. This decomposition does not depend on any 𝕱-compatible swap operations (actually no swap operation was performed yet), only on the symmetric difference of realization *X* and *Y*. So this part of the original proof can be used freely in our current reasoning without any modification. This ordered cycle decomposition singles out ℓ−1 different realizations *H*₁, …, *H*_ℓ−1 of **D**^𝕱 with the following property: for each *j* = 0, …, ℓ−1 we have *E*(*H*_*j*)Δ*E*(*H*_*j*+1) = *C*_*j*+1 if we apply the notations *H*₀ = *X* and *H*_ℓ = *Y*. This mean that $$\begin{array}{r} {E\left( H_{i} \right) = E(X) \bigtriangleup \left( \underset{i^{\prime} \leq i}{\cup}E\left( C_{i^{\prime}} \right) \right).} \\ \end{array}$$ What remains is to design a unique canonical path from *X* to *Y* determined by the circuit decompositions which use the realizations *H*_*j* as *milestones* along the path. With other words, for each pair *H*_*j*, *H*_*j*+1 we have to design the actual swap sequences which turn one milestone into the next one. So, the canonical path under construction is a sequence *X* = *G*₀, …, *G*_*i*, …, *G*_*m* = *Y* of realizations, where each *G*_*i* can be derived from *G*_*i*−1 with one feasible circular *C*₄- or *C*₆-swap operation, and there exists an increasing subscript subsequence 0 = *n*₀ \< *n*₁ \< *n*₂ \< ⋯ \< *n*_ℓ = *m* such that we have *G*_{*n*_*i*} = *H*_*i*. In paper the following result was proved: **Theorem 12** (Section 4 in). *If the designed canonical path system satisfies the three (rather complicated) conditions below, then the MCMC process is rapidly mixing. The conditions are:* (Θ) *For each i* \< ℓ *the constructed path* $H_{i} = G_{0}^{\prime},G_{1}^{\prime},\ldots,G_{m^{\prime}}^{\prime} = H_{i + 1}$ *satisfies that* *m*′ ≤ *c*⋅\|*C*_*i*+1\| *for a suitable constant c*. (Ω) ∀*j there exists a K*_*j* ∈ *V*(𝔾) *s.t.* ${\mathfrak{d}}\left( {M_{X} + M_{Y} - M_{G_{j}^{\prime}},M_{K_{j}}} \right) \leq \Omega_{2}$, *where M*_*G* *is the* **bipartite adjacency matrix** *of G, and* 𝔡 *stands for the Hamming distance of two matrices, finally* Ω₂ *is a small constant*. (Ξ) *For each vertex* $G_{j}^{\prime}$ *in the path under construction the following three objects together uniquely determine the realizations X, Y and the path itself*. *The value of the auxiliary matrix* $M_{X} + M_{Y} - M_{G_{j}^{\prime}}$; *the symmetric difference* Δ = *E*(*X*)△*E*(*Y*); *finally a polynomial size parameter set* 𝔹. The meaning of condition (Ξ) is that these structures can be used to control certain features of the canonical path system: namely their numbers gives a bound on the number of canonical paths between any realization pairs *X*, *Y* which go through any given realization $G_{j}^{\prime}.$ Then condition (Ω) ensures, that the overall number of the used auxiliary matrices is small. So while we determine our canonical paths among any pair *X*, *Y* we have take care for these three conditions. We will describe the construction itself in the following two Sections. # The construction of swap sequences between consecutive “milestones” Now we are going to implement our plan described above. At first we introduce some shorthand. Instead of *H*_*i*−1 and *H*_*i* we will use the names *G* and *G*′. These two graphs have almost the same edge set. More precisely $$\begin{array}{l} {\left( E(G) \smallsetminus \left( C_{i} \cap E(X) \right) \right) \cup \left( C_{i} \cap E(Y) \right) = E\left( G^{\prime} \right)} \\ {\left( E\left( G^{\prime} \right) \smallsetminus \left( C_{i} \cap E(Y) \right) \right) \cup \left( C_{i} \cap E(X) \right) = E(G).} \\ \end{array}$$ Of course *E*(*G*)Δ*E*(*G*′) = *C*_*i* also holds. We refer to the elements of *C*_*i* ∩ *E*(*X*) as *X*-edges, while the others are *Y*-edges. We denote the cycle itself by 𝓒, it has 2ℓ edges and its vertices are *u*₁, *w*₁, *u*₂, *w*₂, …, *u*_ℓ, *w*_ℓ. Since 𝓒 has at least four vertices, therefore we may assume that *u*₁ ≠ *s* (thus *u*₁ is not the center of the forbidden star). Finally, w.l.o.g. we may assume that the chord *u*₁*w*₁ is a *Y*-edge (and, of course, *w*_ℓ *u*₁ is an *X*-edge). We are going to construct the realizations $G_{j}^{\prime}$ one by one. We build our canonical path from *G* toward *G*′. At any particular step the last constructed realization is denoted by *Z*. (At the beginning of the process we have *Z* = *G*.) We are looking for the next realization, denoted by *Z*′. Before we continue the discussion of the canonical path system, we introduce our control mechanism, mentioned in condition (Ω). This *auxiliary* structure originally was introduced by Kannan, Tetali and Vempala in: For any particular realization *G* from *V*(𝔾) the matrix *M*_*G* denotes the *adjacency matrix* of the bipartite realization *G* where the columns and rows are indexed by the vertices of *U* and *W* respectively (Therefore the column sums are the same in each realization, except perhaps at column *s*.) Our indexing method is a bit unusual: the columns are numbered from left to right while the rows are numbered from bottom to the top. (Like in the Cartesian coordinate system.) This matrix is not necessarily symmetric, and elements *M*_*i*,*i* can be different from 0. For example, if we consider the submatrix in *M*_*G* spanned by *u*₁, …, *u*_ℓ and *w*₁, …, *w*_ℓ then we have *M*_*G*(*i*, *i*) = 0 for *i* = 1, …, ℓ, while *M*_*G*(*i*, *i*−1) = 1 (for *i* = 2, …, ℓ) and *M*_*G*(1, ℓ) = 1. (So the first value gives the column, the second one gives the row.) The non-chords between vertices in the same vertex class are not considered at all, while non-chords which are forbidden are denoted by ✠. As it is clear from the previous sentence, we will identify each chord or non-chord with the corresponding position in the matrix. Our auxiliary structure is the matrix $$\hat{M}(X + Y - Z) = M_{X} + M_{Y} - M_{Z}.$$ By definition, each entry of a bipartite adjacency matrix is 0 or 1 (or ✠). Therefore only −1, 0, 1, 2 can be the “meaningful” entries of $\hat{M}.$ An entry is −1 if the edge is missing from both *X* and *Y* but it exists in *Z*. It is 2 if the edge is missing from *Z* but exists in both *X* and *Y*. It is 1 if the edge exists in all three graphs (*X*, *Y*, *Z*) or it is there only in one of *X* and *Y* but not in *Z*. Finally it is 0 if the edge is missing from all three graphs, or the edge exists in exactly one of *X* and *Y* and in *Z*. (Therefore if an edge exists in exactly one of *X* and *Y* then the corresponding chord in $\hat{M}$ is always 0 or 1.) One more important, but easy fact is the following: **Observation 13.** *The row and column sums of* $\hat{M}(X + Y - Z)$ *are the same as row and column sums in M*_*X* *(or M*_*Y* *or M*_*Z*). Next we will determine the swap sequence between *G* and *G*′ through an iterative algorithm. At the first iteration we check, step by step, the positions (*u*₁, *w*₂), (*u*₁, *w*₃), …, (*u*₁, *w*_ℓ) and take the smallest *j* for which (*u*₁, *w*_*i*) is an actual edge in *G*. Since (*u*₁, *w*_ℓ) is an edge, therefore such *i* always exists. So we may face to the following configuration: We call the chord *u*₁*w*_*i* the **start-chord** of the current sub-process and *u*₁*w*₁ is the **end-chord**. We will *sweep* the alternating chords along the cycle from the start-edge *w*_*i**u*_*i* (non-edge), *u*_*i**w*_*i*−1 (an edge) toward the end-edge *w*₁*u*₁ (non-edge)—switching their status in twos and fours. We check positions *u*₁*w*_*i*−1, *u*₁*w*_*i*−2 (all are non-edges) and choose the first chord among them, we call this the **current-chord**. (Since *u*₁ ≠ *s* therefore we never have to check more than two edges to find the first chord, and we need to check two edges only once, since there is at most one non- chord adjacent to *u*₁.) **Case 1**: As we just explained, the typical situation is that the current- chord is the “next” one, so when we start this is typically *u*₁*w*_*i*−1. Assume that this is a chord. Then we can proceed with the swap operation *w*_*i*−1*u*_*i*, *w*_*i**u*₁ ⇒ *u*₁*w*_*i*−1, *u*_*i**w*_*i*. We just produced the first “new” realization in our sequence, this is $G_{1}^{\prime}.$ For the next swap operation this will be our new current realization. This operation will be called a **single-step**. In a realization *Z* we call a chord **bad**, if its current status (being edge or non-edge) is different from its status in *G* (or, what is the same, in *G*′, since they differ only on the chords along the cycle 𝓒). After the previous swap, we have two bad chords in $G_{1}^{\prime},$ namely *u*₁*w*_*i*−1 and *w*_*i**u*₁. Consider now the auxiliary matrix $\hat{M}(X + Y - Z)$ (here $Z = G_{1}^{\prime}$). As we saw earlier, for each position outside the chords in 𝓒 the status of that particular position in *Z* is the same as in *X* or *Y* or in both. Accordingly, the corresponding matrix value is 0 or 1. We call a position **bad** in $\hat{M}$ if this value is −1 or 2. (A bad position in $\hat{M}$ always corresponds to a bad chord.) Since in Case 1 we switch the start-chord into a non-edge, it may become 2 in $\hat{M}.$ (In case if in both *X* and *Y* it is an edge. Otherwise it is 0 or 1, so in that case it is not a bad position.) The current-chord turned into an edge. If it is a non-edge in both *X* and *Y* then the value becomes −1, otherwise it does not become a bad position. After this single-step, we have at most two bad positions in the matrix, at most one position with 2-value and at most one with −1-value. **Case 2**: If the previous case does not apply then the pair *u*₁*w*_*i*−1 is a non-chord, therefore we cannot produce the previous swap. Then the non-edge *u*₁*w*_*i*−2 is the current-chord. For sake of simplicity we assume that *i*−2 = 2, this case is represented in. Consider now the alternating *C*₆ cycle: *u*₁, *w*₂, *u*₃, *w*₃, *u*₄, *w*₄. It has a total of three vertex pairs which may be chords. We know already that *u*₁ *w*₃ is a non-chord. If none of the three positions is a chord, then this is an 𝕱-compatible circular *C*₆-swap—and accordingly to the definitions we can swap it in one step. Again, we found the valid swap *w*₂*u*₃, *w*₃*u*₄, *w*₄*u*₁ ⇒ *u*₁*w*₂, *u*₃*w*₃, *u*₄*w*₄. After that we again have 2 bad chords, namely *u*₁*w*₂ and *w*₄*u*₁, and together we have at most two bad positions in the new $\hat{M}(X + Y - Z)$ with at most one 2-value and at most one −1-value. Finally, if one position, say *w*₂*u*₄, is a chord then we can process this *C*₆ with two swap operations. If this chord is, say, an actual edge, then we swap *w*₂*u*₄, *w*₄*u*₁ ⇒ *u*₁*w*₂, *u*₄*w*₄. After this we can take care of the *w*₂, *u*₃, *w*₃, *u*₄ cycle. Along this sequence we never create more, than 3 bad chords: the first swap makes chords *w*₂*u*₄, *w*₄*u*₁ and *u*₁*w*₂ bad ones, and the second one “cures” *w*₂*u*₄ but does not touch *u*₁*w*₂ and *w*₄*u*₁. So along this swap sequence we have 3 bad chords, at the end we have only 2. On the other hand, if the chord *w*₂*u*₄ is not an edge, then we can swap *w*₂*u*₃, *w*₃*u*₄ ⇒ *u*₃*w*₃, *u*₄*w*₂, creating one bad edge, then taking care the four-cycle *u*₁, *w*₂, *u*₄, *w*₄ we “cure” *w*₂*u*₄ but we switch *u*₁*w*₂ and *w*₄*u*₁ into bad chords. We finished our **double-step** along the cycle. In a double-step in any moment we have at most three bad chords. When the first swap uses three chords along the cycle then we may have at most one bad chord (with $\hat{M}$-value 0 or −1) and then the next swap switches back the chord into its original status, and makes two new bad chords (with at most one 2-value and one −1-value). When the first swap uses only one chord from the cycle, then it makes three bad chords (changing two chords into non-edge and one into edge), therefore it may make at most two 2-values and one −1-value. After the second swap there will be only two bad chords, with at most one 2-value, and at most one −1-value. When only the third position corresponds to a chord in our *C*₆ then after the first swap we may have two −1-values and one 2-value. However, again after the next swap we will have at most one of both types. **Remark 14.** When two realizations are one swap apart (so they are adjacent in 𝔾) then we say that their auxiliary matrices are at swap-distance one. Since one swap changes four positions of the matrix, therefore the Hamming distance of these matrices is 4. Finishing our single- or double-step, the previous current-chord becomes the new start-chord. Then we repeat our procedure. There is only one important point to be mentioned: along the step, the start-chord switches back into its original status, therefore it stops being a bad chord. Thus, even if we face a double- step the number of bad chords never will be bigger than three (together with the chord *w*_*i* *u*₁ which is still in the wrong status, so it is bad), and we have always at most two 2-values and at most one −1-value in $\hat{M}(X + Y - Z).$ When *w*₁*u*₂ becomes the current-chord the last step will switch the last start-chord back into its correct status, hence the last current-chord cannot be in bad status. Finally, when the sweep from *w*_*i**u*₁ to *w*₁*u*₁ is finished we only have one bad chord (with a possible 2-value in $\hat{M}$). This concludes the first iteration of our algorithm. For the next iteration we seeks a new start-chord between *w*_*i**u*₁ and *w*_ℓ*u*₁. Chord *w*_*i**u*₁ becomes the new end-chord. We will repeat our sweeping process for this setup, until all chords are processed. If there was a double-step in the first sweep, then it will not occur again, thus there are never more than three bad chords; at most two 2-values and at most one −1-value. However, if the double-step occurs sometime later, for example in the second sweep, then we face one of the following two cases: the circular *C*₆-swap under consideration is either 𝕱-compatible or not. If it is 𝕱-compatible, we perform the circular *C*₆-swap. This does not change the number of bad chords, except if this swap finishes a current sweep. If, however, the circular *C*₆-swap is not compatible, then there exists a chord in the chord-cycle which is suitable for a swap. If this chord is a non- edge, then the swap corresponding to it produces one bad chord, and at most one bad position in $\hat{M}.$ If this chord is an edge in the current realization, then after the first swap there are four bad chords, and there may be at most three 2-values and at most one −1 value. After the next swap (which finishes the double step) we annihilate one of the 2-values, and after that swap there are at most two 2-values and at most −1-value along the entire swap sequence. When we finish our second sweep, then chord *w*_*i**u*₁ will be switched back into its original status, hence it will not be bad anymore. We apply the same algorithm iteratively. After at most ℓ sweep sequences the entire cycle 𝓒 will be processed. This finishes the construction of the required swap sequence (and the required realization sequence). Meanwhile we also proved the following important observation: **Lemma 15.** *Along our procedure each occurring auxiliary matrix* $\hat{M}(X + Y - Z)$ *is at most swap-distance one from a matrix with at most three bad positions: with at most two* 2-*values and with at most one* −1-*value in the same column, which does not coincide with the center of the forbidden star*. # The Analysis of the Swap Sequences Between “Milestones” What remains is to show that the defined swap sequences between *H*_*i* and *H*_*i*+1 satisfy conditions (Θ), (Ω) and (Ξ) of Theorem 12. The first one is easy to see, since we can process a cycle of length 2ℓ in ℓ−1 swaps. Therefore the derived constant *c* in (Θ) is actually 1. Now we introduce the new **switch** operation on 0/1 matrices with forbidden positions: we fix the four corners of a submatrix (none of them is forbidden), and we add 1 to two corners in a diagonal, and add −1 to the corners on the other diagonal. This operation clearly does not change the column and row sums of the matrix. For example if we consider the matrix *M*_*G* of a realization of **d**^𝕱 and make a valid swap operation, then this is equivalent to a switch in this matrix. The next statement is trivial but very useful: **Lemma 16.** *If two matrices have* **switch-distance** *1, then their Hamming distance is* 4. *Consequently if the switch-distance is c then the Hamming distance is bounded by* 4*c*. We prove that property (Ω) holds for auxiliary matrices: **Theorem 17.** *For any realizations X and Y and for any realization Z on a swap sequence from X to Y there exists a realization K such that* $$\begin{array}{r} {\mathfrak{d}\left( \hat{M}(X + Y - Z),M_{K} \right) \leq 16.} \\ \end{array}$$ Due to Lemmas 15 and 16 it is enough to show that: **Lemma 18.** *Any matrix* $\hat{M}(X + Y - Z)$ *with constant column sums (this does not necessarily hold for the center of the forbidden star) and with at most three bad positions (where there are at most two* 2-*values and at most one* −1-*value) can be transformed into a valid M*_*K* *adjacency matrix with at most three switch operations*. *Proof*. Consider now a given $\hat{M}$ which is not necessarily a valid adjacency matrix of a realization. We show in figures the submatrix in this matrix that describes the current alternating cycle 𝓒. If it happens that *s* ∈ 𝓒 then we choose a submatrix representation such that the center *s* of the forbidden star is in the first column. (We choose this submatrix as an illustration tool, but we still consider the entire matrix to work with.) We know that this matrix contains at most two 2-values and at most one 1-value. All three positions are adjacent to the center *u*₁ of the sweeping sequence, hence they are in the same column. For simplicity we denote the center of the sweep as well the column with *u*. The forbidden positions are denoted with ✠. Any column (except column 1) may contain at most one of them, and any row may contain at most two of them. Finally, in the figures the character ⋄ stands for a character which we are not interested in. That is, it can be 0 or 1 or ✠. We distinguish multiple cases, depending on the occurring of values 2 and −1. **Case 1.** Column *u* has one bad position, which can be −1 or 2, or it has two 2-values. Consider at first the case when $\hat{M}\lbrack uw\rbrack = - 1$. By definition this means that chord *uw* is an edge in *Z* but non-edge in both *X* and *Y*. So vertex *w* ∈ *W* has at least one adjacent edge, therefore the row- sum in its row is at least 1. Therefore there are at least two positions in row *w* with entries 1. They are in column *u*₁ and *u*₂. At least one of them, say *u*₁, differs from *s*. Since the column sums are constant, therefore there exists at least two rows *w*₁ such that $\hat{M}\lbrack uw_{1}\rbrack = 1$ while $\hat{M}\lbrack u_{1}w_{1}\rbrack = 0$ or ✠. However, there can be at most one forbidden position in *u*₁, so at least in one of the rows the entry is 0. Using these positions for the corresponding switch it eliminates the bad position without creating a new one. (See.) Before we continue, we prove an important observation: **Observation** If *w* belongs to the alternating cycle 𝓒 and $\hat{M}\lbrack uw\rbrack = 2$ then row *w* contains at least two 0-values. Indeed, there are *α* forbidden chords in row *w*. Since *w* is in an alternating cycle, therefore *d*(*w*) ≤ \|*U*\|−*α*−1. Therefore the sum of row *w* in $\hat{M}(X + Y - Z) \leq |U| - \alpha - 1.$ But it contains a 2 and it does not contain -1 therefore there are at least two 0’s in it. When the single bad value in $\hat{M}$ is 2 then, due to our previous Observation, in its row there are two 0’s. And with them one can repeat the reasoning which we used about the unique −1-value. Finally, when there are two 2-values which raises a very similar situation. Here we can do the same procedure independently on both rows. In this case, however, we need two switch operations. **Case 2.** Here we assume that there is one 2-value and one −1-value in column *u*. For example $\hat{M}\lbrack uw_{1}\rbrack = 2$ and $\hat{M}\lbrack uw_{2}\rbrack = - 1$. Again, in row *w*₂ there are at least two 1-values. *Case 2a* Assume at first that we have *u*₁ ∈ *U* s.t. $\hat{M}\lbrack u_{1}w_{2}\rbrack = 1$ and $\hat{M}\lbrack uw_{1}\rbrack \neq ✠$. Then the corresponding switch will produce $\hat{M}\lbrack u_{1}w_{1}\rbrack = 1/2$ while the other three positions are 0 or 1. (See.) If now $\hat{M}\lbrack u_{1}w_{1}\rbrack = 2$ then we are back to Case 1, and one more switch eliminates the last bad position as well. So we needed at most two switches. *Case 2b* It can happen, that there are only two 1-values in row *w*₂ and both are facing with forbidden positions in row *w*₁. Then at least one 0 in row *w*₂ faces a chord in row *w*₁. (See) The appropriate switch kills 2 bad chords and can make at most one −1-value. At this point we are either finished or back to Case 1. **Case 3.** Finally suppose that there are three bad positions, two 2-values at positions *uw*₁ and *uw*₂ and one −1-value at position *uw*₃. Now both rows *w*₁ and *w*₂ contain at least two 0’s. If any of them face a 1 in row *w*₃ then an appropriate switch annihilates one 2 and one −1 and does not create new bad position. We are back to Case 1. Altogether we need two switches. If this is not the case then we consider the following: assume that $\hat{M}\lbrack u_{1}w_{1}\rbrack = 0.$ Since the column sums are the same, and we assumed that $\hat{M}\lbrack u_{1}w_{3}\rbrack = 0$ therefore there exists a row *w*₄ s.t. $\hat{M}\lbrack u_{1}w_{4}\rbrack = 1$ while $\hat{M}\lbrack uw_{4}\rbrack = 0.$ Then we can switch this 2-value without making a new bad position. After that we are back to Case 2. Altogether this requires at most three switches. The proof of Lemma 18 is finished. If this is not the case then we consider the following: assume that $\hat{M}\lbrack u_{1}w_{1}\rbrack = 0.$ The column sums are the same, and we assumed that $\hat{M}\lbrack u_{1}w_{3}\rbrack$ = 0 or ✠. Therefore the difference between column sums in *u* and *u*₁ is 1 due to rows *w*₁ and *w*₃, and the difference increase at least 1 for row *w*₂, where against a 2-value in column *u* there is either 1 or 0 in column *u*₁. Therefore there exists at least two further rows, where there is a 1 in column *u*₁ against a 0 or ✠ in column *u*. Since column *u* can contain at most one ✠, one of the rows must contain a 0. Let it be denoted by *w*₄. Hence $\hat{M}\lbrack u_{1}w_{4}\rbrack = 1$ while $\hat{M}\lbrack uw_{4}\rbrack = 0.$ Then we can switch this 2-value without making a new bad position. After that we are back to Case 2. Altogether this requires at most three switches. We finished the proof of Lemma 18. There was no word yet about condition (Ξ) in Theorem 12. We discuss this in the next Section, because the magnitude of parameter 𝔹 heavily depends on. To finish the Theorem 12, let us assume for now that we have the proper upper bound on 𝔹. Then (Ω) in Theorem 12 and therefore the theorem itself is proved as well. Thus, our Markov chain is rapidly mixing as Theorem 10 stated. # Some further technical details of the Sinclair’s method In the last two sections we proved the rapidly mixing nature of our proposed MCMC method on the 1F1S restricted degree sequence problems through a special instance of Sinclair’s method, developed in. However, we need slightly generalize this method in order to finish the proof. Let’s recall that the method takes two realizations, *X* and *Y*, of the same degree sequence. It considers all possible ordered circuit decompositions of the symmetric difference of the edge sets, then it uniquely decomposes each such decomposition into an ordered sequence 𝓒 = *C*₁, …, *C*_*m* of oriented cycles. Based on this latter decomposition the method determines a well defined unique path between *X* and *Y* in the Markov chain 𝔾. To find this unique path the method first defines a sequence of “milestones”. These are different realizations *X* = *H*₀, *H*₁, …, *H*_*m*−1, *H*_*m* = *Y* of the degree sequence where the edge set of any two consecutive realizations *H*_*i*−1, *H*_*i* differ exactly in the edges along the cycle *C*_*i*. (Until this point no swap operation happened.) In the next phase, for any particular *i* = 0, …, *m*−1 the method determines a sequence of valid swap operations transforming *H*_*i*−1 into *H*_*i*—describing a unique path *Z*₀, *Z*₁, …, *Z*_ℓ between *H*_*i*−1 and *H*_*i* in the Markov chain 𝔾. This sequence of course depends on the available swap operations. In these are the usual (bipartite) circular *C*₄-swap operations. In this work these correspond to the restricted swap operations. These operations, while exchanging chords in the realizations along the alternating cycle *C*_*i*, also use some further chords. Therefore the edge set of any *Z*_*j* is not completely contained by *E*(*X*) ∪ *E*(*Y*); there exist a small number of edges in *Z*_*j* which are non-edges in *X* and in *Y*, or non-edges in *Z*_*j* but are edges in *X* and *Y*. If *Z*_*j* is between the milestones *H*_*i*−1 and *H*_*i*, then *C*_*k* for *k* ≤ *i*−1 alternates in *Z*_*j*, and *C*_*i* alternates with a “small error”: there is a very small number of vertices where the alternation does not hold. Sinclair’s method requires this number to be small. In the original application this number is actually one (See, Section 5, (F)(c).) In the original application this number is actually one. Here, as we saw in Section “Milestones”, this number is three: that many bad chords may occur after any particular ReDeSe. As we saw all these chords are adjacent to the same vertex *u*₁. These numbers are used by our method to determine the size of a parameter set 𝔹. This parameter set must have a polynomial size. When we have one bad chord, then it is determined by its end points—there are at most *n*² possibilities for them. This contributes with an *n*² multiplicative factor to the size of 𝔹. When we have at most three bad chords, then they can be chosen in at most *n*⁴ ways: vertex *u*₁ is fixed (*n* different choices), while the other three end points can be chosen at most *n*³ independent ways. Altogether it contributes with an at most *n*⁴ multiplicative factor to the size of 𝔹. This remark finishes the proof for the case of 1F1S restricted swap operations. # 1F1S Restricted Degree Sequence Problem is a Self-Reduced Counting Problem In Computer Science there are two special complexity classes, FPRAS and FPAUS, which are concerned with the approximability of counting problems. One can find detailed definitions for these complexity classes, for example, in. Here we only give a sketchy description of the points that are important to our case. Roughly speaking a counting problem is in FPRAS (Fully Polynomial Randomized Approximation Scheme) if the number of solutions can be estimated fast with a randomized algorithm, such that the estimation has a small relative error with high probability. A counting problem is in FPAUS (Fully Polynomial Almost Uniform Sampler) if the solution can be sampled fast with a randomized algorithm that generates samples following a distribution being very close to the uniform one. It is easy to see that a counting problem is in FPAUS if there is a rapidly mixing Markov chain for which a starting state can be generated in polynomial running time; one step in the Markov chain can be conducted in polynomial running time; and the relaxation time of the Markov chain grows only polynomially with the size of the problem. The Markov chain we defined in the 1F1S problem satisfies all these requirements. Jerrum, Valiant and Vazirani proved that any *self-reducible counting problem* is in FPRAS iff it is in FPAUS. A counting problem is self-reducible if the solutions for any problem instance can be generated recursively such that after each step in the recursion, the remaining task is another problem instance from the same problem, and the number of possible branches at each recursion step is polynomially bounded by the size of the problem instance. Clearly, a graph with prescribed degree sequence can be built recursively by telling the neighbors of a node at each step, then removing the node in question and reducing the degrees of the selected neighbors. However, this type of recursion does not satisfy all the requirement for being self-reducible since there might be exponentially many possibilities how to select the neighbors of a given vertex. On the other hand, the degree sequence problem with a forbidden one factor and one star is a self-reducible counting problem. Indeed, consider the center of the (possibly empty) star, *s* ∈ *U*, and the vertex *v* ∈ *V* with the smallest index for which (*s*, *v*) is a chord. Any solution for the current problem instance belongs to one of the following two cases: The chord (*s*, *v*) is not present in the solution. In this case, extend the size of the star by adding chord (*s*, *v*) to the forbidden set, and do not change the degrees. This is another problem instance from the 1F1S problem, whose solutions are the continuations of the original problem belonging to this case. The chord (*s*, *v*) is present in the solution. In this case, extend the size of the star by adding chord (*s*, *v*) to the forbidden set, and decrease both *d*_*s* and *d*_*v* by one. The new degree sequence is still a bipartite 1F1S restricted degree sequence which is half-regular in class *U* (except, possibly, at vertex *s*), and the solutions of this new problem extended with the previously decided step provide solutions to the original problem. Since the 1F1S counting problem is a self reducible counting problem, and we proved that it is in FPAUS therefore it is also in FPRAS: via our sampling process one can solve the approximate counting problem with high probability. We finish this paper with a short analysis of the connections between our approach and the paper of Jerrum, Sinclair and Vigoda. Their seminal result from 2004 solved the uniform sampling problem of perfect 1-factors of a given graph. As their Corollary 8.1 pointed out this method can be applied for uniform sampling of the set of all possible realizations of a given *f*-factor of a complete graph. It also proves that the problem is in FPAUS therefore in FPRAS as well. Since the restricted degree sequence problem in general is equivalent to the *f*-factor problem, therefore our 1F1S ReDeSe problem is only a special case of the *f*-factor problem, so the JSV result applies to it. This describes the similarity. The important differences lay in the swap operations applied in the JSV method and in the Kannan-Tetali-Vempala Markov chain. In the JSV method a special graph 𝔊 is introduced for the sampling via Tutte’s gadgets. Then the swap operations are working on the graph 𝔊 with the unintended result that for a (sometimes very long) sequence of swaps does not change at all the generated *f*-factor. Combining this issue with the known relative slow mixing time of the Jerrum- Sinclair-Vigoda’s Markov chain, the resulted approach in not suitable for any practical application. Our Markov chain operates in the original graph and each jump provides a new realization of the original degree sequence problem. Therefore our Markov chain is presumably much faster than the JSV chain, furthermore the JSV theorem does not proves the rapidly mixing nature of our Markov chain. Similarly it does not prove that this Markov chain is a self reducible procedure. PLE and IM acknowledge financial support from grant \#FA9550-12-1-0405 from the U.S. Air Force Office of Scientific Research (AFOSR) and the Defense Advanced Research Projects Agency (DARPA). PLE was partly supported by the Alexander von Humboldt-Foundation, when this author visited Universität Hamburg in Fall of 2014. SZK was partly supported by Hungarian NSF, under contract K77476 and NK105645. IM was partly supported by Hungarian NSF, under contract PD84297. LS was partly supported by Hungarian NSF, under contract NK 83726 A preliminary version of this paper can be found as *arXiv* **1301.7523** [^1]: The authors have declared that no competing interests exist. [^2]: Wrote the paper: PLE SZK IM LS. [^3]: Current address: Reáltanoda u 13-5, Budapest, H-1053, Hungary

# 1. Introduction The pandemic induced mayhem including the lockdowns poses challenges to all. If the public is worried about “life vs livelihood,” it also burdens the health care professionals differently. Rising up to the occasion during these trying times, and also the inherent risk of getting the disease for self and their own close families has its own effect on the psyche of a HCP. The psychological burden and trauma inflicted upon the treating health care professionals gets translated into secondary/vicarious trauma. Recent studies on Chinese health care professionals who dealt with the first and largest outbreak of the COVID-19 infections had shown that the frontline health care workers such as doctors and nurses faced depressive symptoms, insomnia and anxiety as compared to non-frontline healthcare workers. The findings are also supported by a meta-analysis and systematic review done by Pappa and colleagues which indicated the prevalence of anxiety, insomnia and depression amongst primary health care professionals. Similar results were also found in a large scale survey done on Indian health care professionals where they found that the health care professionals reported higher rates of burnouts during the COVID era. Secondary Traumatic stress (STS) is a neglected entity experienced by the HCPs during this unprecedented situation. STS is defined as a natural consequent behaviour and emotions resulting from knowledge about a traumatising event experienced by a significant other. It is stress resulting from helping or wanting to help a traumatised or suffering person and it comprises of symptoms including intrusion, avoidance, and arousal. These STS symptoms share similarities with those of posttraumatic stress disorder (*PTSD*), as suggested in the 4th edition of the Diagnostic and Statistical Manual of Mental Disorders \[DSM-IV-TR; APA\]. However, unlike PTSD, STS could be due to indirect contact in a professional context (e.g., caring for a traumatized patient). While dealing with the COVID-19 affected patients; the roles of HCPs evolved drastically, now venturing into areas other than their area of specialization due to the lack of professionals available to keep up with the demand. Besides uncertainties about the disease, its outcomes and more importantly the knowledge of risking self and their families creates occupational stress among HCPs. Reports suggest that in addition to work stress, HCPs has faced new problems such as being verbally and physically assaulted by mobs. A shortage in personal protective equipment (PPE) was also is a significant concern among healthcare professionals. Furthermore, improper training and guidelines regarding PPE raised several concerns. One of the major protective factors among healthcare professionals is their optimistic attitude to cope with the stress related to their profession. ***Optimism*** can be defined as the ability to look at the brighter side of things. Specific skills such as optimism, interpersonal skills, hope, and faith can protect one’s mental health. Lack of literature on exploring the mental health status and possible coping factors of the Indian health care professionals during the pandemic and our own personal experiences led to the conceptualisation of the current study. Hence, the present study aims to report secondary traumatic stress levels, optimism, and mood states experienced by the HCPs within the Indian subcontinent during the COVID-19 pandemic. # 2. Material and method ## 2.1 Participants and procedures For the present cross sectional study, information associated with the level of secondary traumatic stress (STS), optimism and mood state during COVID-19 among health care professionals in India were collected. Due to lockdown and to reduce human contact and transmission risk related with the disease; online platforms such as Google forms and social media were utilized as a mode of data collection. The Google forms were circulated on various groups and social media (LinkedIn, WhatsApp) and invited health care professionals from different cities across India to complete the questionnaire voluntarily. Additionally, we also sent the questionnaire to many health care professionals who had cooperated with us, and used their contact network to spread the questionnaire, utilizing snowballing method. For the safeguarding of data (also mentioned in ethical clearance document); all the data was collected from primary research supervisor’s institutional email address and every two weeks the data was removed and secured in an external hard drive which was not connected with internet and was not accessible to anyone but the primary researchers. The respondents were English proficient health care professionals above 20 and below 65 years of age in India (including doctors, nurses, physiotherapist, lab technicians, dieticians, administrative staff and clinical pharmacist). Before collecting responses, in the consent form and safeguard process for maintaining the anonymity of the data; we stated the purpose of the investigation, and responses were collected only after obtaining the consent. This questionnaire was anonymous. The data collection took place in the months of April & May 2020, which was the initial national lockdown, for first wave of COVID-19 to hit India. The survey began on April 16, 2020, and ended on May 15, 2020, when India was in a complete lockdown period due to outbreak of COVID-19. Due to the nature of Google forms any incomplete questionnaires were not accepted. We collected a total of 2153 questionnaire out of which only 2008 were valid and finally used for analysis. We excluded those observations which were inconsistent or were inappropriately filled and those which were not consented. ## 2.2 Ethics statement The approval was obtained from The Ethics committee of Manipal hospitals, Bangalore (ECR/34/Inst/KA/2013/RR-19). Our investigation process remained anonymous, and no identifiers (such as name, address, email id, phone numbers, name of hospital employed) were collected. Every participant was informed about and understood the purpose of our investigation before entering the study. # 3. Measurements In order to understand the socio-demographic profile of the population, information on individual’s age, gender, marital status, occupation (doctors, nurses, and allied healthcare professionals), years of experience, type of practice (clinic and hospital) and their current state of practice were collected. During the COVID-19 lockdown period, the present study has been conducted during lockdown; therefore, it is assumed that all the doctors irrespective of their specialty where engaged in same duties. The mental health status was assessed using below describes scales: ## 3.1 Secondary Traumatic Stress Scale The Secondary Traumatic Stress Scale (STSS) is a self-report inventory designed to assess the frequency of STS symptoms in professional caregivers. The STSS is a 17-item measure explicitly designed to assess the effects of healthcare providers’ exposure to secondary trauma from patient experiences. Unlike other measures that include items related to burnout or compassion satisfaction, the 17 STSS items correspond to the 17 DSM-IV PTSD symptoms for Criteria B (Intrusion), C (Avoidance), and D (Arousal; American Psychiatric Association). Respondents indicate how often they experienced each symptom in the past seven days on a Likert-type scale ranging from 1 ("never") to 5 ("very often"). In place of assessing Criterion A of the diagnostic guidelines for PTSD (trauma exposure): The STSS use “prompts" that suit professional’s setting, for instance. In order to fit the emergency room environment, the word "client" was changed to "patient" in all relevant items. By replacing Criterion A (trauma exposure) for DSM-IV-TR PTSD with these prompts, the STSS largely mirrors PTSD from a secondary stressor, which is the definition of STS used in the present study. The fact that the STSS closely mirrors the DSM-IV-TR criteria for PTSD allows it to be validly compared—albeit with some caution—to DSM-IV-TR PTSD. In prior research, the STS showed good psychometric properties. The STSS has acceptable psychometrics as measured by convergent (mean *r* = 0.39) and discriminant (mean absolute *r* = 0.07) validity. The STSS has high overall internal consistency based on Cronbach’s alpha values (*α* = 0.93), and acceptable internal consistency for the symptom cluster sub-scales (Intrusion, *α* = 0.80; Avoidance, *α* = 0.87; Arousal, *α* = 0.83). Additionally, its tree- structure model is supported by confirmatory factorial analysis, although the factors are inter-correlated. ## 3.2 Life orientation test-revised Life orientation test revised (LOT-R) is a 10-item scale that measures how optimistic or pessimistic people feel about their future. Respondents use a 5-point rating scale (0 = strongly disagree; 4 = strongly agree) to show how much they agree with 10 statements about positive and negative expectations. These statements include “In uncertain times, I usually expect the best” and “If something can go wrong for me, it will.” The internal consistency (Cronbach’s alpha) ranged between.74 and.78. ## 3.3 Mood (visual analogue scale) Mood visual analogue scale (VAS) (0- extremely sad to 10- extremely happy) is a psychometric response scale which is used to measure subjective characteristics or attitudes and have been used in the past for measuring the multitude of disorders. The Cronbach’s alpha values test-retest reliability for mood visual analogue scale (VAMS) ranged between 0.71 and 0.80. # 4. Statistical analysis For the exploratory analysis, mean and standard deviations of continuous variables and proportions for categorical variables were used to describe the levels of secondary traumatic stress (STS), optimism/ pessimism, and current mood states in the sample based on profession and gender of the healthcare professionals in the study. Secondary Traumatic Stress Scale, Life Orientation Test Revised and Mood Visual Analogue Scale were used to measure the Secondary Traumatic Stress Symptoms (intrusion, avoidance and arousal), optimism parameters & state of mood respectively. Regression analysis was further used to explore changes in secondary traumatic stress, optimism and mood states. Significance of all statistical tests’ were defined as bilateral P\<0.01. SAS university edition was used to analyse data in the study. # 5. Results ## 5.1 General characteristics The number of participants who participated was 2153, of which complete information was available for 2008 (93%) individuals, which is considered as a population for present study. Among the study sample, 1027(51.15%) were females. Mean age was 35.7(± 11.9) years; females (mean 29.7 \[± 8.9\] years) were younger than males (mean 41.9 \[± 11.5\] years). The majority were married (60.2%), percentage of married males (80.2%). Most HCPs were nurses (924, 46%) followed by doctors (611, 30.4%) and the remaining were other allied healthcare professionals. The population is classified in three broad categories based on their clinical roles, viz., doctors, nurses and allied healthcare professional (physiotherapist, dentists, lab technicians, dieticians, administrative staff and clinical pharmacists). In the population, majority (1109, 55.2%) of the respondents were practising at hospitals having ICU facilities, and among them (738, 66.6%) were females. The remaining respondents (899, 44.8%) were practising at hospitals without having ICU facilities, among them majority were males (610, 67.9%). Mean years of experience in the field of HCPs were 11.0\[± 15.8\] years, females were less experienced than males (mean 7.1 \[± 18.7\] years) vs mean 15.1 \[± 10.4\] years). ## 5.2 Secondary traumatic stress The key clinical characteristics of the present study are the STSS and optimism (LOT-R) levels of the HCPs. Secondary traumatic stress (STS) was experienced by 1548 (77%) of the HCPs. The doctors and nurses showed more STS than others HCPs, and STS decreased with increase in the age. In the study sample, on STS Categorisation—among doctors, 11.8% had no STS, and 19% had Severe STS, among the Nurses 20.8% had no STS, and 8.2% had severe STS, among the Allied HCPs 41.4% had no STS and only 7.4% had severe STS. In the study sample, there was a male preponderance for “Mild STS” (above 30%) amongst doctors and allied health care professionals, but among the nurses there was female preponderance (above 40%). In “moderate STS” category, female preponderance (above 20%) was noted amongst doctors and allied health professional and males showed preponderance in the nursing category. In “high STS” category there was no gender bias amongst doctors and nurses, whereas female preponderance was noted among the allied health care professionals. “High” (74, 12.1%) and “Moderate STS” level (129, 21.1%) were also found more among doctors than that of nurses and allied healthcare professionals. The doctors were found to be high on “Severe STS” level (116, 19.0%), followed by nurses (76, 8.2%) and the other allied healthcare professionals (35, 7.4%). In “severe STS” female doctors and female allied health care professionals were noted to be higher than their male counterparts (10%). Mean score on Intrusion Scale, which measures intrusive thoughts related to trauma, flashbacks and recollections was found to be 10.2 (±3.75); of which female reported a mean score of 10.7 (± 3.33), which is slightly higher than those of males (mean score of 9.62 (± 4.08)). This indicated high intrusive thoughts among females as compared to males. On the Avoidance Scale, which measures the attempts to avoid any stimuli or triggers that might be related to the traumatic event, the participants reported an overall mean score of 14.4 (±4.6); mean score of females being 14.6 (±4.3) whereas males mean score being 14.2 (±4.9); indicating both males and female utilizing avoidance as a coping strategy. On the arousal scale which indicates jumpiness, irritability, insomnia, decreased concentration and hyper vigilance the participants reported a mean score of 10.8 (± 3.56); females reporting a mean score of 10.7(± 3.4) and males mean score of 10.8(± 3.7). ## 5.3 Optimism (LOT-R) and perceived mood state Life Orientation Test-Revised (LOT-R) is another key component, which measures the dispositional optimism of an individual. shows that among the HCPs, high Optimism was mostly observed among doctors (244, 39.3%), followed by nurses (247, 26.7%) and allied healthcare professionals (108, 22.8%), whereas, in cases of low optimism category, the order changed i.e., allied healthcare professionals (73, 15.4%), followed by doctors (67, 11.0%) and nurses (86, 9.3%). The perceived mood state of the HCPs was assessed with the help of a mood visual analogue scale (11-point Likert scale; where 0 = extremely sad and 11 = extremely happy); the overall mean for the sample was found to be 5.68 (± 2.26) indicating moderate mood states reported by participants at the time of taking the survey. Gender-wise mean mood VAS score were similar between the groups \[females: 5.71 (± 2.34) and males: 5.65 (± 2.18)\]. In mood VAS among females, more nurses were having “neutral” mood than the doctors and allied HCPs. On the analysis of happiness based on the VAS score, we found nurses to be happier than doctors and allied HCPs. In mood VAS, more allied male HCPs had “neutral” mood as compared to doctors and nurses. To summarise the regression analysis, doctors and nurses showed happier mood when compared to others HCPs. In STS, doctors and nurses showed more STS than others HCPs, and STS decreased with increase in the age. The doctors and nurses had shown higher optimism than others HCPs. Females HCPs experienced higher “sad” mood as compared to males. # 6. Discussion COVID-19 came with a threat package of being highly contagious in nature with rapid spread across the globe, and warranted an unprecedented situation to be faced by the medical fraternity. The current study showed that 77% (N = 1548) of HCPs (doctors, nurses, and allied health care professional) reported prevalence of STS. Severe STS was reported at a higher rate among doctors as compared to nurses and allied HCPs, which is similar to the earlier study findings published during the pandemic. There was a difference in the patterns of responses among female and male participants who showed signs of intrusive thoughts, using avoidance as a coping mechanism and arousal when faced by triggers in the environment. The results show that female health care professionals showed higher levels of secondary traumatic stress (also related to symptoms of post-traumatic stress) as compared to their male counterparts, especially doctors and nurses as compared to other health care professionals. In the other studies conducted in India related to burnout and distress among doctors and nurses during the time of COVID-19, the health care professionals also showed significant burnout due to their direct contact and involvement in their work with pandemic related work and patient involvement. Though no comparative data exist for pre-COVID-19 or pandemic and related secondary traumatization studies in Indian health care professionals, earlier studies suggest that burnout is associated with professional life experienced by doctors and nurses in India. Various Indian researches including a systematic review and meta-analysis (during COVID-19) also throws light on the presence of high levels of stress- related disorder among health care workers such as anxiety, depression, insomnia, hopelessness during the pandemic. A study by Li et al. informed that vicarious traumatization (STS) adversely affected both medical and non-medical staff; also the vicarious traumatization was worse in non-front line medical workers as compared by frontline medical staff. The current study sheds light on the reported mood states along with the traumatic stress and pessimism, experienced during patient care by the HCPs. The results show that neutral moods were recorded across spectrum between both male and female health care professionals. The findings are in line with the study among healthcare professionals during the pandemic conditions in India; where they showed signs of various mood and anxiety disorder like symptoms. The STS and burnouts have been reported higher in other studies as well, where the data collection was during a similar period of COVID-19 spread peak. The results of our study are consistent with the studies done on nursing students during the SARS pandemic. ## 6.1 Suggested intervention As the pandemic peaks, the disease related psychological burden also spirals high in the neglected healthcare providers. Age related variance and marital status contributes to fear of transmission of the disease to the family and job insecurity. Identifying and addressing these mental health issues and ensuring both physical and psychological safety should become the priority for not only front liners but for everyone in the field of medicine. The deleterious effect of the pandemic on the mental health status of HCPs is important to be addressed on a war footing. Early recognition and intervention to tackle these issues would go a long way to prepare the HCPs to cope with this situation and to give their best to the society. High level of optimism helps to cope with pandemic stress and foster lower level of psychological problems. Resources such as psycho-social support, leisure time and improvement in infrastructure adaptations in hospitals could help improve their mental health. No war could be won if warriors are fighting demons within themselves. We propose the following interventions: 1. Early Recognition of the Mental health issues of HCPs and their families. 2. Easy, Free and Confidential access to the counsellors / psychologists/ psychiatrists. 3. Building a peer network within the co-workers to provide a psychological support system at work place. 4. Emphasise on Work—Life Balance. 5. Positive Reinforcement System. # 7. Limitations Irrespective of the large data set and strength of the study, there are certain limitations. Firstly, utilization of the cross-sectional design, lack of homogeneity at various levels, and over-representation of a particular group of healthcare providers, could have played a mediating role in the results, interfering with causality analysis between the variables of the study. Secondly, the participants recruited with the help of Google forms shared on social media with snowballing effect, by virtue of this methodology utilised, an over representation of technology savvy participants could have happened contributing to the bias. We have not excluded patients with prior psychiatric ailments and addictions. Besides, psychological health is influenced by various factors, including the personal and professional situation, besides the situation created by the pandemic, it has increased the workload and safety concerns, other factors such as family support, job stress, disturbed daily activities could have contributed significantly to the overall health and quality of life. # 8. Conclusion This study sheds light on the levels of distress and secondary trauma experienced by healthcare professionals in India during COVID-19 pandemic. Various factors such as the sudden outbreak of the disease, rampant spread, lack of preparedness, uncertain management guidelines, besides risks for self and family, could have been critical intervening factors to create distress and burnout amongst HCPs, making them vulnerable to various mental health and physical health issues during the pandemic. There is immediate need for focus on the secondary traumatic stress experienced by the healthcare provider. This study further emphasises the need for social and administration level support in helping to build better healthcare policies to cater the need of HCPs. This study is a call for Saving the Saviour and a humble request to throw light on the darker side of being HCPs in this current situation. Under current circumstances, it is important to evaluate, understand and prioritize the mental health needs of the health care professionals. Current research highlights the mental health needs of HCPs and calls for the policy makers and administrators to prioritise the mental health interventions for the HCPs, enabling them to not only cope up but also to serve the community at large. This becomes an important study throwing light on the prevailing unaddressed mental health state of HCPs. # Supporting information The authors would like to thank both of the anonymous reviewers and editorial team for their valuable comments and suggestions that helped us to improve the article’s quality. 10.1371/journal.pone.0257429.r001 Decision Letter 0 Inbaraj Leeberk Raja Academic Editor 2021 Leeberk Raja Inbaraj 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 May 2021 PONE-D-21-14477 Prevalence and severity of secondary traumatic stress and optimism in healthcare professionals in India during COVID-19 lockdown. PLOS ONE Dear Dr. Parashar, 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 11 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, Leeberk Raja Inbaraj, MD Academic Editor PLOS ONE Journal Requirements: 1\) 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. 2\) 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 <u><https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_m ain_body.pdf></u> and [https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_titl e_authors_affiliations.pdf](about:blank) 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 more 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 sensitive information, data are owned by a third-party organization, etc.) 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. 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 your tables as part of your main manuscript and remove the individual files. Please note that supplementary tables (should remain/ be uploaded) as separate "supporting information" files 6\) Please amend either the title on the online submission form (via Edit Submission) or the title on the first page of your manuscript (title page) so that they are identical. \[Note: HTML markup is below. Please do not edit.\] 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: 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: 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: Since Doctors and Nurses reported to have high level of stress, more among doctors-author can a make statement about the relationship of specialty of a doctor and experienced stress. Author can share recommendations about management of stress for HCPs,it will be helpful for policy makers to implement. Author can specify the Cronbach alpha of the used tests for the study. Reviewer \#2: Authors have gone for a very novel idea which is very much relevant in present times. There are definitely few grammatical errors which must be addressed. Authors need to mention more Indian studies - few are from Indian Journal of Psychiatry and Asian Journal of Psychiatry. \*\*\*\*\*\*\*\*\*\* 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: **Fazle Roub Bhat \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] 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.0257429.r002 Author response to Decision Letter 0 1 Aug 2021 RESPONSE TO REVIEWERS’REPORT(S) Reviewer 1 Response \(a\) Since Doctors and Nurses reported to have high level of stress, more among doctors-author can a make statement about the relationship of specialty of a doctor and experienced stress. During the COVID-19 lockdown period, all the doctors irrespective of their specialization were involved in COVID duties only. Based on their role in broader perspective the clinical group is classified into three subgroups. \[Page-5, paragraph 2; line 5-8\] \(b\) Author can share recommendations about management of stress for HCPs, it will be helpful for policy makers to implement. Page-13, paragraph 4; line 5-10 \(c\) Author can specify the Cronbach alpha of the used tests for the study. 1. Secondary traumatic Stress Scale (Page-6, paragraph 2) 2\. Life Orientation Test-Revised (Page-6, paragraph 3) 3\. Mood (VAS) (Page-7, paragraph 1) New reference added for Mood (VAS) Cronbach score. 23\. Flynn D, van Schaik P, van Wersch A. A comparison of multi-item likert and visual analogue scales for the assessment of transactionally defined coping. Eur J Psychol Assess. 2004;20:49–58. Reviewer 2 Response There are definitely few grammatical errors which must be addressed. Needful changes have been incorporated at various places in the revised manuscript. (yellow highlights, apart from the changes mentioned in the list). Authors need to mention more Indian studies - few are from Indian Journal of Psychiatry and Asian Journal of Psychiatry. Page-12, paragraph 2 and Page-14, paragraph 4; line 4-10 (multiple new and relevant Indian studies added). 33\. Arslan, G., Yıldırım, M., Tanhan, A. et al. Coronavirus Stress, Optimism- Pessimism, Psychological Inflexibility, and Psychological Health: Psychometric Properties of the Coronavirus Stress Measure. Int J Ment Health Addiction. 34\. Mathur S, Sharma D, Solanki RK, Goyal MK. Stress-related disorders in health-care workers in COVID-19 pandemic: A cross-sectional study from India. Indian J Med Spec 2020;11:180-4 35\. Chatterjee SS, Chakrabarty M, Banerjee D, Grover S, Chatterjee SS, Dan U. Stress, Sleep and Psychological Impact in Healthcare Workers During the Early Phase of COVID-19 in India: A Factor Analysis. Front Psychol. 2021 Feb 25;12:611314. 36\. Singh RK, Bajpai R, Kaswan P. COVID-19 pandemic and psychological wellbeing among health care workers and general population: A systematic-review and meta- analysis of the current evidence from India. Clin Epidemiol Glob Health. 2021 Jul-Sep;11:100737. 37\. Grover S, Dua D, Shouan A, Nehra R, Avasthi A. Perceived stress and barriers to seeking help from mental health professionals among trainee doctors at a tertiary care centre in North India. Asian J Psychiatr 2019;39:143-9. 38\. Banerjee D, Vijayakumar HG, Rao T S. ”Watching the watchmen:” Mental health needs and solutions for the health-care workers during the coronavirus disease 2019 pandemic. Int J Health Allied Sci 2020;9, Suppl S1:51-4 Other comments/ suggestions: 1\. Change in the name from Diagnostic and Statistical Manual of Psychiatric Disorders and DSM-IV”. 2\. Mention the timeline of first national lockdown. 1. Page-3, paragraph 43 lines 6 to “Diagnostic and Statistical Manual of Mental Disorders and DSM-IV- TR”.) 2\. Page-5, paragraph 1, line 8 (The first lockdown was initially stated to be from 25 March 2020 – 14 April 2020 but it was extended post April 14.-in certain states.) 1.Change in table numbering (as table number 1 related to STSS score was added). 1\. Table 1 : Page 6 (new addition) 2\. Earlier Table 1= Current Table 2 (page-8) 3\. Earlier Table 2= Current Table 3 (page-10) 4\. Earlier Table 3= Current Table 4 (page-11) \*Table numbers were also updated respectively in the revised manuscript (wherever mentioned). Note: All of the changes in the revised manuscript are highlighted using YELLOWcolour. 10.1371/journal.pone.0257429.r003 Decision Letter 1 Inbaraj Leeberk Raja Academic Editor 2021 Leeberk Raja Inbaraj This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 1 Sep 2021 Prevalence and Severity of Secondary Traumatic Stress and Optimism in Indian health care professionals during COVID-19 lockdown PONE-D-21-14477R1 Dear Dr. Parashar, 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, Leeberk Raja Inbaraj, MD 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: 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 \*\*\*\*\*\*\*\*\*\* 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: Recommended in terms of good statistical analysis, data support the conclusions,the manuscript is written well. \*\*\*\*\*\*\*\*\*\* 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.0257429.r004 Acceptance letter Inbaraj Leeberk Raja Academic Editor 2021 Leeberk Raja Inbaraj This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 16 Sep 2021 PONE-D-21-14477R1 Prevalence and Severity of Secondary Traumatic Stress and Optimism in Indian health care professionals during COVID-19 lockdown Dear Dr. Parashar: 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. Leeberk Raja Inbaraj Academic Editor PLOS ONE [^1]: The authors have declared that no competing interests exist.

# Introduction Steroid 21-hydroxylase deficiency is the most frequent cause of congenital adrenal hyperplasia (CAH), accounting for more than 90 to 95% of CAH cases. Due to a lack of negative feedback from cortisol, ACTH stimulation increases, shifting the precursors of steroidogenesis toward androgen synthesis. There is a spectrum of clinical forms, traditionally divided into classical and nonclassical (NC) forms. In the classical form, in addition to the manifestations of cortisol insufficiency, female patients usually present with prenatal external genital virilization, and both sexes present with postnatal virilization. Additionally, approximately 70–80% of patients also present with severe impairment of aldosterone production, resulting in hyponatremic dehydration during the first weeks of life. In the NC form, the hyperandrogenic signs begin later in life: children typically present with precocious pubarche, while adolescents and adults present with hirsutism, menstrual abnormalities and/or infertility. Clitoromegaly has also been observed in both children and adults, occurring in up to 7–10% of NC cases. In fact, these classical and NC forms reflect different impairments of enzymatic activity caused by *CYP21A2* mutations. In CAH, there is a good correlation between genotypes and phenotypes; that is, homozygous patients carrying mutations resulting in total/severe (\<7%) and moderate (20–50%) enzymatic activity impairments generally present with the classical and NC forms, respectively. Genotypes predicting the NC form carry mild mutations in homozygosis or in compound heterozygosis with severe mutations. Despite the presence of the mild allele, some studies have reported a modulatory effect of the severe allele on the NC phenotype: these patients presented higher ACTH-stimulated 17OH- progesterone (17OHP) levels, higher basal androgen levels and/or earlier onset of hyperandrogenic manifestations compared to those who are homozygous for mild mutations. However, this hypothesis has been debated in the literature because other reports have not found these correlations. Previously, in our cohort consisting of 114 NC patients, we did not identify correlation among the severity of NC genotypes, age at the beginning of hyperandrogenic manifestations, basal serum androgen levels and the presence of virilizing signs. Interestingly, similar frequencies of both NC genotypes (mild and severe) were observed in asymptomatic female patients and in those with slight clitoromegaly. These data suggested that individual differences in the peripheral androgen sensitivity, modulated by the androgen receptor and 5-α-reductase type 2 genes, could account for this phenotypic variability. Androgens act via the androgen receptor (AR), the gene for which carries a polymorphic CAG tract at exon 1, varying from 8 to 35 repeats in the normal population. Variations in the CAG repeat numbers (nCAG) have been inversely correlated with AR transactivation activity and consequently with androgen phenotypic variability. Shorter tracts have been associated with idiopathic precocious pubarche, increased severity and earlier age onset of prostate cancer, whereas longer CAG tracts have been associated with oligospermic infertility. Dihydrotestosterone (DHT), the main active androgen binding to the androgen receptor (*AR*), is converted from testosterone by the action of 5α-reductase type 2 enzyme. The 5α-reductase type 2 gene can carry two frequent polymorphisms, V89L and A49T, which modify the enzymatic activity and influence the phenotypic variability of androgen-dependent disorders. The A49T variant might play a role in the etiology and progression of prostate cancer, while the V89L variant has been strongly associated with hypospadias risk in children and also with protection against PCOS. Based on these findings, in this study of a noteworthy NC cohort, we evaluated the modulatory effects of androgen receptor and 5α-reductase type 2 gene variants on the hyperandrogenic phenotypic variability of NC patients. # Materials and Methods This study protocol was approved by the Ethics Committee of the Hospital das Clínicas, Universidade de São Paulo, and written informed consent was obtained from all of the patients and/or their caretakers. Most patients were from São Paulo state and presented with late onset hyperandrogenic manifestations and hormonal diagnosis of NC-CAH, which was defined by basal 17OHP levels ≥ 10 ng/mL or by ACTH-stimulated 17OHP ≥ 10 ng/mL at 60 min after i.v. injection of synthetic 1–24 ACTH (0.25 mg). All of the patients had a defined molecular NC genotype, i.e., mutations identified in both *CYP21A2* alleles, and according to these criteria, 114 patients were selected. Fifty-three patients (45 females) presented at early or middle childhood with precocious pubarche (6 ± 1.9 years old), and 50 patients (all females) presented at adolescence or adulthood with hirsutism, menstrual abnormalities and/or infertility as their chief complaint (23 ± 11.3 years old). The other 11 cases (5 females) were asymptomatic and were diagnosed during familial molecular studies. Slight clitoromegaly was defined by a clitoris length \>9 mm in children and \>16 mm in adult females, evaluated by a single examiner. Clitoromegaly was observed in 10 of 114 patients (3 children and 7 adults), but separate urethral and vaginal openings were identified in all of these patients. The mean duration of patient follow-up was 11.0 ± 7.5 years. Clinical and hormonal data from patients were retrospectively obtained from medical records. These data were correlated with nCAG of the *AR* gene and with 5α-reductase type 2 allelic variants, and the following manifestations were analyzed: precocious pubarche, amenorrhea, oligomenorrhea, infertility, hirsutism and clitoromegaly. Precocious pubarche was defined by the appearance of pubic hair before 8 years old in girls and before 9 years old in boys. Amenorrhea was defined by the absence of menstrual periods for at least 3 consecutive months and oligomenorrhea by fewer than 6 menstrual periods in the previous year. Infertility was defined by the inability to conceive within 18 months of unprotected intercourse. Hirsutism was defined by male pattern of body hair distribution and a Ferriman and Galley score (FG) ≥ 8. The severity of hirsutism was classified into 2 groups: mild (≤ 14 FG) and severe (≥ 15 FG). Symptomatic patients were grouped according to the beginning of clinical manifestations into pediatric and adolescent/adult groups. None of the subjects had taken any medication for at least 3 months. ## Hormone Assays Serum 17OHP levels were measured by radioimmunoassay (Diagnostic System Laboratories INC/USA, Webster, TX, USA). Cortisol, progesterone and testosterone levels were measured by immunofluorometric assays (AutoDELFIA, Wallac, Finland). Androstenedione levels were determined by chemiluminescence assay (Immulite 2000, Siemens Health Care, UK). The intra- and inter-assay coefficients of variation varied from 5% to 10%. ## Molecular Studies *CYP21A2* (Gene ID 201910) point mutations were screened using allelic-specific PCR and/or *CYP21A2* sequencing, including of the promoter and intronic regions. Large gene rearrangements (*CYP21A2* deletions and large gene conversions) were screened by Southern blotting and/or MLPA techniques (SALSA P50B CAH MLPA Mix, MRC-Holland BV, Amsterdam, the Netherlands). ### Patient groups Patient genotypes were classified according to the predicted impairment in enzymatic activity observed in the *in vitro* studies of 3 groups: A/C (severe), B/C (moderate) and C/C (mild) nonclassical genotypes. Basically, A alleles carry mutations predicting total or almost total impairment of enzymatic activity: *CYP21A2* deletion, large gene conversion, IV2-2A\>G, p.G110Efs, exon 6 cluster (p.I236N, p.V237E, p.M239K), p.Leu307fs, p.R356W, p.Q318X, p.G424S, p.Arg483fs or the IVS2-13 A/C\>G (I2 splice) mutations. B alleles carry the p.I172N mutation, resulting in 3 ± 7% residual enzymatic activity, and C alleles carry the p.P30L, p.V281L or p.P453S mutations, resulting in 20 ± 60% residual enzymatic activity. ### X-chromosome inactivation analysis Methylation of the *Hpa*II site, close to the CAG repeats, is correlated with X-inactivation. This site is methylated in the inactive X chromosome, and it resists to cleavage by the *Hpa*II enzyme; therefore, a PCR product is obtained only from the inactive X-chromosome. For each DNA sample, two reactions were performed; in one, 2μg of DNA was digested with 20 U of *Hpa*II (CCGG) at 37°C overnight. A second reaction was similar to the previous one, except for the absence of the enzyme. The reactions were stopped by incubation at 96°C for 5 min. Both digested and undigested DNA (100 ng) were used for PCR amplifications of the CAG polymorphic tract of the *AR* gene, as previously described. PCR products were submitted to capillary electrophoresis on an ABI PRISM 310 Genetic Analyzer (Applied Biosystems, Foster City, CA USA) and were analyzed by GeneScan software to determine the sizes of the amplified fragments, which were established from their comparisons with size markers submitted to electrophoresis in the same run. These sizes were correlated with the CAG repeat numbers. The relative inactivation pattern of the AR alleles was calculated using the correction described by Pegoraro et al.. The area of the allele with a smaller repeat number (which is preferentially amplified) was divided by the area of the allele with the greatest number of repeats to determine how many times the first was preferentially amplified. After digestion with *HpaII*, this factor was applied to the area of the larger allele to normalize the level of amplification between the two alleles. We considered the amplification of one allele ≥ 75% as skewed X-chromosome inactivation. The results of X-inactivation analysis were used to generate a mean value that represented differences in the expression of constituent alleles. It was achieved by multiplying each allele in a genotypic pair by its percentage of total expression (100 minus % inactivity) and by totaling the two adjusted repeat values to achieve a new mean value, which we called the X-weighted biallelic mean. ### S*RD5A2* polymorphisms Screening for the V89L and A49T SRD5A2 allelic variants was performed using 3 UI of the restriction enzymes *RsaI* (GT//CA) and *MwoI* (GCNNNNN//NNGC) (New England Biolabs, Beverly, MA, USA), respectively, and 10 μL of the PCR product of exon 1 at 37°C overnight in a final reaction volume of 20 μL. The restriction products were analyzed by electrophoresis on 3% agarose gel. To monitor the size of the fragments generated by restriction enzymes, the molecular weight marker ΦX174/ Hae*III* was used (Invitrogen, Life Technology, Gaithersburg, MD, USA). Each reaction was performed with a known control sample, without polymorphisms. ## Statistical Analysis Qualitative variables were compared using the chi-squared test. Quantitative data, after normality testing, were compared using Wilcoxon’s test or Student’s *t*-test. For the multivariate analysis, modification of the linear regression, which considered different patterns of genetic inheritance (recessive, dominant or codominant), was used. Traditional linear regression, adjusted for covariates of interest, was performed and three results were provided based on recessive inheritance (recessive, dominant or codominant). For the linear regression, we included hormonal data, clinical data, *CYP21A2* genotype, mean CAG repeat number and *SRD5A2* alleles. For univariate statistical analysis, p\<0.05 was considered significant, and 95% confidence intervals were generated that did not include the unit. For multivariate analyzes, we planned Bonferroni’s correction for multiple comparisons; however, because no associations attained statistical significance, this correction was not necessary. All of the statistical analyses were performed using Small Stata software, version 11.1 (2010 StataCorp., College Station, TX, USA). # Results The patients were divided into three groups according to their ages at the beginning of manifestations: pediatric (n = 53), adolescent/adult ≥ 12 years old (n = 50), and asymptomatic (n = 11). The CAG repeat numbers presented a normal distribution, and 88% of women were heterozygous for this number. The mean CAG repeat numbers in pediatric, adolescent/adult and asymptomatic groups were 22.2 ± 3.1 (12–30), 22.6 ± 3.5 (16–31) and 22.4 ± 3.6 (14–28), respectively (p\>0.05). The allelic frequency of CAG repeats was also evaluated among these three groups. The frequencies of shorter CAG alleles (≤ 18 repeats) in the pediatric, adolescent/adult and asymptomatic groups were 15%, 14% and 9%, respectively (p\>0.05). The frequencies of longer CAG alleles (≥ 26 repeats) were 6.6% in the pediatric group, 16% in the adolescent/adult group and 13% in the asymptomatic group (p\> 0.05). Nonclassical genotypes from the A/C (severe) and C/C (mild) groups were identified in 52% and 48% of patients, respectively. Only one patient carried the B/C genotype group, and this patient was excluded from the analysis. Initially, frequencies of CAG alleles were compared among the patient groups carrying the same *CYP21A2* genotype group. Among the patients carrying the A/C *CYP21A2* genotype, frequencies of shorter and longer CAG alleles did not differ between the pediatric and adolescent/adult groups (p\>0.05). Among patients carrying the C/C *CYP21A2* genotype, the frequency of longer CAG alleles was significantly greater in adolescent/adult than in the pediatric group (p = 0.01), whereas the frequencies of shorter alleles did not differ between these groups (p\>0.05). The CAG distribution and period of manifestations according to most frequent *CYP21A2* genotypes were described in. Subsequently, the frequencies of shorter and longer CAG alleles were compared between *CYP21A2* genotypes in patients from the same group according to the beginning of manifestations. The frequencies of longer CAG alleles between pediatric patients carrying the A/C and C/C *CYP21A2* genotypes were 10% and 2%, respectively (p = 0.09); no difference was observed in the frequency of shorter CAG alleles. Regarding adolescent/adult patients carrying the A/C and C/C genotypes, no difference was observed in the frequencies of shorter and longer alleles. Skewed X-chromosome inactivation was observed in 13% of NC females, and its frequency did not differ in female subjects between the pediatric and adolescent/adult groups (23% *vs* 10%, respectively) (p\>0.05). The weighted biallelic CAG mean was also evaluated according to the number of hyperandrogenic symptoms at diagnosis. In asymptomatic patients, in patients with one symptom and in those with two or more symptoms, the mean CAGs were 21.8 ± 2, 21.2 ± 2.9 and 21.6 ± 2.8, respectively (p\>0.05). In the hirsute patients, the weighted biallelic CAG mean was 20.9 ± 2.5, and in those without hirsutism it was 21.7 ± 2.8 (p\> 0.05). Ferriman scores were compared across 33 women who were first evaluated before the use of any topical and/or antiandrogen therapies. In patients with severe hirsutism (Ferriman score ≥ 15), the weighted biallelic CAG mean was 21.5 ± 3.0 and, in those with mild hirsutism (Ferriman score ≤ 14), it was 19.9 ± 2.3 (p\>0.05). The influence of CAG repeats on the presence of virilization was also evaluated and the weighted biallelic CAG mean was compared between patients´ groups with and without clitoromegaly. The weighted biallelic CAG mean was significantly lower in patients with clitoromegaly than in those without it: 19.1 ± 2.7 *vs* 21.6 ± 2.5, respectively (p = 0.007). Eight of ten (80%) patients with clitoromegaly and 14 of 93 (15%) without clitoromegaly carried shorter CAG alleles (≤ 18 repeats) (p\<0.001). There were no differences in serum testosterone levels between adult women with and without clitoromegaly: 138 ± 73 ng/dL and 94 ± 66 ng/dL, respectively (p\>0.05). The V89L-*SRD5A2* variant was identified in 31% of alleles in Hardy-Weinberg equilibrium; 45% of the patients were heterozygous, and 9% were homozygous carriers. Unlike the V89L, the A49T variant was rare in this series (1% of alleles) and was excluded from association analysis. No difference in the frequency of the V89L variant was observed regarding the onset of symptoms, severity of hirsutism and presence of clitoromegaly in patients carrying both *CYP21A2* genotypes. # Discussion The NC form presents great phenotypic variability, even among subjects carrying similar genotypes. Additionally, in the literature, it has not been well defined whether the severe allele, in a compound heterozygous patient, modulates the severity of hyperandrogenic manifestations. Previous studies have reported conflicting results regarding the association between the severity of the NC genotype and the onset of symptoms. In this context, we emphasized two studies with large NC cohorts that found no influence of *CYP21A2* genotypes on the onset and/or severity of hyperandrogenic manifestations. The phenotypic variability of hyperandrogenism in the NC form could result from interindividual differences in peripheral androgen sensitivity, which might arise from genetic variants related to androgen action and/or metabolism. The influence of these variants has been widely evaluated in hyperandrogenic disorders. Some studies have identified that shorter CAG alleles of the AR gene present with greater frequency in women with PCOS in relation to the normal population. However, other studies did not find this association, and it is likely that these discordant findings result from methodological differences, such as those for defining the criteria for X-inactivation skewing and for defining shorter CAG alleles. An influence of the CAG tract was observed on the phenotypic severity of prostate cancer, and a significant correlation between shorter alleles and an earlier age at diagnosis was identified. In this present study, besides the great miscegenation of Brazilian population, we found a normal distribution of CAG repeat numbers in the NC cohort, and no differences were observed compared to our normal population or with Caucasians. It seems more important to evaluate the influence of the CAG tract at the beginning of manifestations in the NC form; we hypothesized that pediatric group would carry a higher frequency of shorter CAG alleles, since they present at younger age. However, the weighted biallelic CAG mean, which reflects both the number of repeats and the percentage of activity of each allele, did not differ between NC patients from the pediatric and adolescent/adult groups. This weighted biallelic mean did not differ, even when we considered the severity of the 21-hydroxylase genotype. We cannot exclude that these negative results could be related to a sample size effect. It is worth emphasizing the higher frequency of longer CAG tracts in the adult group bearing the mild NC genotype compared with the pediatric group bearing the same genotype; consequently, we speculated that this higher frequency of alleles with lower AR activity contributed to the later onset of symptoms in patients carrying similar genotypes. In the same manner, the influence of CAG repeat numbers was previously analyzed at the beginning and with regard to the severity of hyperandrogenic manifestations in the NC form. Ben-Shachar et al., analyzing a group of 119 NC female patients, identified that patients carrying shorter CAG alleles (\< 25 repeats) had a higher frequency of precocious pubarche and precocious puberty. Ibáñez et al., in a cohort of 181 women, found that shorter CAG alleles (≤ 20 repeats) were associated with increased risk for premature pubarche relative to the normal population and also with subsequent development of ovarian hyperandrogenism. Interestingly, Hickey et al. found that shorter CAG alleles (≤ 22 repeats) were associated with infertility in a subset of 122 PCOS patients. In contrast, our study revealed no influence of nCAG on the prevalence of premature pubarche, hirsutism, menstrual abnormality or infertility or in the number of manifestations at diagnosis, such as isolated hirsutism and/or associated with menstrual irregularity. Similarly to our results, it was observed in a large PCOS cohort that CAG repeats and serum testosterone levels were not predict factors of hirsutism and acne. However, we would like to emphasize that our criteria for classifying shorter alleles (≤ 18 repeats) might have been stricter than those used in the studies cited above. To assess the severity of hyperandrogenism, we used the Ferriman score for hirsutism and the presence of an enlarged clitoris, which indicates a stronger phenotype of hyperandrogenism. Association analyses disclosed no influence of the weighted biallelic CAG mean on the severity of hirsutism, similar to the findings of Dasgupta et al., who evaluated PCOS patients. However, the number of hirsute NC patients, who started follow-up in our service without ever having received treatment before, was small. Furthermore, we also speculated that small variations in the CAG repeats within the normal range might not be significant in the modulation of hirsutism. Although this is a large cohort considering nonclassical patients, probably to definitively exclude the influence of CAG repeats in the less severe hyperandrogenic manifestations, besides clitoromegaly, it should be necessary higher number of patients. Interestingly, in our patients with clitoromegaly, basal androgen concentrations were not different from those in subjects without clitoromegaly, but we noted a positive association between shorter CAG alleles and the presence of clitoromegaly, as well as with a lower weighted biallelic CAG mean. This finding reinforced those observed in a previous study conducted in our laboratory in patients with the classical form of CAH, in whom we found an influence of the CAG tract in the phenotypic variability of external genitalia virilization. In men, the influence of the CAG tract on the phenotype of the external genitalia was also reported. Ogata et al. described a patient with 46,XY and with genital ambiguity, carrying a rare allele with 44 CAG repeats, without any other allelic variant in exonic/intronic regions of the *AR* gene and no changes in testosterone synthesis. Interestingly, no differences in CAG allelic distributions were observed between our asymptomatic and symptomatic NC female subjects. Asymptomatic NC females despite the 21-hydroxylase deficiency, presented with normal basal androgen levels through consecutive evaluations. It is likely that this cryptic phenotype is related to genetic variations in androgen synthesis. However, in this study few asymptomatic cases were evaluated, since they were diagnosed during familial study of an index case. We also call attention to the process of gene regulation of the androgen receptor, which is extremely complex and differs in various tissues according to the activity of co-regulatory proteins. These proteins might also play a modulatory role in peripheral sensitivity to androgens and corroborate the phenotypic variability of the NC form. Another protein of importance in the peripheral action of androgens is the 5α-reductase enzyme. The peripheral action of testosterone is boosted by the activity of this enzyme converting testosterone into dihydrotestosterone, which is a more potent androgen than testosterone, largely responsible for the growth and stimulation of hair follicles. The increased peripheral activity of 5α-reductase was considered a predisposing factor for the development of idiopathic hirsutism and precocious idiopathic pubarche. Allelic variants in the *SRD5A2* gene have also been associated with the development of androgen-dependent disorders, such as prostate cancer and PCOS. In a series enrolling 187 PCOS women, the status carrier for the 89L allele, which is known to reduce the activity of 5α-reductase type 2, was associated with a lower likelihood of developing PCOS. However, this variant was not correlated with a lower intensity of hirsutism. The 89L variant was identified among approximately 31% of the alleles in our patients, similar to the normal Brazilian population. In contrast, our association studies have not identified an influence of the 89L allele on the hyperandrogenic phenotype of the NC form. This allele was not protective against the development of premature pubarche, and its frequency did not differ significantly between hirsute and non-hirsute patients and or between those with and without menstrual irregularity. Our NC patients carrying the 89L variant had lower median scores for hirsutism and later onset of manifestations relative to wild-type carriers, but these differences were not significant. Although this variant decreased the activity of 5α-reductase to 30% *in vitro*, it is possible that was not significant *in vivo* for the NC phenotype. It is likely that the 5α-reductase type 1 enzyme is more important to the severity of hirsutism, as previously reported in women in PCOS. In conclusion, we described that the CAG tract of the androgen receptor gene could explain, at least partially, the phenotypic variability in the NC form, but these data must be replicated in other populations. Although the NC form of CAH is a monogenic disorder, our preliminary data suggested that the interindividual variability in the hyperandrogenic phenotype could arise from polygenic interactions. # Supporting Information [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: TASSB. Performed the experiments: VOMM. Analyzed the data: VOMM FSC. Contributed reagents/materials/analysis tools: DBDG. Wrote the paper: VOMM FSC. Collected data and contributed to the statistical analysis of the data: LGG GM. Reviewed the literature and collected data: JAMM. Coordinates the group that carried out the work and revised the manuscript: BBM.

# Introduction Obstructive sleep apnea (OSA) is a prevalent breathing disorder and is considered a major public health issue, affecting 5–15% of the general population and increasing with both body mass index and age (up to at least 60–65 years). OSA is characterized by repetitive narrowing and closure of the upper airway during sleep that results in brain arousal, intermittent hypoxia, negative intrathoracic pressure swings, and increased sympathetic activity. OSA is associated with a reduction in quality of life, daytime sleepiness, traffic accidents, neurocognitive impairment, metabolic, cardiovascular disease and malignancies. The treatment of choice for OSA is the application of continuous positive airway pressure (CPAP) to the patient’s nose or mouth through a mask during sleep at home. This pressure in the mask is transmitted to the pharyngeal area, splinting the collapsible upper airway walls thereby avoiding obstruction. Auto-adjusting positive airway pressure (APAP) devices, which are increasingly being used, are driven by algorithms that measure abnormal sleep breathing events, analyze the patient’s breathing pattern and eventually increase the delivered pressure in response to airway obstruction, or decrease pressure when breathing is stable to increase patient comfort. In theory, APAP devices should be ideal for treating a range of patients with different characteristics, and for effectively treating OSA despite within-night and night-to-night variations in the upper airway collapsibility experienced by each individual patient. However, commercially available APAP devices contain undisclosed proprietary algorithms, and therefore the way that they measure and respond to specific breathing patterns varies. In addition, some APAP manufacturers are introducing new algorithms based on specific patient characteristics. This move towards personalized medicine in the treatment of OSA means greater choice for patients and more variability in APAP algorithms. Therefore, understanding how each device responds to different OSA patterns requires comparative studies using well defined references. Bench testing is a useful method to characterize the response of different APAP algorithms under well-controlled conditions, thus avoiding the biological variability inherent in clinical trials. However, previously used bench test models have been based on subjecting the APAP device under test to a repetitive string of disturbed breathing patterns, without providing a sufficiently wide spectrum of events. These limitations mean that variety in patient characteristics and phenotypes, or the changes that occur during different sleep stages and body positions over the course of a night’s sleep, cannot be taken into consideration. This is particularly relevant given that different subpopulations of OSA patients (e.g. children, men, women, the elderly) exhibit specific traits in their sleep-related breathing disorders. Therefore, the aims of this proof-of-concept study were: 1) to design a new complex and versatile bench test approach for realistically simulating respiratory events throughout the course of the night in an OSA patient, mimicking breathing disturbances across different phenotypes, and 2) to implement a full night example of a female OSA phenotype and use this to compare the responses of several currently-available APAP devices. # Materials and Methods The hardware of our new model was based on a previously described bench test. This fully computer-driven model comprises a servo-controlled pump able to deliver a flow that replicates any breathing waveform stored in the computer. An obstruction valve allows the simulation of controlled obstructive events by imposing mechanical impedances previously recorded in patients with OSA. Two other valves can mimic leaks and mouth breathing, and a loudspeaker-in-box system can superimpose simulated snoring onto the breathing flow. The test bench is equipped with two sensors, one to measure pressure at the simulated patient entrance and one to measure the actual flow generated by the patient simulator. A calibrated leak based on a 4-mm internal diameter (ID) orifice mimics the mask leak (exhalation port) in nasal masks. In previous studies, this system was fed by a collection of disturbed breathing events, such as obstructive and central apneas, hypopneas, flow limitation, mask leaks and mouth expiration. To design the new OSA simulator model we developed a novel approach aimed at realistically replicating a typical night of sleep for a female patient. With this aim, we considerably expanded our library of disturbed breathing patterns anonymously extracted from polysomnography recordings obtained from real OSA patients and we incorporated several new adjustable features into the simulator. Specifically, the new patient model can be set to react to the pressure delivered by the APAP device (PAP-responsive mode) or to reproduce a fixed scenario of disturbed breathing events (Steady mode), depending on the device characteristics being tested. Moreover, the severity of the simulated OSA profile is now fully modifiable by changing the frequency and duration of each breathing event. Various artefacts were introduced into the event spectrum, such as changes in tidal volume and breath rate, to replicate typical events during wake such as irregular breathing, swallowing, moving and talking. By combining these new features, we aimed to create a new OSA model concept model that can realistically replicate a whole night of sleep, including phases of wake, rapid eye movement (REM) and non-REM sleep, and change in body position, each one designed to mimic different characteristics in terms of upper airway collapsibility. For this study specifically, as an example of an entire night of sleep- disordered breathing (SDB), the bench test model was set to simulate the disturbed patterns of a female OSA patient with the following characteristics: long sleep latency (45 min), low positive airway pressures (PAPs) required to overcome obstructive events, high proportion of flow limitation events versus apneas, higher apnea-hypopnea index (AHI) during REM sleep, and only minor positional effects on upper airway collapsibility. The features and structure of this female-specific OSA patient simulation are detailed in. The breathing pattern of the simulated patient depended on the PAP applied by the device under test, with a total duration of 4 hours and 15 minutes. APAP pressure values required to normalize breathing during each stage of the simulation are shown in. The simulated night consisted of programming the different stages described in, starting with 45 minutes of simulated awake stage (sleep onset) followed by a succession of different sleep stages with the features detailed in (e.g. breathing frequency, number and types of respiratory events) and a final awake short period. In this way we were able to model a patient exhibiting different sleep breathing characteristics throughout consecutive sleep stages. Ten different commercially available APAP devices were tested using the new bench test model and the female-specific simulation described above: AirSense 10 (A) and AirSense 10 AutoSet for Her (B) by ResMed; Dreamstar by Sefam (C); Icon by Fisher & Paykel (D); Resmart by BMC (E); Somnobalance (F) and Prisma 20A (G) by Weinmann; System One by Respironics (H); iCH (I) and XT-Auto by Apex (J). Each APAP device was connected with its own tube to the bench model. Default APAP settings were used (minimum pressure 4 cmH₂O, maximum pressure 20 cmH₂O). Each device was tested twice and the results averaged to obtain the final values. # Results The new OSA patient simulator could effectively combine a great variety of SDB elements to mimic the response of the predefined patient type. The responses of the assessed APAP devices to the new female-specific bench test model are summarized in. There was considerable variation among devices, particularly with respect to the mean and maximum nasal pressures applied, and the ability to overcome obstructive events and flow limitation, The residual AHI was calculated as the number of residual obstructive events per hour and the residual flow limitation was measured as the portion of the test in minutes (excluding the initial 45-minute wake period) that the simulated patient remained on flow limitation. Breathing normalization with a residual AHI \<5/h was only achieved with devices A, B and D; devices E, H, I and J were associated with more than five residual events per hour. Pressure changes of each device throughout the whole test are displayed in. Considering the 45-minute wake period, there was significant variation in the behaviour of the different devices. shows the pressure values reached by each tested device at the end of the simulated wake period. Device C did not increase the pressure during wake periods. Three devices (A, B and E) displayed only mild pressure increases (\<2 cmH₂O). Moderate pressure increases (2.5–3 cmH₂O) were displayed by three devices (H, I and J), and significant pressure increases (\>7 cmH₂O) were seen from three devices (D, F and G). Three examples of different responses during the simulated wake period are presented in, together with the flow signal generated by the simulator during the initial awake phase, which consisted of normal breathing with some events inserted simulating flow alterations due to irregular breathing (E) and swallowing (S). Devices A, B and D contain algorithms aimed at automatically detecting sleep onset (for A, B AutoRamp mode and for D SenseAwake mode). Devices A and B showed similar pressure increases with AutoRamp mode turned off, while device D responded with higher pressure increases when the SenseAwake mode turned off. To assess whether the observed variations in pressure during wake had an influence on the results of testing, a subset of devices that showed a moderate to significant pressure increase during sleep onset (D, G, H and I) were retested without the wake phase of the test. In this additional analysis, the responses of the tested devices were relatively similar to the ones in the previous tests that included the 45-minute sleep onset phase. The largest change was seen in device D, where the residual AHI increased from 0.6 to 6 events per hour. # Discussion We successfully developed and carried out a proof-of-concept test of a novel optimized bench model easily adaptable to simulate different SDB patterns found in OSA, including periods of wake, periods representing different sleep stages and phases of more or less severe SDB events. This tool can be useful to objectively evaluate bench test performance of different APAP devices with realistic breathing patterns covering a wide range of patient phenotypes. In its “Steady mode”, the simulator could also assess the capacity of APAP, as well as CPAP, devices to estimate treatment duration and detect residual respiratory events of a fixed predefined disturbed breathing scenario. The presentation and severity of OSA varies greatly depending on patient characteristics such as gender, age, body mass index, and craniofacial structure. Specific patient subgroups have been gaining a lot of attention recently because of their clinical relevance. At one end of the age spectrum, elderly patients tend to present with severe OSA and snoring becomes less common. In addition, the frequency of central events increases, although obstructive events still predominate. In contrast, children with OSA have frequent snoring, restless sleep, mouth breathing, apneas, gasping, and laboured or paradoxical breathing. With the growing trend towards personalized therapy, specific patient breathing patterns will be increasingly studied as manufacturers work to design the most optimal treatment for each phenotype. One good example of this is OSA in females versus males. It is well-known that the polysomnographic features of female OSA are different from those of male OSA. Overall, women have less severe OSA with, on average, a lower AHI and shorter apneas. Women also have more episodes of upper airway events during REM sleep. Body position is far less important for the severity of OSA in women, while OSA severity in men is based more on position than sleep state. Furthermore, women may take longer to fall asleep, but have fewer awakenings during sleep. Regardless of the patient’s gender, there is also significant night-to-night variation in OSA, based on factors such as body posture, sleep stages, and previous drug or alcohol intake. Besides OSA pathophysiology, gender influences also patients’ PAP requirements, as generally female patients require lower pressures. Such considerable variability between phenotypes highlights the relevance of the simulation approach taken in this study. In our optimized bench test we implemented a dynamic pattern (“PAP-responsive”) simulating a female patient phenotype (although an individual male patient may also present with this OSA pattern), which included long periods of flow limitation, low AHI, and short, low-severity obstructive events. Only three of the APAP devices tested were able to achieve full breathing normalization by overcoming all types of disturbed events including flow limitation. Considering the potential for increased flow limitation in female patients, which may lead to breathing disturbances, the effectiveness of treatment in patients presenting with a high component of flow limitation should be carefully examined. Published data comparing different APAP algorithms is scarce, particularly for devices recently launched into the market. Pevernagie *et al* examined two APAP devices and found that the residual apnea-hypopnea index (AHI) was lower during use of one device compared with the other (3.5±5.6/h vs 9.9±31.0/h), and that the amount of snoring during the night was significantly higher with one device. A similar study by Nolan *et al* compared three commercially available devices. The authors found that mean pressure and patient compliance were significantly lower on one of the APAP devices. Differences between algorithms combined with a lack of information regarding how different auto-adjusting devices work has led to the perception that auto-adjusting devices are a ‘black box’ which should be used with caution. In this study, we also found considerable variation among devices in both the magnitude of response to obstructive events, the time taken to increase pressure during disrupted breathing, and device behaviour during the simulated wake period. With the exception of one device, which did not increase the pressure at all, most devices at least slightly increased pressure during simulated wakefulness. Some devices showed quite an intense pressure response during the wake period of the test, with one reaching almost 14 cmH₂O and two reaching 12 cmH₂O. Due to the potential impact this could have on patient comfort, pressure changes during wake periods should be assessed in clinical practice, particularly in patients who report difficulties falling asleep while using PAP therapy or issues with comfort at higher PAP pressures. As stated above, our finding of considerable variability in the response of APAP devices when subjected to the same breathing pattern under well-controlled conditions is in agreement with previous reports. These variations can be attributed to the individual algorithms within each APAP device. Each algorithm analyses flow and pressure to determine whether there is a breathing disturbance, and then initiates the most appropriate response to correct such a disturbance. For instance, it is interesting to note that, as we explained previously, the simulated hypopneas in our model were defined according to specific values of a flow-limitation pattern index initially introduced by Teschler et al. Therefore, it could be possible that automatic CPAP devices set to detect hypopneas using this index, or something similar, could be more suitable for detecting our simulated events than other devices that use other metrics to define and detect hypopneas. Another reason for the observed different response in the automatic CPAP devices tested is that the optimal rate of pressure increase after detection of obstructive events has not been clinically defined. In fact, APAP devices are designed to normalize breathing at a rate which treats actual SDB, avoiding any response to false events, thereby unnecessarily modifying pressure. The results of this bench test have shown that, under well-controlled conditions, there are marked variations in response by different APAP devices, and that there may be high residual AHI or uncontrolled flow limitation in some female patients on some APAP devices. Therefore, all APAP devices should not be considered equal, and efficacy and patient comfort should be carefully examined following APAP initiation. It must be noted that our results are restricted to the specific patterns of disturbed breathing used in this bench test to simulate a specific OSA patient. It is possible that the response of the tested devices would have been different from the ones reported here if SBD was simulated using different patterns or patient phenotypes. In addition, a limitation of this study is that one device of each type was used. Hence, a more complete assessment would require testing of a larger number of each type of device randomly obtained from those available in the market. Finally, it should be stressed that although bench testing is a useful way to investigate the behaviour of different devices, testing outcomes may vary in clinical practice due to the almost unlimited spectrum of events and phenotypes found in real life. Indeed, crucial factors such as changes in loop gain, and upper airway compliance and pharyngeal critical pressure are not considered in our model. Accordingly, bench testing should be considered as a preliminary assessment before clinical evaluation in patients. In conclusion, this study showed that a dynamic bench model tailored to represent specific OSA patient phenotypes, incorporating a variety of disturbed breathing events within the same simulated night, including different degrees of severity along sleep stages, and a period of wakefulness, can be useful to characterize treatment responses of commercially-available APAP devices. This demonstrates that bench testing can be modified to better represent a “real” patient, and that APAP devices can show markedly different responses to the same simulated breathing patterns. Realistically mimicking OSA patients during bench testing is useful as a first step to aid in the understanding of actual APAP device responses observed in the clinical setting, and can be helpful in selecting the device that best meets the individual needs of each patient, thereby improving comfort and increasing adherence to therapy, which is essential for effective treatment and reducing the consequences of OSA. Authors wish to thank Mr. Miguel Angel Rodriguez for his technical support. English language medical writing assistance was provided by Nicola Ryan, independent medical writer, funded by ResMed. [^1]: The work presented here has been partially supported by a research agreement (number 307990) between ResMed Science Centre and Universitat de Barcelona (PI: Ramon Farré). AJ Wimms, D Ramanan, and H Woehrle are employed by ResMed Science Centre. There are no patents, products in development, or marketed products to declare. This does not alter the authors' adherence to all the PLOS ONE policies on sharing data and materials, as detailed online in the guide for authors. [^2]: Conceived and designed the experiments: VI JMM RS AJW DR HW DN RF. Performed the experiments: VI RS RF. Analyzed the data: VI RS RF. Contributed reagents/materials/analysis tools: JMM AJW HW DN RF. Wrote the paper: VI JMM AJW HW DN RF.

# Introduction Symptomatic HIV-associated sensory neuropathy (HIV-SN) is a frequent complication of HIV infection and its treatment. Recent data from sub-Saharan Africa, the region worst affected by HIV, indicate that symptomatic neuropathy affects between 35 and 60% of ambulatory HIV-positive patients. Moreover, the burden of HIV-SN is expected to remain high for the foreseeable future despite changes in the management of HIV. The most salient symptom of the neuropathy is pain, which frequently is moderate to severe in intensity, and has been associated with reduced activities of daily living and physical function, sleep disruption, increased severity of depression and anxiety symptoms, and increased risk of being unemployed. Yet, evidence for managing painful HIV-SN is poor. Amitriptyline, is a tricyclic antidepressant that is effective in treating neuropathic pain, and is widely available in developing countries. However, two published trials found amitriptyline was no more effective than placebo for the treatment of painful HIV-SN. Both trials were stopped on the advice of their respective data monitoring boards because of lack of efficacy before recruitment targets were met. The trial by Shlay and co-workers had poor blinding, and used a 2x2 factorial design with acupuncture and amitriptyline as interventions, which complicated the interpretation of the data. But the second study by Kieburtz and colleagues was a high-quality parallel group study, with three interventions: placebo, amitriptyline and mexiletine. Despite these negative results, amitriptyline is recommended by international and national agencies for the treatment of painful HIV-SN, and is the only recommended first-line treatment for painful peripheral polyneuropathy included on the WHO essential medicines lists. Because of this persistent recommendation of amitriptyline for the management of painful HIV-SN, we conducted a randomized-controlled trial to ascertain whether amitriptyline is a clinically effective analgesic for moderate to severe foot pain in individuals with HIV-SN. This trial differs from the previous two studies in that amitriptyline and placebo were the only interventions studied, simplifying the interpretation of results. Importantly, it included the separate investigation of the efficacy of the treatment in two patients groups; one group on stable antiretroviral therapy, and the other never exposed to antiretroviral therapy (both previous trials included mixed cohorts). This is critical, as there are no pathognomonic features to distinguish the polyneuropathy that develops in HIV-positive individuals before or after starting antiretroviral therapy despite likely differences in the mechanisms underlying the neuropathies. The neuropathy that develops in patients never exposed to antiretroviral therapy is likely to be predominantly immune-mediated, while the neuropathy that develops soon after initiating antiretroviral therapy includes an additional insult by neurotoxic antiretroviral drugs such as stavudine (a drug still in common use in developing countries). It is unclear whether these mechanistic differences may alter responsiveness to therapy, but differences in the analgesic response to lamotrigine have been reported between patients with painful HIV-SN exposed to non-neurotoxic and neurotoxic antiretroviral therapy. It is important to understand whether neuropathic pain caused by *either* mechanism may be amitriptyline responsive, with both likely contributing in many cases of painful, HIV-SN. Previous studies were underpowered for this purpose. Thus, we have conducted a double-blind crossover trial of amitriptyline for analgesia in painful HIV-SN: i) in individuals never exposed to antiretroviral therapy, and ii) in individuals on stable antiretroviral therapy. We report that pain relief achieved by amitriptyline therapy did not differ from that achieved by placebo, irrespective of antiretroviral therapy exposure. # Materials and Methods The protocol for this trial and supporting CONSORT checklist are available as supporting information; see and. ## Trial Design We conducted a randomized, double-blind, placebo-controlled, crossover study comparing the analgesic efficacy of amitriptyline tablets to inactive placebo tablets, with a three-week washout period between interventions. ## Ethics statement Permission to conduct the study was obtained from the Human Ethics Research Committee, University of the Witwatersrand (clearance number: M080709). Written informed consent was obtained from all participants. The trial was registered at the South African National Clinical Trials Register (NHREC#1188, issue date 4^th July 2008; registry is no longer operational), and International Standard Randomized Controlled Trial Number Register (ISRCTN54452526, issue date 17 November 2014). ## Participants The study took place at a single site, Chris Hani Baragwanath Hospital, Soweto, South Africa, between 23 April 2009 and 17 November 2009. Eligible participants were aged 18 years or older, had a confirmed HIV infection, and met the criteria for current symptomatic HIV-SN according to the validated Brief Peripheral Neuropathy Screening Tool. The tool requires the bilateral presence of at least 1 symptom (pain, aching, burning, numbness, or pins-and-needles) and at least 1 clinical sign of neuropathy (reduced vibration sense or absent ankle reflexes) for a diagnosis of symptomatic HIV-SN to be made. Vibration sense was assessed using a 128 Hz tuning fork, which was placed on the interphalangeal joint of each great toe; perception of vibration sense for 10 seconds or less was considered abnormal. Trial participants were required to have moderate to severe pain (score ≥4 on an 11-point numerical pain rating scale) in both feet over the previous three days.. Participants had to be either on stable antiretroviral therapy for more than 6 months (ARV-user group), or antiretroviral therapy naïve (ARV-naïve group). Exclusion criteria were: severe pain from HIV-SN that warranted a change in treatment regimen; already taking amitriptyline or having taken the drug in the previous three weeks; limb amputation; Kaposi sarcoma of the lower limbs; current post-herpetic neuralgia or herpes zoster; pregnancy or the intention of falling pregnant; receiving treatment for tuberculosis; malignancy not related to HIV; major psychiatric disorders; epilepsy; use of monoamine oxidase inhibitors, other antidepressants or anti-epileptic drugs; renal failure requiring intervention; diabetic neuropathy; clinically significant liver failure or a history of liver failure; extreme pain or exhaustion; recent myocardial infarction, arrhythmias or heart block; a history of urinary retention, urinary hesitancy or closed angle glaucoma; and participation in another trial or study. ## Intervention Participants attended six scheduled visits every three weeks over 15 weeks. This comprised six weeks on intervention A (amitriptyline) or B (placebo), three weeks washout and then crossover to six weeks intervention on B or A respectively. Participants took the assigned medication daily during each six- week intervention. At the baseline visit, participants were provided with a three-week supply of medication. Over a maximum of two weeks, dose escalation occurred to tolerance (participants were not able to tolerate perceived side- effects) or effect (participants achieved significant pain relief) every three days, based on telephonic conversation between the participant and a study nurse. The dose escalation schedule was as follows: 25mg, 50mg, 75mg, 100mg, 150mg (maximum dose) formulated as single 25mg tablets for amitriptyline or one, two, three, four or six tablets of placebo. The maximum tolerated dose achieved was taken by each participant for the remainder of the six-week period. At the end of week three, the participants visited the study center, returned unused drugs and were issued with a second container of tablets containing the correct maximum daily dose for each participant for the remaining three weeks of the initial treatment phase. At the clinic visit at the end of week six, participants returned all unused drugs, and were questioned on the number of missed doses. They were also advised on the three-week washout period. The duration of the washout period was calculated based on the pharmacokinetics of amitriptyline. The use of pre-specified rescue medication (acetaminophen, non- steroidal anti-inflammatory drugs, codeine phosphate) was permitted. After the washout period, the above procedures were repeated, but the interventions were switched. ## Outcomes The primary outcome measure was the difference in pain intensity of the feet at baseline and at six weeks of intervention, as measured by participant self- report using an 11-point numerical pain rating scale. Specifically, pain intensity referred to the pain in the feet at the time of each of the six study visits (the convention of asking “average pain” has been shown to perform poorly in this population). Dose escalation and maximum dosage of amitriptyline used was noted, as were side effects and adverse events. The use of rescue medication, including prescribed, over-the-counter and self-administered treatments, as well as traditional medicine and treatments, was recorded. No participants were taking gabapentin, pregabalin, carbamazepine or lamotrigine at enrolment or used these agents during the course of the study as rescue medications. ## Sample Size To detect a clinically meaningful reduction in the intensity of peripheral neuropathy of 2 units on an 11-point NRS, assuming 90% power, 5% level of significance and assuming common variance in the two crossover periods of 2, we estimated that 56 individuals would be required. We assumed a loss to follow-up of 10%, thus increased the sample size to 62 participants in each of the ARV- naive and ARV-user groups. ## Randomization and Masking The random allocation sequence was determined using a randomized block model. For allocation of the treatment sequence, a computer-generated list of alphanumeric identity numbers was used for all 124-treatment packs. EM generated the random allocation sequence, and assignment of interventions was by SM. The active and placebo tablets were identical in appearance. The study doctor, study nurse (responsible for enrolment) and participants were blinded to the intervention. The drug used was Trepiline (Aspen Pharmacare, Port Elizabeth, South Africa). An identical-looking placebo was manufactured by Azochem Laboratories (Roodepoort, South Africa). Placebo tablets contained microcrystalline cellulose and sorbitol. ## Statistical Methods Primary analysis was performed on a per protocol basis. Analyses were conducted separately for each of the two participant groups (ARV-naive and ARV-user), and for all participants combined. Pain intensity scores are reported as mean (SD). Crude statistical analyses of pain scores at the start of each intervention (baseline scores) were carried out using paired and unpaired t-tests, with Bonferroni correction for multiple comparisons. Analysis of the primary outcome included crude comparisons of change in pain score over six weeks of treatment with amitriptyline or placebo using paired t-tests (ignoring period or treatment order effects), and using ANOVA to account for possible period effects and treatment order effects in the model. The ANOVA analyses were repeated using the intention-to-treat population, using the baseline observation carried forward method to interpolate missing values. # Results ## Participant Flow Participant flow is summarized in. One hundred and twenty-four (124) participants were randomized to treatment; 122 completed all 6 study visits and were included in the per protocol analyses. Two participants lost their tablets and missed three and two days of treatment, respectively, their data were included in the analyses. There were no missing data on the primary end-point for 122 participants included in the analyses. ## Demographic Data Participant baseline demographic data are shown in. All participants were of black African descent, 71% were females, and the mean age was 38 years. Amongst the ARV-user group, 62% had been exposed to stavudine. Participants in the ARV- user group tended to be older, had fewer years of formal schooling, and had lower CD4 T-cell counts than participants in the ARV-naïve group. ## Dosage Titration In the ARV-user group there was no difference in titration dose between amitriptyline and placebo; participants were titrated to a median of 2 (IQR: 1–2) tablets of per day (drug dose median: 50mg, IQR: 25–50mg) when administered amitriptyline, and a median of 2 (IQR: 1–2) tablets per day when administered placebo (Wilcoxon signed-rank V = 386, p = 0.59). In the ARV-naïve group, there was no difference in median titration dose between amitriptyline and placebo; participants were titrated to a median of 2 (IQR: 1–2) tablets of per day (drug dose median: 50mg, IQR: 25–50mg) when administered amitriptyline, and a median of 2 (IQR: 1–3) tablets per day when administered placebo (Wilcoxon signed-rank V = 415, p = 0.84). There was no difference between the ARV-user and ARV-naïve groups in the median titration dose for amitriptyline (Wilcoxon rank sum W = 1643, p = 0.24) and placebo (Wilcoxon rank sum W = 1757, p = 0.58). ## Primary Outcome shows pain scores across the two periods of the trial for the ARV-user and ARV- naïve groups, and all participants combined. There were no significant differences in mean pain scores for ARV-users receiving amitriptyline or placebo at the start of week 1 (baseline for period 1) (amitriptyline: 7.9, SD 1.7; placebo: 8.3, SD 1.8; t₍₅₉₎ = -0.85, p = 0.40), or at the start of week 9 (baseline for period 2) (amitriptyline: 5.5, SD 3.3; placebo: 6.0, SD 3.2; t₍₅₉₎ = -0.60, p = 0.55). However when ignoring intervention, there was a significant period effect, such that pain scores at the start of week 9 were significantly less than at the start of week 1 (period 1: 8.1, SD 1.8; period 2: 5.7, SD 3.2; t₍₆₀₎ = 5.26, p \< 0.001). There were no significant differences in mean pain scores for ARV-naïve participants receiving amitriptyline or placebo at the start of week 1 (amitriptyline: 7.8, SD 1.7; placebo: 7.6, SD 1.3; t₍₅₉₎ = 0.34, p = 0.73), or at the start of week 9 (amitriptyline: 5.1, SD 3.3; placebo: 5.4, SD 3.1; t₍₅₉₎ = -0.35, p = 0.73). However, there was a significant period effect, such that pain scores at the start of week 9 were significantly less than at the start of week 1 (period 1: 7.7, SD 1.5; period 2: 5.2, SD 3.2; t₍₆₀₎ = 5.60, p \< 0.001). There were no significant differences in baseline pain scores between the ARV-user and ARV-naïve groups at week 1 (ARV-users: 8.1, SD 1.8; ARV-naive: 7.7, SD 1.5; t₍₁₂₀₎ = 1.45, p = 0.15) or week 9 (ARV-users: 5.7, SD 3.2; ARV-naive: 5.2, SD 3.2; t₍₁₂₀₎ = 0.85, p = 0.40). Ignoring any potential period or treatment order effects, there was no significant difference in the absolute change in pain score over six weeks of treatment with placebo or amitriptyline in the ARV-user group (amitriptyline: 2.7, SD 3.3; placebo: 2.1, SD 2.8; t₍₆₀₎ = -1.13, p = 0.26), the ARV- naïve group (amitriptyline: 2.8, SD 3.3; placebo: 2.8, SD 3.4; t₍₆₀₎ = 0.05, p = 0.96), or all participants combined (amitriptyline: 2.7, SD 3.2; placebo: 2.4, SD 3.2; t₍₁₂₁₎ = -0.72, p = 0.47). Summaries of the ANOVA analyses, which controlled for period and order effects, are shown in. For all analyses, no significant treatment effects (changes in pain scores did not differ significantly between amitriptyline and placebo treatment) or order effects (the sequence in which placebo and amitriptyline were taken did not influence changes in pain scores) were detected. But, significant period effects (pain scores in period 2 were lower than in period 1) and significant time effects (pain scores decreased over time within each period) were detected. Intention-to-treat analysis of the full cohort of 124 randomized participants, using baseline observation (from period 1) carried forward, did not alter the result. To avoid the period effect, and because we found no meaningful differences between the ARV-user and ARV-naïve groups, we analyzed the data as a parallel two-arm study using the data from the first period only (weeks 1 to 6). When data from the ARV-user and ARV-naïve groups were combined we did not detect a treatment effect (amitriptyline vs placebo), but there was a significant decrease in pain intensity in both treatment groups over the six-week period, such that pain intensity decreased from 7.9 (SD 1.7) to 4.9 (SD 3.0) in the group receiving amitriptyline (n = 56), and from 8.0 (SD 1.6) to 5.5 (SD 3.1) in the placebo group (n = 66) (ANOVA results are summarized). We also calculated the number needed to treat to achieve at least 50% pain relief using data from all participants in period 1 and data from period 2 for participants who started week 9 (baseline for period 2) with at least moderate pain (≥ 4 on the 11-point NRS). Forty-seven (47) of 105 participants achieved at least 50% pain reduction over the six-week treatment period when taking amitriptyline, while 42 of 110 participants achieved at least 50% pain relief when taking placebo (NNT: 16, 95% CI: 5.1 to -15.2). Because the 95% CI of the NNT includes negative values, the NNT may be interpreted as: amitriptyline was helpful (compared to placebo), and the number needed to treat is greater than 5.1, or amitriptyline treatment was harmful (compared to placebo), and the number needed to harm is greater than 15.2. We observed no dose-response relationship between the dose of amitriptyline taken and the magnitude of the change in pain intensity over six weeks of treatment. ## Safety and Adverse Events Adverse events are reported in. Treatment was well tolerated at the dosages administered. The three most common adverse events observed during the study were drowsiness, dry mouth and chest pain, which were common to the use of amitriptyline and placebo. Significantly more participants on the amitriptyline arm reported dry mouth compared to when taking placebo. ## Rescue Medication The use of rescue medications in the last week of each six-week treatment period was analyzed. Data from the ARV-user and ARV-naïve groups were combined for these analyses. In brief, 13 participants (11%) in each of the placebo and amitriptyline arms of the trial indicated that had taken rescue medication for their pain in the last week of each intervention (McNemar’s p-value = 1). Despite the low proportion of patients indicating that they had taken rescue medication for pain, the majority of participants’ complete lists of medications they were taking (for any reason), included analgesics and anti-inflammatory agents (data summarized). # Discussion We conducted a randomized, placebo-controlled trial to evaluate the efficacy of amitriptyline for the treatment of moderate to severe pain in participants with HIV-SN. In participants on stable antiretroviral therapy (ARV-user group) and in those who had never been exposed to antiretroviral therapy (ARV-naïve group), there was no significant difference in the magnitude of the pain relief produced by amitriptyline compared to placebo after six weeks of treatment. Treatment order did not influence the primary outcome. A period effect was seen for amitriptyline and placebo treatment following the washout period, with pain intensities not returning to baseline values after treatment withdrawal. Our study is novel in its investigation of the efficacy of amitriptyline in two patient groups: i) an ARV-naïve group, whose neuropathy is presumed to me immune-mediated, and ii) an ARV-user group, whose neuropathy may include an additional iatrogenic insult to peripheral nerves from the antiretroviral drugs. These two groups reflect the two broad classes of patients attending outpatient HIV care clinics, and despite possible differences in the mechanisms underlying the polyneuropathy they develop (immune versus immune and drug toxicity) no studies on analgesic therapies for painful HIV-SN have differentiated between these two groups. Also, we conducted the study in a cohort of African descent, set in a socio-economic environment that is broadly reflective of those encountered by most HIV-infected individuals in Southern Africa; the region worst affected by HIV. Our finding that amitriptyline is no better than placebo in ARV-users and ARV-naïve patients is significant for the clinical management of painful HIV-SN, especially in developing countries where the burden of HIV and therefore HIV-SN is high. Agencies providing clinical guidance need to decide whether amitriptyline, should continue to be recommended in the treatment of HIV-SN. Our findings support those of two previous trials, both of which did not achieve the calculated sample size after being stopped on the advice of their respective data monitoring boards because of lack of efficacy before recruitment targets were met. However, the potential clinical effectiveness of amitriptyline was difficult to assess in the those two trials because enrolment included participants with mild pain, and the Initiative on Methods, Measurement, and Pain Assessment in Clinical Trials (IMMPACT) recommend that pain of moderate to severe intensity be investigated. Furthermore, in the study conducted by Shlay and colleagues the investigators had to deviate from the original study design because of issues related to randomization, and their 2x2 factorial design with acupuncture and amitriptyline interventions complicated interpretation. Thus our trial is the first completed study to investigate the efficacy of amitriptyline in moderately to severely painful HIV-SN. Moreover, it is also the first trial of amitriptyline to show that the agent is not superior to placebo in the painful neuropathy that develops in participants who have not been exposed to antiretroviral therapy. While the average results in the ARV-user and ARV-naïve groups were no additional analgesic effect for amitriptyline above that achieved by placebo, 42% (26/61) of participants in both groups achieved pain relief of at least 2 points (on the 11-point numerical pain rating scale) greater than that achieved when they were taking placebo. Analysis of disease, demographic, and phenotype characteristics that may differentiate these responders from non-responders in each group, did not yield any significant findings, except in the ARV-naïve group, where responders tended to be younger than non-responders ( and Data). A study of the analgesic effects of pregabalin in painful HIV-SN reported that responders were more likely to have increased sensitivity to a pin-prick stimulus compared to non-responders, but unfortunately we only assessed for pin- prick hypoesthesia and not hyperesthesia. The convention in analgesic trials is to measure “average pain”, but we measured “current pain” at the time of each interview as our primary outcome. Our reason for using “current pain” is the poor performance of “average pain” in this population. While “current pain” is susceptible to circadian changes, participants in this study were seen in a six-hour window between 09:00 and 15:00, thus limiting the effect any diurnal variation in pain intensity may have had. Also, reanalysis of the data using “average pain over the past three days” did not change the interpretation of the data ( and Data). We used a crossover design for this study. Whilst period and treatment-order effects have not been noted in other neuropathy studies employing a crossover design we found that the reduction in pain scores continued to be observed between treatment periods, making interpretation of the results difficult. From a pharmacokinetics perspective the three-week wash out period was adequate. Moreover, when we combined the participants in the ARV-user group and ARV-naive groups (exposure status to antiretroviral therapy did not influence treatment effect) and looked only at the first six-week period to obviate the period effect, we still found that amitriptyline was no more effective than placebo in reducing HIV-SN pain intensity. Nevertheless, parallel-group studies may be a better option for future studies to avoid this complication in pain studies in particular. Our trial found that inactive placebo produced similar pain relief to that of amitriptyline. Emerging research reports that the placebo effect is particularly prevalent in chronic pain and appears to be high in HIV-positive participants; in the current study, 38% of participants with moderate pain achieved at least 50% pain relief when on placebo, which is similar to the placebo responder rate of 42% recently reported in a trial of pregabalin in painful HIV-SN. A recent meta-analysis of placebo responder rates in drug trials of painful neuropathy reported that trials on patients with painful HIV-SN show significantly greater placebo responder rates than other causes of neuropathic pain. Indeed, the strong period effect in pain relief we observed may well indicate a persistent placebo response in our group of participants. This sustained pain relief may have been triggered through non-conscious cues, for example continued interaction with investigators. This therapeutic “care effect” has been reported in a variety of clinical situations. To mimic clinical practice, we used a flexible dose study design with dose escalation starting at 25mg per day, and a median dose of 50mg per a day was achieved in both study groups. Although the median dose of amitriptyline achieved was within the dose range recommended by Attal and colleagues, meta- analyses of studies employing amitriptyline for the treatment of neuropathic pain report that the average dose of amitriptyline being taken in trials where amitriptyline was deemed to be superior to placebo was 90 mg per day (range: 65–150 mg per day). Thus, participants in the current study may have been taking too low a dose of amitriptyline to achieve pain relief superior to placebo. We escalated the dose of amitriptyline to tolerance or effect, and certainly, there was a significant decrease in pain intensity over each six-week intervention period. This decrease in pain intensity over time, irrespective of treatment, may have led to suboptimal dose escalation for amitriptyline. Indeed, it is unlikely that side effects were a significant cause of participants failing to escalate the dose of amitriptyline being taken. Moore and colleagues recently reported that 64% of participants taking amitriptyline had at least one side effect (compared to 40% taking placebo), a rate that is far greater than the very low rate of side effects we detected in our study cohort. The most plausible explanation for the low side effect rate in this study cohort is that participants were taking too low a dose of amitriptyline. Alternatively, the low side effect rate in this study may have been caused by malabsorption of the drug or non-adherence to the treatment. We did not measure plasma levels of amitriptyline, so we cannot exclude either possibility, but only two participants were noted to have missed doses when pill counts were conducted at weeks three and six for each intervention period. A possible confounder when assessing tolerability of amitriptyline in HIV-positive individuals is the occurrence of dry mouth and fatigue in up to 72% and 48% of ambulatory HIV- infected individuals, respectively. Thus, study participants may not regard these symptoms as new, leading to under-reporting of side effects. # Conclusions We have shown in this randomized controlled study that amitriptyline, which is widely used and recommended for this indication, is no more effective than inactive placebo for the treatment of moderate to severe HIV-SN pain, irrespective of whether participants were on antiretroviral therapy or not. While we cannot exclude the possibility that a significant treatment effect may have been observed at higher doses of amitriptyline, we believe that our findings are robust, and that the doses attained during drug titration reflect those used in clinical practice in our setting (unpublished data). We believe that our and others’ failure to show efficacy for amitriptyline in a variety of different settings is of significant concern given that amitriptyline is recommended by several agencies for the treatment of painful HIV-SN, and is the only drug recommended first line for the treatment of neuropathic pain that is included on the WHO master list of essential medicines and the essential medicines lists of most developing countries. Further research is urgently required to understand the mechanisms and pathophysiology of HIV-SN and to develop alternative treatment modalities and evaluate preventative strategies. # Supporting Information We thank Duncan Mitchell for his comments on the manuscript, Juliane Weber for technical editing of the manuscript, and Catherine Cherry for statistical assistance and comments on the manuscript. [^1]: The following conflict of interest declarations were made by the authors: SM and EM have no declarations; ND and PRK have received honoraria from Pfizer Inc for educational activities; ASCR is a member of the Scientific Advisory Board of Spinifex Pharmaceuticals and throughout has been engaged on the non-clinical and clinical development of AT2 receptor antagonists, for which he is remunerated via Imperial College Consultants, has conducted contract pre-clinical efficacy research on the AT2 programme for which his laboratory at Imperial College received funding from Spinifex Pharmaceuticals, has share options in Spinifex Pharmaceuticals, undertakes consulting for Imperial College Consultants and in the past 36 months has received fees from Spinifex Pharmaceuticals, Astellas, Medivir, Servier and Allergan, has received consultancy fees from the Wellcome Trust Seeding Drug Discovery Committee, is a Principal Investigator in the European Union funded Private Public Partnership Innovative Medicines Initiative “EUROPAIN” – Joint Undertaking and a number of pharmaceutical companies are also engaged in that project, has received funding through EUROPAIN for research studentships from Pfizer and Astellas, has other recent or current grant/studentship funding from Wellcome Trust (London Pain Consortium), Dunhill Medical Trust, NC3Rs, Westminster Medical School Research Trust, International Association for the Study of Pain, National Institute of Academic Anaesthesia, Derek Butler Trust, Medical Research Council Industrial, Biotechnology and Biological Sciences Research Council and Pfizer/ Christian-Albrechts University of Kiel (Neuropain), and is a member of the England and Wales Joint Committee on Vaccination and Immunisation (Varicella sub-group). These declared conflicts of interest do not alter the authors' adherence to PLOS ONE policies on sharing data and materials. [^2]: Conceived and designed the experiments: ND EM PRK. Performed the experiments: ND SM. Analyzed the data: EM PRK. Wrote the paper: ND EM SM ASCR PRK.

# Introduction Climate change poses a serious threat to species persistence. Many species will experience selection in new directions and at new intensities, and the degree to which species respond adaptively to dynamic and uncertain futures will determine their survival over the coming decades and millenia. Intensification of weather extremes and associated catastrophic disturbances is emerging as one of the most important aspects of climate change and research is advancing from studying the impacts of changes in mean climate values (trend effects) to those produced by changes in the magnitude or frequency of extreme events (event effects). Catastrophes are characterized by their statistical extremeness combined with a short duration relative to the life cycle of the organisms affected; they can disrupt ecosystem, community or population structure and change resources, substrate availability, or the physical environment. Evidence suggests that the frequency and intensity of extreme weather events (i.e. floods, droughts) is increasing in many regions in response to global climate change. Despite the urgent need to advance research on extreme events and catastrophic disturbances, their evolutionary consequences have largely been unexplored. Freshwater salmonids are commonly subject to substantial environmental variability in the form of changes in stream flow at different time scales. Extreme events such as floods, droughts or landslides play an important role in the regulation of population dynamics in salmonids, to the extent that in high- density streams, fish demography and persistence might be largely shaped by extreme flow events. The direct and short-term effects of floods are largely a result of high-water velocities and sediment movement that cause the displacement and death of fish. The impact of floods is expected to be higher in the coming years. According to IPCC predictions, an increase of rainfall is expected in the next 50 years along with an intensification and altered frequency of catastrophic disturbances. Recent advances in the statistical theory of extreme events suggest that large flood events will be more frequent, with return times markedly shorter than expected even in the absence of climate change. Our model system is the stream-dwelling salmonid marble trout *Salmo marmoratus*, an endemism of the Adriatic Sea and its tributaries, currently restricted to Northern Italy, former Yugoslavia (Slovenia, Croatia, Bosnia- Herzegovina) and Albania. Marble trout is considered to be one of the most endangered species in the Adriatic basin. For decades, massive restocking practices have been conducted throughout its natural range by means of introduction of exotic brown trout. Hybridization between marble and brown trout has been so extensive that hybrids now dominate most rivers. Molecular data confirm a high level of introgression in the Po river in Northern Italy, the Soca river system in the Italian/Slovenian border, and recently, the Adige river system in South Tyrol. However, all studies also reported pure populations of marble trout in headwaters of all river systems, persisting above natural barriers (i.e. waterfalls). While those barriers prevent the upstream migration of conspecifics and consequent hybridization, the secluded nature and the small size of the remnant marble trout populations makes them extremely vulnerable to stochastic factors, including environmental events such as floods, droughts or landslides. A project for the conservation and rehabilitation of marble trout in Slovenia started in 1993. Only seven remnant pure (non-introgressed with brown trout;) populations of marble trout remain in the Adriatic basin of Slovenia, persisting in totally isolated headwaters without predators or fishing. Recurrent major floods and debris flows are the most important risk factor for the viability of marble trout populations in Slovenian streams. The trigger of debris flows in the area is extreme precipitation, with records of intense flows going back to the 18^th century, and with a presumable occurrence interval of approximately 50 years (i.e. in the 20^th century major floods were recorded in 1926, 1954 and 1990;). However, an intensification of the frequency of flash floods has been observed in the last decade, with four important flood events recorded in the 1999–2010 period. The aim of this study is to explore the impact of catastrophic weather events on the genetic composition of isolated marble trout populations from the Adriatic basin of Slovenia. As a result of intense precipitation and associated flash flood and debris flows, which causes mass mortalities but does not alter connectivity among isolated streams and basins, low levels of genetic diversity and high genetic substructuring at local geographic scale are expected. In this study, we were particularly interested in the comparison of pre- and post-flood samples, testing shifts in genetic composition and genetic diversity and estimating the possible existence of bottleneck/population declines. # Results A total of 18 out of 24 loci were polymorphic, showing from 2 to 8 alleles per locus. All locations showed low levels of genetic diversity as evidenced by heterozygosities \<0.25 and allelic richness \<2. Many loci showed a discontinuous allele distribution, with missing alleles across the allele size range, and few or no rare alleles. A total of 31 location-specific private alleles were found, representing 49% of the total number of alleles sampled. Across locations, Studenc showed the highest genetic diversity, with intermediate levels found in Zadlascica and the lowest genetic variability found in Lipovscek and Sevnica. When comparing pre- and post-flood samples, a moderate drop in genetic diversity was observed in all four locations in terms of *H*_o and *H*_e and allele richness (AR), mostly due to the loss of rare alleles. However, statistic comparisons between pre- and post-flood values of genetic diversity were non-significant at all locations: Zadlascica (*H*_o: p = 0.714; *H*_e: p = 0.891; AR: p = 0.747), Lipovscek (*H*_o: p = 0.650; *H*_e: p = 0.650; AR: p = 0.396), Sevnica (*H*_o: p = 0.591; *H*_e: p = 0.554; AR: p = 0.551) and Studenc (*H*_o: p = 0.726; *H*_e: p = 0.451; AR: p = 0.999). All loci were at HWE after Bonferroni correction. A neutrality test using LOSITAN suggested no loci to be subject to balancing selection or directional selection. No difference in values of genetic diversity was found between EST- derived microsatellites and genomic microsatellites, including *H*_o (EST-microsatellites: 0.126; genomic microsatellites: 0.195; p = 0.232), *H*_e (EST-microsatellites: 0.138; genomic microsatellites: 0.180; p = 0.464) and AR (EST-microsatellites: 2.87; genomic microsatellites: 2.78; p = 0.964). A highly significant extreme overall genetic differentiation was found among samples (*F*_ST = 0.716; p\<0.001). All pairwise *F*_ST and genetic distances between samples from different locations were high (*F*_ST = 0.649–0.779; D_CE = 0.592–0.839) and statistically significant (p\<0.001;). By contrast, comparison of pre- and post-flood samples from the same location showed low non-significant *F*_ST and genetic distances with the exception of Studenc (*F*_ST = 0.041, p = 0.020; D_CE = 0.021, p = 0.035). Accordingly, comparison of allele frequencies between pre- and post-flood samples using an exact test showed no temporal differences at Lipovscek (p = 0.993), Sevnica (p = 0.450) and Zadlascica (p = 0.985), while significant differences were found at Studenc (p\<0.001) caused by loci CA059136 and Str151. Accordingly, an AMOVA analysis partitioned genetic differentiation significantly among locations (*F*_CT = 0.713; p\<0.001) but not among (pre- and post-flood) samples within locations (*F*_SC = 0.011; p = 0.067). We conducted a Multidimensional Scaling analysis using D_CE at all loci. When plotting the values of the first and second principal components, samples clustered according to location, the Zadlascica location (from a tributary of the Soca river) appeared very different from the rest of locations (from tributaries of the Idrijca river), but so did Studenc despite being 20 km apart from Sevnica. In this sense, when testing Isolation-by-Distance, no correlation was found between genetic and geographic (shortest waterway) distances (r = 0.715; F = 4.19; p = 0.110). Assignment analysis using STRUCTURE confirmed the extreme genetic differentiation among locations and suggested a scenario with K = 4 groups as the most likely (p\<0.001), corresponding to the four sampled locations, Zadlascica, Lipovscek, Sevnica and Studenc. The MSVAR procedure for assessing past demographic history strongly supported a decline in all marble trout populations. All sampled points of log₁₀(*r*) were substantially below zero in all eight samples, suggesting that the past population size was larger than the current population size. All sampled points of log₁₀(θ) were negative, pointing to small current effective population sizes. Using a conservative mutation rate of μ = 10⁻⁴, *N*₁ varied across locations between 3 and 41 individuals, with a maximum upper-bound CI of *N*₁ of 396 individuals. Using higher mutation rates (i.e. 5×10⁻⁴ and 10⁻³), *N*₁ values \<than 1 were obtained, while lower mutation rates (i.e. 5×10⁻⁵) yielded unrealistic *N*₀ values. The time of the bottleneck varied between 31 and 268 generations across locations (or roughly 100–800 years assuming a generation time of 2.7–3 years), with a maximum upper-bound CI of 2300 generations (roughly 7000 years). Bottleneck-detection tests using the sofware BOTTLENECK and M_P\_VAL showed some indications of population contraction at all locations. Garza and Williamson suggested that values of M lower than 0.7 would indicate evidence of a bottleneck, while values above 0.8 would denote no bottleneck history. In our data set, all Zadlascica and Studenc samples plus the pre-flood Lipovscek sample showed M values between 0.518–0.661. The two Sevnica samples plus the post-flood Lipovscek sample showed M values between the 0.7–0.8 limits. # Discussion ## Extreme differentiation among marble trout populations Freshwater fish species show a greater average degree of genetic differentiation among locations than marine species, resulting from the isolation of fish populations among drainages. The marked zoogeography produced by the historical patterns of isolation among drainages is subjected to the continuous remodeling of the river drainage and to climatic fluctuations of which the glacial/interglacial periods of the Pleistocene played a crucial role. Species respond actively to fluctuations in their natural range, with hydrographic networks being used for colonization and for retreat throughout geological times. Higher levels of population subdivision have been found in many salmonid studies reflecting complex genetic structures often resulting from isolation among drainages. In brook charr *Salvelinus fontinalis*, a global *F*_ST of 0.37 was found among 26 populations from La Maurice National Park in Canada located 3–42 km apart. In cutthroat trout *Oncorhynchus clarkii*, Taylor *et al.* found an overall *F*_ST of 0.32 and pairwise *F*_ST values up to 0.45 in populations from British Columbia, while the recent study of Pritchard *et al.* reported a global *F*_ST of 0.41 for Rio Grande populations. In bull trout *Salvelinus confluentus*, Taylor and Costello found an *F*_ST = 0.33 among 20 North-West America coastal locations, while recent papers reported an overall *F*_ST of 0.15 and pairwise *F*_ST values up to 0.31 in Alberta and pairwise *F*_ST values up to 0.66 between lake samples in a 50 km area off Montana. Collectively, the extreme level of genetic differentiation found among Slovenian marble trout populations (overall *F*_ST = 0.716; pairwise *F*_ST between drainages = 0.649–0.779; 49% private alleles sampled) exceeds the highest values reported in salmonid populations found above waterfalls and/or in small streams like the ones in this study. Such extremely high *F*_ST is indicative of complete genetic isolation that has persisted over time. This is concordant with the previous study of Fumagalli *et al.*, which reported a global *F*_ST of 0.66 in the same area using a smaller number of microsatellites. At present, the four locations in our study are completely isolated from each other and from the main river drainage by means of impassable natural barriers (waterfalls) that restrict dispersal abilities. Baloux and Lugon-Moulin argued that even in the total absence of gene flow, extreme genetic differentiation of the magnitude found in our study is not expected due to the high mutation rate of microsatellites, with appearance of new variants counteracting within-population fixation of alleles. However, the low level of genetic diversity observed in all locations (low heterozygosities and low number of alleles) implies a strong impact of local genetic drift, and suggests that, together with absence of gene flow, genetic differentiation among locations may have been exacerbated by recurrent mortalities likely caused by flash flood and debris flows. The combined effects of genetic drift and inbreeding following demographic bottlenecks may have thus contributed to alter distinctly the genetic composition of each population, so that over time different populations ended up having different alleles at each locus as well as different allele combinations from multiple loci, resulting in the observed pattern of extreme genetic differentiation. The MSVAR procedure clearly suggested a population contraction with a reduction in population size of 3–4 orders of magnitude to current effective densities of around 3–41 (CI: 1–396) individuals per location. Bottleneck signatures were observed in all samples. First, most polymorphic loci presented a biallelic pattern with few (or none) rare alleles. Second, a discontinuous allele range was found at many loci. For instance, only alleles 104, 110, 118 and 124 were sampled across locations at locus Strutta58, which suggests that the non- observed alleles have been lost over time. By contrast, samples collected in tributaries of the river Po in Northern Italy showed a continuous allele distribution and a higher number of alleles at five microsatellite loci , suggestive of demographically stable populations with no signatures of bottlenecks. Loci Ssa85 and Str15, which were monomorphic in Slovenian locations (our study), showed 6–9 alleles in Italian locations. Loci Str85, Str543 and Str591, which presented three alleles each in our study, showed higher allelic richness in the Italian samples (7–17 alleles). Pujolar *et al.* also reported a lower genetic differentiation among Italian pure marble trout locations (global *F*_ST of 0.235). The different pattern suggested for Italian and Slovenian populations of marble trout, with signatures of a bottleneck only observed in the latter, might be due to the particular geography and the extreme effect of meteorological events in the region of Slovenia studied. Rainfall data have been acquired since 1961 (ARSO, Environmental Agency of Slovenia). While the annual mean precipitation was 2,400 mm in the 1961–2004 period, showing little variation across years, monthly rainfall showed up to 500-fold variation. Morphological features of the streams and watersheds in the region (i.e. high slopes, narrow walls, stream bed fragmentation, limited flood plain) are largely responsible for the high water flow in the stream after heavy rainfalls, and might explain the massive mortalities caused by flood events. In contrast with the extreme differentiation found at microsatellite loci, one single mitochondrial DNA control region haplotype (MA1) has been reported for marble trout in Slovenia. This suggests that despite the current complete geographic isolation of marble trout populations in Slovenian streams, those populations were interconnected in a relatively recent past prior to the fragmentation of habitats. On the basis of microsatellite data, Fumagalli *et al.* suggested that the pure populations within the Idrijca drainage belonged to an independent river system. Nevertheless, our analysis on a larger number of microsatellite loci does not support the division by river basins, as the two locations from the Idrijca basin in our sampling (Sevnica and Studenc) did not cluster together in the Multidimensional Scaling analysis, and appeared clearly differentiated despite being merely 20 km apart within the same river basin. ## Genetic consequences of serial bottlenecks The analysis of pre- and post-flood samples in our study allowed us to explore the genetic consequences of population bottlenecks. The effects of bottlenecks are directly related to the increase of stochastic events associated with small population sizes, leading in most cases to a loss in genetic variability. Recently, Bouzat suggested that the potential genetic outcomes of demographic bottlenecks can only be assessed when considering replicated bottlenecked populations. In our study, we explored four separate locations that experienced episodes of massive mortalities with consequent reductions in population size ranging from 56 to 77%. Collectively, and taking into the consideration the magnitude of the demographic decline, only a limited genetic effect was observed. A moderate drop in genetic diversity was found in all locations, both in terms of heterozygosities and allelic richness, but all comparisons were statistically not significant. However, a total of 9 alleles were lost locally, in all the cases representing rare alleles with frequencies \<0.05 prior to the flood that were no longer sampled after the flood events. Moreover, bottlenecks resulted in alteration of allele frequencies, particularly secondary alleles, which in some cases experienced a drop of 20–30% in frequency. This was notably observed at Studenc, where genetic composition was significantly different in pre- and post-flood samples, but not at the other populations. The moderate reduction of allelic diversity and heterozygosity might be attributable to the particular demographic history of those populations. In his recent review, Bouzat emphasized the potential role of population history in determining the outcome of demographic bottlenecks. Specifically, one can expect that populations experiencing recurrent bottlenecks might have had their genetic pool already eroded over time, which would decrease the effectiveness of both purifying selection and random allele elimination. This holds particularly true for the Slovenian marble trout populations in our study, which have been repeatedly impacted by severe flood events, with a presumable occurrence interval of 50–100 years. Bayesian demographic analysis using the MSVAR procedure is concordant with a scenario of serial bottleneck episodes that might have been occurring for 150–1340 years. On the question of how do the Slovenian population still maintain some polymorphism despite experiencing flood events for hundreds of years, one possible explanation is that our sampling is not representative of the entire population, and other compartments of the population might keep some genetic variation. One hypothesis could be that young fish (age 0+ and 1+) might be less sensible to flows because usually they are not found in the main stream but in more protected small tributaries, repopulating the stream even if all adults have been removed. However, field observations suggest that eggs and young fish are more vulnerable to floods than adults and that there is no recruitment at all when floods occur after reproduction (Crivelli, unpublished information). Alternatively, an upstream part of the population could be preserved. This hypothesis is supported by tagging data (Crivelli, unpublished information) that show non-tagged individuals in post-flood samples, whereas all pre-flood samples had been tagged, which suggests that the newly-arrived non-tagged individuals came from a compartment of the population located upstream, flushed by the flood. ## Coping with environmental change Our findings strongly suggest that the evolutionary and demographic history of marble trout populations living in secluded Slovenian streams has been shaped by massive mortalities caused by disturbance factors such as flash flood and debris flows. Overall, genetic analysis revealed an extremely high genetic differentiation among remnant pure populations, together with much low levels of genetic diversity in all populations. Effective population sizes estimated using the MSVAR procedure ranged from 3 to 41 (CI:1–396) individuals per population. The low genetic variability found does not seem to affect the viability of the populations so far, as they have persisted up to the present despite recurrent and unpredictable disturbances. To mitigate their ecological impact, fish populations might exhibit several adaptations that result from trade-off among growth, reproduction and survival. According to life-history theory, recurrent floods can have important evolutionary consequences by selecting for life- histories that are synchronized to either avoid or exploit the direct and indirect effects of extreme flows. However, due to the predicted intensification of extreme weather leading to increased floods, heatwaves, droughts and rainfall in the next 50 years, marble trout may not have the sufficient genetic potential for the evolution of life-histories able to mitigate their impact. Recent studies by the European Commission's Joint Research Centre (JCR) have confirmed rising temperatures (around 1.5°C in the last 35 years) and higher levels of precipitation in Slovenia, with a projected increase in the occurrence of extreme rainfall events and flash floods. For 2020, Lehner *et al.* predict major flood events for the Adriatic basin of Slovenia to become more frequent and intense. This might jeopardize the survival of marble trout as the occurrence of consecutive floods in a very short period of time could wipe out the population. This nearly occurred in the Lipovscek stream: a good recovery was observed following the 2004 flood, but in September 2007 a new important debris flood occurred with a mortality of 92.4% that left only 38 fish in the whole population. After a new major flood in 2009 followed by a moderate flood in 2010, the population has been decimated to only 10 individuals, so that at present, the recovery of Lipovscek remains uncertain (Crivelli, unpublished information). Medium flows (10 to 20-year recurrence interval) have also become more frequent with 3–5 moderate flow events observed in the last 10 years in the four populations in our study, and additionally, since 2009 spring floods have been observed for the first time in the region. Collectively, while the Slovenian marble trout populations have coped successfully with recurrent yet unpredictable disturbance events over time, the low genetic variation found in the species makes it difficult to assess its evolutionary potential since a genetically depauperate population might fail to adapt to future environmental change. Moreover, all populations are closed, and there is no potential for spontaneous colonization of new habitats or re- colonization after local extinction. It has been proposed that an effective population size of 500 individuals is large enough to maintain genetic diversity for key history traits. The estimation in our study of effective population sizes of 3–41 individuals is one-two orders of magnitude lower than the threshold of 500 individuals, plus the upper-bound CI of population size does not overlap with the threshold value. This suggests that the Slovenian marble trout populations are beyond the critical population size for genetically secure populations. The future development of an integrated genetic/ecologic model that explores different demographic scenarios with varying degree of intensity and frequency of flood events might help disentangling the role of catastrophic disturbances in determining the genetic structure and genetic diversity of marble trout. # Materials and Methods ## Sampling All animal work was approved by the Ministry of Agriculture, Forestry and Food of Republic of Slovenia and the Fisheries Research Institute of Slovenia. Original title of the Plan: RIBISKO - GOJITVENI NACRT za TOLMINSKI RIBISKI OKOLIS, razen Soce s pritoki od izvira do mosta v Cezsoco in Krnskega jezera, za obdobje 2006–2010. Sampling was supervised by the Tolmin Angling Association (Slovenia). A total of 223 pure marble trout *Salmo marmoratus* individuals were caught using electrofishing at four tributaries of the Soca (1.Zadlascica), Baca (2.Lipovscek) and Idrijca (3.Sevnica and 4.Studenc) rivers in Slovenia. During field work fin clips were collected from anaesthetised individuals that were tagged and immediately released back into the streams. At Lipovscek, Sevnica and Studenc, samples were obtained in September 2004 prior to a flooding event that caused a mortality of 56.4%, 77.6%, and 67.9% in fish ≥ age-1, respectively. Post-flood samples were obtained at all three locations in September 2005. At Zadlascica, samples were obtained in September 2007 prior to a flooding event that caused a 74.9% drop in population size. Post-flood samples from Zadlascica were obtained in September 2008. Number of individuals in post-flood samples were constrained to the high mortalities and consequent low number of survivors, with only 15 individuals left in the case of Zadlascica after the flash flood event. ## Microsatellite amplification Minute sections of tissue from ethanol-preserved finclips were digested in a lysis buffer containing 100 µl TE Buffer, 7 µl 1 M DTT (dithiothreitol) solution pH 5.2 (diluted in 0.08 M NaAC) and 2 µl proteinase K solution (20 mg/ml) for at least 8 hours at 56°C. After incubation at 96°C for 10 min, samples were centrifuged at 13,000 rpm for 11 min, and the supernatant was stored at −20°C. All samples were scored for a total of 24 microsatellite loci. 15 loci were derived from Expressed Sequence Tags (ESTs) previously described by Vasemagi *et al.* and Siemon *et al.* in Atlantic salmon *Salmo salar*. We selected those loci with the highest genetic variation and a positive cross-species amplification in brown trout *Salmo trutta*. Additionally, 9 genomic microsatellite loci isolated and characterized from other salmonids, seven from brown trout (Str73, Str151: ; Str85, Str543, Str491: ; T3-13: ; Strutta58:), one from Atlantic salmon (Ssa85:) and one from marble trout (BFRO001:) were amplified and scored. Microsatellites were grouped in two separate multiplexes in order to reduce polymerase chain reaction (PCR) and genotyping costs. PCR products were obtained in a GeneAmp PCR System 2700 Thermocycler (Applied Biosystems) using the QIAGEN Multiplex PCR Kit. PCR reactions consisted of 2 µl template DNA, 5 µl QIAGEN Multiplex PCR Master Mix, 0.2 µl 10 µM forward and reverse primers, and water up to 10 µl. PCR conditions were as follows: 3 min at 95°C, 35 cycles of 30 sec at 94°C, 90 sec at 57°C and 1 min at 72°C, and final elongation for 5 min at 60°C. PCR products were visualized in 1.8% agarose gels and screened for microsatellite polymorphism using an ABI 3130 AVANT automatic capillary sequencer (Applied Biosystems). Alleles were sized according to a Liz500 (50–500 bp) marker. ## Data analysis Within-sample genetic diversity statistics were assessed by observed (*H*_o) and expected (*H*_e) heterozygosities per locus using GENETIX version 4.05 and allelic richness (AR) using FSTAT. Differences in genetic diversity among samples were tested by one-way ANOVA using STATISTICA version 6.0 (StatSoft Inc.). Deviations from Hardy-Weinberg Equilibrium (HWE), linkage disequilibrium and differences in allele and genotype frequencies among samples were tested using GENEPOP version 3.4. Neutrality of the markers was tested using the software LOSITAN, which implements a *F*_ST outlier detection approach. This method evaluates the relationship between *F*_ST and expected heterozygosity (*H*_e), describing the expected distribution of Wright's inbreeding coefficient *F*_ST vs. *H*_e under an island model of migration with neutral markers. This distribution is used to identify outlier loci that show excessive high or low *F*_ST compared to neutral expectations. Such outlier loci are candidates for being subject to selection. We used the 0.99 criterion in order to minimize false positives as suggested by the authors. Population structure was explored using non-hierarchical and hierarchical *F*-Statistics calculated using ARLEQUIN. First, overall and pairwise *F*_ST values were calculated. Genetic differentiation was also partitioned among locations (4 locations: Zadlascica, Lipovscek, Sevnica and Studenc) and among samples within locations (pre- and post-flood samples). Significance tests were assessed with 10,000 permutation tests. In all cases, significance levels were corrected for multiple comparisons using Bonferroni. Pairwise multilocus comparisons between samples were calculated by Cavalli- Sforza and Edwards chord distance (D_CE) and graphically represented by Multidimensional Scaling (MDS) analysis using STATISTICA version 7.0 (StatSoft). Isolation-by-Distance (IBD) was tested using a Mantel test implemented in GENETIX, by correlating linearized genetic distance (*F*_ST/(1−*F*_ST)) vs. geographic distance (shortest waterway distance measured along the streams between pairs of samples). A model-based clustering algorithm as implemented in the software STRUCTURE was used in order to infer the most likely number of populations in the data. The software organizes individuals into a predefined number of clusters (K) with a given likelihood, which may represent putative groups. The analysis was performed with 1\<K\<8 to account for population substructuring within species, using the admixture model and without a population prior. The most likely K was determined using the criterion of Evanno *et al.* and then used to assign each individual. A burn-in length of 10⁵ iterations followed by 10⁶ additional Markov Chain Monte Carlo (MCMC) iterations were performed. A minimum of 5 runs were performed for each K to check repeatability of results. Historical demographic changes were inferred using the Bayesian coalescent-based approach implemented in MSVAR. Using a strict stepwise mutation model, the procedure provides distributions of the exponential growth rate *r* = *N*₀/*N*₁ (where *N*₀ is the present effective population size, *N*₁ is the effective population size at the time of population expansion or decline), the time since the population started to expand or decline *t*_f = *t*_a/*N*₀ (where *t*_a is the number of generations since the beginning of the expansion/decline) and the genetic parameter θ = 2*N*₀μ. A mutation rate of μ = 10⁻⁴ was used. Rectangular priors were chosen for all parameters, with limits of (−9, +5) for log₁₀(*r*), log₁₀(*t*_f) and log₁₀(θ) and starting values of 1 for θ at each locus, *r* and *t*_f. The analysis was performed for an exponential model of population change. We used 50,000 thinned updates and a thinning interval of 50,000 steps, with an initial 10% discarded as burn-in. Convergence was assessed using Tracer v1.4. Finally, we tested for a recent genetic bottleneck episode with two different approaches. First, we used the software BOTTLENECK, based on the principle that after a recent reduction of effective population size, number of alleles (k) decreases faster than heterozygosity (*H*_e) at polymorphic loci. Thus, in a recently bottlenecked population, the observed gene diversity is higher than the expected equilibrium gene diversity (*H*_eq) which is computed from the observed number of alleles (k), under the assumption of a constant-size (equilibrium) population. We used the Multiple-step Stepwise (TPM) model, which consists of mostly one-step mutations but a small percentage of multi-step changes, and is the recommended model for microsatellite data sets rather than the Infinite Alleles (IAM) or Single-step Stepwise (SMM) models. The proportion of singlestep mutation events was set to 90% (variance = 12%). Observed and expected heterozygosities were compared using a Wilcoxon sign-rank test as suggested by Piry *et al.*. Second, we calculated M, the mean ratio between number of alleles (k) and range in allele size (r), assuming that during a bottleneck episode k decreases faster than r (M_P\_Val;). Hence, the value of M decreases when a population is reduced in size. Average M was calculated across loci and compared with the critical value M_crit estimated after 10,000 simulations and assuming the population to be at equilibrium. In all simulations, three different values of θ were used (5, 10 and 20). A range of mutation models were examined and conservative values were used for p_s (frequency of one-step mutations) and Δ_g (average size of non one-step mutations), p_s = 0.90 and Δ_g = 3.5, respectively. # Supporting Information We thank the Tolmin Angling Association (Slovenia) for support in sampling. [^1]: Conceived and designed the experiments: JMP SV DJ AC. Performed the experiments: JMP. Analyzed the data: JMP LZ. Contributed reagents/materials/analysis tools: LZ AC. Wrote the paper: JMP SV LZ GDL AC. [^2]: The authors have declared that no competing interests exist.

# Introduction Insecticide resistance is an increasing problem that compromises the control of insect pests of medical, veterinary and agricultural impact. An understanding of insecticide resistance mechanisms is essential for the subsequent development of tools and practices that can improve pest control interventions. During the last decades, extensive biochemical, genetic and molecular studies have been conducted to elucidate insecticide resistance mechanisms. Knowledge of the mechanisms underlying target site resistance in major pests to some commonly used insecticides has been established to some extent. The understanding of detoxification/metabolism-based insecticide resistance mechanisms has not kept similar pace, due to the complexity of the involved multi-gene systems and the lack of genome sequence data. However, in a few cases, the molecular basis of metabolism-based insecticide resistance mechanisms was identified. A single P450, CYP6P3, was over-expressed in pyrethroid resistant *Anopheles gambiae* mosquitoes, and it was capable of metabolizing pyrethroids. Karunker et al. showed that the *B*. *tabaci* cytochrome P450 *BTCYP6CM1* is capable of metabolizing the neonicotinoid Imidacloprid, one of the most important insecticides worldwide, and to confer neonicotinoid resistance. These studies have shed light on cases of metabolism-based insecticide resistance mechanisms. However, there is a number of issues which remain unsolved, such as the underlying molecular mechanisms that are responsible for over-expression of detoxification enzymes. In addition, the information on molecular changes responsible for resistance often comes too late, i.e. when resistance has been irreversibly established in pest populations and/or when the active ingredient has already been replaced by others. The use of modern molecular approaches and models for the early identification or even prediction of insecticide resistance mechanisms could improve the management of the phenomenon. *Drosophila melanogaster*, although not a pest species, has been used extensively for insecticide resistance research. Resistance associated with the over-expression of a single P450 gene (Cyp6g1) has been documented for field- derived *Drosophila* lines resistant to Imidacloprid and DDT. Over-expression correlated with the presence of a single insertion of an *Accord* transposable element into the 5′ end of the Cyp6g1 gene has also been reported. A recent study of Cyp6g1 induction in transgenic *Drosophila* showed tissue-specific expression of this gene controlled by two distinct specific enhancers, suggesting that a single mutation event can modulate Cyp6g1 expression. In contrast to field pest populations, which often possess a highly heterogeneous genetic background, the possibility for the generation of single mutations in a known and characterized background would substantially facilitate the identification of resistance-associated changes. Insertional mutagenesis using transposable elements has been an exceptionally efficient method to create mutants in phylogenetically very distant species, including *Drosophila melanogaster*. The transposon *Minos*, a member of the Tc1/*mariner* superfamily, produces stable transformants with high efficiency in different insect species. This allows genome-wide mutagenesis in insects making *Minos* a promising genome-wide transgenesis tool. High-throughput deep sequencing transcription profiling is a powerful approach to provide genome-wide information in a very short time and a cost effective way. This method is classified as an “open” technology, which in contrast to “closed” technologies like microarrays, does not require biological or sequence information of the analyzed organism. Here, by combining a genome-wide insertional mutagenesis screen and next generation transcriptomics, we were able to identify genes involved in Imidacloprid resistance in *Drosophila melanogaster* within a reasonable time frame and at moderate cost. Gene ontology analysis identified several overrepresented functional gene groups that are differentially expressed in the resistant *Drosophila* line. The results of our novel approach were in line with previous findings that showed that the Cyp6g1 gene is mainly responsible for resistance. The deep sequencing information was further explored to identify transcription binding factors or microRNAs possibly associated with the over- expression of Cyp genes, which are implicated in resistance. Genetic mapping placed the resistance locus to the right arm of the second chromosome, within a ∼1 Mb region in which the Cyp6g1 gene is located. # Results During the genome-wide insertional mutagenesis, about 12,900 new TREP insertions were generated. The *D. melanogaster* genome is estimated to contain approximately 13,000 known or predicted genes. Using the information in Metaxakis et al., it can be estimated that in our screen approximately 22% of known or predicted genes of *Drosophila* genome were hit at least once, directly or within 2 Kb upstream and downstream, excluding introns. Flies with new TREP insertions were selected on medium with 3 µg/ml of Imidacloprid (3 times higher than the LC99 of the susceptible line). One female carrying a novel TREP insertion on the X-chromosome and exhibiting high resistance to Imidacloprid, was retrieved during the mutagenesis. Line MiT\[*w*⁻\]3X, originating from the retrieved resistant female, was established. Genetic analysis of the MiT\[*w*⁻\]3X line showed that the resistance trait mapped to the second chromosome (see below) but showed no linkage between the TREP insertion and the Imidacloprid resistance. ## Resistance to Imidacloprid and cross-resistance to DDT The resistance was characterized by determining the susceptibility of the resistant mutant MiT\[*w⁻*\]3R2 (derived from MiT\[*w⁻*\]3X line) homozygous for the second chromosome and the control line iso31 to Imidacloprid. The Imidacloprid resistance of line MiT\[*w⁻*\]3R2 was found to be about 18-fold higher than that of the wild-type line iso31. Mutant MiT\[*w⁻*\]3R2 showed resistance to Imidacloprid also at the adult stage, as well as cross resistance to DDT. ## Biochemical assays In order to assess if there is a contribution of known resistance pathways in resistance of mutant MiT\[*w⁻*\]3R2, the activities of cytochrome P450 monooxygenase, esterases and GSTs were analyzed. Esterase activity was measured using α- and β-naphthol, and GST activity was measured using 1-chloro-2,4-dinitrobenzene. For both enzymes, no significant difference in activity was detected in the resistant line compared to line iso31. Cytochrome P450-dependent monooxygenase activity was determined by analyzing living third instar larvae, following the protocol of Inceoglu et al.. The activity of cytochrome P450 was 3-fold higher in resistant MiT\[*w⁻*\]3R2 third instar larvae compared to third instar susceptible larvae. ## Deep sequencing analysis Transcriptional profiling of the resistant MiT\[*w⁻*\]3R2 line was performed to obtain more information on the involvement of individual genes in the resistance. Deep sequencing yielded 16,344,712 and 16,859,384 mapped reads of 51 bp for MiT\[*w⁻*\]3R2 and iso31, respectively (GSM707197 MiT\[*w⁻*\]3R2 (GSM707197_Resistant_s\_1_READS.txt.gz); GSM707198 iso31 (GSM707198_Susceptible_s\_2_READS.txt.gz)). Alignment of the sequencing reads to the *Drosophila* reference genome (*Drosophila* release 5 sequence assembly Flybase) identified 18,963 transcripts for the susceptible line and 18,967 transcripts for the resistant line (GSE28560_total_number_of_transcripts.txt.gz). Using a minimum difference threshold of 2-fold, a total of 357 transcripts were found to be differently expressed between MiT\[*w⁻*\]3R2 and iso31 (GSE28560_resistant_vs.\_susceptible_UPREGULATED_GENES.txt.gz; GSE28560_resistant_vs.\_susceptible_DOWNREGULATED_GENES.txt.gz). In the resistant line, 150 genes were up-regulated and 207 genes were down-regulated (GSE28560_resistant_vs.\_susceptible_UPREGULATED_GENES.txt.gz; GSE28560_resistant_vs.\_susceptible_DOWNREGULATED_GENES.txt.gz). Gene functional classification analysis by grouping genes based on functional similarities identified three functional groups in the up-regulated genes and two functional groups in the down-regulated genes. The cytochrome P450 genes, proteolytic genes and genes showing peptidase activity were overrepresented in the up-regulated genes. Cuticular protein genes and genes showing peptidase activity were overrepresented in the down-regulated genes. Functional annotation clustering, which groups genes with similar predicted biological functions, identified 10 overrepresented groups in the up-regulated genes and 13 overrepresented groups in the down-regulated genes. Among the functional groups overrepresented in the up-regulated genes, four clusters are connected to peptidase activity and three functional clusters are connected to P450 gene family activity. There were also other functional groups with significantly overrepresented members in the over-expressed genes, like oxidoreductase activity, mitotic sister chromatid segregation, electron carrier activity and response to DNA damage. In the down-regulated genes, groups like nutrient reservoir activity, chitin and aminoglycan metabolic processes, response to bacteria and immune response activity were identified. Eight of the 150 up-regulated genes, were members of the P450 gene family (GSE28560_resistant_vs.\_susceptible_UPREGULATED_GENES.txt.gz). The three highest over-expressed P450 genes, with more than 15-fold expression in MiT\[*w⁻*\]3R2 were Cyp4p2 (100-fold), Cyp6a2 (19.9-fold) and Cyp6g1 (16.3-fold) (GSE28560_resistant_vs.\_susceptible_UPREGULATED_GENES.txt.gz). We performed quantitative real time PCR to validate the expression difference of two representative cytochrome P450 genes (Cyp6g1 and Cyp6a2), already known to play important role in insecticide resistance. Cyp6g1 showed an expression difference between the resistant and susceptible lines of 8.4 (±0.7), while gene Cyp6a2 showed an expression difference of 10.3 (± 1.7) fold. The expression difference of Cyp4p2 was also validated with quantitative PCR, and showed 4.9 (± 0.3) fold elevated expression in the resistant line. ## Chromosomal mapping of the Imidacloprid-resistance locus Mapping of the resistance locus of MiT\[*w⁻*\]3X flies to a chromosome was done with standard genetic tools. Males from Imidacloprid- resistant line MiT\[*w⁻*\]3X were individually crossed with *w*1118iso/Dp(1;Y)y+; nocSco/SM6a females, who carry chromosome 2 balancer SM6a. Progeny from this cross, heterozygous for the second chromosome, was selected on Imidacloprid as described. Both resistant male and female progeny emerged, implying that the resistance locus does not map to the sex chromosome. Resistant heterozygous male progeny carrying SM6a was individually crossed with iso31 (susceptible line) females. Progeny from this cross was also selected on Imidacloprid. None of the progeny carried the chromosome 2 balancer, which places the resistance on the second chromosome. Equivalent crosses were performed between resistant MiT\[*w⁻*\]3X males and *w*1118/Dp(1;Y)y+; TM2/TM6C, Sb1 female flies, carrying two balancers of chromosome 3. This cross showed that there is no correlation between the resistance locus and the third chromosome, confirming that the resistance locus is located on the second chromosome. Line MiT\[*w⁻*\]3R2/CyO, which lacked the *Minos* insertion but carried the resistance trait and a lethal locus was derived from MiT\[w\]3X. The lethality locus was mapped to the right arm of the second chromosome, using the Bloomington Stock Center *Drosophila* deletion kit, between position 49C1-4; 50C23-D2 (8.5 Mb –9.9 Mb;). Recombination mapping placed the lethality locus on the same chromosome arm that harbors the resistance locus. Line MiT\[*w⁻*\]3R2, homozygous for the second chromosome resistance locus, was established during the recombination analysis. MiT\[*w⁻*\]3R2 was derived from the original resistant line (MiT\[*w⁻*\]3X) using *Drosophila* lines with different genetic backgrounds (TREP 2.30 and BOEtTA have a yw background, while \[SM6a, MiT 2.4\]/Sco\] is an iso31 derivative). In order to replace the genetic background of the resistant mutant with that of the susceptible control line, MiT\[*w⁻*\]3R2 was back-crossed with iso31 for 6 generations under selection with 3 µg/ml of Imidacloprid. ## P element mapping In order to narrow down the position of the resistance locus in the MiT\[*w⁻*\]3R2 line on the 2R chromosome, genetic mapping relative to P element insertions was used. The distance between a P element located at ∼0.5 Mb and the resistance locus was determined to be 8.2 cM. The distance between a P element located at ∼6.1 Mb and the resistance locus is 3.5 cM. The distance between a P element located at ∼ 6.5 Mb and the resistance locus was determined to be 2.8 cM. The distance between a P element located at ∼ 11.2 Mb and the resistance locus was determined to be 3.0 cM. These genetic distances were converted into Mb using estimates of local recombination rates according to Fiston-Lavier et al. and Singh et al.. The genetic mapping relative to the P element insertions places the resistance locus roughly between 8 Mb and 9.7 Mb on the right arm of the second chromosome in the MiT\[*w⁻*\]3R2 mutant line. ## In silico analysis A comparison of single nucleotide polymorphisms (SNP) of the deep sequencing data between the resistant line MiT\[*w⁻*\]3R2 and the susceptible line iso31 was done for chromosome 2R. Resistant line MiT\[*w⁻*\]3R2 had been back-crossed with line iso31 under selection with 3 µg/ml of Imidacloprid in order to homogenize the genetic background. The SNP comparison indicates a hybrid origin of the 2R chromosome, where the right half comes from iso31, while the left half comes from a different line, most likely yw. This result indicates a recombination event on 2R, close to the region between 8.5 Mb and 9.9 Mb, to which the lethality was mapped. The position of the lethality, resistance and recombination break point shows that the recombination event occurred between the resistance and lethality loci. The highly over-expressed Cyp6g1 gene (16.3-fold – deep sequencing analysis; 8.4-fold- real time PCR analysis) lies close to the recombination break point to the region into which the resistance locus has been placed. Comparison of the sequences of the Cyp genes differently expressed in the resistant versus the susceptible line showed no sequence changes of the P450 proteins. We also analyzed the flanking sequences of differentially expressed genes for possible common transcription factor binding sites. *In silico* analysis, using the JASPAR database, did not detect common transcription binding factors either for the subgroup of Cyp genes or for all over-expressed genes. Similarly, a search for predicted targets of microRNAs, performed with DIANA- microT version 3.0, in the 3′UTRs of all up-regulated and down-regulated genes, or of just the up-regulated and down-regulated Cyp genes, did not identify any significantly overrepresented common target sites. # Discussion We tested a combined approach of mutagenesis and next generation transcriptomics to study insecticide resistance in the model organism *Drosophila melanogaster*. A *Minos*-based construct was used for genome-wide insertional gene activation mutagenesis. During this screen, an Imidacloprid-resistant *D. melanogaster* female was retrieved. In the resistant line MiT\[*w⁻*\]3R2/CyO lacking the TREP (*Minos*) insertion, which was derived from this retrieved mutant female, both lethality and resistance were detected. In a recombination experiment resistance was separated from lethality and line MiT\[*w⁻*\]3R2 homozygous for resistant second chromosome was established. This line was further analyzed for the resistance mechanism. Cross-resistance of the Imidacloprid-selected MiT\[*w⁻*\]3R2 mutant to DTT suggests metabolic resistance as the mechanism of resistance in this line. Furthermore, biochemical analysis showed increased P450 activity in the resistant line compared to the susceptible line. The Illumina parallel short-sequencing technology was used to obtain total cDNA sequences of the resistant line and of the non-resistant isogenic line iso31 (*w*¹¹¹⁸_iso; 2_iso; 3_iso). This approach was used in order to identify and quantify differences in expression between mutant line MiT\[*w⁻*\]3R2 and susceptible line iso31, covering nearly all *D. melangaster* genes. Out of 357 genes differently expressed in the resistance line, 150 were up-regulated, and 207 genes were down-regulated in comparison to the susceptible line. Gene ontology functional classification of the sequenced transcripts identified a significantly up-regulated P450 family group and two groups of genes coding for peptidase activity in the resistant line. Significantly overrepresented down-regulated groups of genes were cuticular protein genes and other peptidase genes. Deep sequencing analysis detected eight members of the P450 family, Cyp4p2, Cyp6a2, Cyp6g1, Cyp6w1, Cyp4e3, Cyp309a2, Cyp6g2 and Cyp4d14, with elevated expression in the resistant line. Genes encoding glutathione-S-transferases, as well as esterases did not show elevated expression in the resistant line. The most highly over-expressed P450 genes as detected with deep sequencing and confirmed by real time PCR, are Cyp4p2, Cyp6a2 and Cyp6g1 (GSE28560_resistant_vs.\_susceptible_UPREGULATED_GENES.txt.gz). The cytochrome P450 genes play an important role in insecticide resistance, because of their variety and the broad substrate specificity of some P450 genes. We report for the first time elevated expression of the Cyp4p2 gene in a *D. melanogaster* line resistant to Imidacloprid and DDT, although its role in resistance (if any) remains to be elucidated. The detoxification function of Cyp6a2 and Cyp6g1 in *Drosophila* is well documented. Over-expression of Cyp6g1 in *Drosophila* confers resistance to DDT and neonicotinoids, which provides a certain degree of validation in our approach for detecting genes conferring insecticide resistance. The Cyp6a2 is also highly expressed in different insecticide resistant *Drosophila* strains, and the CYP6A2 encoded enzyme can metabolize insecticides. Five other P450 genes (Cyp6w1, Cyp4e3, Cyp309a2, Cyp6g2 and Cyp4d14) detected in the resistant line are over-expressed up to 6-fold. Microarray analysis showed that expression of Cyp6w1 is higher in DDT resistant *Drosophila* strain compared to a susceptible line. Over-expression of the Cyp6g2 gene confers resistance to diazonin and nitenpyram in transgenic *Drosophila*. To date, Cyp4e3, Cyp309a2 and Cyp4d14 have not been implied in insecticide resistance. A number of cuticular protein genes were down-regulated in the resistant mutant compared to the susceptible line. This could occur as a result of the general stress response induced by the up-regulated detoxification system. It is not likely that the down-regulation of cuticular protein genes plays a role in the insecticide resistance mechanism. It would be in disaccord with the fact that reduced cuticular penetration of insecticides can contribute to resistance in some insect species. The identification of a group of 21 up-regulated genes involved in peptidase activity is consistent with the finding that genes coding for peptidase activity are also significantly over-expressed in DDT resistant *Drosophila*. The role of the proteolytic genes and genes showing peptidase activity in insecticide resistance is still poorly understood. There is increasing evidence of involvement of protein metabolism in insecticide resistances of different insect species,. Proteases may be involved in modification of enzyme conformation and protein biosynthesis, in order to meet energy requirements during xenobiotic stress. Other groups of overrepresented members among the up- or down-regulated genes belong to the following categories: oxidoreductase activity, chromosome establishment, organelle localization and cellular response to DNA damage (up- regulated;) and nutrient reservoir activity, response to bacteria, biotic stimulus and immune response (down-regulated;). Down-regulation of genes involved in immune response was not seen in other DDT-resistant *Drosophila* lines. Oxidoreductase activity plays a role in detoxification, while the other biological processes could be an indication of general stress response. The line MiT\[*w⁻*\]3R2, homozygous for the resistance chromosome, derives from the mutant line MiT\[*w⁻*\]3R2/CyO heterozygous for the second chromosome carrying both resistance and lethality. Genetic analysis of the mutant line MiT\[*w⁻*\]3R2/CyO line placed the lethality locus to the region between 8.5 and 9.9 Mb on the right arm of the second chromosome of *D. melanogaster*. Single nucleotide polymorphism analysis between the resistant line homozygous for resistant chromosome MiT\[*w⁻*\]3R2 and the susceptible line iso31 indicates that the recombination event that separated the lethality from the resistance locus occurred in close vicinity to the lethality locus. The three highest up-regulated P450 genes Cyp4p2, Cyp6a2 and Cyp6g1 are also located on the right arm of the second chromosome, but they are not closely linked. Mapping against P element insertions confirmed that the resistance locus lies on the right arm of the second chromosome between 8 Mb and 9.7 Mb. Chromosomal mapping of the resistance in DDT and Imidacloprid resistant *Drosophila* lines placed the DDT resistance locus (*Rst (2) DDT*) in an area that overlaps the interval in which the resistance locus of MiT\[*w⁻*\]3R2 is located. The position of the lethality locus (between 8.5 and 9.9 Mb), together with the SNP analysis and P element mapping, suggests that the resistance locus in MiT\[*w⁻*\]3R2 lies within an interval of less than ∼1 Mb. Interestingly, the highly up-regulated Cyp6g1 (16.3-fold – deep sequencing analysis; 8.4-fold – real time PCR analysis) gene is located within this range. In the mentioned study of Daborn and colleagues the Cyp6g1 is strongly suggested as the main candidate gene responsible for the resistance in DDT and Imidacloprid resistant *Drosophila* lines. The mutation event which causes the resistance in MiT\[*w⁻*\]3R2 remains to be identified. The resistance locus is not linked to an insertion of the transposon used in the screen. It is conceivable that a “hit and run” *Minos* insertion effect might be responsible for the mutation, where the transposon first integrated and then re-excised. In *Drosophila*, *Minos* often leaves behind upon excision either a characteristic six bp “footprint” or a deletion around the site of insertion, both of which can be mutagenic. It has been suggested that mutations of *trans*-regulating factor/s, or of cis-acting elements of some of the Cyp genes are responsible for insecticide resistance in *Drosophila*. A recent report suggests that a single mutation event in a specific enhancer can modulate Cyp6g1 tissue-specific induction in *Drosophila* flies. One might thus speculate that a single mutation event occurred in a *cis*-acting element of the Cyp6g1 gene, increasing the expression of this gene. This in turn could activate a resistance cascade, affecting the expression of other Cyp genes involved in resistance. Alternatively, the mutation might involve a gene encoding a transcription factor or a microRNA which regulates in *trans* the Cyp genes involved. We have so far no evidence for the latter assumption, since an *in silico* search failed to identify common transcription factor motifs regulating the over-expressed P450 genes. The same is true for common predicted microRNA targets in the 3′UTRs. Additional analyses are required in order to pinpoint the exact cause of resistance in the MiT\[*w⁻*\]3R2 mutant. # Materials and Methods ## Drosophila lines *D. melanogaster* stocks were maintained on standard cornmeal-agar-yeast medium at 24°C with a 12-hour light/12-hour dark cycle. We analyzed a *Drosophila melanogaster* line (MiT\[*w⁻*\]3R2) resistant to the neonicotinoid Imidacloprid, retrieved during a transposon *Minos*-based insertional mutagenesis screen. Three transgenic *Drosophila* lines were used for *Minos*-based insertional mutagenesis: TREP 2.30, BOEtTA and an iso31 derivative \[SM6a, MiT 2.4\]/Sco\]. Line 2.30 carries a single insertion of the *Minos* transposon TREP, which contains a minimal promoter driven by the tetO operator. When inserted next to or into a gene, TREP can cause over-expression of the gene in the presence of the tetracycline trans-activator tTA \[50, Kiupakis, Oehler and Savakis, manuscript in preparation\]. Line BOEtTA carries a single insertion of a *Minos* transposon which produces tTA. The mobilization of the TREP construct and generation of flies with new insertions was performed with a standard “jumpstarter” system. Flies with new TREP insertions were selected for insecticide resistance during egg to adult development on medium with Imidacloprid. We aimed to generate highly resistant flies, thus mutagenized flies were selected on a concentration of Imidacloprid 3 times higher than the LC99 of susceptible line iso31 (3µg/ml). Female individuals carrying both TREP 2.30 and BOEtTA construct were scored for resistance. The insertional site distribution of new TREP insertions was estimated according to Metaxakis et al.. The correction for multiple insertions into the same genes was done using the Poisson distribution, assuming the same probability of recovering insertions for all loci. One resistant female with a *Minos* insertion located on the X chromosome was retrieved from the screen and further analyzed. For the genetic analysis, we also used balancer lines for the second chromosome, *w*¹¹¹⁸_iso/Dp(1;Y)y⁺; noc^Sco/SM6a, and for the third chromosome, *w*¹¹¹⁸_iso/Dp(1;Y)y⁺; TM2/TM6C, Sb¹. Isogenic line iso31 (*w*¹¹¹⁸_iso; 2_iso; 3_iso) was used as an insecticide susceptible control line (wild- type). ## Bioassays The insecticide Imidacloprid (98.7%, Bayer CropScience GmbH) was added directly to the standard medium for the screening experiments. Per vial, 50 eggs were added, and egg-to-adult viability was determined. Six different concentrations were used with 8 replicas per concentration. A contact assay was used for Imidacloprid and DDT assay. 35 ml glass vials were coated with DTT (DDT 4,4′– DDT PESTANAL®, SIGMA-ALDRICH) on the inside by evaporating 200 µl of acetone containing the required amounts of DDT. Coated vials were plugged with cotton wool soaked with 5% sucrose. The mortality of 25 flies (1–3 days old) per vial was scored after 24 hours. Each DTT amount was assayed for sex different concentration with 4 replicas per concentration. All toxicology data were analyzed for LC50 (lethal concentration 50), using Probit statistics with SPSS 10.0 for Windows. ## Biochemical assays The activities of esterases, GSTs and cytochrome P450 dependent monooxygenases were determined as previously described. The activity of cytochrome P450 dependent monooxygenases was measured in live third instar larvae. For all assays, activity was measured at 25°C on a SpectraMax M2 microplate reader and quantified with the integrated software SoftMax pro v5. ## Deep sequencing analysis Deep sequencing data were deposited to the GEO site (series record GSE28560). The following URL allows review of the record. <http://www.ncbi.nlm.nih.gov/geo/ query/acc.cgi?token=nbyhvegqykeoanm&acc=GSE28560>. Total RNA from 3 day old *Drosophila melanogaster* adults was extracted using standard Trizol RNA isolation protocol (<http://quantgen.med.yale.edu/>). Preparation of cDNA for sequencing was done with the Illumina mRNA Seq V2 kit (Illumina, Inc). Formation of single molecule arrays, cluster growth and sequencing were done according to the standard protocols from Illumina, Inc. Sequencing was performed with a 2008 Illumina Genome Analyzer, version 2 (GA2). Mapping of the 51 nucleotides (nt) long sequencing reads of both lines, MiT\[*w⁻*\]3R2 and iso31, to the reference genome (*Drosophila* release 5 sequence assembly Flybase) was performed with software RMAP, version 2.05. Genes with 10 or less reads for one line and 50 reads or less for the other line were excluded from further analysis. The minimum difference threshold between lines was set to 2-fold. Analysis of up-regulated and down-regulated genes was performed with the DAVID 6.7 BETA release bioinformatics resources. ## Quantitative Real Time PCRs Total RNA from *D. melanogaster* 1–3 day old adult flies was extracted using same protocol as for deep sequencing analysis. Synthesis of first-strand cDNA and PCR reactions were performed on a Techne TC-412 PCR machine. Synthesis of the First-Strand cDNA was done with the AccuScript® High Fidelity RT-PCR System. Expression of Cyp6g1 and Cyp6a2 was measured relative to the housekeeping ribosomal protein gene Rp49. For this purpose, three sets of primers were designed. In order to obtain products specific for the cDNA templates, primers were designed to span exon-intron junctions. The forward and reverse primer sequences were as follows: Cyp6g1 – 5′ACCCTTATGCAGGAGATTG3′ and 5′TAGGCTGTTAGCACGAATG3′; Cyp6a2 – 5′GTTACTGCCTGTATGAGTTGG3′ and 5′TAGAGCCTCAGGGTTTCTG3′; Rp49 – 5′CGGTTACGGATCGAACAAGCG3′ and 5′TTGGCGCGCTCGACAATCT3′. Quantitative real time PCR was performed using the QIAGEN SYBR green kit with the DNA Engine Opticon TM MJ Research analyzer. Three technical replicates were performed on each of three biological samples. The efficiency of PCR amplification with each gene-specific primer pair was analyzed with five serial dilutions in three technical replicates. Cycling conditions were: 94°C for 5 min, then 37 cycles of 94°C for 30 sec, 52°C for 30 sec and 72°C for 30 sec (plate reading at 78°C, 80°C and 82°C). Data were analyzed with the MJ Opticon Monitor 3.1 analysis software. Calculations were done with software REST-MCS. Additionally, relative expression of the Cyp4p2 in the resistant line was analyzed. Quantitative real time PCR for Cyp4p2 was performed using same protocols as for Cyp6g1 and Cyp6a2 expression analysis. Flies maintained for more than 25 generations on standard medium after deep sequencing and Cyp6g1 and Cyp6a2 expression analysis, were used for Cyp4p2 expression analysis. The forward and reverse primer sequences were as follows: Cyp4p2 – 5′ CTGAAAAGGCATCCTTACGC 3′ and 5′ TTGGGATCGATAACAGGCAG 3′. Quantitative real time PCR was performed on the Bio-Rad CFX analyzer with cycling conditions: 95°C for 2 min, then 35 cycles of 95°C for 15 sec, 55°C for 30 sec and 60°C for 30 sec (melt curve 60 to 95 C, increment 1.0 C). ## Mapping of the lethality We used the Bloomington Stock Center *Drosophila* deletion kit (111 lines) for the second chromosome to map the lethality locus (text S1). Individual crosses were set up between MiT\[*w⁻*\]3R2/CyO flies carrying a balancer of the second chromosome (SM6a/lethality) and the lines from the deletion kit (also carrying a balancer of the second chromosome). The progeny was scored for the presence of the balancer chromosome. ## P element mapping of the resistance locus To narrow down the resistance locus on the right arm of the second chromosome, four lines with P elements insertions were employed (Bloomington *Drosophila* Stock Center, IU; text S2). The resistant MiT\[*w⁻*\]3R2 flies do not carry any visible marker gene, while the flies carrying P element insertions have *w⁺* as phenotypic marker. Resistant flies were mass crossed with flies carrying the P element insertion. Virgin female progeny with red eyes (one chromosome deriving from the resistant line and the other from the P element line) were collected and crossed with iso31 males. For each experiment, 50 female flies, heterozygous for the resistance chromosome were crossed with 25 iso31 males, per replica. Each experiment had eight replicas with a total number of 250 females for each P element line. After 2–3 days, crossed flies were transferred to medium with 3 µg/ml of Imidacloprid. Progeny was scored for recombination events. At least 1000 emerged flies were analyzed per replica. Recombination rates were calculated as the ratios of the total number of recombinant flies over the total number of emerged flies. The distance between the P element insertions and resistance was calculated in centimorgans (cM), from which the physical distance was calculated using estimates of the local recombination rates at the sites of the P element insertions after Fiston-Lavier et al. and Singh et al.. This estimate was not possible for one of the P elements, which is too close (about 0.5 Mb) to the centromere. Here, the recombination rate for the interval between the P element and the average position of the resistance locus, as determined relative to the other three P elements, was calculated. ## In silico analysis A comparison of single nucleotide polymorphisms (SNP) of the deep sequence data between the resistant line MiT\[*w⁻*\]3R2 and the susceptible line iso31 was done for the right arm of the second chromosome. Genomic SNP analysis of the pooled assembly of the resistant and the susceptible strains reads were done with the Gigabayes SNP discovery algorithm (improved PolyBayes algorithm) and MOSAIC algorithm , using all Refseq mRNA transcripts of the dm3 assembly as a reference. A polymorphism probability threshold of 0.9 is used, with alleles requiring a minimal overall coverage of 10 and of 5 for the minor allele. A SNP density track with the number of SNPs in 1 Kb tiling windows was created. The SNP density was visualized with the UCSC Genome Browser on *D*. *melanogaster* release 5 (<http://genome.ucsc.edu/>) and is presented for chromosome 2R. The *in silico* search for overrepresented transcription factor binding sites was conducted using the JASPAR database. All up-regulated and down-regulated genes, as well as the subset of Cyp genes (up-regulated, down-regulated and all), were analyzed for the presence of transcription factor binding sites. The sequence of all genes was retrieved from Flybase (*Drosophila* release 5 sequence assembly). Regions from of 3Kb upstream to1Kb downstream of the gene start and the 3′UTR sequences were analyzed. A survey of predicted targets of microRNAs in the 3′UTR of all up-regulated and down-regulated genes, as well as the subsets of up-regulated and down-regulated Cyp genes, was performed with DIANA-microT (version 3.0). We also compared the sequences of Cyp genes differently expressed in the resistant versus the susceptible line for nucleotide differences. Comparison of the DNA sequences and translation to amino acids were done with the APE software (<http://biologylabs.utah.edu/jorgensen/wayned/ape/>). # Supporting Information We thank Dr Ioannis Ragoussis for advice and help with deep sequencing. We also thank Dr Evangelia Morou for her help on the biochemical assays. [^1]: Conceived and designed the experiments: PK SO CS. Performed the experiments: PK. Analyzed the data: PK SO MR NP AGH CS. Contributed reagents/materials/analysis tools: MR NP JV AGH CS. Wrote the paper: PK SO MR JV CS. [^2]: Current address: Department of Biochemical Engineering and Biotechnology, Indian Institute of Technology Delhi, New Delhi, India [^3]: The authors have read the journal's policy and have the following conflicts: Dr John Vontas 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.

# Introduction One of the primary challenges facing contemporary ecological and evolutionary research is predicting the potential effects of global climate change on populations across broad latitudinal ranges. Of recent concern has been the theoretical prediction that populations expanding into northern climates will experience a rapid loss of genetic diversity –, which may in turn restrict the ability of populations to adapt to new selection regimes and novel stressors. The current global warming event is comparable to past episodes of glacial retreat. Specifically, the rate of air temperature increase in the early Holocene and the concomitant deglaciation are thought to be similar to current trends on the 100-year scale. Given these similarities, a productive natural experiment for exploring the genetic consequences of rapid colonization after a global warming event is to survey the current genetic makeup of populations that have undergone a rapid expansion following the retreat of glacial ice sheets 10,000 to 12,000 years ago. The primary hypothesis for how genetic diversity is patterned across the postglacial landscape states that populations founded after a postglacial expansion should have less genetic diversity as a result of population bottlenecks and/or new selection pressures, resulting in reduced effective population sizes and thus less genetic diversity relative to their historical populations outside the glacial barrier. Many studies have evaluated this hypothesis by comparing genetic diversity along latitudinal gradients spanning the glacial boundaries for various species using neutral genetic markers, including allozymes, , mitochondrial markers, and microsatellite loci. Overall, these neutral marker studies have supported the hypothesis that a loss of genetic diversity is associated with the expansion of populations in a postglacial landscape. Although these studies provide a compelling picture of a population's genetic diversity following postglacial colonization, the true metric of a population's adaptive ability is not neutral genetic diversity per se, but rather heritable genetic variation. This is because heritable genetic variation mediates the rate and direction of population-scale responses to selection. The few studies that have quantified heritable variation within and among postglacial populations have suggested that heritability of ecologically relevant traits may increase, rather than decrease, following rapid population expansion. Hypotheses to explain such an increase in heritability following postglacial colonization include: increased effects of dominance in small colonizing populations , ; the admixing of colonists from different refugia populations ; as well as the potential for directional selection to break down epistasis , and expose heritable variation. To better understand the consequences of postglacial colonization on heritable genetic variation, we quantified levels of genetic variation in pigmentation traits from two populations of common garter snake, *Thamnophis sirtalis*, in South Dakota, United States and Manitoba, Canada. The Manitoba population is located in the Interlake near the town of Vogar. The South Dakota population is located at Lake Traverse on the South Dakota-Minnesota border, approximately 600 km due south of the Manitoba site, and represents the southern extremis of the Laurentide ice sheet, which retreated approximately 12,000 years before present. To assess potential clinal variation we analyzed phenotypic data alone from a smaller dataset from a population in Douglas County in eastern Kansas approximately 500 km due south of the South Dakota population. Pathways of garter snake postglacial migration in the region reconstructed from patterns of mitochondrial variation suggest that northern populations likely originated from populations directly to the south. We focused on pigmentation phenotypes for our study because garter snake color traits are scored easily and are thought to be of adaptive importance in natural populations. Additionally, molecular phylogenic analysis of *T. sirtalis* has found a lack of concordance between neutral molecular trees and those based on color pattern, further suggesting a role for selection in pigmentation evolution. Garter snake color patterns are important for predator avoidance, thermoregulation, and other important ecological functions. In this study, we examined two correlated pigmentation traits and a composite measure of the phenotypes. The first trait is the average area of dorsolateral blotches on individual snakes; the second trait is the pigmentation area of the dorsolateral blotch (the area covered by red pigment), while the composite trait is the ratio of pigmentation area of the blotch to the average area of the dorsolateral blotches. We chose *Thamnophis sirtalis* because it is one of only a few reptile species to colonize extreme northern latitudes ; it has been the subject of extensive ecological, physiological, and genetic research, –; and because garter snakes have proven to be tractable organisms in quantitative genetic studies,. Additionally, prior molecular studies (allozyme and mitochondrial) have documented the expected low genetic diversity in northern populations of *T. sirtalis*, e.g. that all Manitoba populations have only a single mitochondrial haplotype at the cytochrome *b* locus. The primary question we address within this multiple population framework is: Does heritability of adaptive pigmentation traits in the common garter snake decline at northern latitudes? The results of our study suggest that heritability of the three pigment traits is not significantly lower in the postglacial population. The data presented herein further shed light on the effect of rapid colonization on heritable variation in natural populations. # Results ## Repeatability Pearson product-moment correlation coefficients for between-observer scores for this sample were 0.926 (*P*\<0.0001) for blotch area and 0.910 (*P*\<0.0001) for pigment area, both of which are as high, or higher, than repeatability values considered acceptable in previous quantitative genetic studies of garter snakes. ## Trait means Adult phenotypic data was collected from both male (n = 139) and female (n = 44) snakes at the Manitoba site. At the South Dakota site, females were far more numerous than males on the date of collection and the sample was therefore skewed towards females (51 females, 5 males). The Kansas sample had approximately equal representation of sexes (24 females, 20 males). Blotch area and pigment area were normally distributed for all populations. Distribution of the ratio values tended to be skewed to the right, but were nonetheless amenable to testing given the procedures used, which are generally robust to minor deviations from normality. Two-way ANOVAs with main effects of sex and population revealed no differences between sexes for blotch area (*F_1,559* = 0.77, *P* = 0.38), pigment area (*F_1,561* = 0.18, *P* = 0.68) or ratio (*F_1,560* = 2.44, *P* = 0.12). Furthermore, no sex-by-population effect was detected for any trait (blotch area *F_1,559* = 0.61, *P* = 0.43; pigment area *F_1,561* = 1.79, *P* = 0.18; ratio *F_1,560* = 1.72, *P* = 0.19). Because we saw no difference in trait means between sexes, males and females were pooled in all subsequent analyses. All phenotypic color estimates are in units of blotch-to-scale area, i.e. the proportion of integument that was white or red relative to the immediately adjacent dermal scute. All trait means for adults were significantly higher in the South Dakota population relative to the Manitoba population (*P*\<0.001, Tukey). However, the Kansas population had higher means for both blotch area and pigment area than the South Dakota sample (*P*\<0.001, Tukey) but the composite ratio of the two pigmentation measures was significantly smaller in the Kansas population (*P*\<0.001, Tukey) relative to the South Dakota population and not significantly different from the Manitoba population (*P* = 0.97, Tukey). We generated a second set of phenotypic data from 144 offspring from 25 dams originating at the Manitoba site and 653 offspring from 50 dams from the South Dakota site. For the neonate snakes all pigmentation phenotypes were smaller relative to those measured on the adults. Again, all means in the South Dakota population were significantly higher relative to the Manitoba population (*P*\<0.001, Tukey). Kansas neonates had significantly higher blotch area (*P*\<0.001, Tukey) and pigment area than the Manitoba population (*P*\<0.02, Tukey) but the blotch area/pigment area ratios were not significantly different (*P* = 0.72, Tukey). Compared to the South Dakota population, Kansas neonates had significantly higher blotch area (*P*\<0.001, Tukey) but pigment area was not significantly different (*P* = 0.77, Tukey) nor was the blotch area/pigment area ratio (*P* = 0.21, Tukey). ## Genetic variances and heritabilities We obtained genetic variance and heritability data from two populations: 144 offspring from 25 dams originating at the Manitoba site, and 653 offspring from 50 dams from the South Dakota site. Because only small numbers of litters were available in the Kansas population, we were unable to generate usable estimates of the genetic variances and heritabilities for this population. Genetic variances and heritabilities were estimated using a full-sib ANOVA on the neonate offspring from each population. The genetic variance for blotch area was 409.62±131.64 s.e. for Manitoba and 360.07±80.96 s.e. for South Dakota; genetic variance for pigment area was 284.44±129.69 s.e. for Manitoba and 162.35±41.64 s.e. for South Dakota; genetic variance for blotch area/pigment area ratio was 0.0466±0.021 s.e. for Manitoba and 0.0067±0.002 s.e. for South Dakota. Heritabilities for blotch area were 0.72±0.14 s.e. for the Manitoba population and 0.46±0.085 s.e. for the South Dakota population, while heritabilities for pigment area were 0.69±0.26 s.e. for the Manitoba population and 0.41±0.087 s.e. for the South Dakota population. Heritabilities for the blotch area/pigment area ratio were 0.75±0.28 s.e. for the Manitoba population and 0.25±0.07 s.e. for the South Dakota population. A common issue in the comparison of genetic variance estimates among populations using a quantitative genetic design is achieving statistical precision that allows the genetic variance and heritability to be bounded away from zero, and also allows those parameters to be bounded away from the population to which it is being compared. Our design is robust in that we have multiple litters with hundreds of individuals from each population, thus for most of the within- population genetic estimates the 95% confidence intervals are easily bounded away from zero. However, we are unable to bound the estimates from each population away from one another based on their 95% confidence intervals. Although within-population significance testing of quantitative genetic parameters is feasible in wild populations, comparing the same parameters among populations is difficult given the extremely large numbers of litters required. The fact that our estimates overlap could therefore be a consequence of a lack of statistical power, or could reflect areal-world equivalence between the populations. That said, given the robustness of the within-population estimates, our data still strongly contradicts a reduction in genetic variance in the recently colonized Manitoba population relative to the South Dakota population. # Discussion In the present study we evaluate levels of heritable variation in pigmentation between two geographically distinct populations of *T. sirtalis* on either side of the boundary of the Laurentide ice sheet, which retreated approximately 12,000 years ago. We also examine a third population in less depth to gain insight into potential latitudinal trends in trait divergence. We explore the hypothesis that postglacial populations should exhibit reduced levels of genetic variation as a result of population bottlenecks and/or novel selection pressures on these populations relative to the historical populations outside the glacial boundary. We found two interesting patterns from this among-population study. First, there is no evidence of a decline in heritable variation for the pigmentation traits in question in the northern population. Rather, we found heritabilities to be equal in the northern population relative to the southern population. Secondly, we detect a significant geographic pattern in size and coloration of the dorsolateral blotches of both neonate and adult snakes moving from southern to northern populations. Phenotypes that are adaptive and contribute to responses to environmental stresses that vary with latitude often display clinal patterns in nature. We found that dorsolateral pigmentation traits (i.e., blotch area and pigment area) in *T. sirtalis* also display patterns of among-population variation consistent with a phenotypic cline. That is, southern populations express significantly larger mean dorsolateral blotch area compared to populations to the north (i.e., Kansas blotch area\>South Dakota blotch area\>Manitoba blotch area;). Furthermore, this pattern is robust regardless of age (i.e., neonate or adult), which is consistent with the single population ontogenetic effects demonstrated by. Although our data do not allow us to directly identify the mechanism driving this latitudinal pattern, a likely candidate variable is the difference in the thermal regimes experienced by southern to northern populations. Dorsolateral blotches are thought to function in both thermoregulation and predator startle response behavior. Reduction in blotch area results in a concomitant increase in the area of the integument covered by melanophores, and would be expected to increase thermoregulatory efficiency, an important adaptive trait in an ectotherm living in the extreme north. Because reduction in blotch area might be maladaptive with regards to predation, and because high expression of red pigment might be maladaptive with respect to thermoregulation, an evolutionary tradeoff may be playing a role in the maintenance of standing variation in pigmentation. The extent to which dorsolateral blotching mediates an evolutionary tradeoff between predator avoidance and thermoregulation deserves further study. In addition to the pattern observed in the mean phenotypes among populations we also evaluated how heritable genetic variation is patterned amongst the South Dakota population on the southern extremis of the Laurentide ice sheet, and the Manitoba population. We found heritability to be equal in the northern population relative to the southern population. There are multiple hypotheses to explain this pattern of increased variation in postglacial populations, including population admixture or a modification of the genetic architectures underlying the traits after colonization. In the population admixture hypothesis, genetic variation is increased as a result of genetic variation being introduced from colonists from multiple separate glacial refugia. The populations in question are located in a broad zone of contact between wholly red-pigmented subspecies in the west and wholly white-pigmented in the east. In Manitoba, the zone of contact roughly coincides with the boundaries of glacial Lake Agassiz, which blanketed the northern Midwest during the glacial retreat from 12,000 to 8,000 years before present. Even as the Laurentide ice sheet melted away, Lake Agassiz would have nonetheless precluded colonization of terrestrial sites. It is possible that garter snakes colonized ever northward along the east and west shores of the lake from their respective sides of the contact zone as the ice sheet retreated. Lake Agassiz drained over a very short time period about 8,000 years before present, at which time snakes might have rapidly colonized the new landscape from both the east and west, resulting in a patchwork of new populations representing different frequencies of colonists. We know of two sources of evidence that argue against the population admixture hypothesis. Rye explored the hypothesis that Manitoba populations were hybrid products of eastern and western clades. Her mitochondrial data suggest that Manitoba snakes are allied with western groups and show no affinity with eastern clades. Moreover, she delineated a mitochondrial contact zone 800 kilometers to the east of the Vogar den, suggesting that admixture with eastern populations was unlikely. Additionally, Westphal and Morgan found a developmental basis for the white/red color variation found at the Manitoba site. Manitoba *T. sirtalis* were found to express significantly more white blotch area (relative to pigmented blotch area) at birth than adults from the same population, and furthermore were found to increase in red pigmentation during their early development. The South Dakota and Kansas neonates were also found to have reduced pigmentation relative to the adult samples from their respective populations. Although suggestive, neither the Rye or Westphal and Morgan results absolutely refute admixture as a root cause of the increase in heritability. Until further molecular work clarifies the historical dynamics of the contact zone between western and eastern clades of *T. sirtalis*, we cannot completely rule out the admixture hypothesis. An alternative explanation for increased genetic variation in postglacial populations is the rearrangement of genetic architectures in postglacial populations. Under this scenario, populations that show low heritabilities as a result of stabilizing selection in the source population become subject to drift and founder effects in the colonial population, which in turn reduces epistasis, increases dominance effects, and ultimately exposes variability that was masked in the source population. The founder effects combined with novel selection regimes can also increase genetic variability in the short term by favoring new allelic combinations, and thus inflating heritable variation. Where the strength of selection itself is reduced, heritabilities can be higher than under more stringent selection regimes. For example, documented correlations between quality of environment and heritability suggest that heritabilities are lower in poor environments, e.g., where selection is stronger. We have therefore at least two competing selection-based hypotheses for the increased heritability of the traits in question. 1. Reduced selection. Given the paucity of competing reptile species, it may be that *T. sirtalis* in Manitoba has experienced a more favorable environment since colonization, with a concomitant decrease in natural selection and increase in heritability. 2. New selection pressures. New selection regimes in a novel environment have exposed hidden variation or favored new allelic combinations. Although genetic diversity is expected to decrease with distance from the founding population during a stepping-stone colonization event, other factors can rescue genetic diversity. For example, high gene flow between neighboring populations can maintain allelic diversity. In addition, high environmental heterogeneity can reduce genetic diversity in a colonial deme. Because the Manitoba landscape is relatively non-heterogeneous (i.e. has virtually no elevational variation and essentially uniform habitat of aspen parkland), the lack of environmental heterogeneity may have contributed to the high heritability we detected. Finally, new mutations or rare alleles can go to high frequency through a founder event called “surfing,” which occurs at the leading edge of a wave of migration. Due to the rapidity with which the postglacial expansion likely occurred, surfing is a valid hypothesis for the higher heritability of the Manitoba population. Regardless of the specific mechanism, a complex model of colonization with or without selection is likely necessary to explain the results of the present study. Although at present we are unable to convincingly support or refute any explanation for an increase in genetic variance of three color traits in the Manitoba population, our results suggest that, although genetic diversity at some loci may be reduced in postglacial populations of *Thamnophis sirtalis*, heritable variation underlying ecologically relevant phenotypic traits may actually be higher than in the founding populations. Our results are consistent with the few previous studies that have examined heritable variation in postglacial populations. An important focus of future research will be to obtain robust genetic data from both populations to better assess neutral genetic variation in both regions. Analysis of microsatellite loci for *T. sirtalis* has been performed in other regions and is a practical next step. Studies of Fst/Qst can address the role of selection over the postglacial landscape and will be forthcoming from the present authors. Finally, forthcoming fine-scale phylogenetic studies are expected to better resolve the issue of historical admixture in the Manitoba population. Because heritable variation can buffer populations against extirpation by allowing them increased capability to respond to selection events, it is important to understand long-term effects on heritable variation stemming from rapid range expansion if we are to make informed decisions on preserving biodiversity in the face of global climate change. # Materials and Methods ## Ethics statement All work was conducted in strict accordance with the US Public Health Service (PHS) *Policy on the Humane Care and Use of Laboratory Animals*, the US Department of Agriculture's (USDA) Animal Welfare Act & Regulations (9CFR Chapter 1, 2.31), and the United States Government Principles for the Utilization and Care of Vertebrate Animals Used in Research, Teaching and Testing. The work was approved by the respective Institutional Animal Care and Use Committees (IACUCs) at Oregon State University (approval \# 3175); Kansas State University (approval \#2676); and Black Hills State University (approval \#A-008-004). ## Field collection 25 gravid female snakes were collected from a single hibernaculum near Vogar, Manitoba (Latitude: N 50° 56.934, Longitude: W 98° 34.572). A sample of 173 adult snakes (139 male, 44 female) was also collected at the Manitoba site for phenotypic scoring. 50 gravid females were collected from a single hibernaculum on the shore of Lake Traverse near Sisseton, South Dakota (Latitude: N 45° 40.528, Longitude: W 96° 43.747). Two additional gravid females were collected from a single location in Douglas County, Kansas (Latitude: N 38° 56.667, Longitude: W 95° 4.829) (see map), as well as a sample of 44 adult snakes (20 males, 24 females) for phenotypic scoring. Gravid snakes were placed in breathable cloth sacks and transported to husbandry facilities. The Vogar, Manitoba sample was collected in May 2003; the Sisseton, South Dakota and Douglas, Kansas samples were collected in June 2008. We acknowledge that the temporal gap in the sampling of each population is not ideal. However, multiple years of empirical observations by MFW suggest temporal stability of the phenotypic means and variances at the Vogar, Manitoba site. Collection times at all three sites coincided with the advent of breeding season prior to dispersal to foraging habitat. Moreover, seasonal effects on these traits are likely limited, because unlike other squamates, snakes do not exhibit seasonal changes in coloration. Due to the large numbers of snakes available at the Manitoba sites, specimens there were collected by grab sampling of large groups of snakes that were either basking or in mating aggregations. The South Dakota snakes were collected as they emerged from the den. The Kansas snakes were collected under cover boards within 100 m of each other during a one-day visit. ## Rearing Female snakes were housed in dedicated facilities at Oregon State University (Manitoba sample) and Black Hills State University (South Dakota sample) and were reared under identical environmental and feeding regimes. The two Kansas litters were obtained from wild-caught dams, which were reared to parturition by a private breeder under the close supervision of MFW. Prior research on the color traits analyzed below demonstrated that individuals born and reared in a common garden laboratory environment express coloration values consistent with samples from the populations of origin. This concordance strongly indicates that the traits in question are stable with regard to the modest environmental variation that might be present among rearing facilities. Therefore, given the similar environmental and feeding regimes at the three facilities, confounding environmental effects were not expected to arise from minor variations in rearing conditions. Nonetheless, it is widely acknowledged that heritability and genetic variance are environmentally dependent. Thus we cannot definitively rule out environmental effects of breeding facility on our population comparisons, but the magnitude of such effects are likely small enough to permit such comparisons. Gravid females were retained until their litters were born. Within three days of parturition, neonates were weighed, measured snout-to-vent, and scored for color traits associated with the distinctive white and red blotches expressed in the dorsolateral region as in. ## Trait scoring We used a standardized scoring system, described in. The system quantifies both the average size of dorsolateral blotches and the extent to which red pigment is present in the otherwise white blotch. The final dataset is composed of two variables; blotch area, which measures the total area of blotch (irrespective of pigment type) while the pigment area measures the total area of blotch that was pigmented red. All phenotypic color estimates are in units of blotch-to-scale area, i.e. the proportion of integument that was white or red relative to the immediately adjacent dermal scute. Following we calculated a composite trait, which was the ratio of pigmentation area to blotch area. We included this trait in the analysis because the two component traits are genetically correlated and the composite trait captures the phenotypic reality of the relative extent to which red pigmentation is expressed across the dorsolateral blotches. That is, for this composite trait individuals with a score of 1 have blotches that are completely red, while individuals with a score of 0 have blotches that are completely white. Color measurements, as well as snout-vent length, mass, and sex, were taken on individual neonates from each litter from all populations as well as representative wild caught adult snakes from each population. The timing of scoring in the present study corresponds to the first scoring period in. Therefore the genetic variance of the traits was not expected to change appreciably within or among population samples across the 3-day scoring window. ## Repeatability Two personnel were involved in scoring the pigmentation traits; therefore we conducted between-observer repeatability tests using 50 adult snakes from the South Dakota population. 50 adult snakes from the South Dakota population were scored independently by MFW and JM. Scorings were separated by a period of several months, but both measurements occurred during the stable portion of the pigmentation ontogeny that occurs when snakes are adults. Pearson correlation coefficients were calculated in SAS v 9.2 using the PROC CORR procedure to assess the repeatability of scoring between observers. ## Analysis Trait means and standard errors for each population were obtained using PROC MEANS in SAS v 9.2. Trait means among populations were compared by a one-way ANOVA using PROC GLM in SAS v 9.2, followed by multiple comparisons. For the quantitative genetic analysis, we analyzed litter data from only the populations with a large number of litters, that is the Manitoba (25 litters and 144 offspring) and South Dakota populations (50 litters and 653 offspring). Population-specific genetic estimates (i.e. genetic variance and heritabilities) and their associated standard errors were calculated using h2boot. Because neonates express pigment differently than adults we used an age-specific full- sib ANOVA to estimate the genetic parameters, rather than parent-offspring regression. Estimates of genetic variance and heritability were therefore estimated for only neonates and not adults. We used 10,000 bootstrap replicates to estimate standard errors of the genetic parameters. The full-sib model as implemented in h2boot can be found in. Full-sib analysis is subject to overestimations of heritability due to the potential effects of common environment. As discussed above, a previous common garden study of the same traits from radically disjunct populations (Manitoba, CA and California, USA) found them to exhibit strong environmental stability, suggesting maternal and other environmental effects were relatively minor. Nonetheless we again cannot rule out the possibility that environmental factors contributed to our genetic estimates, but consider such effects unlikely to be so profound as to negate our results. # Supporting Information We would like to thank S. J. Arnold for making his snake rearing facility freely available. R. T. Mason and R. Shine gave invaluable advice. We also thank J. B. Westphal, B. Blake and Dan Krull for field collection and snake rearing assistance. [^1]: Conceived and designed the experiments: MFW JLM BES TJM. Performed the experiments: MFW JLM JMB BES. Analyzed the data: MFW JLM BES TJM. Contributed reagents/materials/analysis tools: MFW JLM BES TJM. Wrote the paper: MFW TJM. [^2]: The authors have declared that no competing interests exist.

# Introduction For decades, several types of genetic markers have been used to define populations' boundaries in a multitude of species. An alternative approach has only been recently developed due to better understanding of past climatological conditions (i.e. the potential concordance between former climatological models and present-day genetic structure). Indeed, contemporary genetic patterns are known to be the result of both present-day and historical factors. One of the most fundamental premises is that glaciation and inter-glaciation periods of the Pleistocene have shaped the genetic structure of many contemporary species. Present-day genetic structure of populations of a species inhabiting past refugium show a higher level of genetic diversity than those inhabiting formerly glaciated regions, due to range expansion and genetic drift following deglaciation. In the last few decades, examples of concordance between climatological history and present-day genetic structure have been reliably noted within taxa inhabiting freshwater or land habitats, depending on the availability of climatological information and models. Accurate and precise models of environmental conditions and ice-sheet covers have only recently become available for the North Atlantic, and as a consequence the effect of glaciation on the genetic structure of marine species has only recently been investigated - especially so in the North Atlantic Ocean,. This is also true for the climatological history around Iceland, which has been only recently resolved,. Although phylogeography studies related to anadromous fish such as the Atlantic salmon, sometimes consider the effect of the Pleistocene on contemporary genetic patterns, so far no correlation has been made between the divergence time and climatological history of this species. The Atlantic salmon is a philopatric species exhibiting a complex biological cycle that includes spawning in rivers, a freshwater juvenile phase followed by subsequent oceanic feeding migrations, and a high fidelity to natal rivers. This complex life-cycle is often considered the source of reproductive isolation, which facilitates the evolution and persistence of locally adapted populations. In the last 20 years, genetic studies regarding the Atlantic salmon have flourished, in that they have confirmed that populations are highly structured both between and within rivers systems, and that the associated genetic pattern was usually temporally stable. In Icelandic rivers, the only genetic study performed using allozyme data suggested restricted gene flow among rivers as well as within large river systems. Although the complex biological cycle of the Atlantic salmon has been determined to be the primary source of potential genetic signal, another major factor that could have been the origin of the present-day genetic pattern in the Atlantic salmon is the geological and climatological history of the North Atlantic Ocean. Past geological and climatological history has, indeed, been shown to be at the origin of genetic structure and patterns of several land or marine organisms,. Deglaciation following the Last glacial maximum (LGM) may, therefore, illustrate the process of river colonization, which would in turn partly explain the genetic structure of later generations of Atlantic salmon in Iceland. LGM is usually reported between 21–17 cal. kyr BP (calibrated years before the present),. The objective of the present study was, therefore, to investigate the genetic structure of Atlantic salmon collected from 26 Icelandic rivers using 15 neutral markers, and to assess whether the genetic pattern observed could be attributed to the climatological history of Iceland. # Materials and Methods ## Sampling and genotyping A total of 2,352 salmon parr were sampled by electro fishing from 400–500 m stretches at 26 rivers in 2002 (Ellidaa, (25)) and 2004 (all others). Each sample consisted of 54–94 parr, comprising 3 or 4 year classes ranging from 1 to 5 years old. All samples were collected by the Institute of Freshwater Fisheries with full authority from the Directorate of Fisheries in accordance with Icelandic law on salmon and trout fishing number 61/2006, article 26, subparagraph two. All fish were put to death in accordance with Icelandic law on “*Animal Welfare number 15/1994, article 14*”. Samples were taken from across the species range in Iceland – the lack of samples obtained from the southern and eastern coast is due to the fact that the rivers in this region are mostly turbid, glacial-fed rivers, which are unsuitable salmon habitats. Genotype information was retrieved from 15 microsatellite loci, and both PCR condition and genotyping procedures were performed as described in Olafsson *et al.*. ## Statistical analyses Samples from early life stages contain only the genetic material of successful breeders -often of a single year- and may be biased towards particular families. Full-sibling and half-sibling groups were identified within each sample using COLONY 1.2. After identifying any full sibling groups nestled within half- sibling families, we used a trial and error based method with HWE as a benchmark with which to assess the appropriate stringency level required to eliminate the sampling of siblings' effect. After repeated attempts of various combinations of tolerance level, we found that, by allowing only two full siblings and six half- siblings in each family, we were able to reduce the sib-ship effects sufficiently with reference to HWE. After the reduction of full-and half sibling groups in our samples, genetic variability was assessed as number of alleles, and the observed (*H*_O) and expected (*H*_E) heterozygosities and *F*-statistics were calculated in GENETIX 4.05.2 for each sample. The samples were also tested for conformation to Hardy-Weinberg expectations (HWE) with GENEPOP. Genetic diversity was further quantified using FSTAT 2.9.3 to calculate Nei's unbiased diversity (*H*_S, average expected heterozygosity) of individual samples. The significance of pairwise *F_ST* (for all pairs of populations) was calculated in ARLEQUIN 3.5.1.3 using 10,000 permutations, and a principal coordinates analysis (PCA) was performed in GENALEX 6.1 based on *F_ST* values. To detect loci under directional selection we used the program LOSITAN to generate 100,000 simulated loci, providing an expected neutral distribution of *F_ST* values and an estimate of p-value for each locus. In order to infer genetic ancestry and to identify subgroups that have different genotype patterns, multi-locus genotypes were analysed by a model-based clustering algorithm with STRUCTURE 2.3. Three runs were performed for each K value, ranging from 1 to 26. We chose 500,000 iterations of the Gibbs sampler after a burn-in of 50,000 iterations, applying a model that allows for admixture whilst implementing a correlated allele frequency. To determine the most likely hierarchical level of genetic structure, we plotted values of LnP(D) (the log probability of data) for each K and estimated ΔK statistics, which is based on the rate of change in LnP(D) between successive K values. We also analysed the data using the admixture model with the LOCPRIOR setting, which considers location information. STRUCTURE was also used to calculate the *F* value for the implemented *F*-model for correlated allele frequencies which can be used in a historical inference. The model for correlated allele frequencies developed by Falush *et. al* assumes that, “*K populations represented in our sample have each undergone independent drift away from ancestral allele frequencies at rates parameterized by F1, F2, F3…Fk respectively*”. These F1 to Fk values can be used to make evolutionary inferences (see, page 1570). Evanno *et al.* suggests that STRUCTURE tends to only capture the major structure in the data, although a subsequent STRUCTURE analysis performed on each identified cluster can potentially demonstrate a more intricate population structure within these clusters. This analysis was performed using a modified “hierarchical STRUCTURE analysis” as outlined by Vähä *et al.* and Warnock *et al.*. Calculations were made applying the same parameters for the Gibbs sampler and the burn-in iteration, but with K ranging from one to ten; ΔK statistics were consequently calculated to decide the hierarchical structure at each level. We therefore conducted a hierarchical analysis by performing similar Structure runs on each detected group (K) containing several samples. The estimated number of group (K) was based on ΔK statistics and changes in the pattern of LnP(D) values. A hierarchical analysis of molecular variance (AMOVA) was performed using Arlequin to quantify the degree of differentiation between the post-hoc defined regions according to the Bayesian approach in STRUCTURE, i.e. for K = 2. A consensus neighbour joining tree of pairwise D_A between all samples was calculated and constructed using the software POPULATIONS 1.2.28. Genetic diversity indices, such as *H*s and Allelic richness, were then compared between groups of the uppermost hierarchical structure based on ΔK from STRUCTURE using FSTAT. To assess the potential impact of mutation *vs.* drift on the genetic pattern detected, we performed the allele-size randomization test implemented in SPAGeDi 1.3. The objective of this analysis was to test whether *F*_ST = *R*_ST, where *R*_ST is a stepwise mutation model (SMM)-based measure of genetic differentiation informed by microsatellite allele size variance. This is analogous to *F*_ST, and unbiased with respect to any differences in sample size between populations or any differences in variance between loci. When mutations contribute little to genetic differentiation, estimates of differentiation using *F*_ST and *R*_ST should show similar results. On the other hand, if stepwise- like mutations have contributed significantly to divergence, *R*_ST should demonstrate a larger differentiation than that of *F*_ST. Potential historical signatures in the genetic data were therefore assessed by permutating allele sizes at each microsatellite locus among allelic states (max 20,000 replicates) to simulate distribution of R_ST values (ρR_ST) with 95% confidence intervals (CI). To further clarify evolutionary history, the time of divergence was approximated using two different ABC approaches. First, the time of divergence was approximated using the Approximate Bayesian Computation (ABC) implemented in DIYABC v1.0.4.46. ABC is a Bayesian-based approach where the posterior distributions of the parameters in the model are predicted by replacing the calculations of the likelihood (probability of observed data given the values of the model parameters) by a measure of similarity between observed and simulated data. The ABC method can be described as three sequential steps - the first step (*simulation*) simulates many multilocus data sets with characteristics similar to the observed data set; the second step (*rejection*) compares the simulated data set to the observed data set and retains simulations that are arbitrarily close to the observations whilst rejecting others; the final step (*estimation*) then estimates posterior distributions of parameters through locally weighted linear regression on the summary statistics associated with the retained simulations. The tested scenario was that two populations of size N1 and N2 have diverged *t* generations in the past from an ancestral population of size N1+N2, assuming a stepwise mutation model with no indel mutations and a mutation rate μ = 10⁻⁴. In the first step, 1,000,000 datasets were simulated, and 1% of the simulated data sets that most closely fit the observed data were used to inform the posterior distribution of parameter values. The uniform prior distributions used in the DIYABC analyses ranged from 0–20,000 for both N1 and N2 and 0–50,000 for *t*. Secondly, the time of divergence was approximated using the popABC software where we evaluate a model of an ancestral population of size N_A that splits at time T in the past to give two populations of size N₁ and N₂. Since the time of splitting, immigration occurs at rates m₁ and m₂, where immigrants into one population are drawn from the other population. The mean of mean mutation rate among loci was assumed to 5×10⁻⁴, normally distributed with a standard deviation of 5×10⁻⁴. A mean mutation rate of 5×10⁻⁴ is commonly assumed in demographic models, but is a little lower than pedigree-based estimates for autosomal microsatellites in humans. Five million data points from the joint distribution of parameters and summary statistics were simulated, and the 50,000 points closest to each target set of summary statistics were used for regression-adjustment. # Results ## Genetic variability and siblings A total of 324 individuals (i.e. 13.78% of the total database) were discarded due to sibling effects within the sampling, ranging from 40 in Litla Laxa (12) to none in Laxa Dolum (10). A total of 205 alleles were observed across the 15 loci, ranging from two alleles in Ssa14 and SsaD486 to 32 in SsaD144. The locus SsaD486 is usually considered a continental analysis marker, and one that is therefore poorly suited to small-scale genetic analysis, and not polymorphic within eastern populations of Atlantic salmon. However, it has been suggested to be particularly useful in discriminating between brown trout (*Salmo trutta*), salmon, and hybrids thereof. Hybridization between the two taxa has been widely documented. Based on this locus, two individuals were discarded from our samples as hybrids of salmon and trout, showing the exact allele sizes for SsaD486 as described in Perrier *et al.*. This locus was monomorphic in 18 of the 26 populations, and the frequency of the second allele was 0.0029 overall, and thus we chose to discard it from subsequent analyses. Only one sample deviated from HWE (i.e. the river Hvita Efst (6)); this was mainly due to locus Ssa2216 being out of HWE in that sample. The overall genetic variability of samples, as measured by gene diversity (*H*_S) and allelic richness (*A*_R), ranged from 0.630–0.730 and 6.78–9.23, respectively. ## Genetic differentiation An overall *F*_ST value of 0.057 was observed on a sample level. The degree of genetic differentiation between sampling locations, as estimated by *F*_ST, ranged between 0.006 and 0.140. Highly significant genetic differentiations (Fisher's exact test P\<0.01) were observed between all samples except: Olfusa (20), Sog (16), Blanda (1), and Svarta (18). A PCA based on the *F_ST* is presented in. No signs of directional or balancing selection were identified at the 14 loci among the two populations. The Bayesian cluster analysis and subsequent calculations of ΔK showed that the most likely number of genetically distinguishable groups (*K*) was two, both with (Ln *P*(*D*)±S.D = −95402±462) and without location information (Ln *P*(*D*)±S.D = −95725±922). The first group (Group 1) was composed of samples from the northwest, north, and east rivers: 1; 3–5; 8–11; 13–15; 17–19; 23; 25–26 ( green). The second (Group 2) was composed of samples from the southwest and south of Iceland: 2; 6–7; 12; 16; 20–22; 24 (yellow). The average F-values as outputted by STRUCTURE for the K = 2 runs were 0.032 and 0.060 for the two groups, Group 1 and Group 2 respectively. The population structure revealed by the hierarchical clustering method is outlined in. Initial partitioning of the dataset at the 1^st level using *K* = 2 separated the data into the two previously described groups, Group 1 and Group 2. Further levels of analysis within these and subsequent hierarchical groupings resulted in geographically comprehensive groupings at each level, leading to a total of eight major regional groupings at the 2^nd and 3^rd level of analysis. Branches on the NJ-plot tree of the genetic distance D_A corresponded to the colours of the eight groups identified with hierarchical structure analysis. The clustering assignment results for *K* = (2, 3, and 4), with and without location information, are presented in. The AMOVA between groups detected by the Bayesian clustering method corresponded to the historical origin in the data. The observed level of genetic differentiation (*F*_ST) was 0.072 between the two groups. In addition, when the genetic diversity of Group 1 and Group 2 were compared, no significant trend was observed, although the gene diversity (Hs) and allelic richness (Ar) tended to be slightly higher in Group 1 than in Group 2, (Ar = 8.456 and 8.084, Hs = 0.699 and 0.685, respectively). ## Historical origin The random permutation of different allele sizes among allelic states at each locus revealed that estimates of *R*_ST were significantly larger than the 95% CI range of the ρ*R*_ST values at four microsatellite loci (SsaD144, Ssa171, SSsp2201 and SsaF3), suggesting a mutational component to genetic differentiation. *R*_ST was also larger than the ρ*R*_ST values at SSsp2210, but not significantly. When considering the two groups detected with the Bayesian method, the overall estimate of *R*_ST was also significantly larger than the 95% CI range of the ρ*R*_ST values. In the case of the DIY ABC approach, the simulations were based on the scenario of two populations that were diverged *t* generations ago from a shared ancestral population. The two populations seem to have diverged from 24.9 kyr BP (seven years generation time) to 17.8 kyr BP (five years generation time). In the popABC approach the time of divergence was estimated as 17,454 (8,485–26,523). Other parameters estimated by the popABC are available in. Both of these methods suggest that the genetic structure may have originated in the late Pleistocene, and that the two gene pools could have segregated before the last glacial maximum (LGM) in Icelandic waters. # Discussion ## Sampling of siblings' effect An accurate result from population genetic analyses requires a random, independent sample from each population of interest. However, within a population the genetic patterns can differ among demographic groups, and attention to this factor is important in order to attain an accurate conclusion regarding population genetic structure. Parr are often sampled for population genetic research as they are more accessible and abundant than adult salmon, but parr samples may lead to false conclusions concerning population genetic structure due to the small numbers of successful breeders and biased breeding success. If differences in allele frequencies between adult salmon and parr samples are solely due to skewed sampling of full-siblings, then removing the majority of full-siblings from the sample database should in turn remove these differences. The program STRUCTURE has been shown to be sensitive to samples with related individuals and can in such cases detect genetic structure within a sample where there is none. Therefore it is important to take the sampling of sibling effect into account, and the stringency applied in this study was in accord with the suggestions of Anderson & Dunham. For our dataset, the effect of removing siblings is most apparent in the case of Litla Laxa (12), where the average *F_ST* compared to other samples is 0.027 lower. Removing siblings also caused the genetic distance between the two samples from the Blanda river system to drop from significant to non-significant. ## Genetic pattern, restricted gene flow, and discordance allozyme-microsatellite loci Understanding the distribution and connectivity of populations has been one of the major challenges of recent decades, and has been recognised as a necessity for biodiversity conservation. As such, indirect estimates of gene flow (i.e. the exchange of genes among populations due to the successful migrations of individuals) have received considerable attention in recent years. In marine organisms, populations that have been thought to be homogenous owing to a lack of obvious barriers have been shown to exhibit a clear genetic structure (i.e. restricted gene flow among populations on a small or large geographical scale). The Atlantic salmon exhibits a complex biological cycle involving stages that inhabit both marine and freshwater environments. Several genetic studies have demonstrated its propensity to form distinct and reproductively isolated populations,. In the present study, substantial genetic differentiation was found between all collected samples except Olfusa (20) – Sog (16), and Svarta (18) – Blanda (1). Olfusa and Sog are indeed located in the Olfus river system - the river Sog is a tributary to Olfusa, with 13 km between sampling sites. Svarta is a tributary of Blanda, with only six kilometres between these sampling sites. The geographical proximity of these areas might have precluded any genetic differentiation among the samples collected in these rivers. One noteworthy result from the present study was the high divergence between almost all samples collected, among all river systems, whereas a previous genetic study using allozyme loci and 22 sampling locations identical to the present research found that a total of 52 out of 231 possible pair-wise comparisons were non-significant. This is in concordance with the assumption that microsatellite markers yield a higher resolution power than allozymes when identifying populations, due to the presence of many more alleles. A study by King *et al.* found that Icelandic salmon populations seem to be less diverse than other salmon populations originating from Europe, but more diverse than salmon populations in the USA and Canada (the average gene diversity for the Atlantic salmon populations in Iceland in this study was 0.69, compared to 0.60 in the USA and Canada and 0.74 for the rest of Europe). Although there are only five out of twelve comparable microsatellite markers between the two studies, the same pattern was observed looking only at samples included in King *et al.*. The highest pairwise *F_ST* value between salmon populations in Iceland was 0.14 found between Kalfa (7) and Laxa Adaldal (9). Kalfa appeared to be a very unique sample and was highly differentiated from all other samples with a minimum *F_ST* value of 0.075. If we excluded this sample, the *F_ST* value range in this study was 0–0.116 and was very similar to what was observed within salmon populations in the UK. The hierarchical genetic clustering method revealed a population structure to a relatively fine spatial scale in our population complex. This pattern concurred very well with the D_A distance phylogenetic tree and was geographically relevant, suggesting the effect of restricted gene flow was due to isolation by distance. The consequent predicament was a question of the extent to which to pursue the hierarchical analysis. In this case, we stopped at the second and third level, as we concluded that this sufficiently described the eight geographically major stock units of Iceland. Restricted gene flow can, of course, continue to act within some of the eight major genetic groups, but further subdivision (e.g. the within river structure of the Olfus river system) can be determined using the pairwise *F_ST* comparison. It should be noted that although we reduced the likelihood of false discovery by removing sibling groups from our samples, caution should be applied when interpreting *F_ST* values, as they can be inflated by the effect of family structures between streams or the possibility of low intra-stream heterozygosity. In all, the present study corroborated previous genetic studies performed on Atlantic salmon by suggesting high levels of genetic differentiation between populations both within river systems, and on a larger geographical scale,. Although historical climatological conditions in Iceland could explain the observed genetic pattern (see below), the present results suggest that gene flow tends to be restricted among Icelandic populations of Atlantic salmon. ## When climatology meets genetics The glaciation and interglaciation periods of the Pleistocene have considerably shaped the distribution and connectivity of contemporary populations,. Indeed, one of the major concerns about the genetic structure detected with microsatellite loci is that contemporary levels of differentiation might reflect a historical restriction of gene flow, rather than the current isolation of populations. Emerging evidence has shown historical signals embedded in microsatellite loci, especially in Icelandic waters, in which populations of marine organisms have colonized the ice-free environments following the LGM, examples of which are composed of relatively young populations on an evolutionary time scale. Although the complex biological cycle of the Atlantic salmon may be the primary source of potential genetic signals, the past geological and climatological history of Iceland may in part have been the origin of the contemporary genetic structure of salmon populations. In Iceland, most of the shelf water was covered by an ice-cap during the LGM; although scientists do not fully agree on the exact dates that the coastline of Iceland was deglaciated, they generally do agree upon the order of deglaciation in different areas. It has been suggested that this event started circa 17.5–15.4 cal. kyr BP and that by 15.4–14.6 cal. kyr BP the glaciers subsequently retreated well inside the coastline. The order of events during the ice cap deglaciation of Iceland is demonstrated in, on the time scale from 25 ka BP to the Younger Dryas. Based on these particular studies performed in Iceland, the first viable and ice-free environment for migrating salmon was likely to be located at Breiðafjörður in the northwest of Iceland. Thereafter, the ice-cap progressively retreated from the north and east of Iceland, before the final retreat occurred in the southwest and south of Iceland. Although Iceland is small compared to other areas where a latitudinal gradient in colonization would be expected, the icecap retreat model in Iceland is unusual as it follows a north to south pattern dissimilar to that of most other deglaciating areas, and the residues of this pattern are still visible when examining the modern-day glacial distribution within Iceland. When considering neutral genetic markers the *F*-model assumes that a linkage equilibrium and Hardy-Weinberg equilibrium (HWE) exists within populations, such that genetic drift is the only force acting upon the population. With time, genetic drift will increase the population *F*-value, as was demonstrated by Group 2 having a higher *F*-value than Group 1. One possible interpretation of these results is that Group 2 was possibly colonized from Group 1, or alternatively represents a second colonization event from another “source” population. However, other alternative processes might explain these results such as Holocene bottleneck and a smaller effective population size in the southern region (Group 2). The genetic analyses, i.e. the existence of two main genetic clusters and the higher *F*-value in the south, do not provide any direct support for an initial colonisation in the north but are nevertheless consistent with the ice-cap retreat model in Iceland (and that the northern region was deglaciated first). When considering the deglaciation model of Iceland it is highly probable that pioneer salmon colonized the northern region before the southern region. The first available habitat for salmon in Iceland after the LGM is located in the Northwest region (Breiðafjörður) approximately 17.5 ka BP. From there suitable habitats were consecutively formed as the progression of deglaciation went north and east. The estimated time of divergence (DIYABC and popABC) demonstrated the existence and timing of subdivision between the northern and southern populations, and suggested that pioneer salmon were arriving in Iceland at a time when the only viable habitats were in the northern region (Breiðafjörður). Therefore, when we considered the available deglaciation model for Iceland, the presented genetic results seemed to be consistent with an initial settlement of Atlantic salmon in the northern region of Iceland. The proposed order of settlement offers the possibility of three settlement scenarios: First, that the colonizing salmon inhabit the northern region and subsequently colonize the south. Second, that the colonizing salmon inhabit the north of Iceland and a second colonization event from the same population of origin occurs in the southern area. Third, that the colonizing salmon inhabit the north of Iceland and a second colonization event from a different refugium occurred in the southern area. Although we cannot distinguish between the suggested three settlement scenarios, our results suggest that the glaciation event in Iceland has imprinted the present day genetic structure of Atlantic salmon. Although a three cluster scenario suggesting a colonization process consisting of multiple events cannot be completely ruled out, the current analysis better support the two cluster scenario. The additional within-river groups detected within each original group (Group 1 and Group 2) during our three levels of hierarchical structure analysis, and confirmed by the consensus neighbour joining tree of pairwise D_A which is expected to produce the best branching tree pattern, might reflect more recent historical colonisation of river systems and current drift events due to recent isolation of populations (well after the colonisation). It might therefore represent the step-by-step colonisation of the different river systems, and more recent isolation events. Although Icelandic salmon populations are highly reproductively isolated, there is a possibility of gene flow between the two population groups - the DIYABC method does not take this into account, but this limitation is met by adding an analysis using the popABC program which allows for the modelling of gene flow between groups. Furthermore, minor inter-group gene flow might affect the estimates by underestimating the divergence time, but neither of these properties affect the glacial period in question. ## When history explains the exceptions The rivers Ellidaa (25), Leirvogsa (11), and Langa (23) seemed to be misplaced in these ice-free environment re-colonisation events; however, a closer inspection of Icelandic volcanic history and of contemporary human activity offered explanations for these anomalies. The lava field, Leitárhraun, was formed in a volcanic eruption 5,200 years ago, and stretches from the Bláfjöll down to Elliðavatn lake, where it formed the rootless volcanic cones, Rauðhólar, and from there it ran down to what has become the present river channel of Ellidaa (25) (see insert on). No pioneer salmon in Ellidaa (25) could have survived this natural event, therefore we suggest that a re-colonization event occurred, possibly shifting the river from its presumed geographical origin. Unknown human activity that might have caused this seems very unlikely but cannot be completely ruled out. Regarding Leirvogsa (11), the river was probably inaccessible for salmon in its original state as there were cascades near the estuary. However, after the cascades were artificially made accessible to enhance rod fishery, the river became known for its sea-run brown trout. Today, Leirvogsa (11) is a river with a healthy salmon population since the river became stocked with parr and smolts originating from Ellidaa (25). In Langa, (23) there are two waterfalls at the estuary - Sjávarfoss and Skuggafoss - that were possibly impassable for salmon. Skuggafoss waterfall proved especially difficult for salmon to pass until a fishway was built in the 1950s. In the 1960's and 1970's the river was stocked with salmon parr and smolt, which originated from stocks of various origins. Four additional fishways were installed in waterfalls further upstream, thereby increasing the salmon habitat from 13 km to 26 km. This high level of human intervention could easily have altered the genetic signature of this river from its logical position in the gene pool of Icelandic salmon. ## Conclusion The present study has yielded two important findings. Initially, that microsatellite loci revealed a significant genetic signal amongst the populations collected at 26 rivers distributed across the country, a result which suggested a restricted gene flow common to the sampled populations. Secondly, that the time of divergence between the two primary genetic groups detected was highly consistent with the ice-cap retreat model in Iceland, and that further analyses provide overwhelming evidence of a historical component to the observed genetic differentiation. The individual analysis of population structure strongly supports the presence of two genetically distinct groups, with a general trend of higher genetic variability in the northern areas in relation to that of the southern regions, thus substantiating the expectation of the existence of higher genetic variability in the potential initial area of colonization. The rivers Ellidaa (25), Leirvogsa (11) and Langa (23), however, do not seem to fit this re-colonisation route hypothesis, although the volcanic history of Iceland and the history of human activity for the previous 60 years could be attributed responsibility for the displacement of these rivers in the presented salmon histories. In addition, the additional variation observed within each predominant group (Group 1 and Group 2) was likely due to more recent events of colonisation or population isolation, subsequent to the recolonisation event after the LGM. We therefore conclude that the observed genetic pattern at microsatellite loci for the Atlantic salmon in Icelandic rivers serves as additional evidence of the relative immaturity of Icelandic fish populations, on account of the re-colonisation of this relatively young environment following the Last Glacial Maximum. ## Data archiving Data available from the Dryad Digital Repository: <http://doi.org/10.5061/dryad.60kk5>. # Supporting Information We thank Anna K. Danielsdottir and Sarah Helyar at Matis Ltd. for their comments and discussion. We thank Pall Gunnar Palsson at Matis Ltd. for his assistance with the preparation of figures. We thank Gregory Dixon for his general English editing. [^1]: K. Olafsson, S. Hjorleifsdottir and G. O. Hreggvidsson are employed by Matis Ltd. Matis Ltd. is a non-profit, government owned Research Company. The authors are employed there as research scientists. The authors declare no conflict of interest. [^2]: Conceived and designed the experiments: KO CP SH SG GH. Performed the experiments: KO. Analyzed the data: KO CP SH SG GH. Contributed reagents/materials/analysis tools: KO CP SH SG GH. Wrote the paper: KO CP SH SG GH.

# Introduction Two-dimensional (2D) dosimeters have been studied for many years. Radiochromic films are used as absorbed dose dosimeters, and the most recent type, that is the Gafchromic EBT3 film (Ashland ISP Advanced Materials, NJ, USA), is widely applied clinically. The advantages of the EBT3 film are its high resolution and discoloration via a self-development process. For accurate dosimetry, the EBT3 film must be managed extremely carefully, including its in-batch variation, film uniformity, and scanner dependency. For accurate dose measurement, various background subtraction methods have been proposed. The EBT3 film is a single- measurement dosimeter; thus, multiple measurements require multiple films, and real-time dosimetry cannot be performed. To perform 2D dosimetry while maintaining the advantages of the ionization chamber as a reference dosimeter, 2D ionization chamber arrays, such as MatriXX (Ion Beam Application, Belgium) and 2D-ARRAY seven29 (Physikalisch-Technische Werkstätten, Freiburg, Germany), have been developed. The real-time 2D dose distribution can be obtained as an absolute dose through proper calibration; however, the spatial resolution is relatively coarse. In the case of MatriXX, the spatial resolution is 7.62 mm (center-to-center), and a resolution of 3.8 mm can be obtained by taking a second measurement after shifting the array position. Since a pioneering report by S. N. Boon *et al*. published in 1998, research on 2D dosimetry using scintillator sheets has been actively conducted with X-ray and charged-particle dosimetry. Scintillation detectors developed using inorganic 2D scintillators, organic 2D scintillators, and gas 2D scintillators have been extensively studied because such detectors have a high spatial resolution, reusability, and potential for time-resolved dosimetry because of their capability for real-time signal processing. Conventional scintillation detectors usually consist of a scintillator, mirror, and scintillation acquisition device, such as a charge-coupled device (CCD) and complementary metal-oxide-semiconductor (CMOS) image sensors. The mirror located in conventional scintillation detectors not only reflects the scintillation to the image sensor but also protects the camera from the primary beam. However, installing a mirror at 45° requires a strong and heavy holder to sustain the mirror; thus, the detector becomes heavy and bulky. The underside of the scintillator should be empty because the light has to reach the image sensor. However, this causes a dose error ranging from 0.5 to 2.5% due to the lack of backscatter in 6 MV photon beams. If a mirror is not installed in a scintillation detector system with an image sensor, several problems occur, including geometric distortions, brightness dependency, lack of backscatter, and penumbra distortion. In this study, to build a compact 2D dosimeter, we removed the mirror and developed a time-resolved mirrorless scintillation detector (TRMLSD). Various issues caused by the removal of the mirror were corrected using image processing techniques for the correction of the geometric brightness dependency and by using a convolutional neural network (CNN) for backscatter and penumbra correction. It was observed that the TRMLSD could provide a high-precision 2D dose distribution in a plane perpendicular to the beam axis at 24 frames per second (fps). # Materials and methods ## Design The conventional scintillation detector comprises a scintillator, a mirror, optical fibers, and a camera with an image sensor. If an optical fiber is used to collect scintillation, the collected scintillation is not influenced by the scintillation from the environment. However, a 2D array structure requires a large number of optical fibers with a relatively low spatial resolution because of the thickness and array spacing of each optical fiber. When a mirror is used to reflect scintillation, the scintillation detector becomes larger and heavier. To overcome these shortcomings, optical fibers and mirrors were removed. A camera with a CMOS image sensor was located outside the primary radiation field to protect it from damage due to radiation, and the camera was installed facing upwards at a certain angle. The detector and camera were held in place with a frame fabricated from Foamex plastic. The TRMLSD had a volume of 40.5 × 32.5 × 15.5 cm³, and a weight of 1.74 kg, which was much smaller and lighter than the conventional type of scintillation detector developed in-house (60.7 × 37.0 × 33.5 cm³ and 18.26 kg). All the inner surfaces of the TRMLSD were covered with a dark film to prevent the reflection of light so that the camera could detect only the scintillation generated on the inorganic scintillator plate. Nevertheless, scintillation detectors produce dosimetry errors owing to their structure. The scintillation generated from the scintillator installed on the TRMLSD must be recorded by the camera; therefore, the space under the scintillator is filled with air. This structural characteristic disrupts the electron equilibrium on the scintillator. Hu and Zhu *et al*. published a research paper related to the backscatter correction factor for megavoltage photon beams. In that study, the dose loss at a depth of 5.0 cm (source–axis distance, SAD: 100.0 cm) in a water phantom was approximately 2.25% for a 6 MV photon beam. Moreover, we calculated the effect of backscatter using a Monte Carlo simulation toolkit. The energy of the medical linear accelerator (linac, Novalis Tx; Varian Medical Systems, CA, USA) was 6 MV, the field size was 10.0 × 10.0 cm², the standard source-to-surface distance (SSD) was 100.0 cm, and the water depth was 5.0 cm. The dose loss was approximately 2.5% due to the lack of electron backscatter. ### Scintillator: PI-200 A flat-plate-type PI-200 inorganic 300 × 300 × 0.9 mm³ scintillator (Mitsubishi Chemical, Chiyoda, Japan) was selected because of its higher density and yield ratio than an organic scintillator. The main peak spectrum of the emitted scintillation was approximately 530 nm. The PI-200 scintillator layer was composed of three layers: protective, phosphor, and supporting layers. Two problems exist when measuring the 2D dose distribution using an inorganic scintillator: (1) The PI-200 is a high-Z-material (gadolinium oxide sulfide doped with terbium) scintillator. The dose distribution obtained using the PI-200 is different from that obtained using a scintillator composed of a low-Z material, such as normal tissue and water. (2) Because the homogeneity of the chemical components constituting the three layers is not uniform, the uniformity of scintillation needs to be corrected using a homogeneous light source such as a high-quality large liquid crystal display (LCD) diagnostic monitor and/or multiple exposures with a large standard beam. ### Scintillation measuring apparatus: GoPro HERO5 black To measure the scintillation, a GoPro HERO5 Black (GoPro, CA, USA) camera was used. It is lightweight (118 g) and transmits signals wirelessly, enabling remote on/off control of the signal transmission. The scintillation distribution was converted into an image using the CMOS image sensor in the camera. Therefore, determining the camera parameters that are suitable for accurate measurement of the absorbed dose is important. In particular, the camera parameters related to exposure must be precisely selected; thus, the ISO value was carefully determined through iterative experiments to achieve reproducibility in dosimetry. Conventionally, the ISO value represents the sensitivity of a photographic film to light. However, digital cameras also provide ISO settings. Depending on the ISO value, the sensitivity to light and the relative brightness vary. Therefore, based on the setting of the ISO value, the output factor of a linac measured by the TRMLSD varies. For a 5.0-cm-thick solid water phantom, the brightness differs by a factor of 1.99 between ISO values of 800 and 3200. An ISO value of 800 was finally selected because it provided accurate output factors for the radiation field size in the iterative experiments. The experimental setup for the TRMLSD and EBT3 film was as follows: 249 monitor units (MUs) were delivered with a 10.0 × 10.0 cm² field size to the solid water phantom at a distance of 100.0 cm from the source (SSD was 100.0 cm), and the scintillator or EBT3 film was positioned at a depth of 5.0 cm in the phantom. In, Setup-A represents the scintillator in the TRMLSD setup, and Setup-B denotes the EBT3 film setup. depicts the lateral profiles normalized to the maximum pixel value of each measurement, which changes with the ISO values for the TRMLSD. The profile presented in was calculated using a treatment planning system for Setup-B (TPS, Pinnacle^TM; Philips Medical Systems, CA, USA). The HERO5 Black provides the maximum ISO value of 6400; however, with this value, the dose in the outfield is greater than that of the EBT3 film measurements; therefore, ISO values of 800 and 1600 were tested. Other camera parameters, such as color, white balance, shutter speed, and exposure value, should be kept constant to ensure the reproducibility of the dose measurement. The image resolution in the video-recording mode was determined to be 4K resolution (3840 × 2160 pixels). ## Dosimetry from video Scintillations were recorded in video mode at 24 fps with 4K resolution using the CMOS camera; then, the recorded data were divided into frames. In the case of the step-and-shoot IMRT plan, the beam was only turned on when the leaves of the multileaf collimator were stationary in each of the prescribed segment/subfield positions. Therefore, a time interval existed between the segments. To acquire the 2D dose distribution for each segment and/or field, all frames were reviewed, and the image frames with scintillation signals were grouped into intervals. During preprocessing, the recorded color images were converted into grayscale images. The major wavelength of the scintillator was approximately 530 nm; thus, the recorded image appeared green. The recorded pixel intensity ratios were approximately 10.5%, 51.5%, and 38.0% for the red, green, and blue channels, respectively. The weights of each color channel for the mix were fixed at a ratio of 1:1:1 to conserve the intensity of each color channel. To test the dosimetry capability, a 100 MU of a 6 MV X-ray beam with a 10.0 × 10.0 cm² field was irradiated to a 5.0-cm-thick solid water phantom on top of the TRMLSD (Setup-A). Of the total video-recording period of 75.33 s, the scintillation image was recorded for 14.88 s. shows the scintillation image taken at 50.41 s. ## Frame-based noise-reduction technique Collimator- or phantom-scattered photons interact with the CMOS image sensor during measurement and generate hot pixels in the image, which are instantaneous, high-intensity random noise. The hot pixels were removed using instantaneous characteristics. The frame-based noise-reduction (FBNR) technique monitors all frames over time. It can detect and remove hot pixels by reading each pixel’s value from consecutive frames. If the value of pixel (*x*, *y*) in a certain frame is continuously present in several subsequent frames, the pixel is treated as a scintillation; otherwise, the pixel is treated as a hot pixel. The FBNR technique removes the hot pixels while maintaining the spatial resolution of the image without blurring, thus compensating for the drawbacks of noise-reduction techniques that use conventional filters, such as the median filter or moving average filter. The FBNR technique applied to the image is illustrated in. $$If\ {\sum_{vfn = n}^{n + k}\ {D\left( {vfn,x,y} \right) \geq Threshold\ value,D\left( {vfn,x,y} \right) ≔ 0},}$$ where *D* is the dose distribution measured by TRMLSD, *vfn* is the video frame number of interest for removing hot pixels, *(x*, *y)* are the pixel coordinates in the image, *n* is the scan starting point, and *k* is the number of frames for scanning. The *k* and threshold values, set to 5 and 3, respectively, were empirically determined by considering the computation time. ## Geometric correction for two-dimensional dose distribution The 2D scintillation images obtained by the TRMLSD with the camera included the lens distortion caused by the curvature of the lens and the geometric perspective distortion generated by the angle of the camera lens with respect to the scintillator plate. The 2D dose distribution perpendicular to the beam axis was measured by correcting these two types of geometric distortions with correction algorithms utilizing a chessboard image. A chessboard pattern with squares of area 22.0 × 22.0 mm² was placed at the position of the scintillator. After capturing an image of the chessboard pattern, first, the lens distortion was corrected, and then the perspective distortion was corrected using the ideal image coordinates of the chessboard. Perspective correction transformed the tilted or skewed image into an image that would appear if the scintillation incidence was orthogonal to the camera. The geometric calibration had to be performed only once, and the correction parameters were stored as a geometry calibration file. The geometric correction procedure was applied to the images obtained with 100 MU of 6 MV X-ray irradiation with a 10.0 × 10.0 cm² field on a 5.0-cm-thick solid water phantom placed on the TRMLSD (Setup-A). The geometry- corrected radiation field of 17.6 × 17.6 cm² had a resolution of 2048 × 2048 pixels, which corresponded to a spatial resolution of 0.086 mm. ## Verification of relative brightness variation The pyramid test pattern consisted of 2.0 × 2.0, 5.0 × 5.0, 10.0 × 10.0, and 20.0 × 20.0 cm² fields with 100 MU for each field. The pyramid dose pattern was irradiated to assess the brightness variations because of the absence of a mirror, uniformity of the scintillator, and radiation field size. A solid water phantom with a thickness of 5.0 and 10.0 cm invariance was placed on the TRMLSD. The first step was to correct the change in brightness due to the varying distance between the point of the scintillation source and the camera sensor and the inhomogeneity of the scintillator components at a depth of 5.0 cm. The next step was to test the variation in brightness according to the size of the radiation fields at a 10.0 cm depth in the phantom. The experiments for verification of relative brightness variation were performed in the 100.0 cm SSD setup. ### Dependence of brightness on scintillation-source-point-to-lens distance The flux of light generated from a point source decreases based on the inverse square of the distance. Because the camera is installed at a tilted angle, the distance between each point on the scintillation plate and the pixels of the image sensor varies. Consequently, the brightness of the captured image decreases with respect to the inverse square of the distance. The brightness variation due to both the distance and the chemical inhomogeneity of the scintillator was fixed by obtaining a 2D correction map using a 20.0 × 20.0 cm² open field at a depth of 5.0 cm in the water phantom. The horizontal and vertical profiles of the 20.0 × 20.0 cm² open field shown in were measured using a well-calibrated ion chamber array detector (MatriXX, Ion Beam Application, Belgium). The mean flatness and symmetry of the 20.0 × 20.0 cm² open field were 1.76%, and 0.70%, respectively, and were calculated according to TG-45 protocol. The SSD was set to 100.0 cm, and the dose rate was fixed at 400 MU/s. ### Dependence of brightness on the radiation field size The relative output of the linac varied with the radiation field size because of the change in the interaction of the scattered photons with the medium. The brightness measured by the TRMLSD also tended to increase with increasing radiation field size. However, the output factor of the TRMLSD was different from that measured by the ionization chamber and varied with the ISO value. Thus, the variation in brightness was tested for four different field sizes in the range of 2.0 × 2.0 cm² to 20.0 × 20.0 cm². The dependence of the scintillation intensity on the radiation field size was tested at a depth of 10.0 cm in a water-equivalent solid phantom positioned at 100.0 cm from the source (SSD 100.0 cm). The output factors for different field sizes were measured separately with a 0.13 cm³ ionization chamber (CC13; Scanditronix Wellhofer, TN, USA) at a depth of 10.0 cm in a water phantom with full backscatter. ## Deconvolution for penumbra correction with a CNN When the 2D dose distribution was measured using the TRMLSD with an ISO value of 800 \[M (x, y)\], the penumbra regions were inaccurately measured owing to the ISO setting. Moreover, M (x, y) includes complex errors such as those arising from the lack of backscatter, flatness of the 2D correction map, and optical blurring. To accurately measure 2D dose distributions with the TRMLSD, a 2D dose distribution deconvolution is necessary to recover the true dose distribution \[D (x, y)\] from M (x, y). We hypothesize that the complex errors of M (x, y) could be simultaneously corrected using a CNN consisting of multiple convolution layers, biases, and non-linear activation function. D (x, y) was calculated using a TPS with the following geometric conditions, which are used for routine patient-specific quality assurances for intensity modulated radiation therapy in our clinic: 5.0 cm water depth, SSD of 95.0 cm, and zero-degree gantry angle. In this study, D (x, y) is the ground truth dose distribution for correcting the measurements M (x, y) obtained under the same geometry condition. A CNN model, called PenumbraNet, was designed to recover the dose in the penumbra region of M (x, y). The input was M (x, y), and the output was the corrected 2D dose distribution \[C (x, y)\]. To train PenumbraNet, pairs of M (x, y) and D (x, y) corresponding to a 5.0 × 5.0 cm² open field and static IMRT plans configured with 348 subfields of multiple sites (H&N, lung, liver) were prepared as training data. PenumbraNet comprises an input layer, a hidden layer, and an output layer. M (x, y) enters through the input layer and passes through the hidden layer to reach the output layer. The hidden layer is a combination of convolution filters and biases and is initialized to a random number in the range of 0.000–0.001. In the training process, the hidden layer was optimized to minimize the root-mean- squared error between C (x, y) and D (x, y). The total number of training epochs was 60000, and adaptive moment estimation (Adam) was used as an optimizer. After the optimization process was completed, the optimized parameters of the hidden layer were saved so that they could be independently applied to the test data for TRMLSD performance validation. The test data comprised a 10.0 × 10.0 cm² open field and ten clinical IMRT cases configured with step-and- shoot methods, which were different from the IMRT plans used to train PenumbraNet. The details of PenumbraNet were described in a previous report. ## Validation of TRMLSD performance The performance of the TRMLSD was validated for five different items with 6 MV X-rays from the Novalis Tx. The output of the Novalis Tx was maintained at ± 0.7% of daily quality assurance (QA). To validate the TRMLSD, 6 MV X-rays at a dose rate of 400 MU/s and a gantry angle of 0° were used. Verification was performed for four basic characteristics of the open radiation field: (1) dose linearity, (2) dose reproducibility, (3) dose rate dependency, (4) percent depth dose (PDD), and (5) beam profile. In addition, the accuracy was verified for (6) two simple intensity-modulated fields, and (7) 10 clinical static IMRT cases (step and shoot). All measurements were made with the TRMLSD and converted to a 2D dose distribution using the algorithms developed in-house. In the case of the EBT3 film analysis, background subtraction was performed for accurate dose analysis, and digitization was performed with the 72-dots-per-inch scanning option using a 11000XL EPSON flatbed scanner. The other 2D dose distributions of the clinical case were calculated using Pinnacle^TM (Philips Medical Systems, CA, USA). Analyses were conducted using RIT (RIT Colorado Springs, CO, USA, v6.2) software, and gamma analysis was performed for comparison with the EBT3 film data or D (x, y). Gamma criteria with a 2–4% dose difference and a 2–4 mm distance to an agreement were used. A 5% dose threshold was used. ### Dose linearity, reproducibility, and dose rate dependency For the tests, the SSD was 100.0 cm and the dose rate was fixed at 400 MU/s. The geometry conditions of the TRMLSD and ion chamber were those of Setup-A and Setup-B, respectively. A 6 MV photon beam with a 5.0 × 5.0 cm² radiation field was independently irradiated in a range varying from 25 (20.0 cGy) to 249 MU (200.0 cGy) for the TRMLSD and ion chamber. Linearity test data were used as the brightness versus an absorbed dose calibration curve of the TRMLSD. The reproducibility was verified by irradiating the TRMLSD and the ion chamber five times in each geometric condition with a 6 MV photon beam with 25 MU and 249 MU irradiations. In the case of the dose rate dependency test, irradiation at 249 MU was measured three times with the TRMLSD for different dose rates of 100, 300, and 400 MU/min. Geometric correction, brightness variation correction, noise reduction, and penumbra correction were sequentially applied to analyze the 2D dose distribution. Finally, the brightness data were converted to the absorbed dose by applying a calibration curve. ### Percent depth dose A 6 MV photon beam with a 10.0 × 10.0 cm² field and gantry angle of 0° was applied to verify the capability of the TRMLSD for dosimetry of the absorbed dose at varying depths. The SSD and dose rate were fixed at 100.0 cm and 400 MU/s, respectively. The PDD was measured at depths of 2.0, 5.0, 10.0, and 15.0 cm in the solid water phantoms irradiated with 249 MU at each depth, and the PDD was compared with the measurements from the ionization chamber. The data measured with the TRMLSD and ionization chamber were normalized to the dose at 5.0 cm for comparison. ### Profile verification The profile of a 10.0 × 10.0 cm² open field was measured using the TRMLSD and EBT3 film (Energy: 6 MV, SSD: 95.0 cm, 5.0 cm thick solid water phantom, 249 MU). In addition, the profile measured by the ionization chamber (CC13) was compared. The EBT3 films measured in this study were scanned using EPSON 11000XL. The flatness, symmetry, and penumbra were analyzed for two datasets using the RIT software and were compared according to the AAPM TG-45. ### Two simple intensity-modulated plans To evaluate the capability of the TRMLSD for simple intensity-modulated plan dosimetry, two different simple intensity modulation plans entailing different sizes and doses of rectangular fields smaller than 10 × 10 cm² were used. The preset MUs for the three fields were 249, 50, and 25 MU, respectively. The simple IMRT plans were delivered to a 30.0 × 30.0 × 5.0 cm³ solid water phantom. The SSD and dose rate were set to 95.0 cm and 400 MU/s, respectively. TRMLSD data were analyzed using an in-house program at a time resolution of 1/24 s. Finally, the time-resolved 2D dose distributions were obtained and compared with the data obtained from the EBT3 film. The gamma passing rate was calculated from the 2%/2 mm to 4%/4 mm criteria. The dose threshold was 5% of the maximum dose for each field. ### Clinical IMRT plans The final validation of the TRMLSD was performed for pQA using the test data obtained from two heads and necks (7 fields, 7 fields), two livers (5 fields, 5 fields), two lungs (6 fields, 5 fields), two brains (5 fields, 7 fields), one pelvis (7 fields), and one prostate (7 fields). The scintillation sheets of the TRMLSD and EBT3 film were placed under a 5.0-cm-thick solid water phantom at a 100.0 cm SAD. The gantry was fixed at zero degrees. After the measurements, the per-field TRMLSD measurements were compared with D (x, y) using the gamma analysis method with 2%/2 mm and 3%/3 mm gamma criteria. For the prostate case, the accumulated dose distribution measured by TRMLSD was compared with the results of the EBT3 film using gamma analysis with gamma criteria ranging from 2%/2 mm to 4%/4 mm. The dose threshold was 5% of the maximum dose for each field. # Results ## Brightness dependence correction: 2D correction map The 2D dose distribution was measured accurately by correcting the brightness dependence due to distance and the chemical inhomogeneity of the scintillation with a 20.0 × 20.0 cm² open field. Each pixel value in the images captured by the TRMLSD was normalized to the pixel value of the radiation isocenter, and then, the inverse of each pixel value was saved as a 2D correction map. The correction map was applied by element-wise multiplications of the scintillation distribution images obtained with 249 MU irradiation of a 6 MV X-ray beam (10.0 × 10.0 cm² field, 5.0-cm solid water phantom on the TRMLSD). The result of the brightness dependence correction is shown in. ## Brightness dependence correction: ISO value The ratios of brightness were analyzed for different ISO values and compared with the output factor normalized to a 10.0 × 10.0 cm² field. When the ISO value was set to 800, the dose in the penumbra region was lost; nevertheless, the ISO value was still set to 800 considering the output factor. ## Penumbra correction using PenumbraNet The performance of PenumbraNet for accurate dose measurement in the penumbra region using the TRMLSD was validated by loading the parameters of the optimized PenumbraNet and independently applying them to the test data. depicts the profiles of M (x, y) given by the TRMLSD and Gafchromic EBT3 film, C (x, y), and D (x, y) for the middle of the 10.0 × 10.0 cm² field. The horizontal dose difference from C (x, y) is shown in, while the profiles of M (x, y), C (x, y), and D (x, y) of the field of the static IMRT plan for pelvis cancer are shown in. Finally, shows the horizontal dose difference from C (x, y). The details of PenumbraNet are reported elsewhere. ## Dose linearity, reproducibility, and dose rate dependency Dose linearity, reproducibility, and dose rate dependency are important characteristics of a dosimeter. The response of the developed TRMLSD to the absorbed dose was tested by analyzing the linear relationship between the brightness and the absorbed dose with 6 MV photon beams. The R-squared value of the linear fitting of the measurement data was 0.9998, and the root-mean-square error was 1.010. The brightness measured by the TRMLSD was converted to the absorbed dose using this curve. The reproducibility of the TRMLSD was tested for 20.0 cGy and 200.0 cGy. The mean and standard deviations of each set of five measurements were 19.99 ± 0.15 cGy and 202.69 ± 0.82 cGy for 20.0 cGy and 200.0 cGy irradiation, respectively. The mean and standard deviation of each set of five measurements using the EBT3 film were 20.12 ± 0.54 cGy and 200.37 ± 1.36 cGy, respectively. The dose rate dependency was tested for 100, 300, and 400 MU/min by delivering 200 cGy. The mean and standard deviation of each dose rate for three measurements were 200.55 ± 0.48 cGy, 199.87 ± 0.39 cGy, 199.96 ± 0.93 cGy, respectively. ## Percent depth dose The measurement data for the ionization chamber were normalized to the dose at the maximum dose depth (12.3 mm). The TRMLSD measurements were matched to the percentage at 5.0 cm of the ionization chamber data (85.2%) and are plotted in. The relative errors with respect to the ionization chamber data were −1.51%, 0.19%, and −0.23% with the TRMLSD, and −2.16%, −0.80%, and −1.99% with the EBT3 film at 2.0, 10.0, and 20.0 cm, respectively. The mean of the relative errors between the TRMLSD and EBT3 films was less than 0.64% and 1.65%, respectively. ## Beam profile verification The flatness, symmetry, and penumbra size are presented in. The measured flatness in the cases of the EBT3 film and TRMLSD differed from those corresponding to the ionization chamber by −0.16% and 0.1%, respectively. For symmetry, the difference was 0.2% and 0.3%, respectively, and for penumbra size, the difference was less than 0.5 mm for both cases. ## Simple intensity-modulated fields To evaluate the possibility of clinical application of the TRMLSD, we tested the accuracy of the method for simple intensity-modulated plans. In the case of field type 1, the total measurement time was 131.1 s, and the irradiation periods of the three fields were 32.67, 6.29, and 3.34 s. A gamma comparison was performed for the accumulated dose overtime at 81.6, 110.4, and 127.2 s. The obtained agreements of the 2D dose distribution were 94.72%, 97.84%, and 96.89% with a 3%/3 mm gamma criterion between the TRMLSD and EBT3 film. In the case of field type 2, the gamma passing rates for the accumulated 2D dose distribution for each field were 94.63%, 95.85%, and 96.64%. The gamma passing rates under different criteria are summarized in. ## Clinical IMRT cases The pQA for individual fields was performed for the 10 clinical IMRT cases using the TRMLSD. The TRMLSD analyzed the 2D dose distribution over time. The gamma passing rates were calculated per field for the 10 IMRT cases. The mean gamma passing rates of the subfields were 88.96% and 95.75% for the 2%/2 mm and 3%/3 mm gamma criteria, respectively. The quantitative results are summarized in. For Liver (1), the 2D dose distributions of the TPS, TRMLSD, and gamma index maps with the passing rates of 3%/3 mm are shown in. For the 2D dose distribution of the liver plan, the results obtained using the TRMLSD and EBT3 film were compared with the 2D dose distribution of the TPS using gamma analysis with criteria from 2%/2 mm to 4%/4 mm. The pass rates with the 3%/3 mm criterion were 99.00% and 95.51% for the TRMLSD and EBT3 films, respectively. The gamma passing rates with various criteria for the accumulated fields are summarized in. # Discussion In this study, a compact scintillation detector system without a mirror (TRMLSD) was proposed to perform high-resolution 2D dose distribution measurements using a camera with a CMOS image sensor and a sheet-type inorganic scintillator. The major issues faced by the TRMLSD during the development process are (1) a dosimetric error in the penumbra region due to the ISO value parameter of the digital camera and (2) backscatter due to the structural characteristics of the scintillation detector. The ISO value of a digital camera controls the light sensitivity of the digital imaging sensor. Particularly, this value directly affects the amplifier transistor located before the analog-to-digital converter. Thus, different ISO values cause a significant difference in the measured profile, specifically in the low-dose region, which corresponds to low light signals. In particular, uncertainty in the penumbra region could introduce a significant dosimetric error; this is because the intended dose distribution of an IMRT plan is achieved by irradiating many subfields such that the outfield dose distribution of each subfield actually contributes to the infield dose distribution. Moreover, for a structural reason, the effect of backscatter on M (x, y) was not considered with regard to collecting the scintillation using the camera. Thus, the deep-learning model (PenumbraNet) was developed to simultaneously solve these issues, and its efficacy and accuracy were verified. The cornerstones of this study are as follows: (1) We removed the mirror from the design of the conventional scintillation detector; thus, the volume and weight of the dosimeter were dramatically reduced. In our case, 16.52 kg was reduced. (2) The 2D dose distributions perpendicular to the beam axis were measured with clinically acceptable accuracy for patient-specific IMRT QA configured with a step-and-shoot method owing to the image processing techniques and convolutional neural network. (3) The scintillation collection device used in this study (HERO5 Black camera) was lightweight and wireless; thus, the TRMLSD was a wireless detector. (4) The designed scintillation detector can perform time-resolved 2D dosimetry through frame-based analysis of sequentially recorded scintillation images (video), and thus, can analyze the 2D dose distribution per segment and/or field. However, the following factors may affect the measurement accuracy: (1) The quenching effect has been reported to occur during particle dosimetry, but it is negligible (less than 1%) in photon therapy. (2) A spatial error can occur in the geometric correction process when obtaining 2D planar dose distributions without a mirror. Thus, the accuracy of the geometric calibration was validated using a test plan designed to verify the rotational invariance of the calibration, with dose distributions corresponding to three stripes of different widths (10.0 × 1.0 cm², 10.0 × 2.0 cm², and 10.0 × 4.0 cm²). Two measurements were performed under different conditions to verify the rotational invariance of calibration: one measurement was conducted along the axial direction, and the other by rotating the TRMLSD at an angle of 90°. No directional dependency was proven because the mean gamma passing rate of the two different binary maps with a 1%/1 mm gamma criterion, based on 50% of the maximum value, was 99.85% (0 \< 50% and 1 ≧ 50%). The binary map of the irradiated field corresponds to the field shapes. (3) The method for correcting the brightness variation in 2D, which is generated using a 20 × 20 cm² homogeneous radiation field, could cause an error in dosimetry because of the non-perfect flatness of the radiation field. There is less than about a 1.0% difference in the effective measurement area (17.6 × 17.6 cm²). If a homogeneous light source such as an LCD is used instead of a scintillator sheet, the variations in distance can be compensated for; however, the inhomogeneity of the scintillator sheet cannot be considered. (4) The FBNR technique has the potential to cause a dosimetric error for continuously varying dose distributions. This is because the hot pixel was discriminated using the instantaneous characteristics of the hot pixel. For dosimetry with continuously varying dose distributions such as in dynamic IMRT using the TRMLSD, a median filter and the dark-frame technique can potentially improve hot pixel reduction. (5) The effect of the SSD on the backscatter factor, defined as the ratio of the absorbed dose for a particular backscatter thickness to that in the case of full backscatter under a particular square field size and depth, was not considered in this study. If the TRMLSD is used in a fixed position (i.e., SAD is 100.0 cm), errors can be reduced by employing the aforementioned correction techniques for that specific position. In summary, the basic characteristics of the TRMLSD, including linearity, reproducibility, and dose rate dependency, were verified, and its usability in measuring the PDD for simple intensity-modulated cases was shown to agree well with those of the ionization chamber and EBT3 film. Its capability for clinical step-and-shoot IMRT dosimetry was also verified, proving its good agreement with a TPS. The TRMLSD, which utilizes an action camera that has not been considered for dosimetry before, can not only provide mobility but can also achieve acceptable accuracy in 2D dosimetry with the incorporation of software corrections. Moreover, the TRMLSD can be used as a portable and independent dose-measurement device for QA in IMRT. # Conclusions In this work, we studied the feasibility of using the TRMLSD for the measurement of the 2D dose distribution of a photon beam. To overcome the inherent shortcomings of the design, we have successfully developed and applied techniques, including computer vision techniques and the deep-learning model based on a CNN, for geometric correction, brightness variation correction, FBNR, and backscatter and penumbra correction. The TRMLSD exhibited good performance for clinical IMRT plans configured with the step-and-shoot method. The verified accuracy and time-resolved characteristics of the dosimeter may be useful not only for the quality assurance of machines but also for patient-specific quality assurance for clinical IMRT plans. The TRMLSD can be a dynamic dosimetry tool for clinical IMRT plans for dynamic therapies, such as IMRTs and volumetric arc therapy with X-rays, and scanning methods for proton and carbon therapies in which the beam fluence and characteristics change over time. [^1]: The Samsung Medical Center provided support in the form of salaries for authors S.J Kim, and S.K. Cho and Y. Han but did not have any additional role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript. The specific roles of these authors are articulated in the ‘author contributions’ section. Also the Samsung Medical Center affiliation does not alter our adherence to PLOS ONE policies on sharing data and materials.

# Introduction It is commonly accepted that the evolution of a protein family is the result of large-scale random mutagenesis of amino acids, with selection constraints imposed by their biological functions. Correspondingly most existing computational methods for prediction of functional evolution of protein families are designed based on the statistical analysis of amino acid sequences of the protein family. This type approaches begin from a database of multiple sequence alignment in the protein family, then amino acid frequencies at each sequence position are calculated, which is the fundamental quantity in the statistical analysis of protein evolutionary family. Long time ago scientists had noticed that the individual proteins in a protein family, which perform the similar biological function, may have very different amino acid composition, but they must share the similar three dimensional structure, and keep a stable key structural region. In other words, sharing the similar structural folding pattern is the necessary condition for all members in a protein family. Therefore the structural conservation is more important than the conservation of amino acid composition. The α-amylase protein family is a good example, which has an average sequence length of 420 amino acids. Among the 420 amino acids only 8 to 10 residues are absolutely conservative, and all other residues may be different more or less. On the other hand, the proteins of α-amylase family have a very conservative structure region, TIM (β/α)₈ barrel, and all other structural regions may be different. The differences in biological activity of individual proteins in a family are determined not only by the mutations of amino acids, but also by the structural differences. For example, all types of neuraminidases (NA) of influenza A viruses, which is the drug target of oseltamivir and zanamivir, share the same folding pattern of 3D structures. However, small structural difference at 150-loop in NA subtypes may cause the drug resistant problem. On the other hand, the structural differences at 150-loop of NA subtypes are the structural basis for designing effective drugs against specific subtype of influenza virus. In the previous studies of statistical analysis for functional evolution of protein family, most attentions had focused on the amino acid conservation and mutation. In this study a computational approach, namely structural position correlation analysis (SPCA), is developed to predict mutual correlations of structural segments and positions, and to find the signal communication network in protein family. We expect that the SPCA approach may find applications in protein engineering and in structure-based rational drug design. # Results To test the effectiveness of the SPCA theory and method, developed in this study, the PDZ domain family is selected as a model system, which is a well studied protein family. ## Database of PDZ protein domain The PDZ is a common structural domain found in the signaling proteins of bacteria, yeast, plants, viruses, animals, and human. The PDZ domains consist of 90–100 amino acid residues that adopt a six-stranded β sandwich configuration with two flanking α helices. The structure of PDZ domain 1BE9 and peptide ligand is shown in. In the PDZ domain the target C-terminal ligands bind in a surface groove formed between the α2 helix and the β2 strand at a number of binding sites that determine both ligand affinity and sequence specific recognition. Both the overall three-dimensional structure and most details of ligand recognition are highly conserved in the family despite considerable sequence divergence. The PDZ domains well represent protein binding motifs for which four high-resolution structures of distantly related members exist. These domains help anchor transmembrane proteins to the cytoskeleton and hold together signaling complexes. In this study the multiple structural alignment database consists of 186 3D structures of PDZ protein domains, which are selected from protein data bank (<http://www.rcsb.org/pdb/>). After structural sequence alignment there are 117 residue positions, and after deletion of the unnecessary gaps, the length of database is reduced to 96 positions. The MSA structural alignment of 186 PDZ domains is shown in. ## Position structural displacement matrix Following the procedure described in method section, the standard protein P⁽⁰⁾ and position displacement matrix D^(α)_L×L of the PDZ domain database is built. shows the most frequent amino acids at sequence positions, and shows the average position displacements between the standard protein P⁽⁰⁾ and the proteins of PDZ domains. In the higher frequency represents the stronger conservation of amino acid at the structural positions, and the lower frequency indicates the higher mutation of amino acid at the positions. In the higher displacement represents the larger structural change at the positions, and the lower displacement indicates the stable positions in the structural change. In there are several positions, at which the amino acids have very high frequencies: G at position 18, A at position 50, G at position 59, D at position 60, N at position 66, and G at position 67. These positions are the most conservative positions and listed in. Based on the most conservative positions the position displacement matrix D^(α)_N×L and D^(m)_N×L are built in the second MSA step using Eq.5. After careful observation at, we find interesting complementary relationship between the amino acid frequencies and the structural displacements: the higher amino acid frequency, the lower position structural displacement. All the most conservative positions have very small position displacements, as shown in. Correspondingly in at these positions the structural displacements are small. In at the positions from 25 to 37 the amino acid frequencies are very small. In contrast in the structural displacement at these positions are high. As we know that the amino acid position frequency is the fundamental quantity in the statistical coupling analysis (SCA) – and CMCA (conservation-mutation correlation analysis). According to the complementary relationship between amino acid position frequencies and position structural displacements, we expect that the structural position correlation analysis (SPCA) may provide useful information from different aspects to the functional evolution study of protein family. ## Structural segments of PDZ domains From the position structural displacement matrix D^(α)_L×L of the PDZ domain database and using the Eq.7 to Eq.9, we get the position displacement correlation matrix R^(α)_L×L. From the calculation results we find high correlation coefficients among some continuing sequence positions. The correlation coefficients, higher than 0.60, are listed in. The positions in fall in 6 segments: positions 4 to 7 in segment S1, positions 26 to 34 in segment S2, positions 50 and 51 in segment S3, positions 65 and 66 in segment S4, positions 75 to 83 in segment S5, and positions 90 to 92 in segment S6. For convenience in this study only the segments consisting of 2 or more positions are called segments and numbered as S*_k*. The PDZ domain consists of six β-strands, two α-helices, and eight loops. There are certain relationship between structural segments and secondary structural units. The segment 1 (S1) is located in the loop 1 (L1), the S2 is in the β2 and foreside of loop L2, S3 is in α1, S4 covers part of β4 and part of L6, S5 is basically in α2, and S6 is in β6. The sequence alignment of four PDZ domains (2QKT, 2F5Y, 1G9O, and 1BE9) is shown in. The relationship between 6 structural segments and secondary structural units of PDZ domain database is indicated in. In the 6 structural segments there are 29 positions. Except the 29 positions in 6 segments, the other positions are independent segments (positions). Therefore, in the PDZ domain database the number of segments is K = 73. The segment displacement matrix D^(s)_K×K is calculated using Eq.5. Then the segment displacement covariance matrix C^(s)_K×K and correlation matrix R^(s)_K×K are calculated using Eq.7 to Eq.9, respectively. From the segment displacement correlation coefficients R^(s)_K×K we find the correlation relationship among the structural segments and positions of PDZ domains. As shown in, the displacement of structural segment S2 is intensely correlated with that of the segment S5, and the higher correlation relation between position 37 in β3 and position 78 in α2 is shown in. ## Information abstraction of PDZ domain Some useful information for functional evolution study of PDZ domain family is abstracted from the calculation results of SPCA calculation. The groove between α2 helix and β2 strand is the binding location for peptide ligand. Amino acid mutations and structural changes at these positions play important roles to the functional difference of PDZ domains. As shown in, the structural displacement of S2 (in β2) is intensively correlated with S5 (in α2). The structural correlation between α2 and β2 well illustrates the ligand affinity and recognition specificity of PDZ domains to the peptide ligands. shows the α2-β2 groove of PDZ domains 1BE9 and 2QKT. Experiments found that in α2-β2 groove there are some easily mutative positions: 79 and 81 in α2, and 27 and 28 in β2 (in numbering), which determine the ligand binding affinity and control the shape and physicochemical property of the peptide ligands. In the residues Ala79 and Ala81 (in green) of 1BE9 are replaced by residues Tyr79 and Leu81 (in blue) of 2QKT. The size of Tyr79 and Leu81 of 2QKT are much larger than the Ala79 and Ala81 of 1BE9. Therefore 1BE9 and 2QKT must have very different preferences for peptide ligands. The correlation of amino acid mutations at these positions between α2 and β2 is accompanied by the correlation between structural displacement of segments S2 and S5, hence affects the preference of peptide ligands. The mechanism of distance allosteric interaction in proteins is a challenge and open research topic. The protein functions are not only determined by the interactions between local residues, but also depend on nonlocal and long-range communication between amino acids. For example, allosteric regulation in various proteins, the distributed dynamics of amino acids involved in enzyme catalysis, and information transmission between distant functional surfaces on signaling proteins, all represent manifestations of nonlocal interactions between residues. Long-range allosteric effects that cause the preference change of peptide ligands in the PDZ binding groove were found in several distant positions. In the alignment of four PDZ proteins in the 2QKT is an INAD PDZ domain and belongs to type 5 PDZ. The INAD PDZ domain (PDZ5) exists in a redox-dependent equilibrium, between an oxidized form and a reduced form. In the INAD PDZ an intramolecular disulfide bond covalently links a pair of buried cysteine residues located below the floor of the ligand-binding pocket, as shown in. In 2QKT the disulfide bond is formed between Cys37 in β3 and Cys78 in α2 (in numbering). The positions of Cys37 and Cys78 are corresponding to the residues Ile37 and Ala78 (in numbering) of 1BE9, respectively. The correlation of structural displacement between position 37 and 78 gives a good explanation to the distance allosteric interaction of mutations at β3 to the ligand preference of α2-β2 groove. The interaction between positions 37 and 78 affects the connection between β3 and α2, therefore, causes the structural change of the α2-β2 groove indirectly, hereby change the ligand preference of PDZ domains indirectly. # Discussion Structural conservation is the necessary condition for all members of a protein family, and the local structure differences may be responsible for the functional differences of individual proteins. Taking the structural data into the consideration of statistical analysis for protein evolutionary family certainly can find useful information that cannot be revealed by the amino acid sequence and frequency-based methods. The theoretical implications of SPCA approach are summarized as follows. (i) The standard protein P⁽⁰⁾ of a protein family, in which the position coordinates are the average coordinates of corresponding residues of all proteins and the residues at each position are the most frequent amino acid, keeps the common structural features of the family that are shared by all protein members. (ii) The most conservative positions form the structural core, and the amino acids at the most conservative positions perform the biological activity. The residues at other positions provide the physicochemical environment for the functional residues. The influences of non functional residues to the functional residues are determined not only by the amino acid types, but also by their position displacements. (iii) The position structural displacements between individual protein P*_i* and the standard protein P⁽⁰⁾ are the foundational variables, which determine the bioactivity differences of individual proteins in the family. (iv) The structural segments are the stable structure units of protein family, and the correlation between structural segments (or positions) may conduct signal for distance allosteric interaction. The application example of PDZ domain proves that the structural position correlation analysis (SPCA) is able to find the correlation relationship among the structural segments (or positions) in a protein family, which cannot be detected by the amino acid sequence and frequency-based methods. The functional communication network among the structural segments (or positions) in protein family, revealed by SPCA approach, well illustrate the distantly allosteric interactions, and contains valuable information for protein engineering and protein design study. # Methods Homologous proteins have conservative three dimensional structures that are evolutionarily more conserved than expected due to sequence conservation. The structural position correlation analysis (SPCA) for protein family starts from multiple 3D structural alignment of a protein family. ## Structure alignment and the most conservative positions The database of SPCA is built in a two-step procedure. The first step is a standard multiple structural alignment (MSA) of the protein family. In the standard MSA the α-carbon coordinates of all residues are realigned taking into account their structural similarity. From the initial estimate of the alignment, a new similarity matrix is generated using the relative α-carbon coordinates that result from a multi-body superposition. This matrix is used to realign just these alpha carbon populated chains. This procedure is then repeated until the root mean square distance (RMSD) of the superposition fails to improve. The multiple structural alignment of an evolutionary protein family reveals the structural features of family: all key functional residues are aligned in the same sequence positions, and all key secondary structures (α-helices, β-sheets, and loops) are positioned in the same sectors. After the standard multiple structural alignment the composition of protein family is represented by a binary data matrix A_N×M×L, where N is the number of proteins in the database, M is the types of amino acids (M = 21, including 20 natural amino acids and the gap, which is inserted during the multiple alignment), and L is the length of amino acid sequences (including gaps). In the composition matrix A_N×M×L the element *a_i,k,l* is 1 when the amino acid *k* of protein *i* is at the position *l*, otherwise, it is 0,The amino acid position frequency matrix F_M×L is constructed from the composition data matrix A_N×M×L as follows,The *f_k,l* is a decimal value in region \[0,1\]. The higher value of *f_k,l* means the higher frequency of amino acid *k* at position *l*. In this study the gaps are treated as a special amino acid type numbered by 0, and the 20 natural amino acids are numbered from 1 to 20. The summation of *f_k,l* from *k* = 0 to M is 1. At each position *l* the most frequent amino acid *k* is defined as the amino acid that possesses the largest frequency *f*^(m)*_k,l* at position *l*. The most frequent amino acids {*f*^(m)*_k,l*, *l* = 1,2,…L} compose the amino acid sequence of standard protein P⁽⁰⁾. In the second step of MSA, a set of most conservative structure positions {*l*^(m)*_j*} is selected firstly as follows. If the value *f*^(m)*_k,l* of the most frequent amino acid *k* at position *l* is larger than 0.80 (*f*^(m)*_k,l*\>0.80), the position *l* is the most conservative position. Then the second multiple structural alignment is performed only to the most conservative positions, making the coordinate RMSD of all most conservative positions as smaller as possible. In this way we get the structural database X_N×L, Y_N×L and Z_N×L of the protein evolutionary family for the SPCA calculation, in which the elements *x_i,l*, *y_i,l* and *z_i,l* are the Cartesian coordinates of position *l* in protein *i*. The theoretical consideration of the SPCA database can be illustrated as follows. The residues at most conservative positions are the functional residues, which perform the biological activity. The residues at other positions are non-functional residues, forming the physicochemical environment for the functional residues. The effect of the non-functional residues to the functional residues is determined not only by amino acid types, but also by their structural positions. ## Standard protein of protein family In a protein family the standard protein P⁽⁰⁾ is defined as follows. The amino acid sequence of standard protein consists of the most frequent amino acids at each position, and its 3D structure is the average structure of all proteins in the family. From the SPCA database the structure of standard protein P⁽⁰⁾ of the protein family is calculated using the following equations,where N is the number of proteins in family, L is the sequence length of MSA database, the superscript α indicates the coordinate of α-carbon of residues, and ‘0’ denotes the standard protein. The standard protein is the common representative of the protein family. ## Displacement matrix of structural positions The displacement matrix D_N×L of protein residue positions is derived from the standard protein and the MSA database of the protein family. The element *d_i,l* is the distances between the residue *l* of the standard protein P⁽⁰⁾ and the residue *l* of protein P*_i*. There are two types of displacement matrices. One is the distances between α-carbon atoms of standard protein and proteins of family, D^(α)_N×L, and the other is the distances of residue mass centers between standard protein and proteins of family, D^(m)_N×L. The mass center of residue *l* in protein *i* is computed as follows,where K*_i,l* is the number of atoms in residue *l* of protein *i*, x^(α)*_i,l,k* is the cartesian coordinate of atom *k* in residue *l* of protein *i*, and *m_i,l,k* is the atomic mass of atom *k* in residue *l* of protein *i*. The elements *d*^(α)*_i,l* of α-carbon displacement matrix D^(α)_N×L are calculated using the following equation, ## Reducing unnecessary gaps The SPCA calculation is complicated by the presence of alignment gaps inserted in the multiple structural alignment, which is commonly called indels, indicating a structural region present in some proteins but not in others. The gaps (space positions) may interfere with the results of statistical analysis badly. Before performing the correlation analysis we have to reduce the unnecessary gaps. To do so, the total amino acid position frequencies *q_l* of 20 natural amino acids at each position *l* are needed,In Eq.6 the index *j* for amino acid types runs from 1 to 20, not including the gap. In the amino acid frequency calculation the gap is a special ‘amino acid’ numbered as 0. If the total amino acid position frequency of the 20 natural amino acids *q_l* is less than 20%, the position *l* is deleted from the primer MSA database. Because at the position *l* the gaps are more than 80%, the position *l* is less important for the biological function of the protein family. After unnecessary gaps are deleted, the sequence length L is shorter than that of the primer data matrix. For simplicity, we still use L for the reduced sequence length. ## Position displacement correlation The purpose of SPCA is to find the correlation relationship between structural positions in the protein family. For this purpose we first construct the position covariance matrix C^(α)_L×L from displacement matrix D^(α)_N×L as follows,where and are the average displacements at position *i* and *j*, respectively,Hereby we get the position displacement correlation matrix R^(α)_L×L from the position covariance matrix C^(α)_L×L as follows,where the superscript ‘α’ indicates the ‘α-carbon’, and *r*^(α)*_i,j* is the displacement correlation coefficient between position *i* and *j*. In the same way we can calculate the position displacement correlation matrix R^(m)_L×L using mass center displacement matrix D^(m)_N×L. ## Fragment displacement correlation The secondary structures (α-helix, β-strand, and loop) are the structural units of protein structures. In many cases in the structural change of protein family some residues form a relatively stable segment, especially in some secondary structural units. The position structural displacements of the residues in a stable segment are correlated each other strongly. In order to analyze the structural position correlations among the stable segments, especially in the secondary structural units, it is best to organize the residue positions as structural segments. In SPCA a structural segment is defined as a set of continuing positions with higher mutual correlation coefficients (*r*^(α)*_i,j*\>0.60). The coordinates of structural segments are calculated as follows,where L*_l* is the number of positions in segment *l*, the K is the total number of segments, and superscript ‘s’ indicates the segment. The structural segments are not rigorously consistent to the secondary structural units. Some segments may cover continuing residue positions in two secondary structural units. However, many segments may contain only one residue position. The number of structural segments K must be less than the number of residue positions L of the protein family, K\<L. The displacement matrix D^(s)_N×K, the covariance matrix C^(s)_K×K, and the segment displacement correlation matrix R^(s)_K×K of structural segments can be calculated using the equations Eq.7, Eq.8, and Eq.9, respectively. The displacement correlation coefficient *r*^(s)*_i,j* represents the correlation relationship between segments *i* and *j* in protein family. The computational procedure of structural position correlation analysis is graphically illustrated in. The authors thank the research group of Prof. Rama Ranganathan (University of Texas Southwestern Medical Center) for the PDZ database. [^1]: Conceived and designed the experiments: Q-SD R-BH. Performed the experiments: Q-SD C-HW S-YL J-ZM. Analyzed the data: Q-SD R-BH J-ZM. Contributed reagents/materials/analysis tools: Q-SD C-HW S-YL. Wrote the paper: Q-SD R-BH. Designed software: Q-SD C-HW. Database build: S-YL. [^2]: The authors have declared that no competing interests exist.

# Background Currently the global population is undergoing an epidemiologic transition as individuals’ lifespans increase due to technological and medical advancement, which has resulted in the percentage of people aged 60 years or older being projected to rise from 11% in 2000 to 22% in 2050 on a global level. This demographic transition could particularly impact Asia, which already contains an estimated third of the global elderly population and encompasses countries with rapidly ageing populations, such as Singapore, China and Japan. Within these countries, it is expected that demand for health and social services, including palliative care, may swell as more individuals experience degenerative illnesses associated with ageing, e.g. dementia. Although there is an increasing need for palliative care in Asian countries, there remains a lack of adequate provision of such services in many of them. Even in nations that reportedly have accessible palliative care services, such as Singapore, individuals End-of-Life (EoL) care preferences are rarely realised. A survey on Singaporeans’ attitudes towards death in 2014 revealed that around three quarters of participants preferred to die at home, although around a quarter of deaths take place in these settings. This particular trend could be driven by healthcare professionals (HCPs) and caregivers; lack of knowledge of patients’ EoL care preferences, which can result in them undergoing painful, technologically-intensive and unnecessary treatments. The ‘Living Matters’ Advance Care Planning (ACP) programme was launched at the national level in Singapore in acute care settings in 2011, with the purpose of ensuring that HCP and carers were aware of patients’ treatment preferences. This programme was based on the ‘Respecting Choices’ model, which has an underlying philosophy that patients’ could only fully attain autonomy over their EoL care by articulating a treatment plan in accordance with their personal goals. The ‘Living Matters’ programme develops a treatment plan in a similar fashion to that of the ‘Respecting Choices’ model, wherein participants are encouraged to articulate their treatment preferences through a series of structured discussions with their HCPs and carers focusing on their personal values and beliefs. Resulting preferences are usually documented and endorsed by the responsible physicians in order to guide future HCP involved in care of the patient towards the end of their lives. Family carers attending the ACP conversations are expected to later act as proxy decision makers (PDMs), who may communicate the patients’ goals and treatment preferences should they be too incapacitated to do so, or when the ACP discussion has not been documented. By 2017, over 2000 HCP were trained to conduct ACP conversations, resulting in an estimated 10,000 ACP discussions being completed in Singapore. Although uptake of ACP discussions remains relatively low, a recent evaluation of the ‘Living Matters’ programme indicates that it may improve quality of EoL care by ensuring that patients’ treatment preferences are adhered to. Moreover, findings from evaluative studies conducted mostly in Western countries suggest that ACP has the potential to benefit both patients and caregivers. These studies indicate that ACP enhanced patients’ sense of autonomy and satisfaction with care as it offered HCPs an opportunity to empathetically listen to them. Meanwhile, caregivers displayed improved psychosocial outcomes after the patients’ death in comparison to those who had not undertaken the intervention. It is possible that participation in ACP discussions has a negative impact on dynamics within families. In one study conducted in Denmark, participation sometimes obstructed communication on EoL care because parties involved were overwhelmed or offended by the subject matter. This negative outcome could potentially be amplified within a Southeast Asian context owing to filial piety and pervasive death taboos, which can result in fears that discussion of death could potentially hasten a patient’s demise. Despite these concerns, there remains relatively little research on patients’ and caregivers’ perspectives on ACP in South-East Asian countries with population compositions like that of Singapore, which is mostly ethnically Chinese. Research on the performance of ACP and/or Advance Directives in South- East Asian countries revealed that patients’ receptivity and decision-making processes were markedly different to that of white patients living in Western countries. In contrast to these patients, whose interest in ACP were driven by a desire for autonomy, ethnically Chinese patients’ receptivity were influenced by their families’ acceptance of the intervention. As such, they tended to collectively make decisions on their EoL care in concert with family members, in some cases deferring their decisions on treatment to their physicians. Given that ethnically Chinese patients’ decision-making processes may contrast to that originally envisioned in the ‘Respecting Choices’ model, which emphasises patient autonomy within its underlying philosophy, this paper seeks to examine patients and PDMs experiences of ACP in Singapore. As such, it encompasses factors influencing patients’ receptivity to the programme and the decision-making processes they undertook when choosing their EoL care preferences and a PDM. # Methods This paper draws on data generated from the qualitative component of a largescale national level mixed methods evaluation of the ACP programme in Singapore. The qualitative component was spearheaded by AH, who has extensive experience conducting research on EoL care in Asian countries. It used many forms of data collection, including focus group discussions with HCPs, individual interviews with patients and their designated health spokesperson and ethnographic observation of acute care settings. Hence, this paper focuses on findings generated from interviews conducted with patients and their PDM. ## Sampling Sampling of participants was guided by the Interpretive-Systemic Framework, which seeks the perspectives of a wide range of stakeholders engaged with the programme. In accordance with this framework, patients and PDMs were sought on the basis of their engagement with the intervention and which hospital they receive care from. Respondents were purposively sampled by onsite Principal Investigators attached to 7 tertiary level hospitals, including: Changi General Hospital, Tan Tock Seng Hospital, KK Women and Children’s Hospital, National Heart Centre, Khoo Teck Phuat Hospital, National University Hospital and Singapore General Hospital. These hospitals were representative of acute care settings nationally: there were 8 tertiary level hospitals in Singapore when the data were collected. As the sampling framework had a solid theoretical foundation, the concept of data saturation was not used to guide the number of participants selected for the study. Onsite Principal Investigators attempted to sample 14 patient-PDM dyads who had completed ACP documentation and an additional 14 dyads who had not. However, owing to difficulties in finding patients and caregivers who had not completed documentation, only twenty-eight participants were interviewed. Thirteen of these respondents identified as PDMs, while the remainder identified as patients. ## Data collection Collection of data was supervised by PL, a female research fellow who had extensive experience conducting qualitative evaluations of health and social policy interventions in South-East Asia. She collected data with the assistance of OD, WST, and PVP, all of whom were PhD students receiving training in the fields of psychology or health services research. The interviews were conducted in settings deemed comfortable and convenient for participants, which were usually either in their home or in a clinic. Prior to each individual interview, respondents completed a brief questionnaire recording their demographic details (e.g., age, gender and level of education) as well as the number of ACP conversations they attended. Participants were then interviewed by members of the research team. PL and AH developed two separate interview schedules tailored to reflect how respondents’ role as a patient or PDM influenced their experience of participating in the ‘Living Matters’ programme. Both types of interview schedules were composed of a set of questions and prompts designed to encouraged participants to provide narrative with as little input as possible (e.g., ‘How would you describe your life prior to diagnosis?’). The interview schedules for patients gently guided participants into providing a narrative through focusing firstly on their pathways through treatment, secondly on their memory of the ACP discussion, and thirdly on actions they took after attending the ACP sessions as well as their reflections on their overall experience of the programme. Meanwhile, the interview schedule for PDMs was similar in content but instead focused on their relationship as a caregiver to the patient and their experience of participating in the ACP discussion as a PDM. The interview schedules were pilot tested during training sessions with members of the research team. All participants only attended one interview, which had an average duration of 45 minutes. Members of the research team developed field notes during the interviews that enabled them to collect data with a reflexive awareness of how their status as academic researchers may impact their relationship with the participants. The research team had not interacted with respondents prior to the interview. presents the structure of both interview schedules with sample questions. ## Analysis Audio recordings and transcripts of individual interviews were stored digitally and analysed using QSR NVivo. Transcripts were checked for errors by members of the research team, rather than by participants. Themes emanating from transcripts were identified through a framework analysis approach, in which data is usually reduced through a matrix comparing categories of data or cases. Data was analysed according to phases of analysis identified by Pope et al., which started with the research team familiarising themselves with the transcripts. Following this stage, AH, PL, OD, WST and PVP coded according to themes emerging from the data. They, also, met regularly to discuss the analysis and all emergent themes. Finally, themes and cases were compared through a matrix, which were agreed on by all members of the research team. Trustworthiness and creditability of qualitative data analysis was ensured through a carefully maintained audit trail kept by members of the evaluative team, who meticulously recorded each phase of analysis through memos. Respondents did not provide feedback on the findings. ## Ethics Research for this study received approval from the institutional review boards of Nanyang Technological University \[Ref: IRB-2016-05-023\] and the National Healthcare Group’s Domain Specific Review Board \[Ref: 2016/00603\], which were both situated in Singapore. Prior to each individual interview, respondents were notified of the purpose of the study, selection criteria, their right to withdraw from the study at any point during the discussion and that their confidentiality and anonymity would be maintained. All participants signed an informed consent form prior to commencing the interview, and received a small monetary incentive as a token of appreciation for their time. During the interviews, only members of the research team and participants were present. Details which could possibly reveal respondents’ identity have been anonymised, with each participant being assigned a code indicating whether they were a PDM or patient. The assignation of PAT denoted that the participant was a patient; PDM was shorthand for Proxy Decision Maker. # Results displays participants’ demographic characteristics. Participants had a median age of 55 years old (range: 24–84) and the vast majority were female (68%), married (75%) and of Chinese ethnicity (82%). Most respondents identified as Christian (39%), with the remaining identifying as Buddhist (32%), Muslim (14%) or as having no religious affinity (14%). Participants displayed a high level of education, with around half of the sample reporting having attended university (53%). Almost half of the sample reported that they were either experiencing or caring for a patient with heart disease (43%); while close to a third were dealing with cancer (32%). The remainder of respondents were experiencing or caring for patients with diabetes (11%) or other illnesses (18%), such as osteoporosis. Finally, eleven percent of participants reported that they were dealing with multiple co-morbidities, e.g. heart disease and diabetes. Majority of participants had attended at least one ACP conversation (86%) within which they completed the necessary documentation. Around three quarters of these respondents (74%) had engaged in the programme less than a year prior to data collection. Only 4 participants had started but not completed an ACP conversation. These respondents had not yet conducted the conversation with their designated PDM and, thus, were unable to complete the documentation. outlines the four major themes with relevant subthemes emerging from participants’ narratives. The first theme was **Engagement with Death** that encompassed factors influencing respondents’ active participation in EoL decision-making processes. This theme was influenced by a set of individual experiential, inter-relational and environmental drivers for their initial engagement with ACP. Individual experiential drivers were related to experiences and belief systems pertaining only to the individual; whereas, inter-relational drivers encompassed decisions influenced by concerns for those close to the participants (e.g. relatives). Environmental drivers comprised of wider socio- cultural norms influencing individual’s attitudes towards death as well as referral processes. The second theme, **Formation of Preferences,** encompassed concerns influencing respondents’ choice of care, some of which were related to their personal health and familial care concerns. The third theme, **Choice of Proxy Decision Maker,** covered aspects respondents take into consideration when choosing a PDM. Finally, **Legacy Solidification** considers how ACP is used to ensure the welfare of the family after the patient passes. ## Engagement with death ### Individual experiential drivers Respondents were driven towards engagement with the ‘Living Matters’ program by their *personal beliefs* related to self-autonomy and control. These personal beliefs held implicit assumptions about the nature of mortality; the underlying concept being that death is a ‘natural’ and unavoidable part of life. > “*The day you are born is the day you start dying*. *That’s how I view > things you see*. *So*, *to me*, *death is not something you have to be > afraid of because you already start when you are born”*. > > PDM11, 61 years old, Female Although death was perceived as inevitable, it was understood that the actual time point and circumstances under which one could die remained unpredictable. The circumstances demarcated the quality of an individuals’ death, with a peaceful manner of dying being perceived as swift, relatively painless and hindered by few medical interventions, and an undignified death being perceived as painful and prolonged through unnecessary medical treatments that substantially lowered a patients’ quality of life. Respondents placed importance on ensuring that they died in a peaceful manner, not only to safeguard their own wellbeing, but also to ensure the welfare of those who are emotionally close to them, such as their caregivers, friends and relatives. Under the rubric of their personal belief system the onus is on the patient to set out the conditions necessary for a peaceful death. This required a pragmatic acceptance of death when all options of treatment have been exhausted, which consequently led respondents to prepare for the final days of life: > “*For me*, *you have to be prepared for all this… you don’t know when > God will take you*… *I have to do something to prepare for myself*. *I > prepare something for my husband to know that what I want if in case > at the end of the day I cannot talk or whatever*… *so that they know > what you want*.*”* > > PAT12, 50 years old, Female Another individual experiential driver was their *perceived need* to undertake ACP. A few respondents reported that they were more receptive to the ACP programme after being hospitalised for their condition. This type of experience highlighted to participants that their lives could end earlier than they believed it would; thus, bringing an urgency to the need to organise their lives prior to dying, which may include communicating their treatment preferences to others. > “*The start of this year I had septic shock twice… it was quite > serious*. *So*, *after I survived those two episodes*, *I kind of*, > *realised that I really needed to carpe diem*, *like get as much done > as possible… strike out stuff on my bucket list*.*”* > > PAT1, 24 years old, Female A small number of respondents argued that their personal belief system enabled them to develop a set of *personality traits* that made them more receptive towards the ACP programme than other members of the public, who they believed were unduly influenced by death taboos. These personality traits were ascribed as being ‘open-minded’, independent and ‘positive’. Respondents argued their open-mindedness was evident during their social interactions, wherein they displayed an eagerness to discuss topics related to death; while, their independence was exhibited by their active engagement with their treatment. Finally, the trait that they described as positive was viewed as the ability to maintain an optimistic outlook even when experiencing life threatening illnesses: > “*I’m quite open*. *And I take it very positively… I think it’s a good > thing*...*for me to decide certain things on my own*. *Better than > letting my dear partner to decide for me*.*”*. > > PAT3, 49 years old, female ### Inter-relational drivers Many respondents strongly endorsed concepts underlying ACP, such as patient autonomy and self-determination, after *witnessing a loved one’s demise*. These respondents observed relatives prolonged suffering and lack of mobility, leading them to question the fruitfulness of sustaining their lives. Observing others suffer often shaped participants’ personal beliefs concerning death: > “*You know sometimes when I see the way my sister suffer (from cancer) > right*?.., *Of course you pray and then you ask that she recover*, > *but towards the last was very terrible for her*. *It was terrible > pain*, *very agonizing pain*.... *I don’t think I would like to go > through something like that or see somebody go through like that*. > *And sometimes you think you know it should end*.*”* > > PDM11, 61 years old, female > > “*Unless they (members of the public) see it themselves*. *I mean > frankly speaking–they see somebody suffer*. *I think they will > consider signing it (ACP)”*. > > PDM13, 62 years old, Female ### Environmental drivers The primary environmental driver for engagement with the ACP programme was *referral by staff*, as only three participants had actively sought out a conversation after learning of the program through a relative or engagement with promotional activities. For instance, PAT6 requested an ACP session after learning about the existence of the programme through his nephew, who had previously worked at one of the local hospitals. These findings indicated there remained a lack of awareness of ACP in the general public. Remainder of respondents reported that they were informed of the programme and then later referred to an ACP session by a member of staff at a hospital they were attending. Many of these participants were referred to the ACP programme at a fairly late stage of their illness- there were two cases of referral for a session conducted only with the PDM due to difficulties in communication or the patients not being deemed as cognitively able enough to choose their own treatment preferences. In one such dyad, the PDM attempted to protect her father from stress induced from discussion of topics related to death through forbidding the ACP coordinator from talking to him, which ultimately resulted in him being excluded from future ACP discussions: > “*He (The ACP Coordinator) just go direct and say*, *‘Why you want > (ACP) there*?*’*. *My dad get angry*. *(His wife) called me and said > she didn’t talked to my father this time*. *So that’s why I say*, > *‘No*, *you shouldn’t directly ask my father this kind of sensitive > questions… Because he get upset*. *He’s high blood shoot up on that > day…*. *Very high*. *He got worried*. *And so I say*, *‘You know*, > *anything*, *just talk to the family*.*’ Actually he can’t talk to the > patient direct*. *Because I think the family can just help them to > decide*. *Okay*.*”* > > PDM5, 49 years old, female In other cases, respondents were introduced to the ACP programme by staff, who they had already developed an emotional bond with. A distinguishing feature of these participants was that they had acquired a social network consisting of other patients and HCPs within their acute care setting and were to some extent involved in an informal type of activism. For instance, PAT11 had set up a workshop on ACP for her patient support group. As medical staff already knew these participants well, they were able to discuss fairly personal topics without fear of reproach. One patient stated: > “*For me*, *it’s because I’ve become very close with them (medical > staff)*. *So they know the nurse approach the proper person so that > they brought it up*. *Because for me*, *anything will do*. *So that’s > the time when she (the nurse) asked me–she was very careful*. *She > slowly… explained (ACP) and says*, *‘It’s nothing*. *Just think about > it’*.*”* > > PAT5, 61 years old, female Participants did, nevertheless, contend that the continual existence of pervasive *death taboos* made elderly Singaporeans resistant to concepts underlying ACP as they believed that any mention of death could hasten their demise. Some participants purported themselves to be impervious to such taboos due to their minority religious status either as a Christian or Muslim, which enabled them to be more accepting of death due to their belief in the existence of an afterlife. > “*Asian culture (death) is very taboo*. *But I can swear to you erm… > the 50 something lot or the 60 something lot will have no problems… > deciding and so on*.*”* > > PDM2, 54 years old, Female There were a small number of respondents who used concepts underlying filial piety to justify their endorsement of ACP. These participants would argue that the theoretical foundation of filial piety is for younger relatives to fulfil the wishes of elder family members whilst they are alive and well rather than forcing them to undergo painful and unnecessary treatment to prolong their lives: > “*In Cantonese*, *there is a saying*, *‘…you are filial when they are > alive*, *after they die*, *you don’t need to be filial’”* > > PAT9, 84 years old, female > > “*So I don’t want you all to have that burden where*, *you know*, *the > Chinese have the saying*, *‘Where you let go*, *people comment*. *You > don’t let go*, *you all will suffer’*. *So I said*, *‘No*, *you don’t > have to decide anything at all*. *I decide myself*.*’ That’s all*.*”* > > PAT6, 66 years old, male The aforementioned quotes demonstrated that it is possible to use traditional social norms in South-East Asian cultures to affirm, rather than reject, concepts underlying ACP, such as quality of life. These concepts were endorsed from a collectivist standpoint- participants argued that filial piety can only be fulfilled through abiding by parents’ or elder relatives’ treatment preferences instead of prolonging their suffering through unnecessary treatment. Respondents’ reinterpretation of filial piety could, thus, highlight potential new avenues for the development of culturally appropriate public health campaigns for ACP. ## Formation of preferences ### Personal health concerns Personal health perception, particularly those concerning quality of life, influenced respondents’ formation of preferences. Their *perception of quality of life* was defined as having a minimal level of discomfort and the ability to independently conduct their daily activities. The level of discomfort that they were able to endure were demarcated by their past experiences of pain and illness. This resulted in respondents expressing a wide spectrum of preferences, which spanned from being fairly minimal (i.e. the need to feel ‘clean’) to covering risky medical procedures: > “*I’m saying no to that (tracheostomy)*, *but I’m okay with > intubation… Because*, *I don’t know*. *It (a tracheostomy) seems > really scary… I think it’s just weird to have a hole there…you can see > the scar is quite obvious… But intubation*, *you know*, *can put in a > tube and then take it out and then no one would be the wiser*.*”* > > PAT1, 24 years old, female > > “*Let’s say–okay*, *let’s say you’re bed ridden*. *I say ‘…(at) least > must clean me up la*. *I don’t care what you do*. *You want to poke > me–I’m fine*. *I need to be clean*, *you know*, *shower*.*’”* > > PAT11, 36 years old, female ### Familial care concerns Meanwhile, formation of preferences was shaped by patients’ and PDMs’ concerns for the psychological welfare of those closest to them. These participants were concerned that their loved ones may carry the psychological burden of witnessing their unnecessarily prolonged and painful death, as they had to for their deceased loved ones: > “*We (patients) don’t want to be a burden to them (children)*. *I mean > why prolong*? *For me*, *I always believe in quality life*. *I want > quality life*. *I don’t want quantity*. *Don’t need for that”*. > > PAT14, 71 years old, Female Responsibility for decisions on end-of-life care was framed as an unnecessary psychological “burden” as family members are obligated to make significant decisions on treatment, such as whether to discontinue life support when the patient is in a vegetative state, with little knowledge of what their loved one would have wanted. Participants contended that such decisions could cause distress for family members who are worried their choices could hurt those closest to them. These types of decisions could potentially cause conflict between family members, who may have different perspectives on what types of treatment to follow. The following quotes underscore how treatment preferences were formed with the purpose of alleviating psychological “burden” for family members: > “*There are many cases that when something happen*, *the families are > in conflict sometimes*. *Like this person wants this*, *this person > wants this*, *you know*? *That’s why conflict*, *like uh… the > immediate family and the relatives you know*? *There’s so much > conflict*. *So if you put in writing you know what the patient wants*. > *Because during that period*, *he cannot decide”*. > > PDM6, 54 years old, female > > “*I would say that it’s good*. *Especially if we have*, *ever since > family in intensive care unit or neurosurgery ward*. *Why it’s good is > because there is a guideline given to the family*.*”* > > PAT2, 44 years old, male Few participants actively engaged in behaviours to inform their preferences; only three mentioned that they ‘researched’ their treatment options, which usually involved a search on an internet database. Formation of preferences did not seem to be influenced through *discussion of treatment preferences* with relatives, although their welfare formed a substantial part of patients’ decision-making processes. Instead, patients often chose to inform family members of their preferences once they had settled on them. PAT10 described her discussion of preferences with her PDM, who is her niece, when she said my “niece discussed–I explained to her and she listened, ‘in the future, if there are any changes that I can’t walk or what, I don’t want to count on you’”. Similarly to other patients’ accounts of their discussion of preferences with family members, it is apparent that the conversation was conducted in a linear fashion with the intent of seeking endorsement from those who could be later responsible for their treatment. ## Choice of proxy decision maker ### Values alignment Decision-making processes behind choice of PDM were influenced by perceived alignment of personal belief systems between both parties. Some participants stated that a PDM was chosen on the basis that they held the same attitudes towards death as the patient did. For instance, PDM11 chose her niece as a spokesperson due to their shared belief in Christianity and their lack of adherence to death taboos: > “*We share the same faith*, *number 1*. *Number 2*, *her outlook on > things also not very…for one of the better word*, *traditional…*.*I > don’t think anybody thinks like me because… I tell people*, *‘the day > you are born you are on the road to die’*, *and they will be like*, > *‘how can you say something like that*?*’*. *You know what I mean or > not*? *I don’t know why they are so shocked because logically if you > look at it*, *it’s like that… My niece… (is) very strong in her faith > so I think it’s very easy… (she is) very firm also*, *anybody argue*, > *any other way*. *She will tell them what she dislikes and that’s > it*.*”* > > PDM11, 61 years old, female PDM11’s account of factors motivating her choice of spokesperson also points to perceived personality traits as she argues that her niece firm resolve and honestly imbues her with the ability to execute difficult decisions. A few other participants cited personality traits as a motivating factor in their choice of spokesperson. As with PDM11, these participants believed that their chosen spokesperson held a set of personality traits that imparted them with the ability to fulfil their preferences. For instance, PDM7 argued that she was chosen as a nominated health spokesperson by her aunt as she was ‘strong’ in contrast to her uncle (the patients’ husband) who was deemed as too ‘soft’ and ‘sensitive’ to ‘unplug’. She noted that other family members would be “too sad to even think or do anything to fulfil her command (preferences)”; perhaps, demonstrating that she believed that they lacked stoicism in the face of adversity. There were, nonetheless, some participants who were unable to find a PDM who held a similar personal belief system to themselves. In a few cases, discord between the patient and PDM’s personal belief systems had little influence on implementation of the programme as the proxy decision maker chose to endorse patients’ wishes. For example, the dyad of PDM12 and PAT1 held contrasting religious beliefs, with one describing herself as a Christian, while the other was a self-ascribed ‘free-thinker’. Due to their contrasting personal belief systems they disagreed on treatment preferences- PDM12 preferred to immediately discontinue life support which clashed with PAT1’s wish to wait for 2 weeks. Despite these differences in opinion; PDM12 strongly endorsed her decisions, stating, ‘I think she herself know her conditions’. Other patients did perceive discord between the patient and PDM’s personal belief systems as a potential barrier to fulfilment of their preferences. These participants reported that their chosen PDM were reluctant to fully participate in the ACP session due to difficulties in the relationship or long held death taboos. For example, PAT2 did not conduct the ACP conversation with his wife as they were experiencing marital discord; while, PAT4 reported that during the ACP discussion his wife displayed a reluctance to partake in the conversation, asking, ‘Must we really do this?’. He believed that the ACP conversation frightened his wife and reported that since the session: > ‘*I just don’t want her to be reminded of it again*. *Cause like I > say…she still have traditional view of death*. *So speaking about > death is not so comfortable to her*.*’* > > PAT4, 55 years old, male This quote demonstrates that PAT4 avoids discussion of ACP or even topics related to death in order to protect his wife from potential psychological trauma. Meanwhile, PAT2 expressed the concern that “I will worry if my wife respect what I want. By the time I won’t know already”. These particular cases indicate that some PDMs may need additional psychological support to cope with their own death anxiety or negotiate difficulties within their relationships to the patient. ### Proximity A few PDMs were chosen on the basis of their *availability* to partake in the entirety of the ACP programme from initiation of the conversation to enactment of wishes. These PDMs were chosen in place of other relatives who were often not available to partake in the ACP process due to familial disputes or other commitments. PAT7 chose her close friend, PDM4, as her proxy decision maker as her relationship with her son had been fraught since divorcing her husband. Meanwhile, PAT10 chose her niece, PDM3, as proxy decision maker as her siblings had overriding family commitments: > “*All my brothers are not free*. *They have their own family and work > to take care of*, *I cannot count on them*. *Because I am staying with > my niece–because I told my fourth brother*, *my fourth brother told me > that it is good that I am staying with my niece*.*”* > > PAT10, 50 years old, female In other cases, proxy decision makers were chosen on the basis of their *emotional closeness* to the patient. Their emotional closeness was considered as a product of them being related as a member of the family or partner, which often imbued them with knowledge of patients’ character and belief systems. As PAT6 stated, ‘I think no matter how–your partner is usually the person who should understand you more’. It was, also, forged through shared experiences and prolonged proximity. These shared experiences were described in terms of provision of mutual care and support. PDM7 was chosen as a proxy decision maker by PAT12 as she frequently supported female members of her family in deciphering financial documents, which may have given her the necessary skills to navigate the semi-legalistic ACP process. PDM11 developed a close emotional bond with her niece through caring for her: > “*I have already been making a lot of decisions or helping to make the > decisions that she needs when she’s alive*. *So I*.. *can help with > her decisions… And then even now also*, *more or less*, *and ya that’s > why I nominated \[by\] her*.*”* > > PDM11, 61 years old, female A small number of proxy decision makers were chosen on the basis of their emotional closeness to the patients rather than on alignment of values. As noted earlier, this method of selecting a PDM sometimes resulted in tensions between both parties, which was apparent through proxy decision makers’ reluctance to actively partake in the ACP conversation. Patients chose these PDMs despite concerns that they may not execute their treatment preferences as they were keen to maintain their status within the family: > “*It take two hands to clap*. *I can only tell the boss that there’s a > weaver there*. *I cannot put the weaver into the water*. *Correct*?*”* > > PAT2, 44 years old, Male ## Legacy solidification Both patients and PDMs used ACP with the purpose of securing their enduring legacy after death, which was symbolised through the need to stabilise the family structure. Some participants’ formed preferences in the hope that it would alleviate the financial burden of out-of-pocket expenses related to EoL treatment for relatives. Patients, in particular, displayed a keen awareness of the competing needs of other relatives; arguing that these family members would still need to pay for living costs after they pass away. These patients believed that if there was no possibility that they could survive their treatment, they would be incurring an unnecessary financial burden on those who outlive them: > “*Every one of them needs to work*. *Now*, *if the younger generation > does not have jobs*, *cannot*. *Need to pay for the house*, *need to > pay for the children*, *right*? *And you are lying there doing nothing > everyday*, *need to go is more scary*. *Never mind*, *I have thought > this through*, *need to go*, *then have to go*, *to Heaven is ok*. > *When you are old*, *there is nothing much*, *don’t burden the younger > generation*.*”* > > PAT9, 84 years old, female # Discussion This paper makes a unique contribution to the literature on ACP in South-East Asian countries by examining patients and caregivers’ decision-making processes at each phase of a national-level intervention in Singapore. The findings of this paper differed from those of other studies on ACP and/or Advance Directives that were conducted with ethnically Chinese populations on two fronts. First, these studies identified sociocultural norms associated with pervasive death taboos and filial piety as acting as a barrier to participation in a given intervention. Although participants in this study were keenly aware of these sociocultural barriers, they were also able to point to factors facilitating their acceptance of ACP. One such facilitator was their demographic characteristic of belonging to a minority religious group (Muslim or Christian). These respondents argued that their belief in an afterlife encouraged their involvement in the programme. Moreover, findings pertaining to **Engagement with Death** revealed that participants’ endorsement of the concepts of self-autonomy and control were influenced by their experiences of witnessing their loved ones suffer towards the end of their lives and their own personal beliefs on death and dying. In combination, these factors enabled a few participants to reinterpret the concept of filial piety to endorse elderly patients’ control over their treatment, arguing that it is the duty of children to follow their parents’ wishes rather than to unnecessarily prolong their lives. These findings indicate that it may be possible to encourage acceptance of ACP among ethnically Chinese populations by proposing an interpretation of filial piety that supports elder patients’ needs and dignity. The second way in which this study contrasted from literature on ACP with Chinese populations was that its patients independently chose their own treatment, with family members playing a supportive role. Findings on **formation of preferences** indicated that participants’ decisions were influenced by concerns about personal health perceptions and familial care. The former concern was formed through patients lived experiences of their illness, which often shaped their perception of the level of pain and disability they were able to tolerate towards the end of their life. Meanwhile, the latter concern revolved around alleviation of psychological ‘burden’ incurred by responsibility of caregivers for EoL care. Moreover, results on the **Choice of Proxy Decision Maker** demonstrated that some patients chose a nominated health spokesperson based on their religion and their personality traits, such as stoicism. There were, however, other respondents who were unable to secure a PDM who shared their particular personal belief system owing to familial discord or traditional values around death. In a few of these cases, patients chose PDMs emotionally close to them in order to enable **Legacy Solidification** by maintaining the family unit. Although a few of these patients were anxious that their preferences may not be adhered to, they seemed to be willing to sacrifice their autonomy to ensure long-term stability for their relatives and loved ones. These findings indicate that the ‘Living Matters’ ACP programme may not cater to the needs of patients who are unable to find nominated health spokespersons who fully endorse their treatment preferences. As of 21^st September 2018, the Professional Deputies and Donees scheme was launched, allowing for individuals to hire a paid professional donee to make decisions on their behalf should they lose their mental capacities. This scheme should enable patients to find proxy decision makers who adhere to their documented preferences. It does not, however, cater to patients, whose overriding concern is to maintain their family unit through making one of their loved ones a proxy decision maker to signal their emotional closeness, highlighting a need for extra psychological support for PDMs with anxieties surrounding death. The overall findings of this study have led to the development of the directive decision-making process framework (displayed). This framework delineates personal and sociocultural factors influencing participants’ decision-making processes. Respondents’ continual participation in the intervention is driven by their personal belief system, which acts as a prism through which they interpret religious doctrine and socio-cultural norms according to their own needs. The prism of their personal belief system refracts in light of the ACP process to reveal a set of environmental, individual, and interpersonal concerns at each phase of interaction with the programme. Environmental concerns refer to influences associated with one’s social environment, while individual concerns focus on their self-perception and lived experiences. Finally, interpersonal concerns encompass their relationship to others close to them. The directive decision-making process framework indicates how the Respecting Choices’ model can be adapted to the Singaporean context through an appreciation of how individuals’ personal belief systems can influence their decisions at each phase of the ACP process. Respondents’ initial participation with the programme, or engagement with death, is influenced by environmental, individual, and interpersonal concerns. Individual concerns focus on personal perceptions of death and dying, while interpersonal concerns highlight how witnessing a loved one’s demise may ensure that patients are more receptive to the philosophical foundations of ACP. Environmental concerns refer to death culture in Singapore. As this seems to be the only phase of the intervention affected by environmental concerns, this framework suggests that it may be possible to encourage receptivity to ACP through a culturally sensitive public health programme that promotes acceptance of death and dying. The next phase of the framework, **Formation of Preferences**, was influenced by a combination of patients’ individual and interpersonal concerns. Their individual concerns encompassed their individual dignity, which referred to their sense of personhood and self-worth toward the end of their life; their interpersonal concerns focused on stabilisation of the family unit after passing. The final phase is dominated by interpersonal concerns encompassing alignment of values, discord of values, availability and emotional closeness with one’s PDM. These interpersonal concerns indicate that personal attitudes towards death and dying play as much of a role in patients’ choice of PDM as their perception of the relationship with this individual. In summary, the directive decision-making process framework highlights an underlying weakness in the conceptual framework of ACP, which holds individual autonomy as central to decision making on EoL care. Current research suggests that the concept of individual autonomy can be incongruent with the concept of autonomy in Asian cultures, in which families collectively make decisions. This study highlighted that even when patients are directing the decision-making process, they often do not act as autonomous agents who are untethered to their social context and solely focused on their own wellbeing. Instead, their decisions are often dominated by familial care concerns, as they attempt to solidify the family unit after their eventual death. This finding concurs with that of other studies in Eastern and Western countries, which suggested that patients often make treatment decisions with their loved ones in mind. Hence, this study highlights that the conceptual framework of ACP may need to be adapted to balance patients’ need for autonomy with their wishes to ensure the wellbeing of those they leave behind. ## Limitations This is one of the few studies to seek the perspectives of patients and proxy decision makers who had differing levels of engagement with an ACP programme from a wide array of acute care settings. It, nonetheless, had its limitations, namely, that it was not possible to sample the originally envisioned number of participants owing to difficulties recruiting respondents who had not completed documentation related to the conversation. This was due to records not being kept on patients or PDMs who had not completed the documentation within acute care settings. There were, also a number of patients and carers who were reluctant to undertake the interview due to the controversial nature of the study topic. Hence, it is possible that the research team were only able to sample participants who were already highly motivated and willing to speak about death and dying. # Conclusion This study reveals that many participants made decisions on their EoL care based on their perceived long-term legacy, which resulted in the **Formation of Preferences** and **Choice of Proxy Decision Maker** being used to ensure that patients’ family units remain with a solid financial and psychological foundation after their death. These findings point to possible new avenues for research and policy. In terms of research, difficulties with recruitment of participants for this study highlighted the need for further qualitative and quantitative research on factors influencing lack of receptivity to ACP in South-East Asian countries. Regarding policy, this paper indicated that the ‘Living Matters’ ACP programme needed to offer death education and psychological support to potential proxy decision makers, especially in cases where they are reluctant to approach topics related to death and dying. # Supporting information This study was made possible by the hard work of our research team, including Lok Hang and on-site Principal Investigators who recruited participants. We would also like to acknowledge the vital contribution made by participants, whose experiences were the foundation of our research. [^1]: The authors have declared that no competing interests exist.

# Introduction Hemoglobins (Hbs) became a focus of study early in the history of biochemistry, due to their color and ready availability in red blood cells and muscle. An unambiguous role facilitating respiration through oxygen transport made red blood cell Hb the first protein for which a physiological function was understood. A clear physiological role in combination with facile structural analysis has made oxygen transport Hbs ideal models for those seeking to learn the biophysical details relating protein structure and function. Recent years however, have seen new Hb discoveries that challenge many paradigms in this field, including that of their predominant function as oxygen transporters. For example, it is clear that red blood cell Hb and Mb also play important roles in NO homeostasis. Genomics studies have generated hundreds of new globin sequences and revealed this family of proteins as being nearly ubiquitous among living organisms. Hexacoordinate Hbs (hxHbs) are a structurally unique subset of the Hb superfamily that display reversible coordination of the heme iron ligand-binding site by an intramolecular histidine side chain. Prominent members of this group include neuroglobin and cytoglobin from humans and other animals,, the nonsymbiotic plant Hbs (nsHbs) found in all plants, and the cyanobacterial protein cyanoglobin. The diverse and prevalent distribution of hxHbs among organisms is often coupled with their high cross-species sequence identity, suggesting proteins with critical physiological roles. Current hypothesis suggest they might provide a common mechanism for protecting cells against hypoxia and oxidative chemistry in plants and animals. Despite numerous publications describing the significant structural and biophysical effort directed at these proteins,, a solid assignment of their physiological role(s) remains elusive. While many enzymes exhibit exquisite efficiency and specificity in the catalysis of their reactions, the innate chemistry of the heme prosthetic group is more “promiscuous” in nature, conferring Hbs with several reactivities of potential *in vivo* significance. These vary from reversible binding of diatomic ligands (including O₂, CO, NO, and many anions), to peroxidase activity, and redox reactions. Studies of such reactions have served to focus significant attention on putative functions of newly discovered Hbs involving nitric oxide (NO) binding and/or scavenging. Such studies include the measurement of NO reactions with hxHbs in the oxy, ferric and ferrous protein forms, as well as characterization of peroxynitrite reactions with some ferrous-NO hxHb. Two reaction mechanisms have been proposed for the scavenging of NO by heme proteins. Mb and red blood cell Hb destroy NO using the “NO dioxygenase” (NOD) reaction ( \#1) in which the oxy-Hb rapidly reacts with NO to form ferric Hb and nitrate. In fact, even when ferrous NO-Hb is reacted with O₂, the rate limit to heme oxidation is NO dissociation, which is rapidly followed by O₂ binding to the heme iron and a subsequent NOD reaction (49). Initially it was reported that the bacterial and yeast flavohemoglobins (flavoHbs) use the NOD mechanism in their reactions. Alternatively, it has been proposed that the flavoHbs use an “O₂ denitroxylase” mechanism ( \#4), in which NO binding precedes a reaction between the NO-heme complex and oxygen. In either case, the resulting oxidized heme must be reduced to start the reaction over (\#5). For flavoHbs, rapid reduction is achieved through a flavin- containing reductase domain –. Because of the change in heme oxidation state that accompanies NO reactions with Hbs, reduction reactions have been a recent focus of research. The kinetics of the reduction reaction is not as critical in NO scavenging by red blood cell Hb or Mb, as these proteins are present in vast stoichiometric excess compared to NO. However, with the possible exception of neuroglobin in the retina, hxHbs are present in very low (sub-micromolar) concentrations in the organisms in which they have been identified. Thus, if they were to serve catalytic roles in NO scavenging, efficient mechanisms for reduction would be needed. However, cognate reductases suitable for reduction on flavoHb time scales have yet to be assigned for any hxHb and reduction mechanisms are still unknown. Thus an important question to ask for hxHbs is whether reduction is truly the rate limiting step for catalytic NO scavenging. Ferric Hbs can also bind NO (\#2) and, as bound NO can potentially reduce the heme iron (\#3), ferric NO reactions and NO-reduction must be evaluated for hxHbs as well. The purpose of the research reported here is to systematically compare NO reactions with a representative set of hxHbs to Mb, with the underlying goal of testing whether these reactions distinguish any of the hxHbs in their ability to bind or destroy NO. The importance of this comparison centers on the premise that if we are justified in using *in vitro* reactions to assign NO binding functions in hxHbs, we should expect some characteristic to distinguish a hxHb from Mb, which does not catalyze NO destruction absent a reductase. Here we compare NO scavenging *in vitro* and in flavoHb knockout *E. coli* cells and ferric NO reactions with human neuroglobin (Ngb) and cytoglobin (Cgb), rice nsHb (riceHb1), *Synechocystis* hemoglobin (cyanoglobin, *Syn*Hb), and horse heart Mb. Our results demonstrate that all of these oxyHbs can rapidly destroy NO *in vitro* at a rate equal to re-reduction of the ferric Hb, all bind NO in the ferric form, all but ferric *Syn*Hb are slowly reduced by NO, but only the bacterial hxHb can replace flavoHb function *in vivo*. # Materials and Methods ## Protein production and purification Human neuroglobin (Ngb, GenBank accession number Q9NPG2) and cytoglobin (Cgb, GenBank accession number Q8WWM9), rice nsHb (riceHb1, GenBank accession number O04986) and *Synechocystis* hemoglobin (*Syn*Hb, GenBank accession number BAA17991) were expressed and purified as described previously. Horse heart myoglobin was commercially obtained (Sigma), dissolved in 0.1 M potassium phosphate (pH 7.0) to generate a 2 mM stock, and desalted over a G-25 column. This procedure yielded a sample giving a single band on SDS PAGE with kinetic and spectral properties identical to those previously published. Prior to experiments, all proteins were oxidized with an excess of potassium ferricyanide, which was removed by passage through a Sephadex G-25 column equilibrated in 0.01M potassium phosphate (pH 7.0). The ferredoxin-NADP reductase gene (GenBank accession number AAC76906) was amplified by PCR from *Escherichia coli*. The oligonucleotide sequences were: 5′CCTGGTGCCGCGCGGCAGCCATATGGCTGATTGGGTAACAGGCAAAG–3′ and 5′-TTGTCGACGGAGCTCGAATTCTTACCAGTAATGCTCCGCTGTC-3′. Amplification was done by 35 cycles of 30s at 95°C, 30s at 55°C and 1 min at 72°C. The resulting fragment was then cloned in the pET28a plasmid (Novagen) between the NdeI and EcoRI restriction sites. The His-Tagged protein was expressed and purified as described earlier. ## UV/VIS spectroscopy The NO-ferric spectra were recorded by mixing a deoxygenated ferric Hb solution (5 µM) with a 2 mM NO solution. The NO-ferrous spectra were recorded by addition of a 2 mM NO solution (1mM final) to a sodium dithionite reduced protein solution, in a cuvette previously sparged with N₂. To follow the NO- induced reduction of the protein, spectra were collected over a period of 5h after mixing. First attempts to measure ferric NO binding displayed non-saturated spectrum at low NO concentrations. Therefore, NO binding was measured by equilibrium titration of 5 µM Hb with NO solutions (7 µM to 2 mM). The dissociation equilibrium constants (K_{D NO, Fe3+}) for each protein were extracted using the following equation where F_B is the fraction of NO-bound protein: The saturated NO solution (2 mM) was prepared by equilibrating a buffer solution with NO gas that was first passed through a 20% NaOH solution. All spectra were recorded using a Varian Cary 50 spectrophotometer at room temperature. All solutions were made with 0.1 M potassium phosphate buffer pH 7.0. ## NO dioxygenation and scavenging *in vitro* A multi-port measurement chamber (World Precision Instruments, WPI. Sarasota, Fl) was used to analyze the stoichiometric reaction between oxyferrous hxHb and NO at room temperature. The chamber contained 0.1 M potassium phosphate pH 7.0 and was equilibrated with N₂. The low oxygen concentration was measured using an oxygen electrode (ISO-OXY-2, WPI). At 0% oxygen, the chamber was closed leaving a small dead space (0.5 cm) at the surface of the solution flushed with N₂. Then 20 µM NO was added and the signal was followed using a NO electrode (ISO-NOP, WPI). When it reached a maximum, oxyferrous Hb (20 µM final concentration) was added to the solution and the decay of the NO signal was recorded. Rapid mixing experiments measuring NOD rate constants have been described earlier and were conducted with a BioLogic SFM 400 stopped-flow reactor coupled to a MOS 250 spectrophotometer. Ferric protein was reduced using a system adapted from an earlier study in an Eppendorf sparged with N₂. After 10 min, each ferrous protein sample was added to a gas tight syringe containing 0.1 M potassium phosphate pH 7.0 (262 µM O₂). The oxygen bound protein (1 to 5 µM, after mixing) was confirmed by recording the oxy-ferrous spectrum. The NO solutions were generated by mixing a saturated NO solution (2 mM) with a N₂ equilibrated potassium phosphate buffer (0.1 M, pH 7.0) in gas tight syringes. An anaerobic condition in the NO syringes was maintained using the glucose oxidase-catalse system. Kinetic time courses were collected (20°C) at different NO concentrations (5 to 500 µM) by recording the change in absorbance at a fixed wavelength (413 for Mb, 422 for Ngb/Cgb and 420 for riceHb1). Between five and eight kinetic traces were collected and averaged for each time course. ## Ferric NO binding Rapid mixing experiments measuring NO binding to the ferric protein were conducted with a BioLogic SFM 400 stopped-flow reactor coupled to a MOS 250 spectrophotometer. Two syringes were used: one containing the protein solution (Fe³⁺, ∼5 µM after mixing) and the other containing an NO solution. Reaction concentrations above 1 mM \[NO\] were obtained by mixing a larger volume of the saturated NO solution than volume of Hb. Anaerobic condition in the NO syringes were maintained using the glucose oxidase-catalse system. Kinetics time courses were collected (20°C) at different NO concentrations by recording the change in absorbance at a fixed wavelength (Soret peak). At least three kinetic traces were collected and averaged. The minimal NO concentration was at least 2 times higher than the calculated K_{D NO, Fe3+}. ## Enzymatic reduction of hxHbs The reduction system used in the present study has been adapted from one described in detail previously. First a 1 ml cuvette was flushed for 1 min with 1 atm CO. Then 0.1 M potassium phosphate pH 7.0 equilibrated with 1 atm CO was added through a rubber stopper. After addition of 60 µM NADP⁺ (Sigma, St. Louis, MO), 0.7 U/ml glucose-6-phosphate (G6P) dehydrogenase (Roche, Pleasanton, CA), 1 µM ferredoxin-NADP reductase (*E. coli*) and 5 µM Hb, 3mM G6P was added to start the reduction reaction. Absorbance spectra were collected over a period of 30 min at room temperature. The difference in absorbance between ferric and CO-bound protein was plotted versus time to calculate the initial velocity of the reaction. ## Catalytic NO consumption experiments The multi-port measurement chamber (WPI) containing 0.1 M potassium phosphate pH 7.0 and 3mM G6P, was first equilibrated with a mixture N₂/O₂ to reach an O₂ concentration of 4%, as measured with an oxygen electrode (ISO-OXY-2, WPI). At 4% O₂, the chamber was closed leaving a small dead space (0.5 cm) at the surface of the solution. Then, 10 µM Hb, 60 µM NADP⁺ (Sigma, St. Louis, MO), 0.7 U/ml glucose-6-phosphate (G6P) dehydrogenase (Roche, Pleasanton, CA), and 1 µM ferredoxin-NADP reductase (*E. coli*) was added to the solution to make the ferrous-oxy complex. After 10 min, 40 µM NO was added to the solution and its removal was measured using the NO electrode (ISO-NOP, WPI) at room temperature. The NO consumption rate was calculated from the initial velocity just after NO addition. ## NO scavenging in flavoHb knockout *E. coli* *E.coli* strains AB1157 and AG1000 ((AB1157*Φ* (*hmp-lacZ*)262; *Cm^r*)) were generously provided by Dr Paul Gardner (Cincinnati Children's Hospital Medical Center). All Hbs were cloned into the pANX plasmid between the NdeI and HindIII restriction site except sperm whale myoglobin, which was cloned between the NdeI and BamHI restriction sites. The pANX plasmid is derived from pUC19 and contains the promoter region of the *hmp* gene of *E. coli*. The expression of a gene cloned next to it (via NdeI), will be driven by that promoter and should be comparable to the natural *hmp* gene expression. The generated plasmids were transformed into the *hmp* deficient strain AG1000. As a control, “virgin” pANX and pANX-hmp (provided by Dr Paul Gardner) were also transformed into AG1000. A test tube containing 5 ml LB medium was inoculated with 1% of an overnight culture grown aerobically. No selection was used for the AB1157 strain. For the AG1000 strain, 27 g/ml chloramphenicol was added to the medium, and for AG1000 containing the different plasmids, 50 µg/ml carbenicillin was also added. After inoculation of the test tubes (∼0.02 OD₆₀₀), 3 mM GSNO was added and the cultures were grown aerobically at 37°C with constant agitation. The concentration of GSNO used in these experiments was determined by the minimum level necessary to reveal a clear hmp- phenotype in our reactions. (and is comparable to those used by others in NO challenge experiments). After 14 hours the OD₆₀₀ was measured. As a control, cells were grown without GSNO and in the presence of 3 mM inactivated GSNO (as described in the Supplemental.) Untreated cells were also used for the measurement of NO consumption. The results shown are an average of three independent experiments, repeated once. GSNO was generated by mixing a stoichiometric amount of reduced gluthathione (Sigma) and acidified sodium nitrite for 10 min in the dark. Then the pH was adjusted to 7.4 with 4N sodium hydroxide. The concentration of GSNO was determined using ε₃₃₄ = 767 M⁻¹cm⁻¹. As control, to test if the effect observed is due to NO alone or to other by- products produced during its synthesis, we destroyed the NO component of our GSNO by photolysis. To do this, the GSNO solution (∼0.2 M) was placed in a 4 ml cuvette and photolyzed using a YAG laser (532 nm). After 45 min all GSNO was inactivated, and NO released during that process was neutralized by pure oxygen during a 15 min exposure. These experiments are provided in Supplemental. ## Measurement of hemoglobin expression levels *in vivo* Saturated cultures, inoculated from an overnight culture, were used to measure the expression level for each Hb. For *Syn*Hb, cells treated overnight with 3 mM GSNO were also used. Cells were harvested by centrifugation at 6000 rpm for 6 min. The cell pellet was then resuspended in 0.1 M potassium phosphate pH 7.0 and sonicated (3×30s with ∼5 min interval). After centrifugation to remove the cell debris (20000 rpm for 30 min), the supernatant (50 mg/ml total protein) was used to record CO+sodium dithionite–reduced sodium dithionite difference spectra. The difference spectrum of each protein was corrected by the difference spectrum measured with the supernatant of the strain transformed with pANX alone. The results shown are an average of two independent experiments. ## Measurement of NO consumption *in vivo* The multi-port measurement chamber (WPI) containing air equilibrated 0.1 M phosphate pH 7.0 was used for the aerobic NO consumption, at room temperature, of the different *E. coli* strains. NO (2 µM) was added to the chamber, and when the signal reached a maximum (as measured with the NO electrode (ISO-NOP, WPI)) cells were added (1×10⁷ cells/ml) from an overnight culture. For *Syn*Hb, cells treated overnight with 3 mM GSNO were also used. The rate of consumption was calculated as the time required to consume 1 µM (half the signal) of NO. All rates were corrected for the rate of consumption by the strain transformed with pANX alone. The results shown are an average of six independent experiments, repeated once. # Results ## NO dioxygenation and non-catalytic NO consumption The rate constant for NOD by Mb has been reported previously. It is a very rapid (nearly diffusion limited), bimolecular reaction between NO and the oxy-Hb complex. While peroxynitrite is considered to be an intermediate in this reaction (\#1), its dissociation rate is not limited in Mb over the range of NO concentrations amenable to stopped flow kinetics. All of the oxy-hxHbs react with NO, and NOD by Ngb and a plant nsHb has been reported previously. In the case of mouse Ngb, the reaction displayed similar kinetics at the two \[NO\] tested. The interpretation of this observation was that dissociation of peroxynitrite is rate limiting. Each hxHb investigated here, over a range of NO concentrations, showed saturating kinetics as \[NO\] increased, but the limiting observed rate constant varies between proteins. Ngb and Cgb are similar (k_{obs, NOD} = 360 and 430 s⁻¹, respectively), but riceHb1 is slower (k_{obs, NOD} = 90 s⁻¹). Unfortunately, we were not able to calculate the rate constant for *Syn*Hb as the change in absorbance was very small, resulting in very noisy time courses. ## Ferric NO binding and NO-induced reduction Human Ngb is the only ferric hxHb for which reactions with NO have been published. Binding was very slow, and produced ferrous NO-Ngb. shows the \[NO\] dependence of the observed rate constant following mixing with each ferric hxHb and Mb. Mb binds NO in a bimolecular fashion, with an observed rate constant equal to 70 mM⁻¹ s⁻¹. In the case of each hxHb, time courses for binding also display a linear dependence on \[NO\], but the observed bimolecular rate constants are much smaller (k_obs,NO(Fe3+)). NO reduction of ferric Hbs has been reported extensively for Mb, and as described above for Ngb. Following the NO binding reactions in (particularly those at higher \[NO\]), relatively slow reduction of some of the ferric Hbs by NO was observed. The degree to which this reaction can regenerate ferrous Hb was gauged by calculating the fraction of Hb(2+)-NO present after 30 min. Only Mb, Ngb, and Cgb showed significant reduction over this time period. Time courses for this reaction yielded rate constants (k_red,NO) of 0.03, 0.12 and 0.05 min⁻¹ for Mb, Ngb and Cgb, respectively. For riceHb1, the rate constant was estimated to be 0.006 min⁻¹ (following a 5h time course). Ferric *Syn*Hb showed no significant reduction by NO. ## Catalytic NO scavenging We have demonstrated that each oxy-hxHb has the ability to carry out the NOD reaction with varying degrees of efficiency. However, to scavenge NO catalytically, re-reduction of the resulting ferric heme iron must occur (\#5). In, the contribution of heme iron reduction to NO scavenging is investigated by using an artificial reductase system (ferredoxin-NADP reductase (FdR) from *E. coli*) that reacts with similar effectiveness with each hxHb under investigation. provides an example of the spectral changes associated with reduction of ferric Ngb by this system and in the presence of CO (which serves to trap the reduced Hb). shows time courses for reduction of each hxHb (10 µM) by 1 µM FdR, and the effect of substrate concentration (Hb) on reduction velocity is shown in. The plots have enough curvature to allow calculation of k_cat and K_M for each protein. K_M and k_cat are similar for each Hb, with none deviating by more that 2-fold from the others. demonstrates catalytic NO scavenging in the FdR/Hb system described above. With no Hb present, NO is degraded to 20 µM in ∼5 minutes, and the NO signal reaches the initial baseline value over a period of about 30 minutes. The presence of 10 µM oxy-Hbs accelerates NO removal in a bimodal manner. First, 25% of the signal (10 µM of the NO) is lost rapidly due to stoichiometric NOD. The remaining signal decays linearly back to the background level more rapidly than in the absence of Hb. This rate of NO destruction is attributed to NOD following ferric Hb re-reduction and oxygen binding. In fact, the velocities of NO consumption are nearly identical to the reduction velocities under these experimental conditions (last two columns). Therefore, the ability of each hxHb and Mb to catalytically scavenge NO is directly related to the rate of re-reduction of the heme iron following NOD. ## NO scavenging by hxHbs *in vivo* The only Hbs that are known to be NO scavengers are the bacterial and yeast flavoHbs,, and the *E. coli hmp* (flavoHb) null mutant presents a clear NO- sensitive phenotype under aerobic conditions. The experiments in were designed to test the ability of Mb and hxHbs to substitute for flavoHb in *hmp* knockout cells under similar conditions. The foreign Hbs were introduced into the *hmp* mutant cells (AG1000) on a plasmid (pANX) containing the *hmp* promoter, and cell growth was measured in the presence and absence of NO donated by GSNO. GSNO is a NO-releasing agent with properties comparable to spermine/NO or SNAP, and was chosen as an NO donor because of its documented ability to up-regulate expression of *hmp* – and therefore expression of the Hbs under the control of the *hmp* promoter on the pANX plasmid. As a precaution, to ensure that the effects we observed were due to NO and not a byproduct of GSNO, “NO depleted” GSNO was also used as a negative control (this procedure is described in the section and Supplemental). presents OD₆₀₀ values for uniformly-grown *E. coli* cultures after 14 hours showing a comparison of cultures grown in the absence of GSNO, presence of 3 mM GSNO, or the presence of 3 mM GSNO depleted of NO by photolysis. As expected, the wild type strain (AB1157) and the *hmp* mutant strain (AG1000) expressing flavoHb on the pANX plasmid are able to grow in the presence of GSNO, but the AG1000 strain alone and that carrying the empty pANX plasmid are impaired. However, no protection was observed with the AG1000 strains expressing Mb or other hxHbs except the one expressing *Syn*Hb. In the case of *Syn*Hb, cell growth over this time period was indistinguishable from AG1000 expressing flavoHb. The different strains were not sensitive to inactivated GSNO, indicating that the effects observed in the presence of GSNO are only due to the release of NO and not to other compounds present in the GSNO solution. To investigate whether the results in are attributable to variation in hxHb expression levels in the different strains, expression was measured independently by recording the CO-difference spectrum of each. The level of expression is correlated with the maximum absorbance of the difference between the cell extract (supernatant) with and without addition of CO and sodium dithionite. and report these values. From these data it is evident that expression levels vary significantly, with Ngb being poorly expressed and Mb/*Syn*Hb being expressed at the highest concentrations. Compared to flavoHb, RiceHb1 and Cgb are expressed 2 to 4 times less, and Mb and *Syn*Hb almost 3 times more. NO consumption by these cultures was also measured directly to ensure that it correlates with cell viability. shows rates of NO consumption by strains containing each Hb on the pANX plasmid. Rates of consumption by the hxHbs are at least 5 times slower than flavoHb, even for Mb and *Syn*Hb, which are expressed at the highest levels. It was surprising that cultures expressing *Syn*Hb did not consume NO, as they were viable in the growth experiments presented in, suggesting that *Syn*Hb in the pANX system can protect against NO without consuming it from the media. This discrepancy was investigated by monitoring cell growth as a function of time. In the absence of GSNO, all cultures grew at about the same rate. In the presence of 3 mM GSNO, the cultures expressing Hmp were unaffected, and all others (with the exception of *Syn*Hb) did not grow. Growth of the *Syn*Hb/pANX culture was retarded, but recovered after ∼6 hours to eventually yield the OD₆₀₀ values in that are indistinguishable from those of the Hmp strains. Furthermore, the GSNO treated *Syn*Hb/pANX culture (14h) was capable of more efficient NO consumption, but did not show increased concentration of *Syn*Hb. # Discussion ## Ferric hxHbs do not efficiently catalyze NO destruction It has been proposed that hxHbs might destroy NO through a mechanism that includes binding to the ferric heme iron. In the absence of specific reduction mechanisms, this route is compelling due to the potential for NO to reduce the heme, which could then bind oxygen and go through one cycle of NO-dioxygenase activity to reform the starting ferric Hb complex. One complete cycle would scavenge two molecules of NO using two different chemical mechanisms. The results presented in and, and do not support this hypothesis for any of the hxHbs investigated here. There are at least two kinetic hurdles for this mechanism; both ferric NO binding and NO-induced reduction must be fast. We have demonstrated that binding of NO to ferric hxHbs is significantly slower than to Mb. This is probably due to intramolecular His binding to the ligand binding site, which is enhanced in the ferric oxidation state. The linear dependence of the reaction with \[NO\] combined with small observed second-order rate constants (k_{obsNO, Fe3+}) is indicative of a bimolecular association rate constant for binding to the pentacoordinate complex that is much slower than the rate constants for His binding and dissociation. The combination of small values of k_{obsNO, Fe3+} and low \[NO\] *in vivo* would result in very slow NO association relative to flavoHbs under the same conditions. An additional factor diminishing the likelihood of the ferric NO-binding mechanism is the slow rate of reduction of the Hb(3+)-NO complex. Although this reaction clearly varied between the hxHbs tested, none exhibited rates capable of rapid NO destruction even when \[NO\] is sufficient to saturate the ferric complexes. The fastest reduction was observed in Ngb, where a reaction half-life of ∼6 minutes requires ∼1 mM NO to achieve. This does not support NO reduction of ferric hxHbs as a plausible mechanism for NO scavenging *in vivo* where \[NO\] rarely exceed ∼200 nM. However, it could be sufficient to generate ferrous NO-Ngb for scavenging of peroxynitrite, preventing its deleterious reaction with CO₂, or for O₂-nitroxylase mediated NO scavenging under microaerobic conditions. ## HxHbs as NO dioxygenases In this comparative study of Hb NOD, we find no indication that hxHbs are more efficient in this function than Mb. Instead, we observe a limiting reaction in hxHbs that is probably peroxynitrite dissociation. Hence, Mb would perform better in a NO scavenging role utilizing this reaction mechanism. However, we have also demonstrated that the limiting factor in NO scavenging for all Hbs examined is the re-reduction following NOD. Studies showing that monohydroascorbate reductase increases the NOD activity of barley nsHb, and that the isolated Hb domain from *E. coli* flavoHb is insufficient to protect cells during NO challenge, are in agreement with our results. Thus assignment of NO dioxygenase activity as a physiological function requires the design of experiments that address reduction mechanisms. The *E. coli* flavoHb reductase domain and the cognate reductase identified for *Vitreoscilla* Hb satisfy this requirement in work attributing physiological relevance to their NOD activity. However, in the cases of most non-oxygen transport Hbs (including those investigated here), such reductases have not been identified. For example, a “nitric oxide activated deoxygenase” function has been attributed to *Ascaris* Hb based on *in vitro* experiments using NADPH at pH 6.0 to achieve reduction. In this case, these reaction conditions are known to reduce Hbs nonspecifically, and would likely endow several Hbs including Mb with *in vitro* NO scavenging activity. Thus, as a general phenomenon, *in vitro* scavenging of NO and O₂ by Hbs in the presence of a reduction system may give little insight into true physiological function. The ability of *Syn*Hb to substitute for flavoHb in *E. coli* during NO challenge is the only characteristic that distinguishes any of the hxHbs from the others. The lag in growth following GSNO treatment suggests that an endogenous reductase with activity toward *Syn*Hb is expressed in response to this challenge. The question of whether this reductase has a natural role in NO detoxification in *E. coli*, or whether a homologous reductase is present with similar activity in *Synechocystis*, is currently unanswered. Future studies identifying such a reductase would be taken as support for a NOD function for *Syn*Hb but, more importantly, this observation demonstrates the potential use of *E. coli* AG1000 as a background for reductase screening in co-expression experiments. Co-expression of a reductase (or library containing a reductase) along with a Hb in AG1000 could identify Hb/reductase pairs that are capable of replacing FlavoHb, thus providing targets for identification of cognate reductases for Hbs suspected of providing NOD activity in their native hosts. In conclusion, the present study has evaluated the capacity of hxHbs to serve as NOD enzymes by testing their abilities to react with NO in the oxyferrous and ferric states, to scavenge NO in an artificial reduction system where the rate of reduction is controlled experimentally, and to replace flavoHb *in vivo* under aerobic conditions. Our results demonstrate that these Hbs have a common ability to react rapidly with NO in the oxyferrous state, and that they will subsequently scavenge NO at a rate limited by re-reduction. It is also clear that reaction of the ferric hxHbs with NO are probably not of physiological significance. These results are not contradictory to a role in NO scavenging, but neither do they preferentially support this hypothesis for any particular hxHb. Instead, they serve to focus research in this area on identification of cognate reductases for each Hb within their natural environments. # Supporting Information We thank Dr. Paul Gardner from the Division of Critical Care Medicine, Children's Hospital Medical Center (Cincinnati, OH), for the *E. coli* strains and the pANX plasmid. [^1]: Conceived and designed the experiments: MH JT BS. Performed the experiments: BS. Analyzed the data: MH JT BS. Contributed reagents/materials/analysis tools: MH JT BS. Wrote the paper: MH JT BS. Other: Principal Investigator, edited paper, designed, analysed, interpreted experiments: MH. Performed all experiments, wrote main draft of manuscript: BS. Extensive editorial, conceptual modification to manuscript: JT. Analysis and interpretation of data, experimental design: JT. [^2]: Current address: CBR Institute for Biomedical Research, Boston, Massachusetts, United States of America [^3]: The authors have declared that no competing interests exist.

# Introduction Predicting demographic trends (DT) in the light of emerging complex processes of the 21st Century continues to be an important and open research topic. Understanding developments and the changes in population is critical in assisting governments in targeting policies for the future and saving money for education, public health, retirement, transportation, energy consumption among others. Specifically, DT refers to the changes in the joint distribution between population with time, age or other demographic factors, such as household’s size, health measures, economic status, religious affiliation, education, marriage, etc. Forecasting DT is a challenging task, and remains to be a fundamental concern in both basic and applied ecology. The complexity lies in the DT’S intricate connectivity to the heterogeneous activities of a large group of individuals, and it is impacted by observed and unobserved time dependent factors. Existing methods such as the least square methods and Bayesian inference, in spite of being the most extensively used procedures in estimating and predicting various engineering problems, fail to capture the driving mechanisms of complex processes that shapes DT. There are very few literatures on building optimization models for understanding DT. Typical approaches involve incorporating factors such as environmental, demographic and/or observer-related covariates. However, data to support and verify such techniques is often not readily available as– suggesting that building an optimization model constrained by limited data to characterise DT is fundamentally important with a lot of potential applications. Entropy-based methods, the measure of the uncertainty in random variables, have been successfully applied to many modelling and estimation problems, as seen in. In this paper, we introduce the entropy-based method to estimate DT. We build the model motivated by our empirical observation that the age distribution of population follows an increasing entropy trend. The paradigm is based on minimizing the entropy-based objective function and incorporating some parameters describing the historical trends into the constraints where the dynamic and intrinsic properties can be reflected. We illustrate this procedure by estimating the evolution of demographic distributions over ages and household sizes. Our work involves a three-fold modeling stages. Firstly, an “age- structured population model” based on Leslie matrix is used to predict the age distribution of a country’s population. This makes the modelling of the demographic temporal distributions become possible, as one usually needs to project the age distribution of population into other factors. Secondly, the age distribution of each household size is estimated based on a proposed entropy-based model, where we propose an entropy formulated cost function and incorporate the DT into the constraint conditions. The model applied in this stage is called “individual household size model”. Finally, the age distribution of the country’s population (obtained in stage 1) is projected onto the age distributions of individual household size types (obtained in stage 2), which we refer to as “total household size model”. Note that our estimation does not rely on any observed determinant on the formation of households. The evolution of the household size is estimated based on the historical information and the entropy principle. To compare with existing works, our method predicts DT with limited information. The output is a joint distribution of age and other demographic variables over time. Among its applications will be on policy analysis, economic forecasting and urban planning and so on. For the purpose of illustration, we use the population data from US Census and predict the age DT for each household size in 2010, based on the historical data in 2000 and 2006. The remaining parts of the paper are organized as follows. Section 2 lists the definitions and notations which are used throughout the article. Section 3 presents the three stages for the estimation of DT. The simulation results based US data are illustrated in Section 4 and we conclude the article in Section 5. # Methodology ## Notations In the following, we list the definitions and notations that will be used throughout the article: *t*: The year index..^*T*: Matrix transpose. $\hat{.}$: The estimation of a variable. *A*_*upper*: The upper bound age. *i*: The age index (*i* = 0, 1, …, *A*_*upper*). *P*_*i*(*t*): The population for the people at age *i* (older than *i* but younger than *i* + 1) in the year *t*. *P*(*t*) = \[*P*₀(*t*) … *P*_*i*(*t*) … *P*_{*A*_*upper*}(*t*)\]^*T*: The population vector for the people at all ages in the year *t*. *N*_*i*(*m*, *t*): The population for the male at age *i* in the year *t*. *N*_*i*(*f*, *t*): The population for the female at age *i* in the year *t*. *N*(*m*, *t*) = \[*N*₀(*m*, *t*) *N*₁(*m*, *t*) … *N*_{*A*_*upper*}(*m*, *t*)\]^*T*: The population vector for the male at all ages. *N*(*f*, *t*) = \[*N*₀(*f*, *t*) *N*₁(*m*, *t*) … *N*_{*A*_*upper*}(*f*, *t*)\]^*T*: The population vector for the female at all ages. *D*_*i*(*m*, *t*): The death rate for a male at age *i* in the year *t*. *D*_*f*(*i*, *t*): The death rate for a female at age *i* in the year *t*. *B*_*i*(*t*): The fertility rate for a female at age *i* in the year *t*. *Ratio*_*mf*(*t*): The ratio of the number of newly born boys to girls in the year *t*. *Immig*(*m*, *t*): The male immigrants vector in the year *t*. *Immig*(*f*, *t*): The female immigrants vector in the year *t*. *Emig*(*m*, *t*): The male emigrants vector in the year *t*. *Emig*(*f*, *t*): The female emigrants vector in the year *t*. *m*₀: Total number of household sizes. *j*: The household size index (*j* = 1, 2, …, *m*₀). *k*₀: Number of historical years’ data used in the individual household size. *κ*: An index applied on the historical data for the year *t* − *κ* (*κ* = 0, 1, …, *k*₀). *G*_*n*: The people in the age interval \[0 *A*_*n*\] where *A*_*n* is an upper bound age of this group. *n*: The group number index of *G*_*n* (*n* = 1, 2, …, *n*₀) (as seen). *n*₀: Number of groups (*G*_*n*) in the individual household size. $p_{i}^{j}(t)$: The probability (percentage) that people at age *i* in household size *j* in the year *t*. *p*_*i*(*t*) = \[*p*₀(*t*) … *p*_{*A*_*upper*}(*t*)\]^*T*: Age distribution of the population in the year *t*. $p^{j}(t) = \left\lbrack p_{0}^{j}(t)\mspace{600mu}...\mspace{600mu} p_{A_{upper}}^{j}(t) \right\rbrack^{T}:$ Age distribution of household size *j* in the year *t*. $q_{i}^{j}(t - \kappa) = \left\lbrack p_{0}^{j}(t - \kappa)\mspace{600mu}...\mspace{600mu} p_{A_{upper}}^{j}(t - \kappa) \right\rbrack^{T}:$ Age distribution of household size *j* in the year *t* − *κ*. *Entropy*(*t*): The entropy of the population distribution in the year *t*. $\alpha_{n}^{j}(t + 1)$: A ratio of people in group *G*_*n* to the population in household size *j* in. ${\widetilde{\alpha}}_{n}^{j}(t + 1)$: A parameter defined in. {.}_*i*: The vector that contains the values of the variable {.} by changing subscript *i*. $\omega_{0}^{*}$: A weight of the objective function in. *ω*₁, …, *ω*_*k*₀: The weights defined in and. $\xi_{k}^{j}(t + 1):$ An error term in. $\overline{\xi}:$ The upper bound of *ξ*_*k*(*t* + 1) and $\xi_{k}(t + 1) \in \left\lbrack - \overline{\xi},\mspace{600mu}\mspace{600mu}\overline{\xi} \right\rbrack$. *H*: The hessian matrix. *x*^*j*(*t*): The number of household size *j* (*j* = 1, …, *m*₀) in the year *t*. $X(t) = \left\lbrack x^{1}(t)\mspace{600mu}...\mspace{600mu} x^{j}(t)\mspace{600mu}...x^{m_{0}}(t) \right\rbrack^{\prime}:$ The vector contains the number of each household size. *W*: A weighting matrix in the total household size. *τ*: A parameter in the matrix *W* in the total household size model. *u*: A small positive weight parameter in the total household size. $\hat{F}:$ Predicted weighting matrix by collecting the predicted age distributions of all household sizes. ## Three stages for forecasting the demographic trends summarizes the three stages for forecasting the DT. Stage 1: using an “age- structured population model” to predict the population in the year *t* + 1. Stage 2: using an “individual household size model” to estimate the age distribution for each household size *j* based on data in the historical years where the DT reflected in the previous years can be incorporated into the constraint conditions. Stage 3: Combining the results from Stages 1 and 2, and employing a “total household size model” to predict the number of each household size. We detail in the next subsections each of the three stages shown. ## Age-structured population model: for estimating age distribution of the population We consider the population as a summation of all the organisms of the same group or species, who live in the same geographical area, and have the capability of inter-breeding. Quite frequently, the prediction of demographic temporal distributions is highly linked to the population’s age-structure. Demographic temporal distribution modeling is achievable using the “age-structured population model” since it allows projection of the age distribution into other factors. Assumptions. We apply the Leslie matrix method– that assumes: There is no plague, disaster or war that will lead to abrupt changes in age specific death rate. Statistical variables such as birth rates and birth ratio are slowly changed and predictable. The fertility rate for both local residents and immigrants is the same. All people who are older than *A*_*upper* are in the same age group. Here, we set *A*_*upper* = 90. Problem formulation. We first consider the case without immigration and emigration. In the year *t* + 1, the number of people at age *i* + 1 is $$\begin{array}{cl} {P_{i + 1}(t + 1)} & {= N_{i + 1}(m,t + 1) + N_{i + 1}(f,t + 1)} \\ & {= \left\lbrack 1 - D_{i}(m,t) \right\rbrack N_{i}(m,t) + \mspace{600mu}\left( 1 - D_{i}(f,t) \right)N_{i}(f,t)} \\ \end{array}$$ where *t* and *t* + 1 denote the current year and the next year, respectively, and *i* = 0, 1, …, *A*_*upper* − 1 is the age index. When *i* = *A*_*upper*, we have $$\begin{array}{cl} {P_{i \geq A_{upper}}(t + 1) =} & {\mspace{600mu}\mspace{600mu}\left\lbrack 1 - D_{A_{upper} - 1}(m,t) \right\rbrack\mspace{180mu} N_{A_{upper} - 1}(m,t) + \left\lbrack 1 - D_{A_{upper} - 1}(f,t) \right\rbrack\mspace{180mu} N_{A_{upper} - 1}(f,t)} \\ & {+ \left\lbrack 1 - D_{i \geq A_{upper}}(m,t) \right\rbrack\mspace{180mu} N_{i \geq A_{upper}}(m,t) + \left\lbrack 1 - D_{i \geq A_{upper}}(f,t) \right\rbrack\mspace{180mu} N_{i \geq A_{upper}}(f,t)} \\ \end{array}$$ Let \[*i*₁, *i*₂\] be the age interval that a female has the ability to give birth. Then, *P*₀(*t* + 1) = *N*₀(*m*, *t* + 1) + *N*₀(*f*, *t* + 1) and $$\begin{array}{ll} {N_{0}(m,t + 1)} & {= \frac{Ratio_{mf}(t)}{1 + Ratio_{mf}(t)}\sum\limits_{i = i_{1}}^{i = i_{2}}B_{i}(t)\mspace{180mu} N_{i}(f,t)} \\ {N_{0}(f,t + 1)} & {= \frac{1}{1 + Ratio_{mf}(t)}\sum\limits_{i = i_{1}}^{i = i_{2}}B_{i}(t)\mspace{180mu} N_{i}(f,t)} \\ \end{array}$$ where *Ratio*_*mf*(*t*) is a ratio of the newly born boys (*N*₀(*m*, *t*)) to the newly born girls (*N*₀(*f*, *t*)) at year *t*. Let $$\begin{array}{ll} {P(t)} & {= \left\lbrack P_{0}(t)\mspace{600mu} P_{1}(t)\mspace{600mu}...\mspace{600mu} P_{A_{upper}}(t) \right\rbrack^{T}} \\ {N(m,t)} & {= \left\lbrack N_{0}(m,t)\mspace{600mu} N_{1}(m,t)\mspace{600mu}...\mspace{600mu} N_{A_{upper}}(m,t) \right\rbrack^{T}} \\ {N(f,t)} & {= \left\lbrack N_{0}(f,t)\mspace{600mu} N_{1}(f,t)\mspace{600mu}...\mspace{600mu} N_{A_{upper}}(f,t) \right\rbrack^{T}} \\ \end{array}$$ be the vectors of the population, male population and female population, respectively, for ages between 0 and *A*_*upper* at year *t*. Next, we extend the model to take into account of immigration effects. Let *Immig*(*m*, *t*)/*Emig*(*m*, *t*) and *Immig*(*f*, *t*)/*Emig*(*f*, *t*) be the respective immigrants and emigrants vector for males/females at year *t*. We obtain the “age-structured population model” as follows: $$\begin{array}{cl} {P(t + 1)} & {= N(m,t + 1) + N(f,t + 1)} \\ {N(m,t + 1)} & {= A(t)N(m,t) + \frac{Ratio_{mf}(t)}{1 + Ratio_{mf}(t)}B(t)N(f,t)} \\ & {\mspace{600mu}\mspace{600mu} + Immig(m,t) - Emig(m,t)} \\ {N(f,t + 1)} & {= \left\lbrack C(t) + \frac{1}{1 + Ratio_{mf}(t)}B(t) \right\rbrack N(f,t)} \\ & {\mspace{600mu}\mspace{600mu} + Immig(f,t) - Emig(f,t)} \\ \end{array}$$ where *A*(*t*), *B*(*t*) and *C*(*t*) are the matrices constructed based on Eqs–, and given by $$A(t) = \begin{bmatrix} 0 & 0 & 0 & \ldots & 0 & 0 & 0 \\ {1 - D_{0}(m,t)} & 0 & 0 & \ldots & 0 & 0 & 0 \\ 0 & {1 - D_{1}(m,t)} & 0 & \ldots & 0 & 0 & 0 \\ \ldots & \ldots & \ldots & \ldots & \ldots & \ldots & \ldots \\ 0 & 0 & 0 & \ldots & 0 & 0 & 0 \\ 0 & 0 & 0 & \ldots & 0 & {1 - D_{A_{upper} - 1}(m,t)} & {1 - D_{i \geq A_{upper}}(m,t)} \\ \end{bmatrix}$$ $$B(t) = \begin{bmatrix} 0 & \ldots & 0 & {B_{i_{1}}(t)} & \ldots & {B_{i_{2}}(t)} & 0 & \ldots & 0 \\ 0 & \ldots & 0 & 0 & \ldots & 0 & 0 & \ldots & 0 \\ \ldots & \ldots & \ldots & \ldots & \ldots & \ldots & \ldots & \ldots & \\ 0 & \ldots & 0 & 0 & \ldots & 0 & 0 & \ldots & 0 \\ \end{bmatrix}$$ $$C(t) = \begin{bmatrix} 0 & 0 & \ldots & 0 & \ldots & 0 & \ldots & 0 \\ {1 - D_{0}(f,t)} & 0 & \ldots & 0 & \ldots & 0 & \ldots & 0 \\ 0 & {1 - D_{1}(f,t)} & \ldots & 0 & \ldots & 0 & \ldots & 0 \\ \ldots & \ldots & \ldots & \ldots & \ldots & \ldots & \ldots & \ldots \\ 0 & 0 & \ldots & 0 & \ldots & 0 & {1 - D_{A_{upper} - 1}(f,t)} & {1 - D_{i \geq A_{upper}}(f,t)} \\ \end{bmatrix}$$ Note that the population data we collected allows us to estimate the values of all the above parameters (such as the fertility rates and death rates). These parameters change slowly and are predictable which confirm the validity of our assumption. Thus, the population distribution for the coming years can be predicted based on the age-structured population, and its estimation is denoted as $\hat{P}(t + 1)$ for the year *t* + 1 as shown in later. ## Individual household size model: for estimating age distribution for each household size In this section, we will describe in detail our *individual household size model* that estimates the age distribution of each household size. The model is operated by minimizing an entropy based objective function and using the historical trends as constraints, where both the dynamic and intrinsic properties are reflected. Let *p*_*i*(*t*) be the probability that a person is at age *i* in year *t*. We define an entropy function for year *t* as follows: $$\begin{array}{l} {Entropy(t) = - \sum_{i = 0}^{A_{upper}}p_{i}(t)ln\left( p_{i}(t) \right)} \\ \end{array}$$ where $\Sigma_{i = 0}^{A_{upper}}p_{i}(t) = 1$ and *p_i*(*t*) ≥ 0. plots the entropy of the age distribution based on the population data collected from six countries. In general, the entropy of the age distribution increases monotonically with respect to time in most countries. This observation suggests that we can estimate the age distribution of a particular household size based on entropy concepts. To this end, we divide the household size into *n*₀ types: i.e., 1 person per household, 2 persons per household, …, until *n*₀ persons per household. Let *j* be the household size index and assume that we already have the age distributions for each household size *j* (*j* ∈ {1, …, *m*₀}) in the years *t*, *t* − 1, …, *t* − *k*₀, which are denoted as $q_{i}^{j}(t - \kappa)$ for *κ* = 0,1, …, *k*₀. Let $p_{i}^{j}(t + 1)$ represent the percentage of persons at age *i* in household size *j* in the year *t* + 1. This means we group the people whose ages are above 90 years together. Our objective is to estimate the age distribution $p_{i}^{j}(t + 1)$ in the year *t* + 1 based on the historical data. We group the people from 0 to *A*_*upper* years old into *n*₀ groups, i.e., the groups *G*_*n* for *n* = 1, …, *n*₀, where *n*₀ ≪ *A*_*upper*. The age interval for the group *G*_*n* is \[0, *A*_*n*\] and 0 \< *A*₁ \< *A*₂ \< … \< *A*_*n*₀ = *A*_*upper*. It is easy to see that *G*_*n*−1 ⊂ *G*_*n*. Define $\alpha_{n}^{j}(t)$ as a parameter such that $$\begin{array}{l} {\alpha_{n}^{j}(t) = \sum_{i = 0}^{A_{n}}p_{i}^{j}(t)} \\ \end{array}$$ which means that $\alpha_{n}^{j}(t)$ is a ratio of people in group *G*_*n*, i.e., in the age interval \[0, *A*_*n*\], to the population in household size *j*. Note that ∀*j* ∈ {1, …, *m*₀}, *α*_*n*₀(*t* + 1) = 1 since *A*_*n*₀ = *A*_*upper*. Let $$\begin{array}{l} {{\widetilde{\alpha}}_{n}^{j}(t + 1) = \alpha_{n}^{j}(t + 1) - \alpha_{n}^{j}(t)} \\ \end{array}$$ be the parameter which reflects the percentage change of the ratio $\alpha_{n}^{j}(t)$ from the year *t* to the next year *t* + 1. From here, we build an individual household size model to predict the age distribution $\left\{ p_{i}^{j}(t + 1) \right\}_{i}$ for each household size type *j* where *j* = 1, 2, …, *m*₀, by optimizing the following: $$\begin{array}{cl} {min} & {\sum_{\kappa = 0}^{k_{0}}\omega_{\kappa + 1}\sum_{i = 1}^{A_{upper}}p_{i}^{j}(t + 1)ln\left( \frac{p_{i}^{j}(t + 1)}{q_{i}^{j}(t - \kappa)} \right)} \\ & {+ \omega_{0}^{*}\sum_{i = 1}^{A_{upper}}p_{i}^{j}(t + 1)ln\left( \frac{p_{i}^{j}(t + 1)}{1/A_{upper}} \right)} \\ {s.t} & {\sum_{i = 0}^{A_{1}}p_{i}^{j}(t + 1) \leq \sum_{i = 1}^{A_{1}}q_{i}^{j}(t) + {\widetilde{\alpha}}_{1}^{j}(t + 1)} \\ & {\mspace{600mu}\mspace{600mu}\mspace{ 600mu}\mspace{600mu}\mspace{600mu}\mspace{600mu}\mspace{600mu}\mspace{600mu}\msp ace{600mu}\mspace{600mu}\mspace{600mu}\mspace{600mu}\mspace{600mu}\mspace{600mu} \mspace{600mu}\mspace{600mu} \vdots} \\ & {\sum_{i = 0}^{A_{n}}p_{i}^{j}(t + 1) \leq \sum_{i = 1}^{A_{n}}q_{i}^{j}(t) + {\widetilde{\alpha}}_{n}^{j}(t + 1)} \\ & {\mspace{600mu}\mspace{600mu}\mspace{600mu}\mspace{600mu}\mspace{600mu}\mspace {600mu}\mspace{600mu}\mspace{600mu}\mspace{600mu}\mspace{600mu}\mspace{600mu}\ms pace{600mu}\mspace{600mu}\mspace{600mu}\mspace{600mu}\mspace{600mu} \vdots} \\ & {\sum_{i = 0}^{A_{n_{0}}}p_{i}^{j}(t + 1) \leq \sum_{i = 1}^{A_{n_{0}}}q_{i}^{j}(t) + {\widetilde{\alpha}}_{n_{0}}^{j}(t + 1)} \\ & {\sum_{\kappa = 1}^{k_{0} + 1}\omega_{\kappa} = 1} \\ & {- p_{i}^{j}(t + 1) \leq 0\mspace{600mu}\mspace{600mu} for\mspace{600mu}\mspace{600mu} i = 0,1....,A_{upper}} \\ \end{array}$$ Again, given that the entropy of the population is monotonically increasing with time, we can minimize an entropy based cost function under some constraints by employing the historical data. Compared with, we omit the minus sign “−” such that the model becomes a minimization problem. The upper limit of such entropy as *t* = +∞ is a uniform distribution with a histogram function having a constant 1/*A*_*upper* magnitude. Essentially, there are two parts in this cost function where $\omega_{0}^{*}$ is a small positive weight parameter. The first part is the cross entropy distance (KL distance) between $\left\{ p_{i}^{j}(t + 1) \right\}_{i}$ and the historical data, and the second part is the relative entropy distance between $\left\{ p_{i}^{j}(t + 1) \right\}_{i}$ and population distribution when *t* = +∞. Note that we can never know the value of ${\widetilde{\alpha}}_{n}^{j}(t + 1)$ at the year *t* as we do not know $\alpha_{n}^{j}(t + 1)$. However, it can be estimated from the historical data as: $$\begin{array}{cl} {{\hat{\widetilde{\alpha}}}_{n}^{j}(t + 1)} & {= \sum_{\kappa = 1}^{k_{0}}\omega_{\kappa}\left( \alpha_{n}^{j}(t - \kappa) - \alpha_{n}^{j}(t) \right)/\kappa} \\ & {= \sum_{\kappa = 1}^{k_{0}}\omega_{\kappa}\left( \sum_{i = 0}^{A_{n}}q_{i}^{j}(t - \kappa) - \sum_{i = 0}^{A_{n}}q_{i}^{j}(t) \right)/\kappa} \\ \end{array}$$ where *ω*_*κ* for *κ* = 1, …, *k*₀ + 1 are decreasing weights, which implies that the more recent data is more valued. Let $\xi_{n}^{j}(t + 1) = \frac{{\widetilde{\alpha}}_{n}^{j}(t + 1) - {\hat{\widetilde{\alpha}}}_{n}^{j}(t + 1)}{{\hat{\widetilde{\alpha}}}_{n}^{j}(t + 1)}$ be an error term of the estimation, then we have $$\begin{array}{l} {{\widetilde{\alpha}}_{n}^{j}(t + 1) = {\hat{\widetilde{\alpha}}}_{n}^{j}(t + 1)\left( 1 + \xi_{n}(t + 1) \right)} \\ \end{array}$$ the distribution of $\xi_{n}^{j}(t + 1)$ is known and bounded within $\left\lbrack - \overline{\xi},\overline{\xi} \right\rbrack$. Usually $\xi_{n}^{j}(t + 1)$ can be assumed as a random variable uniformly distributed in $\left\lbrack - \overline{\xi},\overline{\xi} \right\rbrack$. We now have that: Theorem 1. The optimization problem defined in is a strict convex optimization. Proof. Note that the Hessian matrix *H* of the objective function is given by: $$\begin{array}{l} {H = \left\lbrack \begin{array}{cllll} \frac{\sum_{\kappa = 0}^{k_{0}}\omega_{\kappa + 1}}{p_{1}^{j}(t + 1)} & 0 & 0 & {...} & 0 \\ 0 & \frac{\sum_{\kappa = 0}^{k_{0}}\omega_{\kappa + 1}}{p_{2}^{j}(t + 1)} & 0 & {...} & 0 \\ & {\vdots \vdots} & \vdots & \ddots & \vdots \\ 0 & 0 & 0 & {...} & \frac{\sum_{\kappa = 0}^{k_{0}}\omega_{\kappa + 1}}{p_{A_{upper}}^{j}(t + 1)} \\ \end{array} \right\rbrack} \\ \end{array}$$ Since $p_{i}^{j}(t + 1) \geq 0$ for all *i* and *j*, it is easy to see that *H* is a positive definite matrix. On the other side, it is known that the constraints of the optimization problem in the are linear. Therefore, the feasible domain is a convex set. Both the objective function and the feasible domain are convex, hence the problem is a convex optimization. Note that one only needs to find a local minimum point of a convex optimization to obtain the global minimum point. ## Total household size model: for estimating the number of each household size In this section, we build a *total household size model* to further estimate the number of each household size *j* for *j* = 1, 2, …, *m*₀ based on the predicted age distribution of population and age distribution of each individual household size. Here, our objective is to estimate the number of household size *j* for *j* = 1, 2, …, *m*₀ in the year *t* + 1. Let *x*^*j*(*t*) be the number of household with size *j* in the year *t* and denote that $X(t) = \left\lbrack x^{1}(t)\mspace{600mu}...\mspace{600mu} x^{m_{0}}(t) \right\rbrack^{T}$. We hope to estimate the vector $X(t + 1) = \left\lbrack x^{1}(t + 1)\mspace{600mu}...\mspace{600mu} x^{m_{0}}(t + 1) \right\rbrack^{T}$. As mentioned, the first stage is to obtain the estimated total population distribution $\hat{P}(t + 1)$ based on the current fertility rate and death rate. The second stage is then to obtain the estimated age distribution of each household type *j* denoted as ${\hat{p}}^{j}(t + 1)$. Now we estimate the household number distribution by solving the following total household size model: $$\begin{array}{ll} {min} & {(1 - u) \cdot {||}\hat{F} \cdot X(t + 1) - \hat{P}{\left. (t + 1) \right)\left. | \right|}^{2} + u \cdot \left| \middle| W \right. \cdot X(t + 1) - {\left. X(t) \right)\left. | \right|}^{2}} \\ {s.t} & {X(t + 1) > 0} \\ \end{array}$$ where ∣∣.∣∣ is the *L*₂ norm, and *X*(*t* + 1) ≥ 0 means each component of *X*(*t* + 1) is nonnegative, and $\hat{F}$ is a weighting matrix collected from the the predicted age distributions of all household sizes: $$\hat{F} = \begin{bmatrix} {1 \cdot {\hat{p}}_{1}^{1}(t + 1)} & {1 \cdot {\hat{p}}_{2}^{1}(t + 1)} & \ldots & {1 \cdot {\hat{p}}_{A_{upper}}^{j}(t + 1)} \\ {2 \cdot {\hat{p}}_{1}^{2}(t + 1)} & {2 \cdot {\hat{p}}_{2}^{2}(t + 1)} & \ldots & {2 \cdot {\hat{p}}_{A_{upper}}^{2}(t + 1)} \\ \vdots & \vdots & \ldots & \vdots \\ {m_{0} \cdot {\hat{p}}_{1}^{m_{0}}(t + 1)} & {m_{0} \cdot {\hat{p}}_{2}^{m_{0}}(t + 1)} & \ldots & {1 \cdot {\hat{p}}_{A_{upper}}^{m_{0}}(t + 1)} \\ \end{bmatrix}$$ and *W* is a diagonal weighting matrix for different household size type given by $$W = \begin{bmatrix} 1 & 0 & \ldots & 0 \\ 0 & 2^{\tau} & \ldots & \vdots \\ \vdots & \vdots & j^{\tau} & \vdots \\ 0 & 0 & \ldots & m_{0}^{\tau} \\ \end{bmatrix}$$ The above objective function contains two parts with *u* being a small positive weight parameter. The first part is the distance between the estimated age distribution for population and the accumulative of the age distribution for all household sizes. The other part is the weighted distance of the estimated *X*(*t* + 1) (denoted as $\hat{X}(t + 1)$) to *X*(*t*). As there are *j* persons in the household size *j*, we construct a diagonal weighting matrix *W* with a given power *τ* \> 0 in. As shown in Theorem 2, the optimization of is also convex. Theorem 2. The optimization problem defined in is convex. Prof. The proof is similar to Theorem 1. The Hessian matrix of the objective function in is $$\begin{array}{l} {H = (1 - u){\hat{F}}^{T}\mspace{180mu}\hat{F} + uW^{T}W} \\ \end{array}$$ Obviously *H* is a positive definite matrix and we have this theorem holds. # Simulations In this section, we illustrate the procedure we have discussed above using the US’s Census population data. We predict the demographic distribution in the year 2010 based on the historical data in years 2000 and 2006. The prediction is then compared with the actual Census data in the year 2010. We show that the method we described here accurately captures the actual statistics. As mentioned, there are three stages in the estimation: Stage 1: Estimating the age distribution of the population by employing the *age structure based population model* in Section 3A. Stage 2: Estimating age distribution for each household size type by employing the *individual household size model* in Section 3B. Stage 3: Estimating the number of different household size type by employing the *total household size model* in Section 3C. In Stage 1, we collect the population data from the US Census and get the values of all parameters that are required in. By solving this model, we obtain the estimation of the population in the year 2010 based on the the data in the year 2000 and 2006 in. In Stage 2, by letting *ω* = \[0.95 0.025 0.025\] and assuming that the error term bound $\overline{\xi} = 0$, we divide the population into 9 groups (*G*_*n*, *n* = 1, 2, …, 9) and let *A*_*n* = *n*⋅10. By solving the individual household size, the age distributions for the household sizes *j* = 1 and *j* = 2, …, 7 are obtained in Figs –, respectively. It is seen that the individual household size model predicts accurately the age distribution of all household sizes. In stage 3, the numbers of each household size are estimated by solving the total household size. As seen in, the difference between the estimation and the real values is quite close, which again shows the accuracy of our proposed method. In addition, we also look at the cases when the error term bound $\overline{\xi} \neq 0$. Let’s say, $\overline{\xi} = 2$ implying that ${\overline{\xi}}_{k}(t + 1)$ is randomly distributed in the interval \[−2 2\]. By repeating the simulations many times over (200 times in our case), we can verify the robustness of our estimations. We take the household size 1 as an example. As seen in, most of the real values of the age distribution are located inside the red area generated by the estimations. To investigate how the parameters *τ* and *u* affect the performance of the estimation, we set different values for them and see how the error term $Error\left( \hat{X}(t + 1) \right) = \frac{\mathop{\text{||}}\hat{X}\left. (t + 1) - X(t + 1) \right.||_{1}}{\left. ||X(t + 1) \right.||_{1}}$ changes, where ‖.‖₁ denotes the *L*₁ norm. shows the error term against different values of *τ*. We observed that when *τ* is around 0.5, the proposed algorithm achieves its best performance. This is reasonable since the diagonal elements *W*^*T* *W* is just the number of persons in each household size when *τ* = 0.5. On the other hand, shows how the parameter *u* impacts the estimation error. By setting different values for *u* in the interval \[0 0.2\], we observe that the performance of the algorithm provides the best fit when *u* ∈ \[0.005 0.01\]. # Discussion and Conclusions In this paper, we have demonstrated a new method that estimates the development of age and household’s size distributions. The procedure consists of three models in three coupled stages, we referred to as: *the age-structured population model* in stage 1 where the age distribution of countries’ population was predicted; *the individual household size model* in stage 2 where the age distribution of each individual household size was estimated; and *the total household size model* in stage 3 where the number of different household sizes was derived by projecting the age distribution of total population onto the age distributions of individual household sizes. The procedure described here indicates that demographic trends can be accurately estimated using entropy as an optimisation variable, which we believe will be of potential interest to both academics and practitioners alike. We have illustrated and validated the correctness and accuracy of the proposed method using US data. While we have considered age and household size distributions in this article, we note that the method we have demonstrated is general and versatile enough to be extended to other time dependent demographic variables. This work is supported by the Integrated City Planning Programme, SERC, A\*STAR (Grant no. 1325000001), and Complex Systems Programme, SERC, A\*STAR (Grant no. 1224504056). [^1]: The authors declare no competing interests. [^2]: Conceived and designed the experiments: GL DZ YX CM. Performed the experiments: GL SHK HYX NH. Contributed reagents/materials/analysis tools: GZ CM. Wrote the paper: GL DZ YX SHK HYX NH GZ CM.

# Introduction By the year of 2016, it is estimated that a total of 37 million people living with HIV/AIDS (PLWHA) worldwide. With the advancement and availability of HARRT, HIV/AIDS has transited from a lethal to a manageable chronic disease. However, the prognosis differed significantly based upon PLWHA’s socio-economic status (SES). SES is a multi-dimensional and complex measure that has been defined variously in history, from a measurable to an abstract construct that reflects one’s access to collectively desired resources. In the health related studies, SES is usually measured by educational attainment, occupation, and income levels. Existing studies have indicated that low SES may be associated with higher risk of substance use, and the association with mental health was inconclusive. On the other hand, studies have indicated the PLWHA with higher SES would have better prognosis as they usually possessed more intangible (e.g., social support, access to health care) and tangible resources (e.g., medication, food supplies, housing) compared to their peers who were at lower SES. In addition to PLWHA’s prognosis, studies have highlighted the importance of addressing PLWHA’s psychosocial needs and behaviors of substance use, as those with psychosocial comorbidities and problems of substance use are at a higher risk of sexual disinhibition, not adhering to ART, having lower quality of life and encountering other health problems, which might negatively influence the social and individual well-beings. In addition to their psychosocial distress, stigmatized experience is commonly encountered in the life context of PLWHA. Stigma was defined as a discrediting attribute of a given individual as a result of the possession of socially devalued marks. In a continuous social process and complex social interactions including labeling, stereotyping, separation, status loss, and discrimination, the possessor of the socially devalued marks has been stigmatized. Following a similar process, HIV stigma against PLWHA has been developed and maintained within different social and cultural contexts. Scholars have revealed a continuum of devastating consequences (physical, emotional and financial burdens) of HIV stigma against PLWHA in China and other settings. In the current study, we aimed to examine SES stratum specific associations between different types of HIV stigma and psychosocial well-being as well as behaviors of substance use among a group of PLWHA recruited from Chinese settings. Our hypotheses are that HIV stigma is positively associated with psychological distress and problems of substance use among PLWHA, and such impacts may interact with different SES gradients. # Methods ## Study site The current study was conducted in Guangxi Zhuang Autonomous Region (Guangxi) in China. Guangxi is located in the Southwest of China with a total population of 50 million with 32% of Zhuang-ethnic and 8% of other minorities. HIV epidemic has been surging in Guangxi. By the end of 2014, Guangxi has ranked first among all 31 provinces in terms of cumulative HIV seropositive cases. ## Study design Details of the study protocol and the pre-established sampling scheme were documented elsewhere. Briefly, from 2012 to 2013, we collaborated with Guangxi Center for Disease Control and Prevention (Guangxi CDC) to target a total of 29,606 HIV/AIDS cases from the 12 sites in Guangxi, including two cities and ten counties with the highest number of cumulative HIV incidence. We randomly selected approximately 10% of cases from the sampling pool and finally recruited 3,002 HIV patients in the study, with about 10% refusal rate. Among the 3,002 patients, 2,987 (99.5%) of them completed the cross-sectional survey and were included in the current analysis. Specifically about 80% of the included participants completed the questionnaire assisted by the well-trained local CDC staff and health care workers. The rest of the sample completed the questionnaire on their own. Written consent was obtained from all participants. The Institutional Review Boards at Wayne State University in the United States and Guangxi CDC in China reviewed and approved the research protocol and the consent procedure. ## Measurements ### Demographics Demographic information included participants’ gender (male vs. female), age (in years), years of school (in years), ethnicity (Han, Zhuang, or others), religious beliefs (e.g., no-religious, Buddhism, and others), marital status (e.g., never vs. ever married), job categories (e.g., peasant workers, service staff, self-employed, unemployed, and others), residence registration (a.k.a. Hukou, urban vs. rural), residence types of the current place (urban vs. rural) and monthly income (in Chinese currency *Yuan*). Following the guideline from the Department of Health & Human Service (DHHS), we used educational attainment, income levels, job category and residence type as indicators for PLWHA’s SES. People’s education attainment was measured by their reported “years of school”. Two groups (e.g., at least 9 years, and more than 9 years) based upon China’s Compulsory Education Law were generated. PLWHA’s monthly income levels were further categorized into four groups based upon data distribution (e.g. “≤999”, “1000–1999”, “2000–2999”, and “≥3000”). ### Exposure variables We employed the validated Berger HIV Stigma Scale to measure three types of stigma: perceived, internalized, and enacted stigma. Each stigma item was measured by a Likert-type scale (e.g., strongly disagree, disagree, agree, strongly agree) with a higher values indicating a greater agreement with the statement. Perceived stigma (α = 0.91) was measured by a few questions related to awareness of societal norms and prejudicial actions towards PLWHA (e.g., “most people consider PLWHA filthy”). Internalized stigma (α = 0.92) was evaluated by their negative feelings about oneself because of contracting with HIV (e.g., I feel guilty because I have HIV). Enacted stigma (α = 0.63) was measured by questions regarding discriminatory experience that the PLWHA have encountered (e.g., I will lose my job if my sero-status is known by others). The overall Cronbach’s α was 0.927 for the 16-item stigma scale in the current study. For the purpose of data analysis, we divided the total stigma score by its quartiles in order to assess if there were any trends of outcome variables with the increased stigma scores. Outcome variables Indicators for psychosocial well-being included assessments of depression (Shorten version of Center for Epidemiological Studies Depression Scale for Children \[CESD-10\] with 10 item; α = 0.76), anxiety (Zung Self-Rating Anxiety Scale \[SAS\] with 20 items; α = 0.91), HIV management related self-esteem (α = 0.94), resilience (The 10-item Connor-Davidson Resilience Scale with 10 items; α = 0.96), social support (The Multidimensional Scale of Perceived Social Support \[MSPSS\], and The Medical Outcomes Study Social Support Survey \[MOS-SSS-C\] with 28 items; α = 0.98), and coping skills (Family Coping Project Coping Scale \[FCPCS\] with 25 items; α = 0.93). These assessment scales have been validated in previous studies conducted in Chinese settings. Substance use behaviors were measured by questions asking if participants ever had used tobacco, drug or alcohol (yes vs. no) in the past six months. If participants answered “yes” to any of the questions, they would be coded as “ever”, otherwise, “never” was coded. ## Analytical plan Several analytic procedures were employed to assess how stigma affected psychosocial well-being and substance use behaviors. First, we compared the demographic characteristics, psychosocial well-being indicators and substance use behaviors using ANOVA (for continuous variables) and Chi-squares tests (for categorical variables) among PLWHA at different SES gradients. Second, we explored how different types of HIV stigma impacted psychosocial well-being and substance use behaviors using multivariate regression analyses while adjusting for association-specific confounders. In addition, stratified analyses by different SES gradients (e.g., years of education, job category, residence type and monthly income) were further conducted. All data analyses were conducted using Stata 12.0® (StataCorp LP, College Station, Texas, USA). # Results ## Characteristics of participants In the current sample, majority of participants were of Han-ethnicity (70.8%), married (66.5%), without a religious belief (92.2%), having rural residence registration (84.4%). Of these participants, 2391 (80.2%) living in rural areas, 1748(58.5%) working as peasants, 2,584 (86.5%) having at most nine-year education, and more than half (53.2%) having monthly income less than 1,000 yuan (one USD = 6 yuan at the time of survey). ## Psychosocial wellbeing among PLWHA at different SES gradients There were significant differences by SES gradients for most indicators of psychosocial well-being and substance use behaviors. Specifically, PLWHA with longer years of school, higher income levels, living at urban areas and working as “others” showed significantly better psychosocial wellbeing, less drug use behaviors, and less perceived and internalized stigma (*p*\<0.05). ## Impacts of HIV stigma The overall multivariate analyses revealed that each type of HIV stigma is positively associated with psychosocial distress (e.g., depression and anxiety), but negatively related to protective buffers (e.g., resilience and self-esteem) of PLWHA across all SES gradients with enacted stigma having the strongest impact. For instance, compared to perceived (β = 0.20, 95% CI = 0.11, 0.28)) and internalized stigma (β = 0.55, 95% CI = 0.49, 0.62), the magnitudes of having anxiety problems was much higher among PLWHA who reported enacted stigma (β = 3.80, 95% CI = 3.18, 4.41). Similarly, the odds of smoking (aOR = 1.35, 95% CI = 1.04, 1.76) were significantly higher among PLWHA encountering enacted stigma compared to these who reported other stigma types (for perceived stigma: aOR = 0.99 \[95% CI = 0.96, 1.02\], for internalized stigma: aOR = 1.00 \[aOR = 0.97, 1.02\]). The same pattern has been observed for other indicators of psychosocial wellbeing and behaviors of substance use among PLWHA. When we examined the HIV stigma by SES-gradients, our findings indicated that impact of HIV stigma varied by different indicators of SES. PLWHA with longer years of schooling, stigma usually had worse impact on their psychosocial wellbeing status and problems of substance use, while for people with higher income, the impact of the least compared to their peers with lower incomes. For instance, for PLWHA who had shorter years of schooling, the impact of stigma on depression, resilience, coping and social support was at a lower degree compared to people with longer years of schooling, while the impact was stronger for coping and social support among PLWHA with less school education. For the impact on substance use, enacted stigma increased the odds of smoking (aOR = 1.35, 95% CI = 1.01, 1.82) among PLWHA with short years of schooling, but the odds of alcohol use got decreased for PLWHA experiencing perceived (aOR = 0.97, 95% CI = 0.94, 1.00) and internalized stigma (aOR = 0.97, 95% CI = 0.95, 0.99). For PLWHA with more than 3000 yuan monthly income, the impact of stigma on most indicators for psychosocial wellbeing was insignificant. However, among PLWHA with highest income, enacted stigma increased the odds of smoking (aOR = 5.78, 95% CI = 1.77, 18.85) and alcohol use (aOR = 2.40, 95% CI = 1.03, 5.60); internalized stigma increased the odds of drug use (aOR = 1.16, 95% CI = 1.02, 1.33). For PLWHA with less than 999 yuan monthly income, the impact of stigma on psychosocial status was strongest in terms of the magnitude of the odds compared to their high-income peers. For PLWHA living in rural areas, enacted stigma had stronger impact on their depression (β = 1.68, 95% CI = 1.28,2.08), resilience (β = -0.15, 95% CI = -0.22, -0.08), self-esteem (β = -0.20, 95% CI = -0.26,-0.14) and social support (β = -0.11, 95% CI = -0.18, -0.04) compared to people living in urban areas, but not for any substance use behaviors (p\>0.05). On the other hand, perceived stigma increased the coefficients at a greater magnitude for PLWHA living in urban areas on their anxiety (β = 0.32, 95% CI = 0.17, 0.48), depression (β = 0.26, 95%CI = 0.17, 0.34) and coping skills (β = 0.02, 95% CI = 0.00, 0.03) compared with their peers living in rural areas. Only internalized stigma decreased the odds of alcohol use among PLWHA living in both urban (aOR = 0.96, 95%CI = 0.93, 1.00) and rural areas (aOR = 0.97, 95%CI = 0.95, 0.99). Perceived stigma increased the odds of drug use (aOR = 1.06, 95% CI = 1.02, 1.11), but decreased the odds of alcohol use (aOR = 0.97, 95%CI = 0.94, 0.99) among PLWHA living in rural areas. For PLWHA working as unemployed or peasant workers, their psychosocial status was most impacted by HIV stigma. Among unemployed PLWHA, the internalized stigma impacted the odds of anxiety (β = 0.65, 96%CI = 0.43, 0.87), depression (β = 0.33, 96%CI = 0.22, 0.44), resilience (β = -0.06, 96%CI = -0.08,-0.04), coping (β = 0.03, 96%CI = 0.01, 0.04), and social support (β = -0.03, 96%CI = -0.05,-0.01) to the greatest magnitude. While for PLWHA working as peasant workers, their drug use behavior (aOR = 1.06, 95%CI = 1.01, 1.12) was impacted significantly by their encountered perceived stigma, and their alcohol use was negatively affected by their experienced perceived (aOR = 0.96, 95% CI = 0.93, 0.99) and internalized stigma (aOR = 0.97, 95% CI = 0.94, 1.00). # Discussion Our analyses revealed that stigma significantly impacted PLWHA’s psychosocial distress and behaviors of substance use, and the impacts varied by different SES gradients. Specifically, PLWHA with higher education may be more sensitive compared with these who had lower educational attainment. As a result, the magnitude of each type of stigma was higher in the more educated group. On the other hand, PLWHA with higher income may play as a buffer from suffering stigma, therefore, they were least impacted by HIV stigma. However, the enacted stigma had the highest odds on self-esteem, smoking and alcohol use among PLWHA with the highest level of monthly income. Perhaps PLWHA with higher income were more sensitive to their stigmatized experience (e.g., enacted stigma), but less likely to generate negative opinion towards themselves (e.g., internalized and perceived stigma). However, PLWHA with higher income were more likely to report higher likelihood of substance use than their peers with lower incomes. It is perhaps that PLWHA with higher income were more likely to afford their expense of substances compared to people with lower income. Stratified analyses by the SES gradients provided a detailed profile for PLWHA with different SES were impacted by HIV stigma with various magnitudes, although the overall analyses can provide a general direction. For instance, the overall analyses revealed enacted stigma increased the odds of having anxiety problem among PLWHA by 3.8 times after controlling for potential confounders. Our stratified analyses indicated PLWHA with lower education and least income (e.g., \<999 RMB/month) were impacted to the greatest extend by enacted stigma compared to their peers. Although the overall analyses revealed non-significant odds of drinking alcohol among PLWHA who encountered enacted stigma, our stratified analyses indicated that enacted stigma increased the odds of drinking alcohol for PLWHA with monthly income more than 3,000 RMB by 2.4 times, and increased the odds for PLWHA who were self-employed by 1.91 times. If no stratified analyses had been conducted, we may conclude that enacted stigma had least impact on PLWHA’s drinking behaviors. The differences between overall and stratified analyses can be observed throughout other indicators for psychosocial status and substance use in the current study. Such findings aimed to remind health professionals of having a pair of keen eyes to observe subtle differences among PLWHA with different SES backgrounds in order to design tailored interventions. Consistent with existing studies, the psychosocial distress and substance use behaviors were more common among individuals with lower SES or living in lower- graded neighborhoods. People having higher SES usually meant they possess more material (e.g., money, house), human (e.g., skills, knowledge), and social capital (e.g., social support, social network) compared to people with lower SES. PLWHA with higher SES usually had lower mortality rate and better prognosis than their low-SES counterparts. How the SES interacted with the association between HIV stigma and psychosocial problems was complex and required a more sophisticated study design to explore the mechanisms. Our study is the first paper to document the paucity of SES stratified analyses between HIV stigma and psychosocial status among PLWHA in China. A few caveats should be acknowledged when interpreting findings in the current study. First, the nature of cross-sectional design constrained our capacity to make any causal inferences between HIV stigma and psychosocial status as well as substance use among PLWHA. We call for longitudinal studies in future research to examine any potential causality. Second, all data were collected via self- reporting, participants may underreport their substance use or psychosocial status due to social desirability bias. Therefore, misclassification of key variables may result in diluting the effect of stigma toward the null. Although audio computer-assisted self-interview techniques may increase the validity of collected data, the evidence was inconclusive as traditional face-to-face interview may not be uniformly inferior to non-interviewer techniques across all occasions. Third, as the current study was conducted in a southwest region with many minorities, findings in the current study may be subject to limited generalizability to other settings in China. Fourth, due to the dynamic changes of the definition on SES, the way of measuring the SES among PLWHA in the current study may not be able to capture all domains of individual’s social, human and material capital as literatures suggested. However, we followed the guideline suggested by the DHHS for practical and feasible purposes. Last, due to limited space of the questionnaire, we did not collect detailed information on participants’ job categories. PLWHA who have “decent” jobs (e.g., government employees or school teachers) may encounter more HIV stigma compared to PLWHA working as peasant workers or service staff. In future studies, we suggest researchers employ hypotheses-driven strategies to explore how SES impact these studied associations. To our knowledge, it is the first study to assess the stratified impacts of HIV stigma on PLWHA’s psychosocial status and behaviors of substance use by their SES gradients. Findings in the current study served as guideline for health professionals to design tailored interventions among PLWHA in future. For PLWHA with better education, living in urban areas and not working as peasant workers or unemployed, more interventions focusing on psychosocial health are needed. For PLWHA with higher income, more programs on improving coping strategies and reducing their substance use behaviors are desired. Only can we combine individual-level factors with social, historical and biophysical contexts of PLWHA, health professionals can better understand the disease etiology, health, and intervention modes for PLWHA in China. [^1]: The authors have declared that no competing interests exist. [^2]: **Conceptualization:** XL SQ YZ. **Formal analysis:** CZ YL. **Funding acquisition:** XL. **Investigation:** XL YZ SQ ZS YC. **Methodology:** CZ YL XL. **Project administration:** XL SQ. **Resources:** YC ZS YZ. **Software:** CZ YL. **Supervision:** XL. **Validation:** XL YC ZS. **Writing – original draft:** CZ YL XL SQ. **Writing – review & editing:** XL SQ.

# Introduction Seed vigor, an important and complex agronomic trait, is controlled by multiple factors such as genetic and physical purity, mechanical damage, and physiological conditions. Seeds with high vigor can exhibit high germination rates, resistance to environmental stress, and high crop yields. Moreover, high- quality seeds that ensure uniform germination and growth that lead to increased production are important to growers, and seed vigor depends fundamentally on the potential of the seed itself to grow under favorable growth conditions and under adverse stress conditions. The ability to predict seed vigor using an artificial aging test is indispensable for ensuring rapid and uniform emergence of plants and for maximizing potential productivity under a wide range of field conditions. Sensitivity of seeds to artificial aging has been used successfully to rapidly evaluate and predict seed vigor. High vigor seeds germinate normally after being subjected to artificial aging treatments, but low vigor seeds produce abnormal seedlings or die. Several physiological and biochemical processes have been identified that occur during artificial aging of seeds. For example, oxidative damage to DNA and proteins is likely to be involved in seed aging, and the formation of sugar–protein adducts or isoaspartyl residues may be factors contributing to the loss of protein function during artificial aging. In contrast, antioxidants, heat shock proteins (HSPs), and enzymes that repair protein damage may be involved in ameliorating the effects of artificial aging on seed vigor,. Stress-related proteins and enzymes may also play a role in seed vigor. Prieto-Dapena et al. reported that seed-specific overexpression of the sunflower heat stress transcription factor HaHSFA9 in tobacco enhanced the accumulation of HSPs and improved resistance of seeds to artificial aging. Mutations in the rice aldehyde dehydrogenase 7 (OsALDH7) gene resulted in seeds that were more sensitive to artificial aging conditions and accumulated more malondialdehyde than wild-type seeds, implying that this enzyme plays a role in maintaining seed viability by detoxifying the aldehydes generated by lipid peroxidation. A high level of a membrane lipid-hydrolyzing phospholipase D (PLDa1) appeared to be detrimental to seed quality, but attenuation of PLDa1 expression improved oil stability, seed quality, and seed vigor. Lipoxygenases (LOXs) have also been reported to be involved in seed deterioration. Overaccumulation of protein-<span class="smallcaps">l</span>-isoaspartate *O*-methyltransferase (PIMT1, encoding IAMT in *Arabidopsis*) reduced the accumulation of <span class="smallcaps">l</span>-isoaspartyl residues in seed proteins and increased germination vigor. Conversely, reduced PIMT1 accumulation was associated with an increase in the accumulation of <span class="smallcaps">l</span>-isoaspartyl residues in the proteome of freshly harvested dry mature seeds, resulting in heightened sensitivity to aging treatments and the loss of seed vigor under stressful germination conditions. Although environmental conditions during seed formation, harvest, and germination are important, genetic factors also have substantial impacts on seed vigor. Genetic loci associated with seed vigor have been identified in rice, barley, wheat, oilseed rape, and *Arabidopsis thaliana* using artificial aging tests,. In addition, proteome analyses of seed vigor in *A. thaliana* and maize revealed common features in seeds subjected to artificial aging. To our knowledge, only two reports on proteomic characterization of specific proteins associated with seed vigor have been published. The use of artificial aging treatments to map quantitative trait loci (QTLs) associated with seed vigor by linkage analysis in maize has not been reported. In this study, seed vigor experiments and QTL analyses using two recombinant inbred line (RIL) populations and molecular markers were conducted under controlled conditions of seed deterioration to (1) identify QTLs for traits related to seed vigor, (2) integrate the QTLs identified across the two RIL populations to verify true QTLs, and (3) integrate candidate gene analyses with seed vigor QTL mapping across the two populations to assess the roles of candidate genes in natural variations in seed vigor in maize. # Materials and Methods ## Population Development The maize inbred line Shen137 was crossed with two other maize inbred lines, Yu82 and Yu537A, to make two connected F₁ crosses during spring 2005 in Zhengzhou, Henan, China. The two crosses, Yu82 × Shen137 and Yu537A × Shen137, were self-pollinated by the single seed descent method to produce 208 and 212 F₁₀ RILs, respectively, designated as Population 1 (Pop. 1) and Population 2 (Pop. 2), which were used to identify QTLs for traits related to seed vigor. Yu82 and Yu537A were derived from a Chinese Stiff Stalk germplasm, a heterotic group used broadly in China, while Shen137 was derived from a Chinese non-Stiff Stalk germplasm, also a heterotic group widely used in China. ## Artificial Aging Treatment We used an improved method of artificial aging that was proposed originally by Zeng et al.. Sample seeds from the 208 and 212 F₁₀ RILs and their parent lines were treated at 45°C and 90% relative humidity for 0, 2, 4, and 6 days (0 d, 2 d, 4 d, and 6 d, respectively) using a thermostatic moisture regulator. The 0 d treatment was used as a control. ## Germination Experiment and Seed Vigor-related Trait Evaluation The germination experiment followed a randomized complete block design with three replications and was conducted at 28°C in a growth chamber in 2011. Fine sand with a diameter of 0.05–0.2 mm was used as a sprouting bed. The sand was heated at 120°C for 2 h in a high-handed sterilization pan containing a 16 × 8 array of 40-mm-diameter wells. Two seeds were placed on top of 3.5 cm of sand in each well and then covered with 1.5 cm of sand. Fifty seeds of the parental lines and each RIL were selected to ensure good sowing quality. The seeds were incubated in a growth chamber at 28°C, 65% relative humidity, and illumination conditions of 4000 lux and a 14/10 (day/night) photoperiod with three replications. The number of germinated seeds was counted daily. Germination rate data were collected over an 8-day period after sowing and were used for QTL analysis when obvious differences between the parental lines and the RILs were observed. After the 8-day germination period, five plants of each RIL were selected randomly and dried in an oven at 65°C for 48 h and then weighed to obtain root dry weight (RDW) and seedling dry weight (SDW). The germination percentage (GP) was calculated aswhere *n* is the total number of germinated seeds and *N* is the total number of seeds tested. The germination energy (GE) was calculated as the number of germinated seeds on day 4 divided by the total number of seeds. ## Statistical Analysis of Phenotypic Data Analysis of variance (ANOVA) was carried out to estimate genetic variation among the RILs for all of the traits. To normalize the variance, the germination percent was transformed to The frequency distribution of the traits was calculated to test the skewness of the traits toward the parents. Pearson correlation coefficients among the traits were computed using the statistical software package SPSS17.0 (SPSS Inc., Chicago, IL, USA) using the mean values of each trait. ## Construction of Molecular Marker Genetic Linkage Maps and QTL Analysis Young-leaf samples were collected from seedlings of the two connected F₁₀ RIL populations, and genomic DNA was extracted using the CTAB method. Genotype data for single-nucleotide polymorphism (SNP) markers were analyzed using the Genome Studio Data analysis software (Illumina Inc., San Diego, CA, USA), which generates homozygous and heterozygous genotype clusters. In total, 3072 SNP markers were chosen to detect polymorphisms between the two parental lines, i.e., Yu82/Shen137 and Yu537A/Shen137. Molecular genetic linkage maps were constructed with the maximum likelihood mapping function using JoinMap software version 4.0. Analysis of QTLs associated with seed vigor was conducted using composite interval mapping (CIM) with WinQTLcart 2.5 software. The genome was scanned every 2 cM with a window size of 10 cM to exclude control markers around the tested interval. Five control markers were identified by forward and backward regression. An empirical threshold level for declaring a QTL to be significant was an alpha level of 0.05 for each individual phenotypic trait across all traits. A maximum genetic distance of 50 cM was used to group all markers to chromosomes by performing 1000 random permutations. The phenotypic variation and additive effect explained by each QTL were estimated from the values expressed by the QTL peaks obtained from CIM. For the additive effects of the QTLs, positive and negative values indicated that alleles from the normal maize inbreds Yu537A/Yu82 and the maize inbred Shen137, respectively, increased the trait scores. QTLs were named using the following nomenclature: “q” “+”artificial aging treatment days “+”trait abbreviation “+”population code “+”− “+”chromosome number “+”QTL number”. ## 1QTL Integration and Meta-analysis To integrate QTL information for seed vigor-related traits obtained from the two connected RIL populations, the two molecular marker genetic linkage maps were integrated, and consensus QTLs were identified by meta-analysis. The QTLs identified in the two connected RIL populations were projected onto the integrated map using their positions and confidence intervals shared by the two molecular marker genetic linkage maps. Ambiguous markers between the two molecular marker genetic linkage maps were deleted, which improved the accuracy of projection. Meta-analysis was performed on the QTL clusters on each chromosome using BioMercator2.1 software. The Akaike information criterion (AIC) was used to select the QTL model on each chromosome. According to the AIC, the QTL model with the lowest AIC value is considered to be a significant model and indicates the number of meta-QTLs (mQTLs). A mQTL was declared only when it was shared by the two RIL populations. The number of mQTLs that gave the best fit to the results of a given linkage group was determined based on a modified AIC. mQTLs were named using the following nomenclature: “m”+“QTL”+“−” + “chromosome number” + “−” + “mQTL number.” QTL meta-analysis requires independent QTLs for the same trait to be obtained from different plant populations or different environmental conditions. # Results ## Phenotypic Performance of Seed Vigor-related Traits in Maize The Yu82, Yu537A, and Shen137 parental lines, designated P1, P2, and P3, respectively, exhibited clear differences in GE, GP, SDW, and RDW under the various treatment conditions. All four traits showed decreased values for the three parental lines after the 2 d, 4 d, and 6 d artificial aging treatments relative to the control. Using GE as an example, the respective values of GE for P1, P2, and P3 were 0.92, 0.74, and 0.88 for the 0 d treatment (control); 0.80, 0.50, and 0.65 for the 2 d treatment; 0.80, 0.25, and 0.60 for the 4 d treatment; and 0.47, 0.12, and 0.33 for the 6 d treatment. The respective values for GE under the 2 d, 4 d, and 6 d aging treatments decreased relative to the controls by 0.12, 0.12, and 0.45 for Yu82; by 0.22, 0.49, and 0.62 for Yu537A; and by 0.23, 0.28, and 0.55 for Shen137. Similar trends for changes in GE, GP, SDW, and RDW in response to artificial aging treatments were observed in the RIL populations relative to the parents. The values of the four traits among the two RIL populations showed a pattern of continuous distribution around the mean. The range of variability for the traits was large among the RILs and the traits differed substantially in response to the different aging treatments. GE, GP, SDW, and RDW values showed significant positive correlations with each other in both of the RIL populations for the control and aging treatments except between SDW and GP, SDW and GE, RDW and GP, and RDW and GE for the control and 2 d aging treatments. The correlation coefficients of all of the traits were higher for the three artificial aging treatments than for the 0 d control. In addition, phenotypic correlations among different treatments for the same seed vigor-related traits in the two RIL populations were significantly correlated. ## SNP Data Analysis and Genetic Linkage Map Construction Of a total of 3072 SNP markers, 1397 and 1371 were polymorphic between the Yu82/Shen137 and Yu537A/Shen137 parents, respectively. The percentages of missing data in the genotyping of the mapping population across the 1397 and 1371 SNP loci were low at 1.32% and 1.68%, respectively. Statistical tests showed that most of the 2768 SNP markers followed the expected 1∶1 ratio. Ultimately, two genetic linkage maps consisting of all 10 maize chromosomes allocated to 10 linkage groups were constructed based on 1172 and 1139 SNP markers using Joinmap version 4.0 software. The total lengths and average intervals were 1629.61 cM and 1.39 cM for Yu82/Shen137 and 1681.75 cM and 1.48 cM for Yu537A/Shen137. The 2768 SNP loci in the two linkage maps were consistent with the chromosome bin locations in the Maize Genetics and Genomics Database (MaizeGDB) maps. ## Identification of QTLs Associated with GE, GP, SDW, and RDW in the Two RIL Populations Sixty-five QTLs for GE, GP, SDW, and RDW were mapped in seeds of the two RIL populations germinated under the control and artificial aging treatments, yielding 33 QTLs in Pop. 1 and 32 QTLs in Pop. 2. The QTLs mapped to all of the maize chromosomes except for chromosome 8. Individual QTLs explained from 5.40% to 11.92% of the phenotypic variation, while five of the QTLs accounted for more than 10% of the phenotypic variation. ## QTLs for GE Four QTLs for GE were identified on chromosomes 3, 4, and 5 in the 0d control seeds of the two RIL populations with one QTL in Pop. 1 and three QTLs in Pop. 2. The contributions of the QTLs to phenotypic variation ranged from 5.61% to 8.01%. The three QTLs in Pop. 2 were responsible for 20.89% of the phenotypic variation. The positive alleles of *qnGE1-3*, *qnGE2-4*, and *q6GE2-5* were contributed by Shen137. For each RIL in the two populations, molecular markers positioned close to the peaks of the QTLs were categorized by genotypic class (i.e., either Yu82/Yu537A or Shen137), and their respective GE values for seeds germinated under the control treatment were averaged. The respective average GE values in the untreated RILs corresponding to the Shen137 and Yu82 alleles of markers associated with the QTLs were 0.85 and 0.79 for *qnGE1-3*, 0.83 and 0.79 for *qnGE2-4*, 0.84 and 0.80 for *qnGE2-5*, and 0.78 and 0.84 for *qnGE2-3*. Three QTLs for GE were identified on chromosomes 6 and 9 in seeds subjected to the 2d treatment with one QTL in Pop. 1 and two QTLs in Pop. 2. The contributions of the QTLs to phenotypic variation ranged from 5.91% to 7.12%. The two QTLs in Pop. 2 were responsible for 13.97% of the phenotypic variation. The positive alleles of *q2GE1-6* in Pop. 1 were contributed by Yu82, while the positive alleles of *q2GE2-9-1* and *q2GE2-9-2* in Pop. 2 were contributed by Shen137. The respective average GE values corresponding to the Shen137 and Yu82 alleles in the 2d treatment RILs were 0.79 and 0.72 for *q2GE1-6*, 0.73 and 0.75 for *q2GE2-9-1*, and 0.72 and 0.76 for *q2GE2-9-2*. Three QTLs for GE were identified on chromosomes 1 and 3 in seeds subjected to the 4d treatment with two QTLs in Pop. 1 and one QTL in Pop. 2. The contributions of the QTLs to phenotypic variation ranged from 5.82% to 10.67%. The two QTLs in Pop. 1 were responsible for 16.49% of the phenotypic variation. The positive alleles of *q4GE1-3* in Pop. 1 were contributed by Shen137. The respective average GE values corresponding to the Shen137 and Yu82 alleles in the 4d treatment RILs were 0.57 and 0.64 for *q4GE1-3*, 0.63 and 0.57 for *q4GE1-1*, and 0.58 and 0.52 for *q4GE2-3*. Six QTLs for GE were identified on chromosomes 1, 2, 3, 4, and 7 in seeds subjected to the 6d treatment with one QTL in Pop. 1 and five QTLs in Pop. 2. The contributions of the QTLs to phenotypic variation ranged from 6.12% to 10.14%. The five QTLs in Pop. 2 were responsible for 36.91% of the phenotypic variation. The positive alleles of *q6GE2-2*, *q6GE2-4*, and *q6GE2-7* in Pop. 2 were contributed by Shen137. The respective average GE values corresponding to the Shen137 and Yu82 alleles in the 6d treatment RILs were 0.30 and 0.34 for *q6GE2-2*, 0.30 and 0.34 for *q6GE2-4*, 0.31 and 0.35 for *q6GE2-7*, 0.35 and 0.30 for *q6GE2-3-1*, 0.38 and 0.34 for *q6GE2-3-2*, and 0.31 and 0.36 for *q6GE1-1*. ## QTLs for GP Four QTLs for GP were identified on chromosomes 3, 5, and 10 in the 0d control seeds of the two RIL populations with three QTLs in Pop. 1 and one QTL in Pop. 2. The contributions of the QTLs to phenotypic variation ranged from 5.97% to 7.98%. The alleles derived from Shen137 contributed to increased trait values. The respective average GP values corresponding to the Shen137 and Yu82 alleles of markers associated with the QTLs in the untreated RILs were 0.88 and 0.92 for *qnGP1-3-1*, 0.88 and 0.91 for *qnGP1-3-2*, 0.87 and 0.93 for *qnGP1-5*, and 0.86 and 0.88 for *qnGP2-10*. Four QTLs for GP were identified on chromosomes 3, 5, 6, and 7 in seeds subjected to the 2d treatment with three QTLs in Pop. 1 and one QTL in Pop. 2. The contributions of the QTLs to phenotypic variation ranged from 5.99% to 10.47%. Shen137 alleles contributed to increased trait values for the *q2GP1-3* and *q2GP1-5* QTLs and Yu82/Yu537A alleles generally increased the trait values for the *q2GP1-6* and *q2GP2-7* QTLs. The respective average GP values corresponding to the Shen137 and Yu82 alleles in the 2d treatment RILs were 0.82 and 0.88 for *q2GP1-3*, 0.81 and 0.90 for *q2GP1-5*, 0.89 and 0.81 for *q2GP1-6*, and 0.84 and 0.81 for *q2GP2-7*. Seven QTLs for GP were identified on chromosomes 3, 5, 6, and 7 in seeds subjected to the 4d treatment with five QTLs in Pop. 1 and two QTLs in Pop. 2. The contributions of the QTLs to phenotypic variation ranged from 5.99% to 10.47%. Yu82/Yu537A alleles contributed to increased trait values for the *q4GP1-6*, *q4GP2-7-1*, and *q4GP2-7-2* QTLs and Shen137 alleles generally increased the trait values for the other four QTLs. The respective average GP values corresponding to the Shen137 and Yu82 alleles in the 4d treatment RILs were 0.71 and 0.65 for *q4GP1-6*, 0.69 and 0.65 for *q4GP2-7-1*, 0.70 and 0.64 for *q4GP2-7-2*, 0.65 and 0.73 for *q4GP1-3-1*, 0.66 and 0.72 for *q4GP1-3-2*, 0.65 and 0.71 for *q4GP1-5-1*, and 0.66 and 0.72 for *q4GP1-5-2*. Five QTLs for GP were identified on chromosomes 3, 4, 5, and 6 in seeds subjected to the 6d treatment with two QTLs in Pop. 1 and three QTLs in Pop. 2. The contributions of the QTLs to phenotypic variation ranged from 5.74% to 6.96%. Shen137 alleles contributed to increased trait values for the *q6GP1-5* and *q6GP2-4* QTLs and Yu82/Yu537A alleles generally increased the trait values for the three additional QTLs. The respective average GP values corresponding to the Shen137 and Yu82 alleles in the 6d treatment RILs were 0.40 and 0.49 for *q6GP1-5*, 0.41 and 0.46 for *q6GP2-4*, 0.48 and 0.40 for *q6GP1-6*, 0.46 and 0.41 for *q6GP2-3-1*, and 0.45 and 0.39 for *q6GP2-3-2*. ## QTLS for SDW Two QTLs for SDW were identified on chromosomes 3 and 5 in the 0d control seeds of the two RIL populations with one QTL in Pop. 1 and one QTL in Pop. 2. The two QTLs accounted for 6.16% and 5.39% of the phenotypic variation, respectively. The positive allele of *qnSDW2-3* was contributed by Shen137. The respective average SDW values corresponding to the Shen137 and Yu82 alleles of markers associated with the QTLs in the untreated RILs were 0.49 and 0.45 for *qnSDW1-5* and 0.50 and 0.56 for *qnSDW2-3*. Four QTLs for SDW were identified on chromosomes 2, 3, 4, and 10 in seeds subjected to the 2d treatment with one QTL in Pop. 1 and three QTLs in Pop. 2. The contributions of the QTLs to phenotypic variation ranged from 6.29% to 8.29%. The positive alleles of *q2SDW2-2* and *q2SDW2-3* were contributed by Shen137. The respective average SDW values corresponding to the Shen137 and Yu82 alleles of markers associated with the QTLs in the 2d treatment RILs were 0.43 and 0.49 for *q2SDW2-2*, 0.43 and 0.48 for *q2SDW2-3*, 0.88 and 0.85 for *q2SDW1-4*, and 0.48 and 0.43 for *q2SDW2-10*. Four QTLs for SDW were identified on chromosomes 2, 3, 4, and 6 in seeds subjected to the 4d treatment with two QTLs in Pop. 1 and two QTLs in Pop. 2. The contributions of the QTLs to phenotypic variation ranged from 6.00% to 11.92%. The positive alleles of *q4SDW2-2* and *q4SDW2-3* were contributed by Shen137. The respective average SDW values corresponding to the Shen137 and Yu82 alleles of markers associated with the QTLs in the 4d treatment RILs were 0.37 and 0.44 for *q4SDW2-2*, 0.38 and 0.43 for *q4SDW2-3*, 0.40 and 0.36 for *q4SDW1-4*, and 0.39 and 0.35 for *q2SDW1-6*. Three QTLs for SDW were identified on chromosomes 2, 4, and 6 in seeds subjected to the 6d treatment with two QTLs in Pop. 1 and one QTL in Pop. 2. The contributions of the QTLs to phenotypic variation ranged from 5.40% to 8.68%. The positive alleles of *q6SDW1-4* and *q6SDW2-2* were contributed by Shen137. The respective average SDW values corresponding to the Shen137 and Yu82 alleles of markers associated with the QTLs in the 6d treatment RILs were 0.29 and 0.24 for *q6SDW1-6*, 0.24 and 0.29 for *q6SDW1-4*, and 0.29 and 0.35 for *q6SDW2-2*. ## QTLS for RDW Four QTLs for RDW were identified on chromosomes 1, 3, 7, and 10 in the 0d control seeds of the two RIL populations with two QTLs in Pop. 1 and two QTLs in Pop. 2. The contributions of the QTLs to phenotypic variation ranged from 5.67% to 7.71%. The positive alleles of *qnRDW2-3* and *qnRDW2-7* in Pop. 2 were contributed by Shen137. The respective average RDW values corresponding to the Shen137 and Yu82 alleles of markers associated with the QTLs in the untreated RILs were 0.45 and 0.42 for *qnRDW1-1*, 0.44 and 0.42 for *qnRDW1-10*, 0.33 and 0.40 for *qnRDW2-3*, and 0.34 and 0.39 for *qnRDW2-7*. Four QTLs for RDW were identified on chromosomes 2, 4, 7, and 10 in seeds subjected to the 2d treatment with one QTL in Pop. 1 and three QTLs in Pop. 2. The contributions of the QTLs to phenotypic variation ranged from 6.67% to 9.22%. The positive alleles of *q2RDW2-2* and *q2RDW2-4* in Pop. 2 were contributed by Shen137. The respective average RDW values corresponding to the Shen137 and Yu82 alleles of markers associated with the QTLs in the 2d treatment RILs were 0.28 and 0.33 for *q2RDW2-2*, 0.28 and 0.33 for *q2RDW2-4*, 0.49 and 0.45 for *q2RDW1-7*, and 0.33 and 0.28 for *q2RDW2-10*. Two QTLs for RDW were identified on chromosomes 1 and 2 in seeds subjected to the 4d treatment with one QTL in Pop. 1 and one QTL in Pop. 2 contributing 6.67% and 7.99%, respectively, to the phenotypic variation. The positive allele of *q4RDW2-2* in Pop. 2 was contributed by Shen137. The respective average RDW values corresponding to the Shen137 and Yu82 alleles of markers associated with the QTLs in the 4d treatment RILs were 0.22 and 0.29 for *q4RDW2-2* and 0.19 and 0.15 for *q4RDW1-1*. Six QTLs for RDW were identified on chromosomes 1, 2, 4, 5, and 6 in seeds subjected to the 4d treatment with five QTLs in Pop. 1 and one QTL in Pop. 2. The contributions of the QTLs to phenotypic variation ranged from 5.60% to 7.88%. The positive alleles of *q6RDW1-4* and *q6RDW1-5* in Pop. 1 and *q6RDW2-2* in Pop. 2 were contributed by Shen137. The respective average RDW values corresponding to the Shen137 and Yu82 alleles of markers associated with the QTLs in the 6d treatment RILs were 0.13 and 0.11 for *q6RDW1-1-1*, 0.13 and 0.11 for *q6RDW1-1-2*, 0.13 and 0.11 for *q6RDW1-6*, 0.10 and 0.14 for *q6RDW1-4*, 0.10 and 0.13 for *q6RDW1-5*, and 0.16 and 0.21 for *q6RDW2-2*. ## mQTL Analysis The integrated genetic map for the two populations contained 1712 SNP markers and was 1712.6 cM long with an average of 1.00 cM between markers. Meta-analysis was performed to identify mQTLs that were associated with the observed variation in the seed vigor traits. Eighteen mQTLs were identified from the 65 identified initially based on variation in the four seed vigor traits. Sixty-one of the initial QTLs (93.85%) were integrated into the mQTL regions. The 18 mQTLs mapped to all chromosomes except for chromosomes 8 and 9. On average, each mQTL included 3.39 QTLs with a range of two to six QTLs for one to four traits. The initial QTLs included in mQTL5-2 were all detected based on one trait (GP) in the 0 d control seeds and in seeds subjected to the three artificial aging treatments. The QTLs in mQTL3-2, mQTL3-3, and mQTL3-4 were identified based on one trait under three treatment conditions; the QTLs in mQTL2 were identified based on two traits under two conditions; the QTLs in mQTL1-1, mQTL1-2, mQTL3-1, mQTL6-2, and mQTL7-2 were identified based one trait under two conditions; and the QTLs in the remaining mQTLs were identified based on two traits under one condition. Note that the initial QTLs with values of R²\>10% were integrated into four mQTLs: mQTL2, mQTL3-2, mQTL3-4, and mQTL5-2. # Discussion Seed vigor depends on the physiological and genetic potential of the seeds and on the effects of environmental stress conditions that they are exposed to during germination. For maize, plant stand establishment and performance are strongly influenced by seed vigor. Producing high-quality seeds to stabilize crop yield poses a challenge for crop breeders, and a key to meeting this challenge is the elucidation of the molecular mechanisms underlying seed vigor traits. Little research has been published to date on the use of artificial aging treatments to clarify the molecular mechanisms associated with seed vigor traits in maize. In this study, two sets of connected RIL populations were evaluated for four seed vigor traits after seeds were subjected to three artificial aging treatments during germination. ## Phenotypic Variation among Parent Inbred Lines and their RILs after Different Aging Treatments Because germination is affected by seed vigor, many researchers have used the germination rate as an indicator of seed vigor. To predict the ability of seeds to perform under a wide range of field conditions, an artificial aging test was used in this study. The results showed that GE and GP decreased gradually with increasing duration of the artificial aging treatment and that the germination rates differed between the parental lines and the RILs with artificial aging. In addition, correlations between the results from the untreated controls and the seeds subjected to artificial aging treatments were significant, indicating that the artificial aging test provided a valid assessment of seed vigor. ## Genotypic Differentiation in Two Populations Reflects Adaptation to Artificial Aging Conditions The selection of maize lines for adaption to climatic conditions has resulted in differences in seed vigor in various inbred lines. These differences in vigor are reflected by differences in responses to artificial aging conditions. Only one common QTL identified in this study was located near the chromosome region associated with GE at bin 3.02, but multiple population-specific QTLs were identified even though the two populations shared Shen137 as a common parent. Specifically, five QTLs on chromosomes 1, 3 (2), 5, and 6 that were contributed by the parental line Yu82 were identified in Pop. 1 and five QTLs on chromosomes 2, 3 (2), 4, and 7 that were contributed by the parental line Yu537A were linked in Pop. 2. Among the 18 mQTLs, 9 mQTLs included initial QTLs for GE on chromosomes 1, 3, 4, 5, 6, and 7; 8 mQTLs included initial QTLs for GP on chromosomes 3, 4, 5, 6, and 7; 8 mQTLs included initial QTLs for SDW on chromosomes 2, 3, 4, 5, 6, and 10; and 10 mQTLs included initial QTLs for RDW on chromosomes 1, 2, 3, 4, 6, 7, and 10. The *GP1-3* and *SDW2-3* QTLs identified in Pop. 1 and Pop. 2, respectively, were detected in seeds subjected to control, 2 d, and 4 d treatments, but not the 6 d treatment. The results suggested that the loci associated with seed vigor played an important role during the early stages of seed deterioration caused by artificial aging. The progressive decrease in germination vigor depends on the duration of artificial seed aging. This may be due to the diverse negative effects of reduced protein synthesis on processes such as maintenance, repair, and normal resumption of metabolism and cell cycle activity, efficiency of detoxification, efficiency of signaling pathways, and/or the production and secretion of metabolites and plant hormones. Oxidative stress has also been shown to be a causal factor in aging processes. The extent of oxidative damage to nucleic acids, lipids, and proteins has been shown to increase with age. In contrast, the *GP1-5* QTLs in Pop. 1 were all identified in seeds subjected to control, 2 d, 4 d, and 6 d treatments, with contributions to phenotypic variation of single QTLs ranging from 6.52% to 10.67%. These results indicated that the QTLs confer resistance to high-temperature and humidity conditions, which may deserve further study in marker-assisted selection (MAS). A cucumisin- like Ser protease gene (At5g67360) mapped to the *QTL1-5* interval. Increased protease activities were observed in potato seed-tubers, wheat seeds, and maize seeds during artificial aging, which resulted in a substantial reduction in soluble and storage proteins. Notably, *SDW2-2* and *RDW2-2* QTLs in Pop. 2 were identified only with the 2 d, 4 d, and 6 d treatments, indicating that these loci strongly influence plant stand establishment after seed deterioration caused by high-temperature and humidity conditions. ## Synthesis of Initial QTLs across Two Populations Subjected to Different Artificial Aging Treatments We examined the genetic correlations among traits by mQTL analysis. Co- localization of QTLs for the traits associated with seed vigor might indicate pleiotropy and/or tight linkage. In this study, 16 mQTLs (e.g., mQTL6-2, mQTL4-3, mQTL2, and mQTL3-2) included two to six initial QTLs for two to four of the seed vigor traits. These initial QTLs might indicate pleiotropy and/or tight linkage. For mQTL2, mQTL3-2, and mQTL3-4, an initial QTL with a major effect (R²\>10%) was identified for at least one treatment condition, and corresponding candidate genes mapped to the mQTL intervals except for mQTL2. The chromosome regions for three mQTLs with high QTL co-localization might be hot spots of important QTLs for seed vigor traits. Validation of potential candidate genes from the three mQTLs would provide a reliable and feasible strategy for QTL cloning. ## Associations between QTLs and Candidate Genes in Maize Two proteomics studies in maize, and one study in *Arabidopsis* identified 28, 40, and 83 differentially expressed proteins, respectively, that were related to artificial seed aging. We explored the association between mQTLs and these differentially expressed proteins in maize using a bioinformatics approach. Twenty-three differentially expressed proteins were located at bin loci of 13 mQTLs. The proteins had functions related to responses to stress, molecular chaperones, hydrolase activity, energy, cell growth/division, protein targeting and storage, signal transduction, translation, protein metabolism, amino acid metabolism, and other processes. Basavarajappa et al. suggested that seed aging affected metabolism and energy supply in maize, while Ching and Schoolcraft showed that seed storage protein content decreased with seed aging. Various toxic compounds such as reactive oxygen species (ROS), aldehydes, and methylglyoxal can accumulate during seed aging, possibly resulting in reduced seed viability and decreased germination rates. Some proteins in aged maize seeds might interfere with signal transduction in responses to stress caused by heat shock, desiccation, peroxide, salt, or abscisic acid (ABA). Rajjou et al. showed that seed aging treatments strongly increased the extent of protein oxidation (carbonylation), which might cause loss of function for seed proteins and enzymes and/or enhance their susceptibility to proteolysis. ## The Glycolytic Pathway is Affected during Artificial Seed Aging The functions of eight candidate genes located within the mQTLs identified in seeds subjected to artificial aging treatments were associated with the glycolytic pathway. The functions of two of the candidate genes were associated with stress responses. A stress-responsive glyoxalase family gene (226532762) mapped in the mQTL1-1 interval that was associated with RDW and an aldehyde dehydrogenase gene (45238345) mapped in the mQTL6-1 interval that was associated with SDW and GP. The functions of the remaining six candidate genes were associated with energy processes. These genes included an ATP synthase F1 subunit alpha gene (327195227) that mapped in the mQTL3-2 interval associated with GE and GP; a glyceraldehyde 3-P dehydrogenase gene (At3g04120) that mapped in the mQTL4-2 interval associated with GE and RDW; a V-type (H+)-ATPase domain gene (326509331) that mapped in the mQTL4-3 interval associated with GE, SDW, and RDW; a phosphoglucomutase gene (At1g70730) that mapped in the mQTL6-1 interval associated with SDW and GP; a 3-phosphoglycerate kinase gene (226247007) that mapped in the mQTL6-2 interval associated with GE, GP, SDW, and RDW; and an isocitrate lyase gene (AT3G21720) that mapped in the mQTL7-2 interval associated with GP and RDW. These key proteins are responsible for glycolysis during seed aging, which plays a major role in the maintenance of the intracellular redox state and seed vigor. Our results showed that seeds were subject to oxidative stress during artificial aging treatments and mounted a protective response through modification of the glycolytic pathway. ## Protein Metabolism is Major a Factor in Seed Vigor during Artificial Seed Aging Previous studies showed that that protein metabolism plays an important role in seed germination during artificial seed aging and that several metabolic functions are affected, including protein folding, protein translocation, thermotolerance, oligomeric assembly, and switching between active and inactive protein conformations. Rajjou and Debeaujon demonstrated that simultaneous impairment of these functions was closely linked to the loss of seed vigor and Prieto-Dapena et al. showed that transgenic seeds overexpressing a heat stress transcription factor exhibited increased accumulation of HSPs, resulting in improved resistance to artificial aging treatments in seeds. The preservation of a robust stress response and protein turnover mediated by HSPs is a requirement for all organisms. Eight candidate genes identified within mQTLs identified in this study were functionally associated with protein metabolism. Four of the candidate genes had functions related to general protein metabolism or translation, including an elongation factor 1-g2 gene (AT1G57720) that mapped in the mQTL4-3 interval associated with GE, SDW, and RDW; an elongation factor 1B-g gene (AT1G09640) that mapped in the mQTL6-1 interval associated with SDW and GP; a calreticulin 1 gene (AT1G56340) that mapped in the mQTL7-1 interval associated with GE and RDW; and an Asp aminotransferase gene (At5g19550) that mapped in the mQTL3-3 interval associated with SDW and RDW. The other four candidate genes encoded HSPs including a hypothetical ACD ScHsp26-like gene (242056533), an HSP16.9 gene (195605946), an HSP17.2 gene (162459222) that mapped in the mQTL3-1 interval associated with GE and GP, and an hsp20/alpha crystallin family protein gene (At5g51440) that mapped in the mQTL3-4 interval associated with GE and GP. The proteins with chaperone activities encoded by the candidate genes (Hsp26, hsp20, HSP16.9, and HSP17.2) are favored targets for oxidation, presumably because they act as shields to protect other proteins against ROS damage. Other chaperone proteins such as calreticulin (AT1G56340) were also oxidized in deteriorated dry seeds. In addition, five candidate genes identified within mQTLs identified in this study were associated with protein modification and signal transduction functions. These genes included a ubiquitin E2 gene (At1g45050) involved in protein modification and a calcium-dependent protein kinase (CaMK) gene (302810918) involved in signal transduction, both of which mapped in the mQTL3-2 interval associated with GE and GP; a gene containing a predicted RING-finger domain (224143836) involved in protein targeting and storage that mapped in the mQTL3-3 interval associated with SDW and RDW; a CAAX prenyl protease 1 gene (226508796) with hydrolase activity that mapped in the mQTL4-2 interval associated with GE and RDW; and a cucumisin-like Ser protease (At5g67360) that mapped in the mQTL5-2 interval associated with GP. The RING-finger domain- containing protein (224143836) and the protein-degrading ubiquitin E2 protein (At1g45050) play key roles in the ubiquitination pathway. CaMK (302810918) is active in protein phosphorylation through the transfer of phosphate from ATP to protein substrates. Protein phosphorylation plays a key role in many signaling pathways such those involved in responses to stress caused by cold, heat shock, desiccation, peroxide, salt, or ABA. The regulation of CaMK in maize seeds subjected to artificial aging might alter signal transduction. ## Other Pathways Involved in Seed Vigor during Artificial Seed Aging Embryo cells undergo active division and expansion during seed germination. Seeds germinate much more slowly after artificial aging treatments, suggesting that these processes related to cell growth might be affected. Two candidate genes had functions related to cell growth and division. A predicted cyclin- dependent kinase (CDK) A gene (224061823) mapped in the mQTL1-4 interval associated with RDW and GE and a MEK homolog1 gene (162459414) mapped in the mQTL3-3 interval associated with SDW and RDW. In addition, a lipoprotein gene (195658029) involved in lipid metabolism mapped in the mQTL4-2 interval associated with GE and RDW. CDK and MEK homolog1 play pivotal roles in the regulation of the eukaryotic cell cycle. For example, inappropriate activation of CDK can lead to apoptosis. In conclusion, the QTLs and the candidate genes identified in this study provide valuable information for identifying additional quantitative trait genes. The alleles for seed vigor could be useful targets for MAS to produce germplasm with improved resistance to artificial aging treatments. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: YHC ZPH. Performed the experiments: ZPH LXK SLG HYL RFZ ZZZ ZZR LKZ. Analyzed the data: JZ HHS LD. Contributed reagents/materials/analysis tools: YHC ZPH. Wrote the paper: ZPH. Revised the paper: YHC LXK.

# Introduction Coronary CTA is currently the first choice for non-invasive diagnosis of coronary artery disease (CAD). This method has high diagnostic accuracy , but its accuracy is influenced by heart rate, rhythm, and other factors. The width coverage of 320-detector CT is 16cm, so a single CT snapshot can cover the whole heart. By avoiding overlapping scanning, the radiation dose received by the patient can be reduced. Additionally, 320-detector CT shortens acquisition time. Because scanning can be completed shortly after bolus injection of a contrast agent, the amount of contrast agent used can also be reduced. Although the time resolution of 320-detector CT is not higher than that of dual-source CT, the former can overcome this shortcoming through the use of multi-sector reconstruction, and with two-sector acquisition time resolution can reach 87.5 ms. Currently, prospective ECG-gated 320-detector CTCA has very high diagnostic accuracy for subjects with heart rates below 65 bpm, and radiation dose is also low–. Because the technique works best at lower heart rates, patients with heart rates above 65 bpm can be given an oral-blockade (50–100 mg metoprolol) one hour before a CT examination. Arrhythmia, however, poses additional challenges to maximizing image quality while reducing radiation dose. Some reports have shown that prospective ECG-gated CTCA is inappropriate for 64-detector CT and dual- source CT when scanning subjects with heart rates above 70 bpm or significant heart rate change (e.g. arrhythmia),. In clinical settings, ventricular premature beat is very common. Whether physiological or pathological in nature, it negatively impacts scanning results. If premature beats occur when scanning, it disrupts the prospective scheduled scanning program and will result in more severe motion artifacts. However, if we pause scanning during the premature phase and acquire data during the next normal phase or additionally scan the next R-R interval, then we can achieve fair results. In this study, prospective ECG-gated 320-detector CT is adopted to scan patients with ventricular premature beats. Patients with heart rates below 65 bpm were placed in a control group. Both groups underwent coronary angiography to evaluate the diagnostic accuracy of 320-detector CTCA for significant (≥50%) stenosis and access the technique’s ability to maintain good image quality while reducing radiation dose. # Materials and Methods ## Objects This study was approved by the Ethics Committee of Harbin Medical University. Data from 110 patients undergoing 320-detector CT coronary angiography in our hospital from October 2009 to June 2011 were collected (72 males and 38 females, aged 38- to 86-years-old, mean age 59-years-old). These patients were divided into two groups, those with ventricular premature beats (60 cases) and those with a normal rhythm and a heart rate below 65 bpm (50 cases). The criterion of patients with ventricular premature beat is that the acquisition range was just in time to catch ventricular premature beat during prospective ECG-gated scanning. 50 cases were detected during medium-field (M-field, FOV = 320 mm) scanning, and 10 cases were detected during small-field (S-field,FOV = 200 mm) scanning. Patients in the later group were found to have a normal rhythm and a heart rate below 65 bpm through M-field scanning. All the patients underwent CAG within one month. The exclusion criteria for both groups were that patients (1) were unable to complete the breath-hold action, (2) had a heart rate above 65 bpm and a normal rhythm, (3) suffered from arrhythmias other than ventricular premature beat, (4) were allergic to iodine contrast agents, (5) had severe renal insufficiency, and (6) had decompensated cardiac insufficiency. All patients with heart rates above 65 bpm were administered Betaloc (50–100 mg) orally. The detailed basic parameters of the subjects are shown in. This study was approved by the local Ethics Committee and informed consent forms were signed by all patients. ## CTA and Scanning Parameters A Toshiba Aquilion One 320-detector CT scanner was used, with detector collimation at 320×0.5 mm and coverage in Z direction set at 16 cm. A prospective ECG-gated technique and arrhythmia rejection algorithm (Toshiba Medical Systems, Nasu, Japan) were adopted. Patients with no allergic reaction to the contrast agent were placed on the scanning bed, connected to the ECG monitor, positioned, and given breathing training. A normal saline line was administered and checked to ensure that the pressure was suitable and the venous catheter was unblocked. The ECG leads (standard II ECG leads were used) were adjusted to optimize the distinction between R waves and other waves, making the R wave, which triggers prospective ECG gating, easily identifiable, at the same time observing in electrocardiogram whether ventricular premature beat appears. Scanning field of view(FOV) was selected according to the size of subjects’ hearts. Available scanning FOV included medium field (M-Field, FOV = 320 mm), and small field (S-field, FOV = 200 mm). The range was generally from the trachea carina to the heart bottom, covering the whole heart. Manual trigger scan mode was adopted, with the monitoring level at the midline level in the scanning field. Each patient’s ECG was observed during scanning to check whether premature beats occurred. Premature beats were avoided during scanning so that image quality was preserved. After injecting the contrast agent, the scanning began when the left ventricle was light and the right ventricle was dark. The relevant parameters were as follows: different tube voltage and tube current was set according to patient body mass index (BMI). A non-ionic contrast agent, iobitridol, was administered in a (370 mgI/ml) 40–60 mL dose. The flow rate was generally 6.0 mL/s, and after injection a flush with 30 mL of normal saline was performed. Prospective gating CTCA of all the patients with heart rates \<65 bpm were performed from 70% to 80% of the R-R interval. The acquisition range for patients with ventricular premature beats and heart rates ≥65 bpm were expanded to 30%–80%, and the best time phase was then selected for evaluation. ## Data Processing After scanning, the machine automatically generated a volume image with a relative time phase of 75% and automatically reconstructed a volume image of the absolute time phase according to the collected data. If the images were not good, ECG would be used to edit and select the period phase with a longer and fixed R-R interval for reconstruction. Shorter or non-fixed period phases would be removed. A Vitreal FX post-processing workstation was used to restructure and analyze the coronary artery. Restructuring modes included volume render (VR), curved planar reformation (CPR), and maximum intensity projection (MIP). Through the software’s automatic measuring function, the degree of coronary artery stenosis was determined through the following calculation: (%)  =  \[(inner diameter of normal vessel proximal to stenosis segment + inner diameter of normal vessel distal to stenosis segment)/2– inner diameter at the narrowest position of stenosis segment\]/\[(inner diameter of normal vessel proximal to stenosis segment + inner diameter of normal vessel distal to stenosis segment)/2\] × 100%. Segments with an obtained stenosis degree ≥50% were compared and analyzed with CAG. ## Image Evaluation and Analysis Method According to the classification criteria issued by the American Heart Association (AHA), the coronary artery is divided into 15 segments. CTCA image quality is divided into four grades: 4, excellent, no artifacts; 3, good, mild artifacts; 2, acceptable, moderate artifacts, but a diagnosis is still possible; and 1, cannot evaluate, severe artifacts, a diagnosis is not possible. All 15 segments of the coronary artery with a luminal diameter ≥1.5 mm were scored. Two reviewers (with more than 3 years’ experience in cardiac CT each) who were blinded to the CT protocol and the results of CCA independently assessed the reconstructed images. After their separate reading sessions, the discrepancy was resolved during a third session, in which the reviewers read images together at the same time to reach a consensus. ## CAG A Philips FD 20 digital flat-panel angiography machine and 5–6F Guiding Catheter were used. After injecting the non-ionic iodinated contrast agent, left coronary angiography with four projection positions and right coronary angiography with two projection positions were conducted. The obtained images were stored, and measuring software was used to calculate the degree of coronary artery stenosis. The method of calculation was the same as used in CTCA image processing. ## Statistical Analysis SPSS 17.0 statistical software was used to analyze the collected data. Age, heart rate, radiation dose, and other variables were expressed with mean and standard deviation, and an independent sample t test was used for comparison. With CAG as the reference standard, sensitivity, specificity, PPV, and NPV of CTCA stenosis were evaluated respectively at both the patient and segment levels (every segment of every vessel). Then the Pearson chi-square analysis were used to analyze the statistical differences of specificity, sensitivity, PPV, NPV between the two groups. Pearson correlation coefficients were used to analyze the correlation between heart rate and radiation dose in the premature beat and control groups. All analyses use a two-sided test, and *P*\<0.05 were set as a threshold for statistical significance. # Results ## CTCA Accuracy At the patient level, sensitivity, specificity, PPV, and NPV in the premature beat group were 100%, 94.59%, 86.66%, and 100% respectively. Analogous values in the control group were found to be 100%, 95.24%, 80%, and 100% respectively. At the segment level, sensitivity, specificity, PPV, and NPV in the premature beat group were 92.55%, 98.21%, 88.51%, and 98.72% respectively, while those in the control group were 95.79%, 98.42%, 90.11%, and 99.28% respectively. Between the two groups, specificity, sensitivity, PPV, NPV was no significant difference. In the premature beat group, 10 coronary artery segments had false positive(FP) results and 7 segments had false negative(FN) results. In the control group, 9 coronary artery segments had false positive results and 4 segments had false negative results. ## Image Quality M-field scanning coronary segment image scores of 4, 3, 2, and 1 for 50 cases in the premature beat group account for 602 (94.80%), 28 (4.41%), 5 (0.78%), and 0 cases respectively. Those same scores in the control group account for 615 (94.61%), 31 (4.76%), 4 (0.61%), and 0 cases respectively. The difference between the two groups had no statistical significance (P\>0.05). In both groups, scores of 3 and 2 were caused by higher heart rate and poor breath holding. The segments with scores of 1 were not evaluated. ## Heart Rate, Dose, and Scanning Field Selection The mean heart rate (77.20±12.07 bpm) and radiation dose (14.62±1.37 mSv) in the premature beat group were higher than the mean heart rate (58.72±4.73 bpm) and radiation dose (3.08±2.35 mSv) in the control group (P\<0.05). Heart rate and radiation dose in the premature beat group were significantly correlated at a level of confidence (both sides) of 0.01 (Pearson correlation coefficient  = 0.699, P\<0.001). Heart rate and radiation dose in the control group, however, displayed no such no correlation (Pearson correlation coefficient  = 0.048, P\>0.05). In the premature beat group, based on heart size, 10 subjects should have received S-field scanning, but the scanning field was artificially changed to M-field. These patients were compared to 10 patients with S-field scanning in terms of radiation dose and image quality. The results of this comparative analysis showed that the radiation dose in S-field scanning is significantly higher than that in M-field scanning, while there was no difference in image quality. # Discussion ## CTCA Diagnostic Accuracy It can be seen from the results that sensitivity, specificity, PPV, and NPV in both groups are high, and there is no significant differences between the two groups. CTCA can accurately diagnosis CAD, and is particularly effective in excluding CAD. With the increasing strength of post-processing technology, we can observe the position, degree, and range of lesions from all angles and directions. After straightening the coronary artery at the lesion site, we can more accurately observe the degree and range of stenosis. We can also more accurately guide the selection of stent positioning in clinical settings. Before stent positioning, CAG is not sensitive enough to detect mild stenosis caused by small plaques. Because of this limitation, if the range of stent positioning is not sufficient, incidence of restenosis will be significantly higher. Therefore, multi-detector spiral CT has been the first choice for clinical diagnosis and treatment of CAD. Currently, the main factor affecting diagnostic accuracy is the artifacts caused by heart rate and breathing. Additionally, the lumen covering caused by high- density and large-area calcified plaques and metal stents also increases the severity of artifacts, and as a consequence impacts the ability to correctly access the degree of stenosis. Among those patients investigated by this study who had premature beats, 10 coronary segments gave false positive results. Of those, two were at the left anterior descending (LAD), three were at the left circumflex (LCX), and two were at the right coronary artery (RCA). In these cases calcified plaques and a metal stent leading to lumen caused an over- estimation in the true degree of stenosis. The remaining three false positives (two at the LCX and one at the RCA) were the product of motion artifacts that affected assessment accuracy. Seven false negative results were due to poor image quality and calcified plaques. In the control group, nine coronary segments gave false positive results. Of those, five were at the LAD and four were at the LCX. These false positives were caused by large areas of calcified plaque that resulted in an over-estimation of the true degree of stenosis. In the four false negative results, two were at the LCX and two were at the RCA. These were the result of artifacts being mistaken as plaques. ## Image Quality Along with the rapid development of multi-detector spiral CT, 320-detector (640-slice) spiral CT can be seen as the highest level of multi-detector CT. A single scan can cover the whole heart, avoiding the stair-step artifacts that are associated with in 64-detector CT. In addition to the reconstruction of more accurate 3D images and CTA images with higher spatial resolution, it also allows for some high-tech applications. For example, arrhythmia scanning software can intelligently identify premature beats, and when checking subjects with premature beats, the software can maximize image quality while reducing radiation dose. If ventricular premature beats are encountered when scanning the first R-R interval, the equipment will automatically stop scanning and collect data at the next period phase. If ventricular premature beats are encountered when scanning R-R intervals after the second interval, the equipment will continue scanning until the end of the next R-R interval. It is very important to ventricular premature beats, and they are abnormal in only one cardiac cycle, so by omitting the R-R interval with premature beat and scanning normal R-R intervals, the necessary period phase intervals can be collected. This is especially important for patients with heart rates below 65 bpm. This is because when a person’s heart rate is below 65 bpm, the equipment will collect a 70%–80% interval of a cardiac cycle. If scanning cannot be stopped during this period, the interval collected would not truly represent 70%–80% of a cardiac cycle. As a consequence, high-quality images cannot be reconstructed. However, 320-slice CT can automatically collect the data from 30%–80% of the next cardiac cycle, thereby avoiding scanning failure and ensuring high image quality. ## Radiation Dose In order to increase time resolution, when scanning the equipment’s beat value should correspond to patient heart rate within a certain range. The equipment’s default ranges are as follows: \<65 bpm, 1 beat; 65–80 bpm, 2 beats; 81–117 bpm, 3 beats;118–155 bpm, 4 beats; and \>156 bpm, 5 beats. Therefore, the higher the beat value, the longer the acquisition time and higher the dose of radiation. Pearson correlation coefficient analysis shows that radiation dose and heart rate in the control group had no statistically significant correlation (correlation coefficient = 0.048). This is likely because heart rate range in the control group (45 to 64 bpm) was narrow, and scanning was completed in a single heart beat (half scan in 1 beat). Only differences in individual body mass index lead to differences in tube voltage and tube current. Therefore differences in radiation dose between individuals in this group has no statistical significance. However, in the premature beat group, heart rate range was larger (55∼106 bpm), so radiation dose and heart rate showed a moderate correlation (correlation coefficient = 0.699). The radiation dose in the premature beat group was significantly higher than that in the control group because when an individual’s heart rate is higher better images can usually be reconstructed at the end of systolic period. For this reason, acquisition was performed from 30% to 80% time phase in R-R intervals in the premature group. Conversely, in the control group only 70%–80% time phase was collected. As a result, the radiation dose (14.62±1.37 mSv) in the premature beat group was about five times higher than the dose (3.08±2.35 mSv) in the control group. The size of scanning field is adjusted in accordance with the size of the patient’s heart. As a rule, the smaller the scanning field, the higher the spatial resolution of images. When S-field is used in scanning, the equipment’s default reconstruction mode is for brain, and higher spatial resolution is required. When M-field is used in scanning, the default reconstruction mode is for body, and the required spatial resolution reduces, so the radiation dose correspondingly reduces as well. Usually heart size precludes the use of L-field scanning. In the premature beat group, based on heart size, 10 subjects should have received S-field scanning, but the scanning field was artificially changed to M-field. These 10 subjects were compared against 10 subjects who underwent S-field scanning. This study found that there was no statistical difference in image quality or other variables between these two groups. Only the radiation dose in S-field scanning was significantly higher than that in M-field scanning. Considering the health of the patients, the control group in this study all underwent M-field scanning. The artificial change of scanning field in this group is not compared and analyzed. However, in theory, M-field scanning can preserve image quality while significantly reducing radiation dose. ## Conclusions In summary, by taking the advantage of a 320-detector CT scanner’s ability to compensate for premature beats, and then manually adjusting scanning field size to the M-field range, high diagnostic accuracy can be maintained while limiting radiation dose. [^1]: Conceived and designed the experiments: BS TZ. Performed the experiments: TZ JB WW DW BS. Analyzed the data: WW DW. Contributed reagents/materials/analysis tools: TZ JB. Wrote the paper: TZ JB BS. [^2]: The authors have declared that no competing interests exist.

# Introduction Human cytomegalovirus (HCMV) is a classic example of a group of herpes viruses, which is found universally throughout all geographic locations and socioeconomic groups, and infects 50% of adults in developed countries. Although HCMV does not cause clinical disease in immunocompetent individuals except as a mononucleosis- like illness which is observed in a small number of infected individuals, HCMV infection is important to the following three high-risk groups: 1) unborn babies with an immature immune system, 2) people who work with children, and 3) immunocompromised people such as organ transplant patients and HIV-infected individuals). Epidemiological studies have shown that 15%–30% of unborn babies who acquire congenital HCMV infection display a variable pattern of pathological sequelae within the first few years of life that may include hearing loss, vision impairment and mental retardation. It has been estimated that in the US alone, each year 8000 newborns have health problems as a results of congenital HCMV infection, with each child costing the US health care system more than \$300,000. Based on the cost and human suffering that would be relieved by reducing the disease burden associated with HCMV infection, the development of a vaccine to prevent HCMV infection or disease was assigned the highest priority, together with vaccines for HIV, TB and Malaria, by the Institute of Medicine (USA) in 1999. It is now well documented that both humoral and cellular (including CD4⁺ T cells and CD8⁺ T cells) immune responses play an important role in the control of HCMV infection and disease. Therefore a formulation based on viral antigens that activate both humoral and cellular immunity is crucial for a successful HCMV vaccine,. During the last 30 years, various strategies, including whole virus, subunit vaccines based on recombinant gB protein, vector vaccines expressing immunodominant antigens (gB protein, pp65 and/or IE-1 protein), DNA vaccine and dense bodies have been developed, and some of these formulations have shown encouraging results in preclinical studies and can even induce HCMV-specific immune responses in some clinical studies. However, none of these vaccines have shown convincing clinical efficacy in the control of HCMV infection or disease, and a clinically licensed HCMV vaccine is still not available. In recent years, increasing evidence has shown that HCMV-specific immune responses are not restricted to gB, pp65 and IE-1 antigens as previously understood, but are directed towards more than 70% of the HCMV reading frames. Therefore, a vaccine which can induce a broad repertoire of HCMV-specific immune responses in different ethnic populations is likely to provide more effective protection against virus-associated pathogenesis. To achieve this goal, we have designed a novel chimeric vaccine based on a replication deficient adenovirus which encodes 46 HCMV T cell epitopes from 8 different HCMV antigens, restricted through multiple HLA class I and Class II alleles, as a polyepitope. This polyepitope was covalently linked to a truncated form of HCMV-encoded gB antigen which allowed the expression of the HCMV polyepitope and gB proteins as a single fusion protein. Pre-clinical evaluation of this recombinant polyepitope vaccine in HLA A2 transgenic mice (referred to as HHD-2) and humans showed that this formulation is capable of inducing pluripotent cellular and humoral immunity *in vivo* and also readily recalls and expands HCMV-specific CD8⁺ and CD4⁺ T cells. # Results ## Immunisation of HHD-2 mice with Ad-CMVpoly and/or Ad-gB vaccine induces multiple antigen-specific cellular and humoral immunity Recent studies on the immune regulation of HCMV in healthy virus carriers and transplant patients have clearly indicated that long-term protection from viral pathogenesis is critically dependent on the induction of cellular immunity which is directed towards multiple viral antigens expressed during different stages of HCMV infection. In the first set of experiments we specifically designed our vaccine strategy to induce a cellular immune response against multiple antigens of HCMV using the polyepitope technology. HLA A2 transgenic mice (referred to as HHD-2) were immunised (intramuscularly) with an adenoviral vector encoding 46 HLA class I and II-restricted T cell epitopes as a polyepitope (referred to as Ad-CMVpoly; 7.5×10⁸ pfu/mouse;). Ten days after immunisation, *ex vivo* T cell reactivity to the HLA A2-restricted peptide epitopes (pooled; see) was assessed by ELISPOT technology. It is important to mention here that the virus dosage for vaccination was selected based on our preliminary studies where a range of varying doses were investigated (data not shown). Splenocytes were used as responder cells for the detection of epitope-specific T cells. Data presented in shows that we consistently observed a strong HCMV epitope-specific T cell response following Ad-CMVpoly vaccination. Analysis of T cell responses to the individual epitopes within the polyepitope sequence indicated that during primary immunisation the dominant T cell response was directed towards the VLE epitope derived from IE-1 antigen, although subdominant responses towards other HLA A2-restricted epitopes NLV (pp65), RIF (pp65), VLA (IE-1), IIY (IE-2) AVG (gB) was also detected. Recent studies have raised some concerns on the use of adenoviral vectors in humans as the pre-existing immunity to adenovirus may compromise the efficacy of these vaccine formulations. To explore this issue, Ad-CMVpoly immunized HHD-2 mice were rested for one or three months and then immunized with the Ad-CMVpoly vaccine (7.5×10⁸ pfu/mouse). Although secondary immunization of mice after one month of vaccination showed very minimal increase in the T cell response (data not shown), a 3–5 fold increase in the T cell response was observed in HHD-2 mice vaccinated after three months of their primary vaccination. More importantly, following secondary immunization, a small but significant increase in the subdominant responses was observed in some animals, while the T cell response towards VLE epitope remained the most dominant component of overall response. It was interesting to note that the hierarchy of these T cell responses in HHD-2 mice was very similar to that observed in HLA A2-positive healthy virus carriers. This observation was co-incident with the dramatic decline of antibodies against adenovirus vector 2 to 3 months after immunisation with Ad-CMVpoly. ## Combination of Ad-CMVpoly and Ad-gB induces long lasting memory cellular and humoral immune responses It is now firmly established that although T cell responses play an important role in controlling persistent HCMV infection, humoral immune responses also contribute significantly in controlling primary HCMV infection and as well reduce viral load by neutralizing the extra-cellular virus. To ensure that our vaccine strategy can induce both cellular and humoral immune responses, we immunised HHD-2 mice with the mixture of Ad-gB (7.5×10⁸ pfu/mouse) and Ad-CMVpoly (7.5×10⁸ pfu/mouse), and anti-HCMV specific cellular and humoral immune responses in immunised animals were evaluated at different time point by ELISPOT and ELISA respectively. Data presented in shows that this vaccination strategy induced both CD8⁺ T cells and gB-specific antibody responses. The levels of CD8⁺ T cells induced by this co- immunisation strategy were comparable to those seen with Ad-CMVpoly alone and were detectable at reasonably high levels on day 75 post-immunisation. The levels of gB-specific antibody responses were maintained at high levels by day 75 post-immunisation, although a small reduction was observed when compared to the levels observed on days 10 and 25 respectively. In contrast, the antibody response showed significant increase in the virus neutralization capacity by day 75 post-immunisation which was co-incident with the antibody avidity maturation. It is important to mention here that this increase in neutralization capacity was not due to antibody isotype switching, which was consist with previous studies. These observations suggested that co-delivery of HCMV polyepitope and gB vaccine can induce long lasting memory T cells immunity and antibody responses. ## Covalent linking of HCMV polyepitope with extracellular gB induces pluripotent T cell and antibody responses Although co-immunisation with Ad-gB and Ad-CMVpoly induced both humoral and cellular immune responses against HCMV, delivery of this formulation in a human setting may face significant regulatory constraints. To overcome this potential limitation, we constructed another recombinant adenovirus expressing the extracellular domain of gB and HCMV polyepitope as a single polypeptide (referred as Ad-gBCMVpoly). HHD-2 mice were immunised with the Ad-gBCMVpoly vaccine (7.5×10⁸ pfu/mouse) and both humoral and cellular immune responses were evaluated at the indicated time points. Data presented in shows that immunisation with Ad-gBCMVpoly vaccine induced a long-tem memory CD8⁺ T cell response towards the HLA A2-restricted epitopes from HCMV. Furthermore these animals also showed strong gB-specific antibody response and similar to the data presented in, the levels of gB-specific antibody dropped by day 75 post immunisation. A significant increase in the neutralizing activity of the antibody response was observed, which was co-incident with avidity maturation. On the other hand, there was no antibody isotype switching at different time point after immunisation. These observations clearly demonstrated that covalent linking of the gB with the polyepitope sequence does not impair the immunogenicity of each of the components of the vaccine. To further characterize the T cell responses induced by Ad-gBCMVpoly vaccine, we next assessed whether immunisation with Ad-gBCMVpoly result in the differentiation of antigen-specific T cells into fully functional effectors. A number of recent studies have demonstrated that the production of TNF-α in addition to IFN-γ by T-cells is a characteristic of greater differentiation and can enhance protection against infectious pathogens, and the translocation of CD107a from intracellular lysosomal and endosomal compartments to the surface of CD8⁺ T cells is a positive marker of degranulation, a requisite process of perforin-granzyme mediated killing function of CTLs. We assessed the level of TNF-α and/or CD107a expression by IFN-γ expressing CD8⁺ T-cells using intracellular cytokine assays. Data presented in shows that following *ex vivo* stimulation with HCMV epitopes, CD8⁺ T cells from these mice showed strong IFN-γ expression and a large proportion of these T cells also expressed TNF-α and/or CD107a. Furthermore, after *in vitro* stimulation with individual HCMV peptides, these HCMV peptide-specific CD8⁺ T cells could be expanded and expressed both IFN-γ and TNF-α. ## Protection against quasi-virus challenge following immunisation with Ad-gBCMVpoly Having firmly established the immunogenicity of Ad-gBCMVpoly vaccine, the next set of experiments was designed to determine protective efficacy of this vaccine. Due to the species restriction, we challenge immunised HHD-2 mice with recombinant vaccinia encoding HCMV antigens (gB and IE-1) to evaluate the protective efficiency of the Ad-gBCMVpoly vaccine. Data presented in shows that HLA A2 mice immunised with Ad-gBCMVpoly vaccine showed significant reduction in the virus load following challenge with Vacc.gB and Vacc.IE-1. This reduction in the virus load was highly antigen-specific as the vaccinated or naïve animals challenged with Vacc.TK- or Vacc.gB, Vacc.IE-1 respectively showed minimal reduction in the viral load. Although Ad-gBCMVpoly immunized mice showed better protection against Vacc.gB when compared to Vacc.IE-1, this better protection was not due to anti-gB antibodies as Vacc.gB was not neutralized by serum from immunized animals (data not shown), but due to gB-specific CD4⁺ T cell responses. Nevertheless, the anti-gB humoral response should play an important role in human as it induces HCMV neutralizing antibodies. As expected, the reduction in the Vacc.IE-1 virus load in Ad-gBCMVpoly immunised mice was co- incident with the induction of VLE-specific CD8⁺ T cell responses. It is important to note that Ad-gBCMVpoly immunised mice challenged with Vacc.gB or Vacc.IE-1 showed significantly higher humoral and T cell responses respectively when compared to mice challenged with Vacc.TK-. We also assessed the level of TNF-α expression by IFN-γ expressing CD4⁺ and CD8⁺ T-cells using intracellular cytokine assays. Data presented in shows that following stimulation with gB protein or HCMV IE-1 epitope, a large proportion of CD4⁺ and CD8⁺ T cells from these mice showed strong co- expression of IFN-γ and TNF-α. ## Expansion of multiple antigen-specific human CD8⁺ and CD4⁺ T cells following stimulation with Ad-gBCMVpoly Another important aspect of the current study was aimed at exploring the potential efficacy of Ad-gBCMVpoly to recall memory T cell responses from healthy seropositive individuals. PBMC from healthy donors were stimulated with irradiated autologous PBMC-infected with Ad-gBCMVpoly. Following stimulation these T cells were assessed for antigen specificity using intracellular cytokine assays. Data for the gB-specific CD4⁺ and CD8⁺ T cell responses are summarised in, while the T cell responses towards the epitopes within the polyepitope sequence are presented in &. To identify the gB-specific T cell responses we used an overlapping set of peptides based on the gB sequence from Ad169 strain of HCMV. This analysis showed that following stimulation with Ad-gBCMVpoly, more than 88% of the individuals showed expansion of gB-specific CD4⁺ T cells. These T cell expansions raged from 2–36% of the total CD3⁺ CD4⁺ T cells. CD8⁺ T cell responses directed towards gB epitopes were detected in 70.5% donors which ranged from 2–15% of the total CD3⁺ CD8⁺ T cells. T cells from each donor recognized multiple gB epitopes and most of the donors demonstrated a selective expansion of gB-specific CD4⁺ or CD8⁺ T cells. Analysis of the T cell responses towards the epitopes within the polyepitope sequence revealed that there was a rapid expansion of CD8⁺ T cells following stimulation with Ad-gBCMVpoly which recognized multiple epitopes restricted through a number of HLA class I alleles. In most cases, dominant CD8⁺ T cell expansions directed towards 2–3 different epitopes was observed; whilst in other donors (e.g. D9, D10 and D13) strong T cell reactivity towards more than five epitopes was observed. In vitro testing of these T cells also showed that these cells expressed high levels of CD107 and efficiently recognized HLA-matched HCMV-infected target cells (data not shown). These observations were also confirmed by ex vivo stimulating the PBMC from healthy virus carriers with Ad-gBCMVpoly. A representative data presented in clearly shows that ex vivo stimulation of PBMC rapidly stimulated HCMV epitope specific T cells and these cells showed strong expression of IFN-γ. Although the polyepitope sequence was predominantly based on CD8⁺ T cell epitopes, two previously mapped CD4⁺ T cell epitopes were also included in this sequence. As expected, a strong expansion of CD4⁺ T cells specific for these epitopes was observed, however unexpectedly, we also detected low to medium levels of expansion of CD4⁺ T cells which showed reactivity against HLA class I-restricted CD8⁺ T cell epitopes. A careful analysis of these CD8⁺ T cell epitopes revealed that many of these sequences overlapped the CD4 epitopes mapped recently by other investigators. # Discussion The data presented in this study provides a highly efficient strategy for the prevention of HCMV disease in different clinical settings ranging from congenital infection to primary or reactivation of the virus in immunosuppressed adults. The importance of HCMV as the leading infectious cause of mental retardation and other abnormalities such as deafness in children has been emphasized by its categorization by the Institute of Medicine as a Level I vaccine candidate \[i.e. most favourable impact–saves both money and quality- adjusted life years\]. Immunocompromised individuals such as transplant recipients and HIV-infected individuals with CD4 counts below 50/µl are also impacted by HCMV infection and this virus is regarded as the most important viral pathogen affecting transplantation, including both solid organ transplant and allogeneic hematopoietic stem cell transplant recipients. Extensive studies over the last decade on the immunobiology of HCMV infection has provided detailed insight into the immune regulation of persistent HCMV infection in healthy virus carriers and individuals with HCMV-associated diseases. Based on these observations, a number of attempts have been made to design a prophylactic vaccine for the control of HCMV infection. The first series of attempts focussed on the use of an attenuated form of the virus as a vaccine, however, disappointing results coupled with the regulatory problems associated with the live attenuated HCMV vaccine prompted investigators to switch to the recombinant subunit approach. Although the subunit vaccine delivery systems and modalities based on HCMV encoded antigens such as gB, pp65 and IE-1 have failed to result in a licensed clinical product, interesting pre-clinical (based on animal models) and clinical data continues to accumulate demonstrating that subunit vaccination has a protective effect against congenital transmission. It is now firmly established that long-term latent HCMV infection is very efficiently controlled by virus-specific CD4⁺ and CD8⁺ T cells. Perturbation in the regulation of T cell control often triggers reactivation of HCMV and development of HCMV-associated diseases. The concept that a vaccine based on T cell-mediated control would be effective in controlling HCMV diseases grew out of the pioneering work conducted by Riddell and colleagues, who showed that adoptive transfer of donor-derived virus- specific T cells alone were sufficient to reduce the incidence of HCMV disease in allogeneic hematopoietic stem cell transplant recipients. Over the last few years there has been a series of attempts to develop a highly tailored vaccine strategy designed to induce T cell immunity against pp65 and/or IE-1 antigens or defined T cell epitopes from these antigens. While these strategies provided specificity and safety, their application at the population level are rather limited. Thus other approaches which target multiple antigens might be an advantage by providing wider coverage in different ethnic groups. Furthermore, inclusion of a virus neutralization component in the vaccine formulation has been argued by many investigators, especially in the context of congenital HCMV infection. Indeed the chimeric vaccine developed in this study induced high avidity humoral responses and cellular immunity with a single formulation and provided wider coverage through the inclusion of multiple T cell epitopes restricted through a range of HLA class I and II alleles. Our initial studies with a mixture of adenoviral vectors encoding HCMV polyepitope sequence and gB protein showed that it was possible to induce both humoral and cellular immune responses without compromising the immunogenicity of individual components of the vaccine. Taking into consideration these observations, we designed a chimeric vaccine in which the encoding sequence for the extracellular domain of gB was covalently linked with the polyepitope sequence. Extensive studies with this formulation provided further evidence that co-delivery of gB and the polyepitope as a single polypeptide was highly efficient in generating neutralizing antibodies responses and virus-specific CD8⁺ and CD4⁺ T cell responses in a murine model and healthy virus carriers. Subsequently, we employed an experimental animal model system to determine whether immunisation of HLA A2 transgenic mice with Ad-gBCMVpoly is capable of reducing infection with a recombinant vaccinia virus expressing HCMV antigens (i.e. gB and IE-1). These mice not only showed induction of a strong CD4⁺ and CD8⁺ T cell response following immunisation but also acquired strong resistance to virus infection. Interestingly, Ad-gBCMVpoly immunized showed better protection against Vacc. gB when compared to Vacc.IE-1, which suggested that gB-specific CD4⁺ T cell responses could also inhibit Vacc.gB virus. Another important outcome of this study was the longevity of the immune responses induced by the chimeric HCMV vaccine which is particularly critical for the vaccine designed to control congenital infection/disease where long-term memory response over multiple years would be essential. Although the studies outlined here does not allow any firm conclusions on the efficacy of the chimeric polyepitope-based vaccine in humans, it does clearly show that a formulation based on gB and HCMV T cell epitopes can be used as immunogens to induce efficient humoral and T cell responses in vivo. It is important to stress here that a polyepitope-based vaccine for HCMV has a number of advantages over the traditionally proposed vaccines, which are based on either full-length HCMV antigens or synthetic peptide epitopes. There is now convincing evidence that polyepitope proteins are extremely unstable and are rapidly degraded by the proteasome dependent pathway as a result of their limited secondary and tertiary structure. The rapid degradation of these polypeptides dramatically enhances endogenous presentation of peptide epitopes through the class I and II pathway. On the other hand the full-length HCMV protein antigens are unlikely to be degraded rapidly and may also initiate various intracellular signalling events leading to the interference of presentation of epitopes from other antigens. Finally, the polyepitope-based vaccine is likely to overcome any potential problem of reinfection with different strains of HCMV and unique HLA types in different ethnic groups of the world. There is an emerging argument that HCMV vaccine efforts should focus on the development of formulation(s) which are designed to limit or prevent HCMV related diseases rather than to prevent infection itself. This contention is supported by extensive studies in humans which revealed that although the immune responses generated during natural HCMV infection are unable to clear the latent virus, this response is sufficiently competent to keep the virus under control and restrict virus replication. Furthermore, in immunocompromised patients such as HIV-infected individuals and transplant recipients, HCMV related pathogenesis is generally due to reactivation rather than primary infection. Considering the limited efficacy of the currently available HCMV vaccine formulations in protecting against infection in preclinical and clinical studies, we propose that a vaccine to limit or prevent HCMV related disease rather than infection itself is more realistic in the near future. # Materials and Methods ## Construction of recombinant adenovirus encoding HCMV polyepitope, gB and gB-HCMV polyepitope fusion protein The amino acid sequence of the 46 contiguous HLA class I and class II-restricted T cell epitopes were translated to the nucleotide sequence using human universal codon usage. Oligonucleotides (102–107mer long) overlapping by 20 base pairs and representing the polyepitope DNA sequence, were annealed together by using Splicing by Overlap Extension and stepwise asymmetric PCR. The final PCR product was cloned into pBluescript II KS⁺ phagemid (Agilent Technologies, Melbourne, Australia) encoded a Kozak sequence, Start methionine followed by 46 contiguous HLA class I and class II-restricted epitopes. The HCMV sequence encoding glycoprotein B (gB) was amplified from the AD169 virus stock by PCR using gene specific primers. This PCR product was designed to encode gB sequence from the alanine residue at position 31 to valine at position 700 with the deletion of the signal sequence. Following amplification the DNA was cloned into pBluescript II KS⁺ phagemid and confirmed by DNA sequence analysis. For the expression of the gB-HCMV polyepitope fusion protein the recombinant HCMV polyepitope insert was excised from the pBluescript II KS⁺ phagemid and cloned into the gB pBluescript construct. The assembly and production of the recombinant Ad5F35-based adenoviruses was completed in three stages using a highly efficient, ligation-based protocol of the Adeno-X System (CLONTECH, Palo Alto, CA) (See). Firstly, inserts were excised from each of the constructs in pBluescript II KS⁺ phagemid using Xba I/Kpn I restriction enzymes and cloned into the pShuttle expression vector. Following amplification in E.coli, the expression cassette from pShuttle was excised using I-Ceu I/PI-Sec I homing enzymes and cloned into an Ad5F35 expression vector. The recombinant Ad5F35 vector was transfected into human embryonic kidney (HEK) 293 cells, and the recombinant adenoviruses (referred to as Ad-CMVpoly, Ad-gB and Ad-gBCMVpoly) were harvested from the transfected cells by successive freeze-thawing cycles. ## Synthesis of Peptides Peptides, synthesized by the Merrifield solid phase method, were purchased from Chiron Mimotopes (Melbourne, Australia), dissolved in dimethyl sulphoxide, and diluted in serum-free RPMI 1640 medium for use in standard T cell assays. Purity of these peptides were tested by mass spectrometery and showed \>90% purity ## Animals and immunisation HLA A2 transgenic mice (referred to as HHD-2), were maintained under conventional conditions the animal facility at the Queensland Institute of Medical Research. These mice are knocked out for β2 microglobulin and H-2D^b and transgenic for a chimeric HLA-A2.1 with the α3 domain derived from H-2D^b to allow interaction with murine CD8 and a covalently attached human β2 microglobulin. These mice were immunised with varying doses of plaque forming units (PFU) of recombinant viruses (Ad-CMVpoly, Ad-CMVgB and Ad-gBCMVpoly) and HCMV-specific humoral and cellular immune responses were evaluated at various time points. Protocols were approved by QIMR animal ethics committee. ## ELISpot assay The ELISPOT assay was used to detect HLA A2-restricted HCMV epitope-specific T cells following stimulation with synthetic peptide(s) as described previously. Briefly, 2×10⁵ responding cells were incubated in triplicate with each peptide epitope (1 µg/ml) for 18 to 20 hrs in 96- well Multiscreen HA filtration plates (MAHA S4150, Millipore, Bedford, MA) coated with anti-IFN-γ monoclonal antibody (Mabtech AB, Nacka, Sweden). After incubation, the plates were extensively washed with Phosphate buffered saline with 0.5% Tween 20 and incubated with a second biotinylated anti-IFN-γ mAb followed by the addition of streptavidin conjugated alkaline phosphatase. Cytokine producing cells were detected as purple spots after a 30-min reaction with 5-bromo-4-chloro-3-indolyl phosphate and nitro blue tetrazolium. Spots were counted automatically using image analysis software. T cell precursor frequencies for each peptide epitope were based on the total number of cells and the number of spot forming cells (SFC) per well (average of 3 wells). Epitope-specific spots were calculated after subtraction of the number of spots in control wells consisting of cells without added peptide (average of six wells). ## Intracellular Cytokine Staining Splenocytes from immunised mice or T cells from human donors were incubated for overnight at 37°C with HCMV peptide epitopes (1 µg/ml), or stimulator cells either pre-coated with HCMV peptide epitopes (1 µg/ml) or infected with recombinant vaccinia virus encoding HCMV antigens, in growth medium. Brefeldin A (BD Pharmingen, San Diego, CA) was added during the last 5 hour-incubation. For CD107a staining, anti-CD107a antibody was added one hour before the adding of Brefeldin A. These cells were then washed and incubated with PerCP-conjugated anti-CD8, FITC conjugated anti-CD4 and Allophycocyanin-conjugated anti-CD3 at 4°C for 30 mins. Cells were washed, then fixed and permeabilised with cytofix/cytoperm (BD Pharmingen) at 4°C for 20 minutes. Cells were then washed in perm/wash (BD Pharmingen), incubated with anti-IFN-γ and anti-TNF-α mAbs (BD Pharmingen) at 4°C for 30 mins, washed again with perm/wash, resuspended in PBS and analysed on a FACS Canto. ## Expansion of HCMV specific T-cells from healthy donors using Ad-gBCMVpoly A panel of 17 human volunteers were recruited for this study. Each volunteer was asked to sign the consent form as outlined in the institutional ethics guidelines. For the expansion of specific T-cells, peripheral blood mononuclear cells (PBMC) were co-cultured in multi-well tissue culture plates in growth medium with either PBMC (2,000 rad) infected with Ad-gBCMVpoly (MOI of 10∶1) at a responder to stimulator ratio of 2∶1. On day 3, and every 3–4 days thereafter, the cultures were supplemented with growth medium containing recombinant IL-2 (kindly donated by NIH AIDS Research & Reference Reagent Program). These T-cell cultures were assessed for HCMV epitope-specific reactivity on days 10–17. ## ELISA assay for anti-gB and anti-adenovirus antibody Serum anti-gB or anti-adenovirus antibody titres were evaluated by ELISA as previously described. Briefly, PVL microplate 96-well plates (MP Biomedicals, Sydney, Australia) pre-coated with recombinant HCMV gB protein or adenovirus were incubated with serially diluted serum samples for 2 hours at room temperature. After washing with PBS-Tween-20 (PBST), plates were incubated with HRP-conjugated sheep anti-mouse Ig antibody (murine samples) or HRP-conjugated sheep anti-human Ig antibody (human samples) for 1 hour. These plates were washed and incubated with 3.3′, 5.5′-tetramethylbenzidine substrate solution (PanBio, Brisbane, Australia) and the OD at 450 nm was analysed using an ELISA reader. The isotypes of anti-gB antibodies in serum samples were determined by ELISA as described above using the mouse monoclonal antibody isotyping reagent kit (Sigma, IS02-1 kit, Sydney, Australia) according to the manufacturer's protocol. Antibody avidity was evaluated as previously described. Briefly after incubation of plates with serum samples as described above, 5 M Urea (in PBST) was then added to half of the wells for dissociation and the other half received PBST without urea. After incubation for 30 min, plates were washed with PBST and incubated with HRP-conjugated sheep anti-mouse Ig antibody (murine samples) or HRP-conjugated sheep anti-human Ig antibody (human samples) for 1 hour and completed using the standard ELISA. The avidity indices were calculated as the ratio of the OD values with urea divided by the OD values without urea and expressed as a percentage. ## CMV microneutralization assay The neutralizing activity of the anti-gB antibody response in vaccinated animals was assessed as described previously. Briefly serum samples were initially incubated at 56°C for 30 minutes to inactivate complement, followed by serial dilution (25 µl/well) with DMEM medium in 96-well “U” bottom plates. In each well an equal volume of HCMV Ad169 was added and incubated at 37°C for 2 h. This virus was then transferred to infect monolayer of human fibroblast MRC-5 cell culture in 96 well flat bottom plates with 80–90% confluence. After 2 h, plates were washed with DMEM and 200 µl DMEM with 10% FCS were added to each well and then incubated at 37°C for 16–18 h. After incubation, cells were fixed in 100% methanol, incubated with peroxidase blocking reagent (Chemicon, S2001) and then reacted with mouse anti-CMV IE-1/IE-2 monoclonal antibody (Clone MAB810, Chemicon) followed by HRP-conjugated sheep anti-mouse Ig (Chemicon, AP326P). Finally cells were stained with DAB+ substrate (Chemicon, K3467) according to manufacturer's protocol. The numbers of nuclei with brown colour staining were counted using inverted microscope. The neutralizing titre was calculated as the reciprocal of sera dilution that gave 50% inhibition of IE-1/IE-2-expressing nuclei. ## Vaccinia virus recombinant Recombinant vaccinia constructs encoding HCMV antigens IE-1 (Vacc.IE-1), gB (Vacc.gB) and a negative control vaccinia virus construct made by insertion of the pSC11 vector alone, which is negative for thymidine kinase (Vacc.TK⁻), have been previously described. ## Protection assay HHD-2 mice were intramuscularly immunised with the indicated vaccine on day 0, followed by intraperitoneal challenge with recombinant vaccinia virus expressing different proteins from HCMV antigens (Vacc.IE1 or Vacc.gB) at a dose of 10⁷ pfu/mouse on day 21. Mice were then sacrificed 4 days later, spleens collected to evaluate epitope-specific T cell response by IFN-γ ICS assay, ovaries collected to determine vaccinia virus load by plaque assay on monkey fibroblast CV-1 cells, and sera collected to evaluate anti-gB Ab titres by ELISA. To determine vaccinia viral titres, monolayers of CV-1 cells in a 6 well flat bottom plates were incubated for 2 h at 37°C with serially diluted ovary lysates. After incubation, 2 ml of RPMI1640 medium supplemented with 2% FCS and 0.75% methylcellulose was added to each well and incubated for further 3 days. After three days, plates were washed with PBS and stained with crystal violet solution (Sigma, HT901) at a working concentration (0.1% crystal violet in 15% ethanol) for 30 min and the number of plaques were counted using standard procedures. The authors wish to thank Dr Judy Tellam for assistance in preparation of the adenovirus constructs. [^1]: Conceived and designed the experiments: RK. Performed the experiments: JZ MR LC CS. Analyzed the data: JZ MR LC CS RK. Contributed reagents/materials/analysis tools: LC. Wrote the paper: RK. [^2]: The authors have declared that no competing interests exist.

# 1. Introduction Tuberculosis (TB) continues to be a major cause of morbidity and mortality in many low and middle-income countries. Global TB incidence is estimated to be 10.1 million, and mortality 1.3 million. Diagnosis of pulmonary TB (PTB) is based on the detection of *Mycobacterium tuberculosis* (MTB) in the sputum by smear microscopy for acid-fast bacilli (AFB) which is a widely available, simple, and inexpensive tool. The standard treatment for PTB includes 2 months of therapy with isoniazid, rifampicin, pyrazinamide and ethambutol (intensive phase), followed by 4 months of treatment with isoniazid and rifampicin (maintenance phase). This treatment regimen is considered curative for drug sensitive PTB. Response to TB treatment is monitored by follow-up sputum smear microscopy at 2 and 5 months. Diminishing numbers of AFB to smear-negative status during treatment is considered an indication of treatment success, and the sputum smear conversion is considered as a reliable marker for successful treatment. However, it has been shown in a recent study that viable cultivable bacilli were detected in 6% of patients by culture despite successful sputum smear conversion. These findings highlight the need for improvement in monitoring treatment response by developing a sensitive, specific, and rapid surrogate marker for quantifying TB other than the gold standard of culture. It is also not feasible to perform culture in routine practice for monitoring treatment response in the low-resource high TB-endemic setting due to the need for extensive laboratory resources and the long turn-around time. Furthermore, a large proportion of patients with PTB are sputum-smear negative, and physicians tend to treat patients without bacteriological confirmation if the patients come from well-known endemic areas of TB and have a high suspicion of the disease. Monitoring therapeutic response is thus crucial in these cases to enable timely response if a change in treatment is required. The regression of clinical findings does not always correspond to treatment success, and radiological findings often take longer for complete regression. Although routine biomarkers, such as C-reactive protein and elevated leukocyte count and erythrocyte sedimentation rate, decrease with satisfactory TB response, they may not be raised in the chronic milder forms of TB, making them not applicable for treatment monitoring in all forms of TB. There is hence a need for more reliable biomarkers to assess the short and long-term treatment response. Peripheral blood mononuclear cells (PBMCs) are generally used as a model system to investigate immune response in infectious diseases such as TB, where the cell-mediated immune response is mainly responsible for disease processes. Given that PBMCs are the primary cells that play a crucial role in the disease process, their profile is expected to change after containment of MTB by the treatment. The aim of the study was to conduct proteomic profiling of the PBMCs isolated from patients with PTB as a means of exploring the immunopathological changes consequent to TB treatment and the potential of using this information to develop a biomarker for monitoring therapeutic response. # 2. Materials and methods ## 2.1 Study design and setting This was a pilot study conducted among adult patients with confirmed PTB who were participating in another larger progressive cohort study on the validation of a new diagnostic test in Zanzibar, Tanzania. Zanzibar is a semi-autonomous region of the United Republic of Tanzania, with a population of around 1.3 million. In recent years, the total number of TB cases has significantly increased in Tanzania from 62,180 cases in 2015 to 75,845 cases in 2018, representing a 22% increase. The notification rate of new and relapse TB cases has also increased from 128 per 100,000 population in 2015 to 138 per 100,000 population in 2018. In contrast, the number of TB cases co-infected with HIV has decreased by more than one-third between 2015 and 2018. Patients were recruited at Mnazi Mmoja Hospital, which is the only tertiary care referral hospital in Zanzibar, from August 2014 to September 2015. TB was confirmed bacteriologically by a positive MTB culture and/or MTB detected by the Xpert^® MTB/RIF assay (Cepheid Inc, USA) in at least one patient sputum specimen. Patients were excluded if they did not give consent or had received anti-TB treatment in the last 12 months. Blood samples were collected from each patient at three timepoints: at baseline before starting the treatment "0M", at 2 months (during the intensive phase of treatment) "2M", and at the end of 6 months of treatment "6M". The study was conducted according to the principles of the Declaration of Helsinki and approved by the Regional Committee for Medical and Health Research Ethics of Western Norway (REK Vest), and the Zanzibar Medical Research and Ethics Committee (ZAMREC). All patients provided written informed consent. ## 2.2 Preparation of proteins from PBMCs and protein digestion Blood samples (4 mL) were collected using BD Vacutainer^® CPT^™ (cell preparation tubes with sodium heparin). Blood samples were then centrifuged following manufacturer’s instruction, and PBMCs were collected. The PBMCs were subsequently washed thoroughly to remove the plasma, dextran, and other components of the gel in the CPT^™ tube. The first wash was done with TBS by diluting 1:5. The whole tubes, which were about 6 mL each, were filled up. Then, the samples were centrifuged at 400 × g at room temperature for 10 minutes, to make sure the cells were pelleted. The supernatant was removed. The second wash was done by adding another 4 mL of TBS in the tube with a cell pellet. The pellet was gently dissolved by pipetting in and out a few times. Afterward, another centrifugation at 400 × g at room temperature for 5 minutes was performed to make sure that the cells were pelleted. The supernatant was removed. The third wash was done in a similar manner to the second wash. Following this, red blood cells were lysed by using a red cell lysis buffer with 150 mM of ammonium chloride. Approximately 2.5 mL of the lysis buffer was added to the cell pellet; then, the tubes were shacked to dissolve the pellet. The tubes were subsequently incubated for 10 minutes at room temperature, and centrifugated at 400 × g at room temperature for 5 minutes. Finally, the supernatant was decanted. PBMCs pellets were extracted into a 100-μL lysis buffer consisting of 0.1 M Tris/HCl (pH 7.5), 0.1 M dithiothreitol (reducing agent), and 2% SDS. The samples were subsequently transported to the University of Bergen. The proteomics analysis was done at the Proteomics Unit of the University of Bergen (PROBE). Protein concentrations were measured using Direct Detect^® (Merck KGaA, Darmstadt, Germany), which is an infrared-based biomolecular quantitation system that provides accurate and precise results despite the presence of SDS. Proteins were digested using the FASP method as described by Hernandez-Valladares et al.. In brief 20 ug of protein were reduced with 0.1 M dithiotreitol (DTT) and heated to 95 °C for 5 min. The proteins were alkylated with 50mM iodoacetamide (IAA). Buffer exchange was performed in a Microcon-30 kDa Centrifugal filters (Millipore, \#MRCF0R030) using 8 M urea in 0.1 M Tris–HCl pH 8.5, freshly prepared). Trypsin was dissolved in 50mM ammonium bicarbonate and added to the samples in a 1:25 ratio, samples were incubated at 37 °C for 16 h. Desalting was done using Oasis HLB 96-well μElution plate (2 mg sorbent per well, Waters \#186001828BA). ## 2.3 LC/MS method 5 ug of digested peptides were pressure-loaded onto an HPLC column (Acclaim^™ PepMap^™ 100 C18, 3 μm, 75 μm × 2 cm, Thermo Fisher Scientific, Bremen, Germany), with trapping and desalting carried out at 5 μL/min for 5 minutes using 0.1% of trifluoroacetic acid. Analytical separation was carried out with Acclaim^™ PepMap^™ 100 C18 (3 μm, 75 μm × 50 cm, Thermo Fisher Scientific, Bremen, Germany) at a flow rate of 270 nL/min. The elution gradient was run using mobile phase A (0.1% of formic acid in water) and B (100% ACN). Tryptic peptides underwent a 20-minute isocratic elution with 80% buffer B followed by another 20-minute isocratic elution with 5% buffer B. Total gradient time was 4h. The reason for using a 4h gradient was to increase the number of identified proteins. As peptides were eluted from the HPLC column, they were electrosprayed directly into a linear quadrupole ion trap-orbitrap mass spectrometer (LTQ-Orbitrap Elite^™, Thermo Fisher Scientific, Bremen, Germany). The mass spectrometer was operated in the data-dependent acquisition mode to automatically switch between full-scan MS and MS/MS acquisition. Instrument control was through Tune 2.7.0 and Xcalibur 2.2. The mass spectrometric data was acquired in positive ion mode, with an 1,800-V ion spray voltage, no sheath and auxiliary gas flow, and a capillary temperature of 260°C. Survey full-scan MS spectra (from m/z 300 to 2,000) were acquired in the Orbitrap with a resolution of 240,000 at m/z 400 (after accumulation to a target value of 1e6 in the linear ion trap with the maximum allowed ion accumulation time of 300 ms). The 12 most intense eluting peptides above an ion threshold value of 3,000 counts and charge states of ≥2 were sequentially isolated to a target value of 1e4 and fragmented in the high-pressure linear ion trap by low- energy CID with a normalized collision energy of 35% and wideband-activation enabled. The maximum allowed accumulation time for CID was 150 ms, with an isolation window of 2 Da, an activation q value of 0.25, and an activation time of 10 ms. The resulting fragment ions were scanned out in the low-pressure ion trap at a normal scan rate and recorded with the secondary electron multipliers. One MS/MS spectrum of a precursor mass was allowed before dynamic exclusion for 40 seconds. Lock-mass internal calibration was not enabled. ## 2.4 Data management and analysis Four softwares were used for this study: MaxQuant v1.5.5.1, Perseus v1.5.6.0, SPSS v25, and Excel 2019. The raw files from the LC-MS/MS were analyzed using MaxQuant version 1.5.5.1 and the integrated Andromeda search engine. The fasta file version was Sprot_Human_20432entries_20190903.fasta. Moreover, for both proteins and peptides, the maximum FDR was set to 0.01. MaxQuant maps the sequences of detected peptides and uses these peptide levels to determine the identified protein level. Since the protein levels likely vary between samples due to minor differences in handling and analysis, normalization of protein levels is essential. Consequently, label-free quantification (LFQ) algorithms within MaxQuant, MaxLFQ, were used to create the normalized protein intensities. They were normalized in relation to the levels of common proteins in a sample. To construct a relative scale, LFQ uses the signal intensity and the number of observations of commonly observed peptides. This is used to assign new, normalized intensities of peptides, along with an absolute scale of summed-up peptide intensities, LFQ intensities. The LFQ algorithm is incorporated into the search engine of MaxQuant and produces two distinctive data outputs: samples without standardized levels and the same samples with LFQ corrected levels. While the unnormalized spectra–"Intensity"–have been merely used to detect the presence of proteins within a sample, the LFQ values–"LFQ intensity"–were used for statistical analysis. The normalized data from MaxQuant were saved in a.txt file. The file was uploaded to Perseus version 1.5.6.0, and "LFQ intensities" were selected as expression data. Potential contaminants, reverse hits, rows only identified by site, and empty rows were removed from the matrix. The different samples were then grouped into "0M", "2M", and "6M", and a matrix was generated. The intensities values were transformed to log2 values, and gene annotations were uploaded for Homo sapiens. S1 Fig in shows the unsupervised hierarchical clustering of all samples. The variability between samples was very low as shown in. S1 Table in provides information about the standard deviation and coefficient of variation of proteins within samples and missing values. S2 Table in provides information about the standard deviation and coefficient of variation of proteins between samples. S3 Table in shows the Pearson correlation values between samples. S2 Fig in illustrates clustering of Pearson correlation values. S3 Fig in shows the sample distribution of all samples. A histogram was made from each sample were the log2 LFQ intensity was plotted against counts. To compare the differing expression of proteins detected between the groups, ANOVA was carried out using permutation-based FDR, with the number of randomizations set at 250 which is the default setting in Perseus and the FDR at 0.05. The data were normalized on the protein level with Z-scoring prior to hierarchical clustering. In addition, we performed a mixed linear model analysis with correction for multiple testing for comparison (S4, S5 Figs). R-scripts and data output can be found in the S4 Table in and. To visualize the results, a heat map was created to evaluate the significantly differently proteins’ levels between the groups. The data were not imputed. The significantly expressed proteins of interest were further analyzed with IBM SPSS Statistics 25. Besides one-way ANOVA, the Tukey’s range test was also carried out. After running one-way ANOVA and Tukey’s range test in SPSS, box plots were constructed for each of the significantly expressed proteins to illustrate the mean and the median Log2 intensity differences of these proteins detected in the PBMCs of PTB patients at different treatment time points (0M, 2M and 6M). A Principal component analysis were performed on all samples in Perseus using Benjamini-Hochberg FDR 0.05. The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository with the dataset identifier PXD029634. ## 2.5 Protein interaction and pathway analysis The STRING database (STRING v.11.0; [www.string-db.org](http://www.string- db.org)) was used to identify known and predicted functional networks and to predict protein-protein interactions. Gene ontology (GO) annotation ([www.geneontology.org](http://www.geneontology.org)) was conducted to classify proteins based on biological process (BP), molecular function (MF) and cellular component (CC) using the Protein Analysis Through Evolutionary Relationships database ([www.pantherdb.org](http://www.pantherdb.org)). Pathway analysis and biological reactions were performed using Reactome version 72 ([www.reactome.org](http://www.reactome.org)). # 3. Results ## 3.1 Patient characteristics presents the characteristics of the study participants. Overall, 8 patients with bacteriologically confirmed PTB were enrolled into this pilot study. All patients had a positive sputum smear at 0M. After 2 months of receiving standard TB treatment (isoniazid, rifampicin, pyrazinamide, and ethambutol), 6 patients had negative sputum, while 2 patients continued to have a positive sputum smear, which turned negative at 5 months. At 6M, 6 patients had a negative sputum smear, 2 patients were lost to follow up at the end of the study, and one blood sample at 2M was not analyzed. ## 3.2 Proteins changing significantly under and after treatment In total, 3,697 proteins were quantified across the samples using the LFQ shotgun proteomics analysis in Perseus. After removing potential contaminants, reverse hits, rows only identified by site, and empty rows from the matrix, 3,530 proteins were detected. ANOVA of the log2-transformed LFQ spectra performed on the PTB samples at 0M (n = 8), 2M (n = 7), and 6M (n = 6) detected 12 differentially expressed proteins, of which 7 were downregulated, and 5 were upregulated at 2M and 6M. As illustrated in, which showcases the hierarchical clustering of the 12 differentially expressed proteins, three main clusters of proteins were identified, and each cluster matched precisely to the treatment progress grouping. shows the principal component analysis of all samples showing the clustering of proteins according to the treatment groups. The 7 downregulated proteins identified in the PTB samples at 2M and 6M were heat shock 70 kDa protein 1A/1B (HSPA1B/HSPA1A), heat shock protein 105 kDa (HSPH1), heat shock protein 90 alpha family class A member 1 (HSP90AA1), lipopolysaccharide binding protein (LBP), complement component 9 (C9), calcyclin-binding protein (CACYBP), and protein transport protein Sec31A (SEC31A). LBP was reduced to 5- and 6-folds at 2M and 6M respectively, while the fold reduction in the rest of proteins was between 1.76 to 0.25. All these proteins decreased significantly at 2M, and further treatment lead to a relatively lesser decrease. The 5 upregulated proteins were SEC14 domain and spectrin repeat-containing protein 1 (SESTD1), leucine-rich repeat-containing 8 VRAC subunit D (LRRC8D), homogentisate 1,2-dioxygenase (HGD), NEDD8-activating enzyme E1 regulatory subunit (NAE1), and N-acetylserotonin O-methyltransferase- like protein (ASMTL). ## 3.3 Functional enrichment and pathway analysis of the proteins changing significantly with treatment illustrates the protein-protein interaction network analysis consisting of the 12 differentially expressed proteins identified in the PTB samples. Further analysis using GO annotation revealed that the 12 differentially expressed proteins were preferentially associated with MF, binding (carbohydrate derivative binding, organic cyclic compound binding, drug binding, heterocyclic compound binding, small molecule binding, ion binding, protein binding, lipid binding). As well as activity (hydrolase activity, ligase activity, oxidoreductase activity) were the main affected MFs. CC proteins were mainly associated with the cytosol, the endocytic vesicle lumen, the ficolin-1-rich granule lumen, the aggresome, the perinuclear region of cytoplasm, and the cytoplasmic vesicle part. In terms of BP the proteins were involved process (cellular process, metabolic process, multicellular organismal process, multi-organism process, developmental process, immune system process), biological regulation, response to stimulus, cellular component organization or biogenesis, and localization. Regarding protein classes, the main classes of identified proteins were metabolite interconversion enzymes (HGD and ASMTL), protein modifying enzymes (NAE1 and CACYBP), membrane traffic proteins (SEC31A), scaffold/adaptor proteins (LRRC8D), and chaperone proteins (HSP90AA1). A reactome pathway analysis found that the following pathways were highly represented: scavenging by class F receptors, cellular response to heat stress, attenuation phase, HSP90 chaperone cycle for steroid hormone receptors, regulation of heat shock factor 1-mediated heat shock response, interleukin-4 and interleukin-13 signaling, and innate immune response. # 4. Discussion In this study, we have shown for the first time the exploration of the proteome of PBMCs for biomarker discovery to predict response to treatment in TB. In total, 3,530 proteins were quantified across the samples, and 12 differentially expressed proteins were identified in patients with PTB. Based on our results, we speculate that the testing of these 12 proteins by using routine laboratory assays early during treatment will ensure the timely management of patients not responding to anti-TB treatment. This is of great value where anti-TB treatment is started without bacteriological confirmation. LBP is a 60-kDa serum glycoprotein, which plays a role in the innate immune response and antibacterial defense through the activation of neutrophil- producing reactive oxygen species that can kill bacteria. In a small 2013 study conducted among 36 children with TB, LBP was found to be a marker of innate immune system activation. In line with our findings, a study from Uganda using serum proteomics in 39 patients with PTB identified LBP as an important serum biomarker associated with PTB treatment response, as its concentration significantly decreased between baseline and 2 months of therapy. The 5-fold decrease in the levels of LBP after treatment that was found in the present study makes it a suitable candidate for further investigation as a potential biomarker for therapeutic monitoring as early as 2 months after treatment. C9 is a part of a complementary membrane attack complex/perforin domain, and is also a marker of innate immune system activation. In a recently published quantitative proteomics study aiming to identify specific protein signatures in sera of active TB patients and their household contacts, C9 was found to be highly accumulated in the serum of active TB patients. However, a significant change in the levels of this protein with treatment has not been shown in earlier studies. The 1.76-fold decrease in its levels after 2 months of treatment implies its role in monitoring therapeutic response. HSPs are numerous cell proteins involved in the homeostasis of proteins. In TB infection, HSPs exhibit different functions, including the activation of toll- like receptors which in turn activates pro-inflammatory signals, eliciting immune responses. These proteins have been evaluated as a tool for TB diagnosis, and a potent vaccine candidates. However, to the best of our knowledge, no study has evaluated HSPs as markers of treatment response in patients with TB. Besides LBP, C9, HSP70, and HSP90, the other differentially expressed proteins have not been previously reported to be associated with either TB diagnosis or treatment response. Thus, our novel data contribute to a further understanding of the complexity of changes accompanying TB treatment. The proteins increasing in response to treatment imply their role in the protective immune response against TB. Other protein biomarkers, including soluble intercellular adhesion molecule 1, soluble urokinase plasminogen activator receptor, and procalcitonin have demonstrated significant decrease in levels following treatment of PTB. Several studies using whole blood transcriptome analysis have shown significant changes in response to receiving TB treatment; In 2012, 320-transcript signature were significantly diminished in response to treatment. Another study in 2017 had noticed 5-gene signature correlated to TB treatment. The present study did not identify any of these biomarkers. This could be due to the difference in the study of RNAs in the transcriptome studies and proteins in our study, and all RNAs may not be translated into the proteins. Furthermore, only PBMCs were used in our study, as many relevant proteins would be present only in the plasma. Thus, the best approach would be to combine the PBMCs and the plasma proteomics for a comprehensive biomarker discovery for TB treatment monitoring with an increased specificity and high predictive value. Interestingly, our data highlighted an enrichment of GO terms related to cellular and metabolic processes, as well as binding and catalytic activity. This is consistent with the results of several recent quantitative proteomics studies from China conducted among TB patients with or without HIV. This demonstrates that differentially expressed proteins identified in PBMCs from patients with TB have multiple biological functions that require further investigation. Our study is strengthened by the use of an HPLC column, which is associated with minimal disruption of the native condition of the samples, simple procedure, reproducible results, and high capacity. A second methodological strength in the present study was the use of MS with Orbitrap Elite^™, characterized by a high resolution as well as high scan speeds. More importantly, the major strength of this study is the detailed mapping of the PBMC proteome. Most proteomic TB studies have focused mainly on serum or plasma as the primary source of sampling. However, using PBMC samples instead of plasma samples for proteomic profiling has several advantages. First, PBMCs can be obtained relatively easily from routinely collected blood samples and thus provide direct access to physiologically important immune proteins without the well-known analytical complexities of the presence of highly abundant proteins in native human plasma. Second, in contrast to proteomic analyses of plasma samples, proteomic profiling of PBMCs can detect low-abundant proteins from blood which can represent valuable biomarker candidates. Third, PBMCs have been found to be significantly richer as a source of biomarkers compared to plasma. In an experimental study comparing the PBMC proteome to the plasma proteome obtained from blood as the same source sample, the number of proteins identified in PBMCs as a cellular compartment of blood (4,129 proteins) was more than double the amount of proteins reported in plasma (1,929 proteins). Hence, PBMCs as a blood- derived cellular sample represents a valuable sample for TB biomarker studies, and both PBMC samples and plasma samples should be used for a comprehensive proteomic analysis, as these two sample types have been found to encode different proteins. Despite the novelty of our findings, there are several limitations to our study, including a small sample size, and the omnipresence of pre-analytical variability. The lack of suitable controls such as non-responders to treatment make it difficult to distinguish if the differentially expressed proteins represent the host response to the anti-tuberculosis drugs and the toxicity of this treatment rather than the specific response to treatment. Nevertheless, our findings provide a platform for future investigation into the use of biomarkers from PBMCs to assess treatment efficacy in TB. Due to the high number of biomarker candidates identified in the discovery phase by unbiased proteomics, and the costs of assay development and validation, a prioritized selection of the differentially expressed proteins should be performed in future studies using ELISA or western blot based on the fold-change between baseline and post- treatment (as the proteins with the highest fold-change might be the most attractive biomarkers) and relation with TB pathogenesis. In conclusion, proteome analysis of PBMCs can be used as a novel technique to identify potential biomarkers to assess treatment efficacy in patients with PTB. Overall, 3,530 proteins were identified based on LC-MS/MS-based label-free quantitative analysis, and a total of 12 proteins were found to be significantly affected by PTB treatment. The novel proteins elucidated in this work may provide new insights for understanding TB pathogenesis, treatment, and prognosis. Further studies are however needed with a larger sample size and controls to validate our results. # Supporting information The authors would like to thank all doctors, nurses and other staff members at Mnazi Mmoja Hospital, Zanzibar and the Zanzibar Integrated HIV, TB and Leprosy Control Programme for supporting the study, as well as Professor Harald Wiker at the Department of Clinical Science, University of Bergen for his advice in the development of the methodology protocol. The authors would also like to thank Dr. Bram Burger at the Department of Population Health Sciences at the University of Newcastle, UK, for help with statistics. AFB acid-fast bacillus ASMTL N-acetylserotonin O-methyltransferase-like protein BP biological process C9 complement component 9 CACYBP calcyclin-binding protein CC cellular component GO gene ontology HGD homogentisate 1,2-dioxygenase LFQ label-free quantification LRRC8D leucine-rich repeat-containing 8 VRAC subunit D MF molecular function MTB *Mycobacterium tuberculosis* NAE1 NEDD8-activating enzyme E1 regulatory subunit PROBE Proteomics Unit of the University of Bergen PTB pulmonary tuberculosis SEC31A protein transport protein Sec31A SESTD1 SEC14 domain and spectrin repeat-containing protein 1 TB tuberculosis [^1]: The authors have declared that no competing interests exist. [^2]: ‡ AB and EB shared first co-authorship on this work.

# Introduction Myofibres have the capacity to be remodeled to best meet their functional and metabolic demands. Changes in physical activity can initiate a remodeling process toward increased (hypertrophy) or decreased (atrophy) muscle mass. Through protein signaling pathways, integration of chemical, mechanical and bioenergetic signals change the genetic expression patterns for cell size, function and metabolic processes. Physical activities that incorporate unaccustomed eccentric contractions are typically associated with high levels of muscle damage, inflammation and delayed onset muscle soreness (DOMS),. Eccentric contractions, identified by a lengthening while under tension, create an insult to myofibres which may be characterized by: damage to the sarcoplasmic reticulum, t-tubules and structural proteins , the presence of muscle protein in blood, Z-line streaming, soreness and fatigue. Murine and rodent research has often indicated an attenuation of exercise induced membrane damage, structural proteins, and inflammation, along with enhanced satellite cell activation, with exposure to the sex hormone 17β-estradiol (E2). This reduced muscle damage and improved recovery may result from potential antioxidant, membrane stabilizing, or gene regulation properties of E2. As the most abundant estrogen, E2 exerts estrogenic properties that affect the differentiation, growth and function of reproductive, skeletal, neural and muscular tissues. However, human studies do not consistently support the effectiveness of E2 to attenuate exercise-induced muscle damage, mostly reporting similar values for CK efflux, inflammation and loss of muscle function when comparing men and women. Recent microarray analyses have uncovered novel transcriptional programs that coordinate the regeneration and repair of damaged muscle following eccentric exercise. These studies have indentified clusters of genes representing important mechanisms for recovery and adaptation that include the regulation of inflammation, growth, stress response proteins, and membrane biosynthesis. Similar analyses have also described sex differences in the expression levels of genes involved in metabolism and growth inhibition that may result from variations in body composition and hormone content. Specifically, women express a higher abundance of mRNA for several genes involved in fat metabolism that include trifunctional protein β and lipoprotein lipase, and greater expression of the negative regulators of the anabolism growth factor receptor-bound 10 and activin A receptor IIB. The effect of E2 administration on the transcriptome expression profile of skeletal muscle following a single bout of intense eccentric exercise has not yet been evaluated. In this study we used microarray analysis to identify how global mRNA abundance is altered by E2 supplementation in men after 150 eccentric contractions. We hypothesized that the anti-oxidant and membrane stabilizing properties of E2 would attenuate the amount of muscle damage experienced, thereby modifying the expression of mRNA species involved in membrane homeostasis, growth and stress management. Furthermore, we hypothesized that the use of gene array analysis would allow us to detect novel genes relevant to the hypertrophic growth signaling stimulated by intense exercise. # Materials and Methods ## Ethics statement All participants were given an information sheet describing all of the testing procedures before providing written consent to participate. The study conformed to the standards outlined in the *Declaration of Helsinki* and was given approval by the Research Ethics Board of McMaster University (05–438). ## Subjects and anthropometrics Eighteen young healthy men volunteered as participants in this study. All subjects were pre-screened to ensure that they were healthy, fit and had not regularly participated in resistance exercise in the preceding 6 months. Body composition was measured using dual energy x-ray absorptiometry (DEXA) scans (GE Lunar Prodigy, Fairfield, CT). Thigh muscle cross-sectional area was calculated using anthropomorphic measurements of mid-thigh circumference and skinfold thickness to control for potential differences in total work completed. The subject demographics were (mean ± SD): age, 21±2 y; height, 181±5 cm; weight, 76.9±12.8 kg. ## Supplementation protocol Subjects were assigned in a randomized, double-blind manner to either a control (CON, N = 9) or experimental (EXP, N = 9) group. CON subjects consumed 400 mg glucose polymer (Polycose; Abbott Laboratories, Ross Division, St. Laurent, Quebec, Canada) for 10 days. EXP subjects consumed ∼300 mg glucose with 1 mg E2 (Estrace; Shire BioChem, Inc., St. Laurent, Quebec, Canada) for 2 days followed by 2 mg E2 for 8 days. We have previously used this protocol to increase serum E2 concentrations to levels seen during the luteal phase of the menstrual cycle. Glucose and E2 tablets were concealed in gelatin capsules. On the morning of the ninth day, subjects reported to the laboratory and performed the exercise protocol. Supplementation continued until the day of the final biopsy and blood collection to maintain serum E2 concentrations throughout the collection protocol. Subjects in both groups were instructed to take one pill at the same time each day and return any unused pills. All subjects reported 100% compliance. ## Exercise protocol and tissue collection Muscle damage was induced with a previously developed eccentric exercise protocol. Approximately 2 weeks before the exercise protocol, subjects were given a familiarization session with a Biodex isokinetic dynamometer (System 3, Biodex Medical Systems Inc., Ronkonkoma, NY). On the testing day, following a short warm-up (10 min of light cycling), subjects were seated in the dynamometer with their right leg strapped to a lever arm. The lever arm was programmed to extend their leg to 150° of flexion (where 180° is full extension) at a moderate speed (30°/s), then flex their leg to 90° of flexion at a faster speed (120°/s). Subjects did not have to contract maximally during the extension phase. During the flexion phase, subjects were instructed to attempt to maximally resist flexion of the knee (i.e. voluntary ‘maximal’ contraction) against the descending lever arm throughout the entire range of motion. The complete test consisted of 15 sets of 10 repetitions, each set separated by 1 minute of rest. Prior to each tissue collection, subjects abstained from any other form of physical exertion (within 72 h), avoided alcohol (within 48 h), ate their habitual diet (within 48 h), and abstained from caffeine (within 12 h). Each subject consumed a 350 Kcal defined formula diet (57% carbohydrates, 15% protein and 28% fat) two hours before each muscle biopsy and did not eat again until after the final biopsy of each session was taken. These nutritional and activity controls were taken to ensure that the muscle damage would be the only variable to differentially affect the outcomes between biopsies. Muscle biopsies were taken from the vastus lateralis of the control (left) leg during the familiarization session (baseline, BL) and after 8 days of supplementation (post supplementation, PS) and the exercised (right) leg 3 hours (3H) and 48 hours (48H) after exercise, in anatomically distinct sites approximately 6 cm apart. The post exercise collection times were chosen because they represent two distinct phases of recovery from muscle damage. The muscle biopsies were quickly dissected of fat and connective tissue, sectioned into RNase-free cryovials (∼30 mg/piece), flash frozen with liquid nitrogen and stored at −86°C until analysis. Blood samples were drawn from the antecubital vein into heparinized tubes at the same collection times, placed on ice, centrifuged at 1750 g at 4°C for 10 min and stored at −20°C future analyses. ## Blood hormone and enzyme concentrations Serum E2 (Fertigenix-E2-EASIA, Biosource Europe S.A, Nivelles, Belgium) and testosterone (Fertigenix-TESTO-EASIA, Biosource Europe S.A, Nivelles, Belgium) concentrations were measured by enzyme amplified-sensitivity immunosorbent assays (EASIA) according to manufacturer's specifications using BL and PS blood collections. Serum lactate dehydrogenase (LDH) activities were measured in BL, 3H and 48H blood collections with a colourimetric LDH quantification assay (K726-500, Biovision Research Products, Mountain View, CA) according to manufacturer's specifications. All hormone and enzyme measurements were done in duplicate. ## RNA extraction The total RNA was extracted from the frozen skeletal muscle biopsy as described previously in detail by our group. Briefly, ∼30 mg of skeletal muscle was homogenized on ice in 2 mL of Trizol Reagent (Life Technologies, Cat. No. 15596, Gaithersburg, MD). The homogenate was incubated for 10 min at room temperature, followed by phase separation using 200 µL of chloroform and precipitation of the total RNA from the aqueous phase using 500 µL of isopropyl alcohol. The RNA pellet was then washed three times in 75% ethanol and re-suspended in 15 µL DEPC-treated water, aliquoted, and stored at −86°C. The concentration and purity of the RNA was determined using a UV spectrophotometer (Shimadzu UV-1201; Mandel Scientific, Guelph, Ontario) at the absorbance of 260/280 nm. Measurements were done in duplicate and had an average coefficient of variation (CV) of \<10%. The average purity (OD₂₆₀/OD₂₈₀) of the samples was 1.7 before DNase treatment. RNA integrity was assessed in a randomly chosen subset of samples using agarose gel electrophoresis, and the OD ratio of 28S to 18S rRNA was consistently greater than 1 for each sample. ## DNase treatment Prior to microarray chipping and real time quantitative RT-PCR analysis, the isolated RNA samples were treated with DNA-free™ recombinant DNase I (Ambion Inc, Austin, TX) according to the manufacturer's instructions to remove any potential genomic DNA contamination. ## Microarray analysis The resulting total RNA samples were further assessed for integrity prior to chipping using a Nanodrop Spectrophotometer and the Agilent Bioanalyzer Nano Chip System. Samples which passed this initial quality control assurance step were then amplified one round, using an Illumina TotalPrep Kit (Ambion) to generate cDNA then cRNA according to the manufacturer's instructions. This was again assessed for quality by using the Nanodrop and Bioanalyzer as described above. Labeled cRNA samples that passed this second round of quality control were then hybridized to Human Ref-8 BeadChips (Illumina) according to the manufacturer's instructions (approximately 23,000 genes), using equipment specified by the manufacturer (Illumina). Briefly, 850 ng biotin-labeled cRNA in 11.3 µl nuclease-free water was adjusted to 34 µl through the addition of 22.7 µl of 5∶3 HybE1 buffer/formamide. The sample was heated at 65°C for 5 min, allowed to cool to room temperature, and then immediately added to a single array of an 8-array Human Ref-8 BeadChip. Once all 8 samples were added to each BeadChip, it was sealed in a Hyb Cartridge and incubated for 16 h at 55°C with rotation in an Illumina hybridization oven (rotation setting 5). Following overnight hybridization, BeadChips were moved to a slide rack and serially washed using gentle rotation in glass staining dishes filled with a) 250 ml Illumina Wash Buffer×5 min, b) 250 ml 100% ethanol×10 min, c) 250 ml Illumina Wash Buffer×2 min. BeadChips were then blocked for 10 min in 4 ml Block E1 buffer (Illumina), followed by staining for 10 min in 1 µg/ml Streptavidin-Cy3 conjugate (GE Healthcare) in Block E1 buffer. Stained BeadChips were finally washed using gentle rotation in a glass staining dish filled with 250 ml Illumina Wash Buffer×5 min. BeadChips were dried by centrifugation at 280 *g* for 4 min and stored in a light-tight box until reading. ## Array reading Processed arrays were read using a BeadStation array reader (Illumina) according to the manufacturer's instructions. ## Gene ontology analysis In the lists of genes that were significantly differentially expressed with exercise in our study, we carried out gene ontology (GO) analysis to determine the relative enrichment of genes with common or related functionalities to gain insight into biological processes mediated by E2 or exercise. This was carried out using the web interface driven GoMiner tool using an FDR of 5%. Genes were also referenced to their biological functions and canonical pathways with Ingenuity Pathway Analysis (IPA) software. This software identifies functions and pathways most significant to the data set in two ways: the number of differentially expressed genes included in a pathway or function and calculation of a *p*-value using a Fisher's Exact Test to determine the probability of the association of the data set. ## Real-time RT-PCR analysis Changes in gene expression relative to baseline values were measured using real- time reverse transcription-polymerase chain reaction (RT-PCR). Regulator of calcineurin 1 (RCAN1) and capping protein (actin filament) muscle Z-line, alpha 1 (CAPZA1) were selected for analysis because of their roles in growth and sarcomerogenesis. Hemeoxygenase 1 (HMOX1) was chosen for analysis because of its role in stress management. The selected housekeeping gene was β2-microglobulin. Its constant expression following eccentric exercise has been shown in previous work , and was confirmed for the current study. The efficiencies of all primers were tested and determined to be greater than 98%. The primer and probe sequences for these genes can be found in. RT-PCR was completed using a TaqMan® real-time method. The primers and a probe to each target gene were designed based on the cDNA sequence in GenBank (<http://www.ncbi.nlm.nih.gov/sites/entrez/?db=gene>) with primer 3 designer (<http://frodo.wi.mit.edu/primer3-0.4.0/input.htm>). All target gene probes were labeled with FAM at their 5′ ends and BHQ-1 at their 3′ ends. Duplex RT-PCR was performed on an iCycler real-time PCR system (Bio-Rad Laboratories, Hercules, CA) in the One-step TaqMan® RT-PCR Master Mix Reagents (Roche, Branchburg, New Jersey) according to the manufacture's instruction with target gene primers, target probe, housekeeping gene primers and housekeeping gene probe in the same reaction. Determination of significant gene expression change was done as previously described. The genes of interest were normalized to the housekeeping gene, β2-microglobulin by following the standard method. Briefly, C_T values of the housekeeping gene were subtracted from the C_T values of the gene of interest giving a δC_T. This is equivalent to the log₂ difference between endogenous control and target gene. Values were then normalized to baseline, δδC_T. All samples were run in triplicate, fluorescence emission was detected using FAM and Tamra filters, and C_T was automatically calculated. ## Western blotting Muscle biopsy samples were homogenized and prepared for polyacrylamide gel electrophoresis using methods previously described. Briefly, frozen skeletal muscle tissue samples (25–35 mg) were hand homogenized in 25 µl of phosphate buffer (50 mM Kpi, 5 mM EDTA, 0.5 mM DTT, 1.15%KCl (w/v)) per milligram of tissue. A protease inhibitor cocktail (Sigma, St. Louis, Missouri) was added to the phosphate buffer immediately prior to use at a ratio of 1∶1,000. Samples were centrifuged at 600 *g* for 10 min at 4°C and the supernatant aliquoted for analyses. Protein concentrations of each sample were determined using the method described by Lowry et al. Samples were loaded on 10% SDS-polyacrylamide gels and transferred to a PVDF membrane. Membranes were blocked with 5% BSA (wt/vol) in Tris-buffered saline with 0.1% Tween (vol/vol) (TBST) and incubated in primary antibody: RND3 (Abcam, Cambridge, MA; ab50316, 1∶1000); FOS (Abcam; ab16902, 1∶1000); total p38MAPK (Cell Signaling Technology, Danvers, MA; no. 9212, 1∶1000); p38MAPK (Thr180/Tyr182) (Cell Signaling Technology; no. 9215, 1∶1000); total GSK-3β (Cell Signaling Technology; no. 9315, 1∶1000); GSK-3β (Ser 9) (Cell Signaling Technology; no. 9323, 1∶1000), β-actin (BD Biosciences, Mississauga, ON; no. 612657, 1∶10000). After washing in TBST, membranes were incubated in either HRP- linked anti-rabbit or anti-mouse IgG secondary antibody (Amersham Biosciences, Piscataway, NJ; no. NA934V, 1∶6000), washed with TBST and developed using ECL (Amersham Biosciences; model no. RPN2106). Membranes were exposed to x-ray film (Biomax XAR; Kodak, Rochester, New York) which were then scanned with a Dell 920 scanner at 300 DPI and saved in TIF file format. Using Image J v1.40 g software (National Institutes of Health, Bethesda, Maryland), background noise was removed and bands in the region of interest were selected for analysis. Individual profile plots were generated and area under the curve measured in arbitrary units (AU). ## Statistical and bioinformatics analysis Student's unpaired t-tests were used to determine differences in subject characteristics and total work. A 2-way repeated measures ANOVA (supplementation protocol × time) was used to assess differences in LDH, E2 concentration, testosterone concentration, protein levels and the linear 2^−δδCT data set for gene expression measured with RT-PCR using computerized software (Statistica, Statsoft). When statistical significance was achieved, Tukey's honestly significance difference post-hoc test was used to determine the significance among the means. STATISTICA for Windows 5.0 (Statsoft, Tulsa OK) was used to perform t-tests and ANOVAs. The threshold for significance was set at P≤0.05. Data are presented as mean ± SEM unless otherwise indicated. Gene array data analyses were done comparing baseline, post supplementation, 3 and 48 hours post exercise using simple paired t-test on log₂ expression ratios. Genes were ranked by p-value and the inference reported following adjustment for multiple testing using the FDR and the Benjamini and Hochberg method. Among those genes with an adjusted q-value (based on FDR) of \<0.05, we used hierarchical clustering (based on the HOPACH algorithm) to find groups of genes with similar profiles across the subjects. # Results ## Subject and work characteristics CON and EXP groups were not different in age, weight, height, body fat percentage, average thigh cross-sectional area or total work completed. All subjects completed the required 150 eccentric contractions. ## E2 and testosterone concentration were affected by supplementation with E2 Following 8 days of supplementation serum E2 concentration increased by 146% (P\<0.001) and testosterone concentration was reduced by 26% (P = 0.01) in the EXP group. Both hormone concentrations remained unchanged in the CON group. ## Eccentric exercise induced muscle damage The appearance of the muscle protein LDH in serum is an indirect indicator of muscle membrane damage. LDH activity was elevated 13.8% (P\<0.05) 48 hours after exercise. The EXP values did not differ from the CON values at any time. ## Microarray data identifies altered mRNA expression during recovery from eccentric exercise that is not affected by E2 Eccentric exercise significantly increased the early mRNA expression of 310 genes at 3H. DNA microarray analyses did not identify differential mRNA expression of any gene as a result of E2 at any time (we could not reject the global null that E2 was independent of mRNA expression of all genes represented on the chip). For this reason, microarray data at each timepoint was collapsed between groups, increasing the sample size to 18 subjects. Mean fold change for genes with ratios greater than 1 at 3H was 2.3±0.1. In addition, 301 genes had ratios less than 1 at 3H with a mean fold change of 0.8±0.01. By 48H, all genes had expression values that were not different from baseline. The complete data set is freely available at GEO (accession no. GSE19062). Of the genes differentially expressed at 3H, we identified 25 that participate in two signaling cascades for muscle growth and adaptation: ras homologue gene family, member A (RHOA) and nuclear factor of activated T-cells (NFAT). Other regulators of muscle growth and remodeling not directly involved in RHOA or NFAT signaling that were also highly induced included: ATF3, MYC, XIRP1, HBEGF and DNAJB4. IPA identified several biological functions related to muscle growth and remodeling identified by the number of differentially expressed genes included in the function and the calculation of a P-value using a Fisher's Exact Test with a threshold for significance set at P≤0.05. They included: cancer (P\<0.05, 233 molecules), gene expression (P\<0.05, 146 molecules), cell assembly and organization (P\<0.05, 69 molecules), cell morphology (P\<0.05, 59 molecules) and skeletal and muscular system development and function (P\<0.05, 50 molecules). In the same manner, IPA also identified three canonical pathways related to gene expression and growth/proliferation: ILK signaling (P = 0.005, 16/186 molecules), RAN signaling (P = 0.005, 4/23 molecules) and PI3K/AKT signaling (P = 0.006, 12/136 molecules). ## RT-PCR analysis provides additional genes involved in actin dynamics and regulation of RhoA and NFAT signalling Targeted real time RT-PCR was conducted on several genes selected *a priori* for their involvement in recovery from skeletal muscle damage. Previous work using microarray analysis and RT-PCR identified the expression of novel genes following a similar eccentric protocol that are likely involved in the recovery and adaptation to damaging exercise. Confirming the identification of two of its transcript variants in the microarray, RCAN1 mRNA content was highly elevated at 3H (16.6-fold, P\<0.001). Also induced was HMOX1 at both 3H (3.9-fold, P = 0.009) and 48H (3.5-fold, P = 0.002). Expression of CAPZA1, a regulator of the growth of actin filaments, was increased at 48H (1.8-fold, P = 0.04). ## Signaling proteins and protein content are affected by eccentric exercise Phosphorylated p38MAPK and GSK-3β negatively regulate NFAT by promoting its export from the nucleus. Phosphorylation (Thr¹⁸⁰/Tyr¹⁸²) of p38MAPK was significantly lower at 3H (0.77-fold, P = 0.07) and 48H (0.73-fold, P = 0.005) as a result of exercise with no effect of E2. Phosphorylation (Ser⁹) of GSK-3β was not affected by either exercise or E2. Two species highly expressed at 3H by the microarray were selected for western blot analysis. RND3 was significantly higher at both 3H (1.34-fold, P = 0.02) and 48H (1.39-fold, P\<0.01). FOS was significantly higher at 48 hours following exercise (1.16-fold, P\<0.05). # Discussion The sex hormone E2 displays anti-oxidant and membrane stabilizing properties that could protect skeletal muscle from the effects of exercise induced muscle damage and influence genetic expression patterns. Using microarray, real time RT-PCR and protein analyses, we have identified that 8 days of E2 supplementation did not affect the myofibre transcriptome in men. However, a single bout of eccentric exercise did induce differential mRNA transcription in the hypertrophic signaling pathways RhoA and nuclear factor of activated T-cells (NFAT), changes in the phosphorylation status of related signaling proteins and the protein quantities of two of the upregulated genes. Of the genes affected early after exercise, one of the greatest inductions was observed in the novel actin binding protein striated muscle activator of Rho signaling (STARS). This protein is a muscle-specific transducer of cytoskeletal signaling in cardiac and skeletal muscle that responds to calcineurin activation and biomechanical stress. STARS stimulates growth through a mechanism requiring actin polymerization and Rho GTPase activation, increasing serum response factor (SRF)-mediated gene transcription. Originally identified in cardiac muscle, STARS mRNA content increases more than 3-fold following the hypertrophic signaling of pressure overload. More recently, Lamon *et al*. identified a 3.4-fold increase in STARS mRNA in human skeletal muscle following 8 weeks of resistance training. Our measurement of more than a 10-fold increase suggests that this gene is very important for the early signaling for growth and remodeling following eccentric exercise. A downstream target of STARS associated with skeletal muscle hypertrophy and adaptation is the Rho GTPase, RhoA. Rho GTPases are a family of small signaling G proteins that interact with effector proteins to regulate actin cytoskeleton, cell cycle progression and gene transcription. These molecular signals switch to an active GTP bound state under the control of Rho GEFs (guanidine exchange factors), and return to their inactive GDP bound state by Rho GAPs (GTPase- activating proteins). For the first time, our array analysis has identified increased expression of two GEFs (ARHGEF7 and ARHGEF12) and reduced expression of a GAP (ARHGAP24) that specifically regulate RhoA. These gene expression modifications, if translated into altered protein quantities, could increase the potential for RhoA activation. Although increased activity of RhoA protein is necessary for myogenesis induction, it must be downregulated before myotube formation can proceed. This is achieved by RND3, another negative regulator of RhoA activity whose upregulation is an essential step of myoblast fusion,. Our damaging exercise protocol resulted in an early induction of this gene at 3H and elevated protein content by 48H. Cell culture experiments have identified that in the presence of growth factors, RND3 mRNA increases of ∼1.7-fold result in greater protein content within 30 h. This in turn inhibits RhoA activity and promotes myotube formation and elongation. Once activated, RhoA signaling is associated with myogenesis and actin remodeling in various cell types through regulation of genes that include DIAPH1, CORO1C, FLNB and CAPZA1. Through SRF activation, RhoA and STARS also mediate the induction of actin proteins ACTA2, ACTN1 and members of the AP1 transcription factor complex: FOS, FOSB and JUND. Each of these genes was induced by eccentric exercise at 3H, as identified by our gene array and targeted real time RT-PCR analysis and our data indicate that the increased transcription of FOS was effectively translated into protein, increasing levels significantly by 48H. Although the number of studies investigating these genes after exercise is few, some support can be found for the upregulation of select downstream targets. Following eccentric exercise, the largest induction occurs with the transcription factor FOS 2–8 hours post exercise (23 to 38-fold increases). Resistance training results in a 2.7-fold increase in ACTN1 after 8 weeks. Thirty minutes of high intensity running increased the expression of FOS (7.0-fold) FOSB (17.8-fold) and JUND (7.6-fold). Given that the mRNA levels for STARS, associated regulatory and transcription factors and downstream targeted genes were all significantly elevated 3 h after exercise, it appears that STARS signaling through a RhoA/SRF pathway is important for early skeletal muscle remodeling following damaging exercise. A second calcineurin influenced signaling pathway identified in our microarray analysis to be transcriptionally active was nuclear factor of activated T-cells (NFAT). NFAT proteins exist in the cytoplasm of cells in a phosphorylated and inactive state. The influx of calcium following sustained contraction or damage increases the binding of calcineurin to NFAT, dephosphorylating conserved serine residues and promoting translocation of NFAT into the nucleus. Once inside the nucleus, NFAT cooperatively binds to DNA with transcription factors AP1 and MAF initiating the transcription of prohypertrophic genes. Our microarray analysis also identified a significant induction of the genes NFATc1 and MAF at 3H. Along with the greater expression of the AP1 complex components FOS, FOSB and JUND, an increased abundance of NFATc1 and MAF could improve signaling by NFAT. The NFAT pathway interacts with mitogen activated protein kinases (MAPK) and glycogen synthase kinase-3β (GSK-3β) for coordination of the hypertrophic response. p38MAPK and GSK-3β act as a negative regulators of cardiac hypertrophy, rephosphorylating NFAT and promoting its export from the nucleus. Although GSK-3β mRNA content was higher in the microarray, its activity did not change and would not have affected NFAT nuclear residence. GoMiner analysis identified MAPK regulation as a transcriptionally active biological process through the induction of four related kinases (MAPKAPK2, MAPK6, MAP3K6 and MAP3K8), three related phosphatases (PP2CA, DUSP8 and DUSP16) and one surface receptor (ADRB2). Western blotting confirmed that p38MAPK phosphorylation status was lower after exercise, reaching significance by 48H. Lower activity of p38MAPK would assist in the transcriptional activity of NFAT, and may also occur to inhibit the induction of apoptosis and necrosis. It should be noted that molecules that function in the recovery and repair of skeletal muscle through other mechanisms were also induced by our exercise protocol. Similar to other reports, ATF3, MYC, XIRP1, HBEGF and DNAJB4, were highly up-regulated (6.1 to 28.4-fold) early after exercise. In addition, the altered mRNA content of members of the signaling pathways ILK, RAN and PI3K/AKT identified them as being transcriptionally active. Their respective functions in actin cytoskeleton remodeling, transport across the nuclear envelope for gene expression and protein synthesis relate to the top biological functions returned by gene ontology analysis. Interestingly, the biological function that contained the greatest number of molecules was cancer (233 molecules), likely due to similar alterations to the regulation of cellular growth experienced with both exercise and cancer. Maintained muscle contractions and exercise-induced damage to the sarcoplasmic reticulum and sarcolemma may result in the accumulation of excess calcium, known as Ca²⁺ overload. As a signaling molecule, Ca²⁺ binds to calcineurin, activating both STARS and NFAT. Unrestrained calcineurin activity can be regulated in skeletal and cardiac muscle by the inhibitors regulator of calcineurin 1 (RCAN1, aka MCIP1, DSCR1), and hemeoxygenase 1 (HMOX1). Increased mRNA content of RCAN1 was identified in microarray and targeted real time RT- PCR, confirming a previous gene expression profile that also identified an increase following exercise (3.8-fold). RT-PCR also identified significant increases in HMOX1 at both time points after exercise, similar to the 8-11-fold induction following 5 days of resistance training. Together, the upregulation of these two calcineurin inhibitors identifies the importance of regulating the elevated calcineurin activity that occurred following unaccustomed eccentric exercise. The primary mechanism through which estrogens influence the growth, differentiation and function of tissues occurs via the estrogen receptors ERα and ERβ. As transcription factors, active homodimers and heterodimers of ERα and ERβ bind to estrogen response elements (EREs) in nuclear and mitochondrial DNA, increasing the transcription of target genes. In a nongenomic manner, E2 also interacts with a number of proteins that include endothelial nitric oxide synthase (eNOS) and MAPK, altering the signals that modulate cellular differentiation, migration and survival. Our observation that the myofibre transcriptome was unaffected by E2 is surprising. Although women have higher E2 concentrations than men, protein quantities of ERα and ERβ are similar and our supplementation protocol successfully increased circulating E2 to the level seen during the luteal phase of the menstrual cycle in healthy women. This suggests that the differential expression between men and women of genes involved in metabolism and growth regulation, may result from factors beyond E2 alone; factors that may include variations in body composition, X-chromosome genes, and/or other sex hormones (i.e., progesterone). These results indicate that E2 supplementation does not affect the transcriptional pattern in skeletal muscle following eccentric exercise in men. However, the stress of a single bout of exercise induced a transcriptional response in two signaling pathways, STARS/RhoA/AP1 and NFAT/AP1, providing important insights for future research into the early hypertrophic response. [^1]: Conceived and designed the experiments: LM MAT. Performed the experiments: LM SB. Analyzed the data: LM SM AH. Contributed reagents/materials/analysis tools: SM MAT. Wrote the paper: LM. Biopsy collection: MAT. Collected muscle biopsies: SB. [^2]: The authors have declared that no competing interests exist.

# Introduction Labor, in particular, delivery is a sensitive and vulnerable time in a woman’s life. Every woman has the right to woman-centered healthcare that is safe, effective, timely, respectful, and free of violence and discrimination during pregnancy, labor, and childbirth. Maternity care is a service that focused on improving maternal and newborn health outcomes during pregnancy, childbirth, and the postpartum period. It includes monitoring the mother’s and baby’s well-being, health education, and assistance during childbirth. Every year, around 140 million births occur worldwide, the vast majority of which are vaginal births with little difficulty for women and their newborns. Pain, anxiety, threat, and exposure to the circumstance are the causes of women’s vulnerability during labor and delivery. Despite significant advances in maternal and child health, there is still a high rate of maternal and neonatal deaths globally. Poor childbirth care contributes directly and indirectly to 82 percent of this problem. Governments are striving to improve the quality of clinical care provided to women throughout pregnancy, and childbirth to achieve the global maternal mortality ratio target of 70 per 100,000 live births by 2030. Following mounting evidence of mistreatment of women during pregnancy and childbirth around the world, the WHO declared the prevention and elimination of disrespect and abuse during childbirth by implementing the Respectful Maternity Care (RMC) initiative. RMC is one of the WHO’s eight dimensions for quality maternal and newborn health care, and it refers to care that includes the right to access friendly, abuse-free, timely, and discrimination-free maternal health care, along with privacy, confidentiality, equality, informed consent, and autonomy. It is a strategy that will be put in place to encourage positive interpersonal relationships between women and health care providers and workers throughout labor, delivery, and the postpartum period. This notion advocates for good staff attitudes, behaviors, and accountability that contribute to women’s contentment with their birth experience in a sustainable way. Currently, the change from home delivery to hospital birth has increased access to lifesaving care for difficulties, but it has also generated new challenges, such as facility overcrowding, an excess of procedures, mistreatment, and over- medicalization. Timely, respectful, and consensual obstetric care, is not the norm in many healthcare settings around the world, especially in developing countries like Ethiopia. Despite a recent dramatic increase in the number of skilled providers and health facilities in Ethiopia, the uptake of prenatal care, skilled delivery service, and postnatal care remain at only 68%, 28%, and 17%, respectively. Even though numerous circumstances contribute to low healthcare utilization, it is becoming evident that poor service quality and provider mistreatment are among the reasons why many women are unable to seek maternal, neonatal, and child health (MNCH) services. Several studies have revealed that women’s expectations of how they would be treated at health facilities may have a substantial impact on where they prefer to get maternal health services, notably childbirth. In 2016, the Ethiopian government launched its Health Sector Transformation Plan(HSTP), which aims to promote Compassionate and Respectful Care (CRC), with an emphasis on RMC, to improve maternal and newborn health outcomes. Although the target has not yet been met, this plan emphasizes the need of achieving 90 percent skilled birth attendance and lowering the maternal mortality ratio (MMR) from 420/100,000 live births in 2015 to 199/100,000 live births by 2020. As per small-scale studies conducted in Ethiopia, the prevalence of RMC lies between 12.75% to 77%. Factors affecting the receipt of RMC during childbirth were the place of delivery, time of delivery, ANC uptake, planning status of index pregnancy, educational level, and facing obstetric complications. Improving the quality of care through enhancing RMC has been highlighted as the most important intervention for lowering maternal and newborn mortality by laying the path for skilled delivery. Understanding the prevalence and determinants of RMC can help to improve the effectiveness of RMC initiatives and may have a beneficial impact on the uptake of MNCH services. Although several epidemiological studies on the magnitude and determinants of RMC in Ethiopia have been conducted, the results have been inconsistent and varied. Existing studies have also been small-scale or limited by locality, which might also make drawing equivocal conclusions and evidence at the national Prevalence harder. Therefore, such disparities may be inadequate for policymakers and planners to intervene, demanding an assessment of the pooled estimates. Combining information from multiple data sources can enhance estimates of health-related measures by using one source to supply information that is lacking in another. Hence, this systematic review and meta-analysis aimed at estimating the pooled prevalence of RMC and its determinants at the national level. The study’s findings will help policymakers and program planners build appropriate interventions to enhance the prevalence of RMC, which is one of the four pillars of HSTP. # Methods ## Study design While conducting this systematic review and meta-analysis, the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) standards for literature search method, study selection, data extraction, and result reporting were followed. To establish the inclusion and exclusion criteria, the PEO (Population, Exposure of interest, Outcome) technique was used, which was adapted from the JBI 2017 review guideline. ## Eligibility criteria ### Inclusion criteria 1. Population: Women who experienced a childbirth 2. Exposure of interest: Maternity care (prenatal, skilled delivery, and postnatal) cares 3. Outcome: Receiving respectful maternity care (RMC). 4. Study designs: All crossectional studies reporting the prevalence of RMC and its determinants were considered. 5. Study setting: Community- and facility-based studies conducted in Ethiopia were considered. 6. Publication status: Both published and unpublished studies were considered, and if a study appeared in multiple reports, the most comprehensive and up-to-date one had been used. 7. Language: Articles published in the English language were considered. 8. Year of publication: All publications reported before June 30, 2022, were taken into account. ### Exclusion criteria - Systematic reviews, case series, commentaries, conference abstracts, letters to editors, technical reports, qualitative studies, and other opinion publications were excluded. - Studies that were not fully accessible after two emails with the primary/corresponding author were excluded since assessing methodological quality in the absence of the full text was problematic. - Studies that were not explicitly addressed to RMC, such as those studies conducted on CRC in general outpatient department patients, were not taken into account. - As potential duplicates, studies conducted in the same area during the same study period were excluded. ## Search strategies The studies had to have been published in English before June 30, 2022. Initially, a comprehensive search of studies was done by using PubMed/MEDLINE, Google Scholar, Science Direct, Scopus, ProQuest, Web of Science, Cochrane Library, and Direct of Open Access Journals. The following keywords were used for the database search: “Respectful”, “Woman-Centered”, “Dignified”, “Friendly”, “Non-Abusive”, “Compassionate”, “Non-discriminatory”, “Maternity”, “Maternal”, “Prenatal”, “Antenatal”, “Delivery”, “Childbirth”, “Postnatal”, “Care”, “Maternal Health Care”, “Health Service”, “Level”, “Magnitude”, “Prevalence”, “Determinants”, “Associated Factors”, “Predictors”, “Ethiopia”, and “Ethiopian”. To connect those keywords, Boolean operators (AND and OR) and truncation were employed. The following key search terms were used ("Respectful"\[All Fields\] OR "Woman-Centered"\[All Fields\] OR "Dignified"\[All Fields\] OR "Friendly"\[All Fields\] OR "Non-Abusive"\[All Fields\] OR "Compassionate"\[All Fields\] OR "Non-discriminatory"\[All Fields\]) AND ("Maternity care"\[All Fields\] OR "Maternal care"\[All Fields\] OR "Prenatal care"\[All Fields\] OR "Antenatal care"\[All Fields\] OR "Delivery service"\[All Fields\] OR "Childbirth"\[All Fields\] OR "Postnatal care"\[All Fields\] OR "postpartum Care"\[All Fields\] OR "Maternal Health Care"\[All Fields\] OR "Health Service"\[All Fields\] OR "Maternity care"\[All Fields\]) AND "Ethiopia"\[All Fields\]. Gray literature searches via Google scholar, Google searching, and Addis Ababa and Jimma University institutional repositories supplemented the electronic database search. ## Study selection process All identified studies were imported into the EndNote XI library and checked for duplication. After removing duplicate articles, three authors (AH, DW, and AT) extracted all articles independently at the title, abstract, and full text. A fourth author (BB) independently reviewed 20% of the removed studies and compiled the screened articles, and any inconsistencies were settled by discussion. Finally, 16 studies were considered for systematic review and meta- analysis. ## Data extraction The data were extracted using a Microsoft Excel spreadsheet. Two authors (AH and DW) separately extracted the important data using a pre-setted and piloted data extraction form. The data extraction format comprised the primary author’s name, publication year, study year, study design, study area, study setup, sample size, response rate, data collection technique, the proportion of RMC, and adjusted odds ratio(AOR) with their 95% confidence interval. ## Risk of bias in individual studies The methodological quality of included studies was assessed by using Joanna Briggs Institute (JBI) Critical appraisal checklist for prevalence studies. Two reviewers independently rated the quality of the included studies (AH and BB). There are nine parameters in the evaluation tool and each parameter has equal weight. (1) Was the sampling frame appropriate to address the target population? (2) Were study participants sampled appropriately? (3) Was the sample size adequate? (4) Were the study subjects and the setting described in detail? (5) Was the data analysis conducted with sufficient coverage of the identified sample? (6) Were valid methods used for the identification of the condition? (7) Was the condition measured in a standard, reliable way for all participants? (8) Was there appropriate statistical analysis? (9) Was the response rate adequate, and if not, was the low response rate managed appropriately? Each item was assessed as either low or high risk of bias. The evaluators assigned a score of ’0’ if the study met each specific parameter and a score of ’1’ if it did not. A composite quality index was computed and the risk of bias was graded as low (0–2), moderate (3 or 4), or high (≥5). Articles with low and moderate risks of bias were considered for this systematic review and meta-analysis. ## Measurement of the outcome of interest The primary outcome variable of this systematic review and meta-analysis was the prevalence of RMC in Ethiopia, which was determined using the pooled prevalence. The secondary outcome variable was RMC determinants, which were estimated using a pooled AOR with 95 percent CIs. RMC is a universal human right that must be provided to all childbearing women in every health system and is measured by four performance standards (friendly care, timely care, discrimination-free care, and abuse-free care). When those women received all four performance domains, they were considered to have received RMC. ## Statistical methods and analysis The data extracted from a Microsoft Excel spreadsheet were exported to the STATA^TM 16 statistical software, where all statistical data analyses were undertaken. First, Higgins I-square (I²) statistics and the Cochran’s-Q test were used to determine the presence of statistical heterogeneity across included studies. Heterogeneity was classified as low, moderate, or high when the values of I-square were \<25, 50–75, and \>75%, respectively. Accordingly, significant heterogeneity was detected \[I² = 98.5%, p-value\<0.001\]. Thus, a random-effects meta-analysis model with the DerSimonian-Laird method was used to determine the pooled prevalence of RMC. The adjusted Odds Ratios(AOR) from eligible studies were extracted, along with their 95% CIs. The pooled AORs were computed using a random- or fixed-effect model. Finally, forest plots were used to display the pooled estimates for RMC and its determinants, along with their respective 95% confidence intervals. ## Publication bias The presence of publication bias was visually checked by using funnel plots, and a symmetrical, large inverted funnel revealed that the likelihood of publication bias was less likely. Statistical methods such as Egger’s and Begg’s tests were used to supplement visual assessment, and a p-value \<0.05 indicate the likelihood of publication bias. ## Additional analyses ### Subgroup analyses and heterogeneity Subgroup analyses were performed based on geographical regions, residence, study year, and publication year. To identify potential sources of heterogeneity across studies, a univariate meta-regression analysis was performed with sample size, publication years, and study years as covariates. ### Sensitivity analysis To assess the influence of a single study on the overall pooled prevalence of RMC, sensitivity analysis was performed using a random-effects model. # Results ## Study selection A total of 1599 studies were found through all searches and 1078 records were duplicates and were thus removed. The remaining 521 studies were eligible for screening. Based on the title and abstract screening, 478 studies were excluded, having left 43 full articles. Again, 27 studies were removed (twelve owing to insufficient data, seven failed to state the outcome of interest clearly, two case reports, and six were qualitative studies). Finally, 16 studies were considered for this systematic review and meta-analysis. ## Characteristics of included studies Sixteen studies with a total of 6354 study participants were considered. All of the eligible studies were cross-sectional in design. The studies were carried out between 2013 and 2021. All of the included studies collected data through face-to-face interviews with a pre-tested, interviewer-administered questionnaire. Studies conducted in Addis Ababa (n = 173) and South Nations, Nationalities, and Peoples’ Region(SNNPR) (n = 783), accounted for the minimum and maximum sample sizes, respectively. In terms of the distribution of the studies across the geographical region, five studies were from Amhara, five from Oromia, two from Addis Ababa, two from SNNPR, one from Harari, and one from Benishangul Gumuz. When it came to the risk of bias in the included studies, the majority (14) had a low risk, with the remaining two having a moderate risk. ## The pooled prevalence of RMC in Ethiopia Because the prevalence estimate varied across studies with significant heterogeneity (I² = 98.50%; P\<0.001), we used a random-effect model with a DerSimonian and Laird method. The overall pooled prevalence of Respectful maternity care in Ethiopia was found to be 48.44% (95% CI: 39.02–57.87). Regarding each component of RMC, 78.53 (95% CI: 72.57, 84.48) and 68.95 (95% CI: 64.52, 73.38) of women received discrimination-free and timely care, respectively. Only half, 49.99 (95% CI: 32.46, 67.52) of women got friendly care. ## Subgroup analyses Subgroup analyses were conducted by region, study year, and publication year. Accordingly, the highest prevalence of RMC was reported in Oromia and Amhara regions, 58.01%(95% CI: 42.44, 73.58), and 51.31%(95% CI: 41.10, 61.53), respectively. On the other hand, the lowest Prevalence of RMC was reported in Benishangul Gumuz, 12.65% (95% CI: 9.45, 15.90). In addition, we performed a subgroup analysis based on the year when the studies were conducted. Accordingly, the pooled prevalence of RMC was 41.15% (95% CI: 28.85–53.44) for studies conducted before 2020 and 68.83% (95% CI: 55.30–72.37) for studies conducted in 2020 and after. ## Heterogeneity and publication bias A univariate meta-regression analysis was run using study-Prevalence characteristics (publication year and sample size) as a cofactor to identify the possible source of heterogeneity across the included studies. However, heterogeneity was not explained by sample size (P = 0.582), and the publication year (P = 0.448). The funnel plot was used to visually examine publication bias, and the effect estimates were asymmetrical, indicating that publication bias was unlikely. Furthermore, we checked the presence of publication bias statistically by running Egger’s regression test and an adjusted Beggs rank correlation test and the p values were 0.51 and 0.86, respectively. All of these indicate that the presence of publication bias in this study was unlikely. ## Sensitivity analysis A sensitivity analysis using a random-effects model was carried out to detect the effects of a single study on the overall meta-analysis estimate. As a result, there is no evidence that a single study influenced the pooled prevalence of RMC. ## Determinants of RMC Thirteen variables were extracted from the included studies to identify determinants of RMC. As significant determinants of RMC, six variables were identified namely giving birth during the day, planning status of the last pregnancy, having adequate ANC visit, experiencing an obstetric complication during the last delivery, and receiving service from health care providers who were trained on CRC. The influence of CRC training on RMC was assessed by using the findings of three studies. Women who received maternal health services from CRC-trained healthcare providers were 4.09 times more likely than their counterparts to receive RMC \[AOR = 4.09, 95% CI: 1.73, 6.44\]. Moreover, we used four studies to assess the relationship between having ANC and receiving RMC during childbirth. Accordingly, women who received adequate ANC have a 2.34 times greater chance of receiving RMC than their counterparts \[AOR = 2.34, 95% CI: 1.62, 3.06\]. The effect of time when childbirth took place was assessed using the findings of five studies. A fixed-effect meta-analysis of AORs revealed that the odds of receiving RMC were 2.61 times higher for women who gave birth during the daytime as compared to those who gave birth at night \[AOR: 2.61, 95% CI: 1.92, 3.31\]. As per the findings of four studies, the pregnancy planning status at the time of childbirth had a positive association with RMC. The likelihood of receiving RMC was 4.43 times higher for those mothers with planned pregnancies as compared to women with unplanned pregnancies \[AOR: 4.43, 95% CI: 2.74, 6.12\]. Finally, a negative association was identified between having obstetric complications and RMC. Those women who sustained any obstetric complication had 54% less likely to get RMC than those women who didn’t face any obstetric complication \[AOR: 0.46, 95% CI: 0.30, 0.61\]. # Discussion Ethiopian HSTP-I and HSTP-II advocate ensuring equitable and timely delivery of quality health care (reliable, patient-centered, and efficient) to all in need. Safe motherhood must include respect for women’s basic human rights, such as autonomy, decency, feelings, preferences, and priorities, in addition to the prevention of illness or death. Currently, WHO recommends providing RMC per the human rights-based approach to reducing maternal and newborn morbidity and mortality by improving women’s pregnancy and childbirth experiences and addressing inequities in MNCH care access. Given the importance of RMC in ensuring the quality of MNCH services, assessing its status at the national Prevalence allows for a better understanding of its potential contribution to the national and global accomplishment of HSTP and SDG, respectively. Hence, this systematic review and meta-analysis was aimed at determining the Prevalence of RMC and its determinants in Ethiopia. The estimated pooled prevalence of RMC in Ethiopia was 48.44% (95% CI: 39.02–57.87). Accordingly, the Prevalence of RMC was higher as compared to findings of a systematic review and meta-analysis in India(28.7%) and a study conducted in East and Southern Africa(30%). In addition, the current finding was higher than primary studies conducted in Pakistan(2.6% and 0.5%), Peru(2.6%), Tanzania(30%), and Nigeria(2%). This could be due to the Ethiopian government’s effort since 2020 to create CRC healthcare providers as one of the pillars of HSTP-I and–II. Furthermore, there is an increasing commitment and interest in implementing the CRC initiative at the national prevalence through the provision of CRC training to over 27, 000 health leaders and health workers across the country, with a particular emphasis on those who work at MNCH service delivery points. On the other hand, the finding was lower than primary studies conducted in Brazil (81.7%), Mexico (72.3%), India(76%), and Kenya(80.0%). The disparity in these findings could be attributed to differences in the methodologies and tools used to measure RMC, socio-cultural and economic differences, study period, and organizational factors such as a shortage of health facility to population ratio. Furthermore, it was significantly lower than the Ethiopian government’s stated goal of increasing CRC to 90 percent by 2025 in the HSTP-II. Also, the current finding was found to be low as compared to a finding obtained by direct observation of 16 model health facilities in Ethiopia(60.4%). Hence, the government should work to ensure an adequate number and mix of quality health workforces who are motivated, competent, and compassionate to enhance RMC. In addition, due consideration should be given to the development of a short-term training manual that improves the awareness and practice of RMC among health workers at service delivery points. The Ministry of Health should implement a multi-pronged strategy, starting with the enrollment of students in health science programs and the efficient administration of currently employed health professionals. As per subgroup analysis results, the highest and the lowest Prevalence of RMC was reported in the Oromia, 58.01%(95% CI: 42.44, 73.58) and Benishangul Gumuz region 12.65% (95% CI: 9.45, 15.90) respectively. These differences could be attributed to regional disparities in the number of HCPs and health facilities, with the region with the lowest prevalence being one of Ethiopia’s emerging regions with the lowest MNCH service coverage. Furthermore, the low prevalence could be attributed to the small number of studies included in this meta- analysis, which included only one study from the Benishangul region. Furthermore, in the last 2–3 years, the Benishangul Gumuz region has been one of the most insecure, with frequent conflicts that have resulted in the displacement of civilians and healthcare providers due to security concerns. As a result, there may be a lapse in stringently monitoring the MNCH program, resulting in low RMC. Furthermore, from a sub-group analysis studies conducted since 2020 had the highest Prevalence of RMC, 63.83% (95% CI:55.3, 72.37), compared to studies conducted before 2020. The possible justification could be the Ethiopian government’s emphasis on addressing a low uptake of maternal health service utilization by improving RMC through various measures, particularly in the previous two years. The measures taken were, creating model professionals in each health facility, advocacy campaigns through mass media, and enacting a Patients’ Rights and Responsibilities law. Furthermore, the development and implementation of a generic curriculum in pre-service education, as well as the establishment of well-functioning 16 CRC incubation centers, including national referral and regional hospitals, may have contributed to the good progress of the RMC Prevalence over the last three years. Another objective of this systematic review and meta-analysis was to identify the most important factors that affect the Prevalence of RMC. Accordingly, receiving service from CRC-trained health care providers, having adequate ANC visits, planning status of the last pregnancy, giving birth during the daytime, and experiencing an obstetric complication were identified as determinants of RMC. According to the current systematic review and meta-analysis, receiving MNCH services from CRC-trained providers increases the likelihood of receiving RMC. The finding was supported by studies conducted in India, Sanford, USA, and Tanzania. It is widely acknowledged that CRC training is vital for gearing up MNHC providers to offer human-centered care, serve patients ethically and with respect, keep taking a professional oath, and promote providers to provide clients with satisfactory service quality. Besides that, the training may influence HCPs’ knowledge, motivation, and attitude toward the RMC, which will have a significant positive impact on its provision. As a result, managers in the health sector need to emphasize on the provision of CRC training for health care providers, with a due consideration paid to those who work at maternity service delivery points. The Federal Ministry of Health should collaborate with the Ministry of Education to incorporate CRC issues into the acting curriculum in order to familiarize newly emerging health care providers with the RMC. In addition, women who received adequate ANC had a greater chance of receiving RMC. This finding was in tandem with studies conducted in Kenya and Tanzania. The possible justification could be that women who had adequate ANC visits had a better chance of acclimating to the health facility setup and developing close relations with the HCP. As the evidence showed that having adequate ANC may result in a change in the dynamic between provider and client, which may increase the likelihood of receiving RMC. All of these are essential in ingraining trust in the facility’s services, which resulted in RMC. Furthermore, women with planned pregnancies were more likely to receive RMC than those who had not. This could be because women with planned pregnancies are more likely to receive prenatal care services in the same facility where they will give birth, facilitating their interaction with health professionals and ultimately leading to RMC. Furthermore, using MNCH services helps women become acquainted with the service providers, reduces depression, and increases the mother’s attitude toward the care as respectful. Evidence indicated that planned pregnancy increases women’s contentment and they acknowledge the service provided as reverent. On the other perspective, HCPs should be aware that it is their responsibility to treat all birthing women equally, regardless of their pregnancy planning status. Rather than mistreating women who have had unplanned pregnancies, it would be advisable to focus on preventing those very pregnancies through the provision of contraception. In addition, this meta-analysis discovered that women who gave birth during the day had a higher chance of receiving RMC than those who gave birth at night. The finding was supported by studies conducted in Kenya. This could be because there are more healthcare providers during the day than at night, when only one health worker may be assigned to duty in health centers. Furthermore, the way senior health workers and managers monitor health care providers during the day may be good, which gives rise to the delivery of RMC. At the national level, the majority of health facilities had infrastructural problems, such as a lack of electricity, and the range of this problem may be lowered during daytime childbirth, enhancing the likelihood of receiving RMC. On the other hand, the tendency to receive low RMC during the night shift may be explained by the low staff number -to obstetric cases that require nighttime maternity care services (i.e labor starts for most women at night time). Furthermore, health providers may become tired during the night due to workload, and they may not act normally because they are awake from sleep, all of which may result in physical or verbal abuse of the parturient. This may be an implication for human resource managers to assign an adequate number of HCPs to the night-shift duties to reduce workload. Finally, women who experienced obstetric complications were found to have a lower likelihood of receiving RMC. Studies conducted in India and Tanzania corroborated the current finding. This could be because women who experienced complications during labor are more likely to develop postpartum blues and depression, which can impede and lower the process and prevalence of receiving RMC. In addition, complicated labor necessitates frequent and meticulous follow- up, which exhausts the provider and may result in service abandonment. Furthermore, those women are admitted and stay in health facilities for an extended period with little or no support, and they may perceive the service as unwelcoming, which may result in underreporting of RMC. Furthermore, there are several dimensions of D&A that could theoretically be associated with complicated birth (e.g., unconsented care, lack of information and choice, lack of respect for values and preferences, exclusion of choice companion, lack of privacy), and all of these could result in a low prevalence of RMC. Regarding the strength, this was the first systematic review and meta-analysis of its kind in Ethiopia to assess the prevalence of RMC and its determinants. It could help policymakers and managers at all levels to improve the quality of MNCH, which is one of the HSTP and SDG agendas. However, due to some of the limitations listed below, the findings should be interpreted with caution. First, the search only included articles published in English. Because of the nature of the study design, the majority of the studies considered were cross- sectional, making it difficult to establish a cause-effect relationship. Furthermore, the studies were limited to six regions, which may limit the generalizability of the findings. Finally, because of the scarcity of comparable systematic reviews and meta-analyses, we were compelled to discuss some of our findings, with primary studies conducted outside of Ethiopia. # Conclusion As per this meta-analysis, the Prevalence of RMC in Ethiopia was low, suggesting that more emphasis is needed to plan and implement intervention measures. The pooled prevalence of receiving RMC varied across geographical regions and study periods. Accordingly, receiving service from CRC-trained health care providers, having ANC visits, pregnancy planning status, giving birth during the daytime, and experiencing an obstetric complication were identified as determinants of RMC. Managers in the health sector need to give due emphasis to the provision of CRC training for healthcare providers, who work at maternity service delivery points. Stakeholders in the health sector need to work to increase the uptake of prenatal care to improve client-provider relationships across a continuum of care. Human resource managers should assign an adequate number of HCPs to the night-shift duties to reduce the workload among obstetric providers. Due emphasis needs to be given to those women who developed an obstetric complication through continuous follow-up. # Supporting information We would like to thank Wachemo University, College of Medicine and Health Sciences, for providing us with free internet access while we were conducting this research. We would like to express our gratitude to all of the authors of the studies included in this systematic review and meta-analysis. AOR Adjusted Odds Ratio CRC Compassionate, Respectful and caring FMOH Federal Ministry of Health HCPs Health Care Providers HSTP Health Sector Transformation Plan JBI Joanna Briggs Institute LMICs Low and Middle-Income Countries MNCH Materna, neonatal and child health PRISMA Preferred Reporting Items for Systematic Reviews and Meta-Analysis RMC Respectful maternity care SNNPR South Nations and Nationalities People of the Region WHO World Health Organization 10.1371/journal.pone.0277889.r001 Decision Letter 0 Kassa Zemenu Yohannes Academic Editor 2022 Zemenu Yohannes Kassa 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 Oct 2022 PONE-D-22-19516Towards the quality of maternal and newborn health care: The level and determinants of respectful maternity care during childbirth in Ethiopia: A systematic review and Meta-analysisPLOS ONE Dear Mr Aklilu Habte, 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 16 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, Zemenu Yohannes Kassa, Msc 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 upload a new copy of Figures 3 and 4 as the detail is not clear. Please follow the link for more information: [" ext-link-type="uri" xlink:type="simple"\>https://blogs.plos.org/plos/2019/06/looking-good-tips-for- creating-your-plos-figures- graphics/"](https://blogs.plos.org/plos/2019/06/looking-good-tips-for-creating- your-plos-figures-graphics/) <https://blogs.plos.org/plos/2019/06/looking-good- tips-for-creating-your-plos-figures-graphics/> 3. We note that this manuscript is a systematic review or meta-analysis; our author guidelines therefore require that you use PRISMA guidance to help improve reporting quality of this type of study. Please upload copies of the completed PRISMA checklist as Supporting Information with a file name “PRISMA checklist". Additional Editor Comments: Dear Mr Aklilu Habte, Academic editors’ comments The topic of the manuscript is interesting. Nevertheless, the reviewers raised several concerns: considering this point, I invite authors to perform the required major revisions. \# You should modify the title “Determinants of respectful maternity care during childbirth in Ethiopia: A systematic review and Meta-analysis” You should give line numbers across the manuscript \#Abstract The background is too long, make shortened. 1\. The first and second sentences, it is difficult to understand. Write clearly and understandable way. 2\. Methods from when to June 2022? 3\. In the abstract abbreviation does not recommend (CRC and HCPs). 4\. As per this meta-analysis, the level of RMC in Ethiopia was low (48.44 percent), suggesting that more emphasis is needed to plan and implement intervention measures. This sentence needs modification and avoid the word level. What is your ground to say low? 5\. You should forward your recommendation based on your pertinent findings. \#Introduction 1\. The introduction is too long, you should be focused and addressed your objectives, what is known and what is not unknown. This article is similar to your manuscript <https://pubmed.ncbi.nlm.nih.gov/30760318/>. \#Methods 1\. Population: Women in the reproductive age group (15-49) Your population is childbirth, not reproductive age 2\. study settings are either facility based on community-based Result 1\. Why do you exclude qualitative studies? Why not synthesise evidence from these studies? Discussion It is too long. You should discuss your pertinent findings, how and why this result comes, and the limitations. \[Note: HTML markup is below. Please do not edit.\] 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 Reviewer \#3: Yes Reviewer \#4: Yes \*\*\*\*\*\*\*\*\*\* 2\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes Reviewer \#2: Yes Reviewer \#3: Yes Reviewer \#4: 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 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: Yes 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: This is an important and methodologically sound article, which will be of great interest and practical utility to a local audience as well as a global one. Nevertheless, there are some minor revisions that are necessary to clarify significant ambiguities stemming mostly from language issues, and the paper would be strengthened overall by good copyediting to improve the quality of the writing and thus the clarity of the content. Please find my inputs below: Abstract In the abstract (but not the body of the paper) there is a typographical error: “DerSimonian Laired”. CRC: this abbreviation appears throughout the paper, in reference to some aspect of the Ethiopian government’s Health Sector Transformation Plan that emphasizes “compassionate, respectful care”. It is unclear if this is a specific training program with a defined curriculum, learning objectives and outcome measures, such that it could be replicated with similar results. These details are important as it emerges as a significant variable in the logistical regression reported. If so, the program should be described and cited; from a language perspective it should appear with first letters capitalized. If not, I would suggest citing the HSTP, and adding some discussion and calls for future research to identify what elements of that program are effective in strengthening RMC. Introduction What is Reference 3 and how is it relevant to the point? There are two rights- based frameworks, White Ribbon and Khosla et al. that I suggest should be cited here. Maternity care includes more than monitoring. See ILO ISCO-8 classification of occupations for midwifery professionals for a concise, yet comprehensive list of responsibilities. Re. the following sentence, “Although several epidemiological studies on the magnitude and determinants of RMC in Ethiopia have been conducted, the results have been inconsistent and varied,” 1\) I am not aware of studies that expressly measure the prevalence of RMC; many measure the prevalence of Disrespect and Abuse (DA)/mistreatment. Does this study derive the prevalence of RMC from studies that aim to measure DA, and if so by what methodology? Or are these all studies that specifically measure RMC, and if so, how was RMC defined and operationalized in these original studies? Was it the same? 2\) I suggest reading and referencing here the study by Sando et al.: <https://reproductive-health- journal.biomedcentral.com/articles/10.1186/s12978-017-0389-z> Methods, Measurement of the outcome of interest Similar comment to the above: There are so few studies that measure RMC and there have been no standard definitions for RMC to my knowledge other than this paper (<https://pubmed.ncbi.nlm.nih.gov/34598705/>) that I am concerned whether the studies truly measured prevalence of RMC or whether they looked rather at DA/mistreatment and extrapolated RMC. RMC cannot be construed as the simple absence of DA, although this becomes nuanced. It would be important to understand how the authors calculated the prevalence of RMC in the base studies. Re. the following sentence, “RMC is a universal human right that must be provided to all childbearing women in every health system and is measured by four performance standards (friendly care, timely care, discrimination-free care, and abuse-free care),” What is the citation for this? Or was this the study definition? If so, it should be explicitly stated and operational definitions provided. Characteristics of Included Studies Re. the sentence, “All of the included studies collected data through face-to- face interviews with a pre tested, interviewer-administered questionnaire,” in what setting? In what timeframe relative to birth? See Sando et al. for comparison of prevalence by data collection setting and timing. Determinants of RMC In this section, there are some language issues that obscure the meaning of the results presented. Re. the following sentence, “As significant determinants of RMC, six variables were identified: giving birth during the day, planning status of previous pregnancy, having ANC visit, experiencing an obstetric complication, and receiving service from health care providers who were trained on CRC”: 1\) Does this refer to the current pregnancy, or a previous pregnancy resulting in a previous birth? I think it means the current pregnancy preceding the current birth. Please clarify. 2\) Does this refer to any ANC or adequate ANC? I believe from reading the whole paper it refers to adequate ANC, but this should be clarified. These are significantly different. 3\) Again, if CRC is a significant determinant, it should be defined and cited at first mention in this paper and its essential elements described. 4\) Most importantly, the directionality of the association between obstetric complications and RMC is NEGATIVE. This MUST be clarified. The way that these four significant factors are lumped together and jointly described as determinants of RMC is very misleading and confusing. The study results demonstrate that obstetric complications are a determinant of reduced odds of RMC, the opposite of the three other variables. This distinction should be made very clear every time it is mentioned. Page 13: Same comment: The fact that the association between RMC and obstetric complication is not conveyed clearly in the summary descriptions above and this is a very important distinction from the other factors that are positively associated. This must be made explicit in each instance it is mentioned. Discussion Page 15: Re. this sentence, “As per subgroup analysis results, the highest and the lowest level of RMC was reported in the Oromia, 58.01%(95% CI: 42.44, 73.58) and Benishangul Gumuz region 12.65% (95% CI: 9.45, 15.90) respectively, while it was lowest in.” Correct significant typos and missing words here. Page 15: Re. this sentence, “Accordingly, receiving service from CRCtrained health care providers, having ANC visits, pregnancy planning status, giving birth during the daytime, and experiencing an obstetric complication were identified as determinants of RMC” : Again, this is incorrect and misleading. OB complication is NOT a determinant of RMC but a barrier to RMC or risk factor for low RMC. Page 16: Re. the following sentence, “The possible justification could be that women who had adequate ANC visits had a better chance of acclimating to the health facility setup and developing close relations with the HCP. All of these are essential in ingraining trust in the facility's services, which resulted in RMC\[39, 53\]” please see my comment: Since provider behaviors are the basis of RMC, an explanation that centers the change in provider behavior toward clients who had adequate ANC, or a change in the dynamic between provider and client might be mentioned here. Otherwise, is the hypothesis that women's attitudes or perceptions changed if they had adequate ANC? The WHO Bulletin definitions of DA by Freedman et al might be interesting to consult here (<https://apps.who.int/iris/handle/10665/271621>). Page 16: Re. the following sentence, “Rather than mistreating women who have had an unplanned pregnancy, it would be recommendable to focus on preventing those very pregnancies through the provision of contraception,”: A more woman-centered way to express this might be, "assisting women to meet their need for contraception"... Page 17: Re. “This could be because women who experienced complications during labor are more likely to develop postpartum blues and depression, which can impede and lower the process and level of receiving RMC.” This statement needs a reference citation. Re. “In addition, complicated labor necessitates frequent and meticulous follow- up, which exhausts the provider and may result in service abandonment,” see my comment: There are a number of dimensions of DA that could be associated with complicated birth theoretically (e.g., unconsented care, lack of information and choice, lack of respect for values and preferences, exclusion of companion of choice, lack of privacy...) therefore, this merits further discussion and literature search/citations. Reviewer \#2: Dear PLOSE One team of editorials, thank you for giving me the chance to review the manuscript entitled "Towards the quality of maternal and newborn health care: The level and determinants of respectful maternity care during childbirth in Ethiopia: A systematic review and meta-analysis". Reviewer Comments to the Author This study gives very important results regarding the level and determinants of respectful maternity care during childbirth. However, in a few areas, here are my comments. General Comments Why do you review the articles on RMC only from Ethiopia? The abstract is many worded, Abbreviations are used in the abstract section. need correction What is the unique characteristic of the quality of maternal and newborn health care? What is the level of quality care? The determinants of respectful maternity care are not mentioned clearly in the introduction section as mentioned in the results. (e.g., friendly care, timely care, discrimination-free care, and abuse-free care). Make sure that all the elements of the background section are fulfilled. Describe in a sequential way what is known and unknown and what gaps you want to fill with your study. The introduction, results, and conclusion should be in line with the research objectives. The topic can be refined ( or make short) I think the population will be pregnant women. Reproductive age group is a vague term for RMC. The inclusion of data from unpublished studies can itself introduce bias. Insufficient citation, particularly in discussion for safe interpretation. Use correct tense, grammar, sentence, spelling, paraphrase, consistency…etc needs correction . Reviewer \#3: Dear Editor/ authors Despite writing nicely I felt some issues in the manuscript. I suggest addressing these issues to accept for publishing it, My suggestions/ comments are as follows. Abstract or summary section RMC measurement method is not clearly defined with specifying measurement scale (count, ordinal continuous, or binary).  The study claimed the use of the random effect model to analyze 43 (some places 38) studies.  Studies use a random effect model in panel data. This study evaluated mostly the results of cross section studies. The cluster variable (whether year, region, or something else) of this study is not clear.  If the studies were clustered, was that sample enough or statistical analysis? Missing full form of AOR.  Introductions section: The writing of the introduction section is too long but it missed explaining vital things.  In the last paragraph of the introduction section, the current knowledge of RMC requires further elaboration to justify. The message of the statement "the results have been inconsistent and varied"  is inadequate. Method section An illustration of the data screening process in the figure would make the paper more appealing to readers. Please refer to other meta-analysis-based papers in the health sector.  Results  I suggest placing most figures and tables in the main body. Readers find it difficult to follow materials in supplements and appendix. The determinant variables are a vital part of this study. Presenting the variables in the main body instead of S4 file would increase the values of this paper.   Discussion: This study benchmarked with meta-studies of different countries. I am doubtful whether the dates of the publications are of similar times.  Conclusion section: Hardly a few findings are generalized in the conclusion section. Most of the space is used for recommendations.  I would avoid the strong word "should" to write recommendations. Reviewer \#4: The systematic review and metaanalysis on the determinants of RMC was good. It would be worthwhile for policymakers to plan and improve childbirth care. The following changes are required in this manuscript: 1\. Abstract: Go through lines 4-7 and it's better to remove from the abstract and include in the introduction section. \- used the terms "prevalence" or "incidence" of RMC instead of "level of RMC" \- Remove the last two lines of conclusion in the abstract: " A due empahsis. 2\. Introduction: The introduction is too long, make it clear and to the point, focusing on research questions. I suggested including studies related to variables that affect RMC, the prevalence of RMC in ethopis health facilities or community- based facilities, and any differences in the prevalence of RMC in different sectors. The existing evidence of mistreatment and abuse or other components of RMC Then fill in the gaps with the study in the last paragraph. 3\. Study selection process: include how many articles for SR and metaanalysis. 4\. Incusion criteria: study design: make clear regarding reporting the level of RMC, I suggested to replace the level of RMC throughout the study by Prevalence. 5\. Could you explain which threshold of p value has been used for statistical significance when using the Cochrane Q test to determine statistical heterogeneity? 6\. Results Table 1 suggest to write Prevalence of RMC in heading Table 2: wirite components of RMC instead of domains of RMC. explain its details in the results section. Explain sensitivity analysis based on..(Fig 6); elaborate these information in the result section. 7\. Discussion: -The discussion was so long. Could you please focus on the main objective of the study? Look for the first paragraph of the conversation (you can make it very brief). -Discussed regarding components of RMC. -Could you include some other critical factors that influence components of RMC or the overall prevalence of RMC?  Include the current study's strength in the last paragraph before the limitation. \_Provide references in discussions ection line start...On the other hand, the tendency to receive low RMC during the night shift may be explained by the low staff number -to- obstetric cases -Provide reference for line start..... In addition, complicated labor necessitates frequent and meticulous follow-up, which exhausts the provider and may result in service abandonment \- Check reasons for this and reference (this might not be the case during child birth)..line start from...This could be because women who experienced complications during labor are more likely to develop postpartum blues and depression, which can impede and lower the process and level of receiving RMC. 8\. Conclusion: Include some information regarding the strength of evidence and write some of the geographical differences in RMC prevalence in Ethiopia. You already recommended removing the duplication of information in the discussion section. focus on the main findings of the study. \*\*\*\*\*\*\*\*\*\* 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 Reviewer \#4: **Yes: **Rojana Dhakal, School of Health and Allied Sciences, Pokhara University \*\*\*\*\*\*\*\*\*\* \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] 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.0277889.r002 Author response to Decision Letter 0 5 Oct 2022 Attached as "Response to reviewers" in the submission system. 10.1371/journal.pone.0277889.r003 Decision Letter 1 Kassa Zemenu Yohannes Academic Editor 2022 Zemenu Yohannes Kassa This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 31 Oct 2022 PONE-D-22-19516R1The prevalence of respectful maternity care during childbirth and its determinants in Ethiopia: A contemporaneous systematic review and Meta- analysisPLOS ONE Dear Dr. Habte Hailegebireal, 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 15 2022 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <plosone@plos.org>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript: A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'. A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'. An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols>. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at [https://plos.org/protocols?utm_medium=editorial-emailutm_source=authorlettersut m_campaign=protocols](https://plos.org/protocols?utm_medium=editorial- email&utm_source=authorletters&utm_campaign=protocols). We look forward to receiving your revised manuscript. Kind regards, Zemenu Yohannes Kassa, Msc 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: Dear Dr Hailegebireal, Thank you for considering PLOS ONE. I assessed your revision as Editor I am pleased to inform you that it is potentially publishable in PLOS ONE and I invite authors to perform the required minor revisions. in the title, you should remove contemporaneous. you should revise the grammar. line 49 rewrite again,line 287,288,325,337, 366, 384, 460 and 470 grammar error. \[Note: HTML markup is below. Please do not edit.\] 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 \#3: (No Response) Reviewer \#4: 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 Reviewer \#4: Yes \*\*\*\*\*\*\*\*\*\* 3\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#3: Yes Reviewer \#4: 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: No Reviewer \#4: 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 Reviewer \#4: 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: (No Response) Reviewer \#4: The review was good and highlights the major components of respectful maternity care in Ethiopia. I suggest checking for spelling errors in manuscripts, tables, and figures. \*\*\*\*\*\*\*\*\*\* 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 Reviewer \#4: **Yes: **Rojana Dhakal \*\*\*\*\*\*\*\*\*\* \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] 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.0277889.r004 Author response to Decision Letter 1 31 Oct 2022 General comment and suggestion I assessed your revision as Editor I am pleased to inform you that it is potentially publishable in PLOS ONE and I invite authors to perform the required minor revisions. Response: I thank you for the time and effort made to review the manuscript in detail, for your constructive comments, and for the opportunity to revise and resubmit. After completion of the suggested edits, the revised manuscript has benefitted from an improvement in the overall presentation and clarity. We have highlighted the document to indicate any changes in words, phrases, or sentences. Comment 1: In the title, you should remove contemporaneous. Response: We have amended the title as per your suggestion and highlighted it on the “title page” of the “Revised manuscript with track changes” Line 2-3, Page1. Comment 2: line 49 rewrite again, with lines 287,288,325,337, 366, 384, 460 and 470 grammar error. Response: thank you for your comment. As per your suggestion, we have tried to correct those sentences in the above-mentioned lines. All the corrected versions were highlighted throughout the revised version of the manuscript. 10.1371/journal.pone.0277889.r005 Decision Letter 2 Kassa Zemenu Yohannes Academic Editor 2022 Zemenu Yohannes Kassa 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 Nov 2022 The prevalence of respectful maternity care during childbirth and its determinants in Ethiopia: A  systematic review and Meta-analysis PONE-D-22-19516R2 Dear Dr. Hailegebireal, 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, Zemenu Yohannes Kassa, Msc Academic Editor PLOS ONE Additional Editor Comments (optional): Reviewers' comments: 10.1371/journal.pone.0277889.r006 Acceptance letter Kassa Zemenu Yohannes Academic Editor 2022 Zemenu Yohannes Kassa 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 2022 PONE-D-22-19516R2 The prevalence of respectful maternity care during childbirth and its determinants in Ethiopia: A systematic review and Meta-analysis Dear Dr. Habte Hailegebireal: 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. Zemenu Yohannes Kassa Academic Editor PLOS ONE [^1]: The authors have declared that no competing interests exist. [^2]: ‡ FE, BB and AG also contributed equally to this work.

# Introduction Pressure garment therapy (PGT) has been used to reduce scarring following burn injuries, resulting in improvements in erythema, scar height, and scar pliability\[–\]. Unfortunately, a major obstacle of PGT is low patient compliance. A survey of 100 adult pressure garment users found that less than half were in complete compliance with the therapy. Another study conducted on the use of tubular compression bandages found an average of 30% of the study population reported poor compliance. In addition to observing poor compliance within the study population, outcomes were correlated with patient compliance. Patients who were compliant had a higher probability of a “good” clinical outcome (58%), whereas non-complaint patients had a substantially lower probability of exhibiting good clinical outcomes (16%).. Therefore, therapeutic outcomes could be improved by increasing patient compliance. On average, patients are instructed to wear the pressure garments for 1 year following burn injury, though the recommended time frame varies widely from as little as 6 months to as long as 2 years. An early study that used adhesive sponges and elastic bandages to apply pressure following burn injury suggested that the therapy was effective even when discontinued after only 4 months. The recommended time frame for PGT is often based on the physician’s experience, and there has never been a controlled experimental study to suggest the necessary length of time that pressure garments should be worn, or to examine the effects of early removal. Because the garments are reported to be uncomfortably hot and itchy, patients may be more likely to wear them if the therapy could be discontinued after a shorter period of time. The goal of this study was to observe the effects of removing pressure garments four months after therapy was initiated. Red Duroc pigs that had been undergoing PGT for 17 weeks after burn injury were used in the experiment. Half of the wounds continued PGT for an additional 12 weeks, while the other half was released from pressure. Scar characteristics including scar area, roughness, depth, and scar biomechanics were assessed and tracked over time. # Materials and methods ## Animal care and scar formation All experiments were conducted using female red Duroc pigs following a protocol approved by the Ohio State University Institutional Care and Use of Animals Committee (Protocol \#2014A00000072). Anesthesia was initiated with Telazol (Zoetis, Florham Park, NJ) and maintained with isoflurane. Following shaving, alternating scrubs of 2% chlorohexidine and 70% isopropyl alcohol (Butler Schein, Columbus, OH) were used to clean the dorsal trunk. Prior to wounding, a 1x1 cm grid was tattooed on each side of the pig to track skin surface area increase with time. Thermal injuries were utilized to create full-thickness, cutaneous defects as scarring in porcine skin was previously reported to be exacerbated in burn injuries and to more closely mimic human hypertrophic scar compared to scars resulting from excisional injuries. To create full-thickness burn wounds, a 1” x 1” custom metal stylus was heated to 200°C and held against each side of the pig (N = 4 pigs) for 40 seconds. The burn eschar was excised and the wound bed grafted with split-thickness skin that was harvested from the dorsum using a Zimmer Air Dermatome (Zimmer, Warsaw, IN) and meshed 1.5:1. Saline soaked surgical sponges (Hydrasorb®, Carwild Inc. New London, CT) were packed into the wounds and secured with sterile spandex and skin staples (Henry Schein, Melville, NY). A fiberglass cast (3M Helathcare, St. Paul, MN) was formed over the back of the pig to protect the healing wounds. The cast was held in place with Vetrap^TM (3M Healthcare) and Elastikon® (Johnson & Johnson, New Brunswick, NJ). NOVAPLUS Fentanyl patches (Watson Pharmaceuticals, INC, Parsippany, NJ) were applied for pain management and removed 3 days post wounding. One week following grafting, bandages and bolsters were removed. ## Pressure application and removal Pressure was applied circumferentially around the dorsum using custom, adjustable pressure garments (Powernet; Darlington Fabrics, New York, NY) fabricated with Velcro™ to allow for pressure adjustments. Powernet fabric was selected as it was previously shown to exhibit lower amounts of fatigue and higher levels of applied pressure for a given reduction in circumference than Dri-Tek Tricot fabrics. Pressure was initiated within one week of grafting and was maintained over the wounds at 20 mm Hg. Polyurethane foam was used to ensure even pressure over concave areas, and pressure was checked using a Kikuhime pressure sensor (MediGROUP, Melbourne, Australia). Pressure beneath the garments and directly over the scars was checked once per day and adjusted if not at the targeted 20 mm Hg. Every three days, garments were removed and replaced with freshly laundered garments. 17 weeks after the initial wound creation, pressure was released from half of the pressure treated wounds and maintained in the other half for an additional 12 weeks, resulting in three treatment groups: controls (no pressure applied throughout the experiment), continuous pressure group (pressure applied within 1 week of grafting and maintained for a total of 29 weeks) and a pressure released group (pressure applied within 1 week of grafting for 17 weeks, then removed for the final 12 weeks). Each pig contained both treatment groups and controls with sites randomized (N = 8 scars per group). ## Scar area Immediately after burn excision and at 1, 3, 5, 11, 17, 21, 25 and 29 weeks after grafting, the boundaries of the scars (n = 8 per group per time point) were traced using a transparent film. The tracings were scanned with a ruler in the field of view, imported into ImageJ, and the scar area was quantified. The area at each time point was normalized to the area of that scar at the initial time point (week 0, post excision) to obtain a percent area for each scar. Average % area ± standard error of the mean (SEM) is reported. Growth of the pigs was calculated by tracking the increase in surface area using the tattooed grid. The area of six, initially 1 by 1 cm squares was measured on each pig at each time point, normalized to the area of that square at week 0, and averaged to obtain a percent increase in surface area due to the growth of the pig. ## Scar roughness Surface roughness of the scars was quantified using ten-point mean roughness, or Rz. Aquasil Ultra XLV dental impression material (DENTSPLY Caulk, Milford, DE) was applied to the scars to create a negative impression (or mold) of the surface topography of the scars. Molds from weeks 17 and 29 (n = 8 per group) were filled with Phase2Gel dental alginate (Accu-Cast Dental, Bend, OR) to create a positive impression of the surface of the scars. The alginate was removed and cut into slices. The cross sections of the slices were imaged with a ruler in the field of view, and the pictures were imported into ImageJ. Rz was obtained by measuring the depth of the 5 largest peaks in each tracing and averaging them. Three representative cross sections were used and averaged for each scar. Average Rz ± SEM is reported. ## Scar height Molds from weeks 17 and 29 were also used to measure the volume of the scars (n = 8 per group except Continuous Pressure group at week 29 where mold was too warped to collect reliable volume measurement). The negative molds were filled with Phase2Gel dental alginate until they were flush with the level of surrounding uninjured skin. The molds were weighed before and after being filled with alginate. The difference in the weight was used to calculate the volume by using the density of the dental alginate. The average height of the scars above uninjured skin was then calculated by dividing the volume of the scar by the scar area at that time point. For scars that were below the level of the surrounding normal skin, a positive replica of the mold was created. The mold was circled with fast dry acrylic latex caulk (DAP Products, Inc, Baltimore, MD) to include both the scar and the surrounding normal skin. After the caulk dried, casting resin (Alumilite Corp. Kalamazoo, MI) was poured into the mold and allowed to harden. The resulting positive resin mold was then weighed, filled with alginate, and weighed again to obtain the negative volume of skin below the surface of the normal skin. Dividing this volume by the area of the scar at that time point resulted in how far the surface of the scar was below the surrounding normal skin. These distances are reported as average scar height from normal skin ± SEM. ## Scar biomechanics At week 17 post-injury (prior to removal of PGT in the release group) and at the final time point (week 29), *in vivo* biomechanics were assessed using a torsional ballistometer (Dia-Stron, Inc.; Clarksburg, NJ). The ballistometer drops a small probe onto the skin from a known distance, k, and measures the rebound bouncing of the probe until it comes to rest. The initial penetration of the probe into the skin, or the indent (mm), can be used as a measure of the hardness of the skin. Softer, more pliable skin will result in deeper penetration of the probe, and a greater indent. K-values between 1.15 and 1.45 mm were considered reliable thus leaving an n = 8 for all groups at all time points with the exception of the Pressure Released group where n = 6. Average indent ± SEM is reported. ## Scar morphology At weeks 17, 21, 25 and 29, six mm biopsies were taken of the scars for cryosectioning. For each biopsy, a line was drawn across the scar in the cranial-caudal direction to ensure that all samples were sectioned in the same orientation. Half of the biopsy was snap frozen with the other half prepared for histological analysis. Biopsies were first sectioned in cross-section to assess whole tissue structure. To examine collagen orientation, the epidermis was then removed from the frozen tissue by coarse sectioning, the dermis was sectioned *en face* at 7 μm and sections were stained with hematoxylin and eosin. Images were collected of sections 60–75μm below the dermal-epidermal junction with the dorsal axis of the tissue aligned with the top of each image (n = 6 samples per group per time point). At weeks 17 and 29, additional biopsy punches were collected, fixed in 10% formalin (Sigma-Aldrich, St. Louis, MO) for 72 hours and embedded in paraffin. The samples were sectioned at 10 μm and stained with Masson’s Trichrome (Sigma-Aldrich, St. Louis, MO) and Picrosirius Red (Electron Microscopy Sciences, Hatfield, PA) to visualize scar morphology. A polarized light microscope (Zeiss Axioskop Widefield LM, Oberkochen, Germany) was used to image Picrosirius Red stained sections. ## Statistics A One-way Analysis of Variance (ANOVA) was used with a Tukey’s post-hoc test (JMP, SAS Institute Inc., Cary, NC). For pairwise comparisons, a student’s t-test was performed. Pig identification was blocked for random effects and statistical significance was considered at p \< 0.05. # Results ## Scar appearance Pressure treatment resulted in scars that were smooth with less contraction compared to control scars that became rough and elongated. When compression was released after week 17, scars became contracted, rough, and raised, similar to the control scars. Scars that received pressure throughout the experiment maintained a smooth appearance, with shapes more similar to the original square wound. Scar area was quantified as a measure of the contraction of the scars. From week 3 to week 17, scars that received PGT had significantly greater scar area compared to the control scars but were significantly smaller than normal skin (p \< 0.001). By week 17, scars undergoing PGT were an average of 148% of the original wound area, compared to control scars that were an average of 100%. Note that the uninjured skin area measured at week 17 was on average 230% of the original area due to the substantial growth of each animal, thus all scars exhibited significant contraction. When pressure was removed, the scars rapidly contracted, and at four weeks after pressure removal the pressure released group had significantly smaller area than the continuous pressure group, 142% and 176% respectively. For the rest of the study, the continuous pressure group continued to have significantly greater scar area compared to both the pressure released and the control groups (p \< 0.05) with no statistical difference in scar area between the control and pressure released groups. Scars treated with pressure had significantly reduced scar height and Rz (surface roughness) compared to control scars (p \< 0.01; Figs). Scars receiving PGT tended to be below the level of surrounding normal skin and were smooth, while control scars were raised \~1 mm above normal skin and were rougher. After pressure was removed from the pressure released group, the scars increased in height (p \< 0.001) and became rougher, similar to the control group, while the continuous pressure group remained smooth with reduced height. ## Biomechanics Compression treatment resulted in scars that were more pliable on average, indicated by an increase in the indentation of the ballistometer probe compared to control scars. After pressure was removed, the pliability of the pressure released group decreased significantly (p \< 0.01), while the pliability of the continuous pressure and control groups remained relatively constant. By the final time point, the continuous pressure group had a significantly greater indent, and thus greater pliability, compared to the pressure released and control groups (p \< 0.01). ## Scar morphology Trichrome stained histological sections showed thicker, denser collagen matrix (blue stain) in control scars at week 17 compared to scars that were undergoing PGT. Pressure treated scars showed a larger portion of the dermis that contained a greater amount of cell cytoplasm (red stain) compared to control scars. By week 29, the collagen content and density of all treatment groups increased, indicated by the darker blue stain, though scars in the released and control groups appeared to have a denser matrix than scars that received continuous pressure. Picrosirius Red staining at week 17 revealed that the collagen fibers in the control group tended to be thicker, compared to the thinner collagen fibers in the pressure treated groups that were oriented parallel to the surface of the scars. At week 29, fibers in the continuous pressure group continued to be thin and parallel, while the pressure released group became much thicker. Fiber thickness in the control group also increased from weeks 17 to 29. *En face* sectioning of the scar showed that scars treated with PGT contained fibers that were organized in a cross-hatch pattern whereas the collagen within the control scars was highly aligned in the dorsal-ventral direction. When PGT was halted in the pressure released group, a substantial increase in collagen alignment was observed within 4 weeks of release with fibers largely oriented in the dorsal- ventral orientation as observed in the control group. At week 29, alignment in the released and control groups was still observed; however it was obscured slightly by the increase in fiber thickness. # Discussion During the maturation of hypertrophic scars, collagen synthesis is elevated compared to normal skin and normal scars, and can remain elevated for 2–3 years after injury. During this time, there is an increase in collagen fibril thickness\[–\] and number of cross links. Picrosirius Red staining revealed a marked increase in the thickness of collagen fibrils in the control scars that occurred between weeks 17 and 29, indicating that the scars were not mature and still undergoing high rates of collagen synthesis. When pressure garments are removed, there is an increase in the biomechanical forces applied to the scars due to the tension of the surrounding skin. Prior studies have shown that applying uniaxial tension to reconstituted collagen *in vitro* results in decreased degradation of the collagen fibrils compared to relaxed control samples when exposed to either collagenase or matrix metalloproteinase-8. The collagen matrix in both studies was acellular, suggesting that a physical change in collagen morphology, driven by the mechanical environment of the dermis, is capable of altering collagen accumulation within the scar. We hypothesize that when PGT is halted and the compressive forces removed, tension is quickly re-established within the scar, altering the balance between collagen degradation and deposition. Additionally, it has been reported that fibroblasts on aligned matrices up- regulate collagen synthesis compared to randomly aligned matrices. As the release of pressure resulted in increased collagen alignment in the current study, it is likely that fibroblasts within this group produced more collagen (as evidenced by the increase in scar thickness from week 17 to 29). This increase in collagen deposition coupled with a greater resistance to degradation imparted by the mechanical tension on the scar may accelerate the scar thickening and contraction process. It is possible that maintaining pressure until the scar is fully matured, and collagen synthesis has begun to decrease, may prevent this disparity between synthesis and degradation, and allow the benefits of PGT to be preserved. The pressure released group in the current study showed rapid scar contraction after pressure removal; within 4 weeks the scars had significantly decreased area compared to the continuous pressure group. Wound contraction is often attributed, in part, to the differentiation of fibroblasts into myofibroblasts. An increase in myofibroblast differentiation, indicated by expression of α-smooth muscle actin (α-SMA), has been correlated with increased wound contraction in porcine burn wounds. Multiple studies have shown that fibroblasts cultured *in vitro* differentiate into myofibroblasts in response to increased stress\[–\], with significant increases in α-SMA observed in as little as 6 hours following loading. Conversely, a study conducted on human hypertrophic scars undergoing PGT found a significant decrease in α-SMA after 3 months of PGT compared to scars that were not treated with pressure. Removing pressure on the immature scars in the current study could have resulted in an increase in strain and subsequent differentiation of fibroblasts, resulting in the rapid contraction exhibited in this study. # Conclusion PGT therapy resulted in scars that were smoother, less raised, and less contracted after four months compared to control scars. However, removing pressure prior to scar maturation resulted in a rapid increase in scar contraction and significant scar thickening resulting in scars that were not statistically different than scars that had never been treated with PGT. Scars that received uninterrupted pressure continued to maintain observed benefits. These changes in scar properties after pressure cessation support the need to continue PGT for greater lengths of time. Additional studies will be needed to determine the duration of PGT required to achieve maximum benefit without regression after pressure removal. # Supporting information The authors would like to thank the Campus Microscopy and Imaging Facility at OSU and The Ohio State University Laboratory Animal Resource staff. The authors would also like to thank the Microscopy and Imaging Special Shared Facility at the Shriners Hospitals for Children- Cincinnati. [^1]: The authors have declared that no competing interests exist.

# Introduction Slow transit constipation (STC), an intractable constipation, is usually characterized by a heavily delayed colonic transit up to true colonic inertia. This is a very prevalent motility problem, but its aetiology has not been elucidated as yet. Even so, compelling evidence has revealed that interstitial cells of Cajal (ICC) contribute to pathogenesis of STC. ICC is found between the nerve endings and smooth muscle cells in the gastrointestinal (GI) tract. It is recognized not only as pacemaker cells for gastrointestinal movement but also as mediators of neuromuscular transmission. Studies have demonstrated that ICC density in the colon of patients with STC significantly decreased compared with those of normal patients. Therefore, a decreased number of ICC might lead to absence of slow wave activity, thereby affecting the contractile response and causing delayed transit in STC patients. In this regards, amelioration of ICC function using drugs may be essential to treat STC. Total glucosides of paeony (TGP) is extracted from the root of Paeonia lactiflora pall, which has been used for cramp and pain for over 1,500 years in traditional Chinese medicine. TGP, which contains paconiflorin (\>90%) and other components such as hydroxyl-paconiflorin, paeonin, albiflorin, benzoylpaeoniflorin and so on, was approved by State Food and Drug Administration (SFDA) of China (Approval No. H20055058) to enter the market as a disease-modifying drug for rheumatoid arthritis(RA) in 1998.A series of previous studies have demonstrated the most adverse events of TGP were gastrointestinal tract disturbances, mostly mild diarrhea, and no adverse events following hepatic, renal, or hematological tests were reported\[–\]. However, its side-effect of leading diarrhea caught our attention. Whether TGP is involved in mediating intestinal motility remains uncertain. This study is important for investigating the intestinal promotion function of TGP and its functional effect on ICC. # Materials and Methods ## Animals The experimental protocol was approved by the Animal Welfare Committee of Zhejiang Chinese Medical University, Hangzhou, China (SYXK2008-0115). The animal housing facility is a barrier housing facility, conforming to the national standard of China (Laboratory Animal-Requirements of Environment and Housing Facilities) (GB14925-2001). Female Munich-Wistar rats, weighing 160±20g, were obtained from Laboratory Animal Research Center of Zhejiang Chinese Medical University, China. ## Drugs and Reagents The Rat vasoative intestinal peptide (VIP) enzyme linked immunosorbent assay (ELISA) kit and the P substance (SP) ELISA kit were bought from Cusabio Biotech Co., Ltd (Wuhan, Hubei, China). The nitric oxide (NO) and the nitric oxide synthase (NOS) kit were purchased from Nanjing Jiancheng Bioengineering Institute (Nanjing, Jiangsu, China). In this study, primary antibodies used for western blot included rabbit anti-SCF and anti-β-actin antibodies (Santa Cruz Biotechnology, CA, USA). cDNA synthesis Kit (TaKaRa Bio, Siga, Japan) was used for reverse transcription. C-kit andβ-actin primers used for reverse transcription polymerase chain reaction (RT-PCR) were designed and synthesized by Sangon Biotech (Shanghai) Co., Ltd (Shanghai, China). Compound diphenoxylate was purchased from Changzhou Kangpu Pharmaceutical Co., Ltd, Changzhou, China (lot number: 1010004). TGP capsules were purchased from Ningbo Liwah Pharmaceutical Co., Ltd, Ningbo, China (lot number: 110705). The extract was obtained with the following procedure. The root of white peony material was extracted with 70% aqueous ethanol, 2 h each time. Then, the extracts were combined and evaporated to dryness under reduced pressure. The residue obtained from the combined extract was dissolved with water. After filtration, the aqueous solution was extracted three times with ethyl acetate successively each time, and then evaporated to obtain the crude extract. Then the ethyl acetate extract was chromatographed on polystyrene resin with water and 20% ethanol, respectively. Portions of 20% ethanol were collected and evaporated to get the final extract. According to the information supplied by the medical corporation and the results of HPLC analysis, the TGP mainly contained paeoniflorin(90.42%), hydroxyl-paeoniflorin(0.24%), paeonin(0.93%), albiflorin (0.14%), benzoylpaeoniflorin(3.44%), and some trace substances(about 4.8% total). The amount of each substance was calculated based on the peak area of HPLC. ## Experimental Design The atropine-diphenoxylate was used to induce constipation in rats according to previous reports. A total of 30 rats were randomly divided into 3 groups (n = 10 per group): normal control group (NC), model control group (MC) and group of STC rats treated with TGP (MC+ TGP). NC: treated with normal saline (10ml/kg); MC: treated with atropine (0.1mg/kg)–diphenoxylate (10mg/kg); MC+ TGP: treated with atropine (0.1mg/kg)—diphenoxylate(10mg/kg) and TGP (0.18g/kg). Except the NC rats, others were orally administered by atropine-diphenoxylate once daily for 14 consecutive days to induce constipation. The TGP was suspended in water and was orally administered once daily from day 15 to day 28 in the MC+TGP group, while normal saline was given gastric gavage at 1 ml/100g in the NC group and MC group. ## Observation of the Fecal Volume and Moisture Content The animal faeces were collected and counted during the experimental process. The wet weight (A) of faeces was recorded immediately after collection, and dry weight (B) of faeces was recorded after they were dried in drying oven for 3h. The moisture content was calculated as follows: the fecal moisture percentage (%) = (A-B)/A\*100%. ## Blood Sample and Tissue Collection After 14 days of treatment, rats were fasted for 12h but allowed free access to water. Each rat was administered with 0.2 ml of standard charcoal meal (5% activated charcoal suspended in 10% gum acacia) orally. Rats were then anesthetized 30 min later by intraperitoneal injection of sodium pentobarbital (50 mg/kg) and placed on a temperature-regulated table. Blood samples were collected and centrifuged at 3500 rpm for 15 min to obtain serum. The intestine was carefully inspected and the distance traversed using the charcoal meal from the pylorus was measured for all the groups. The colon was removed immediately and flushed with normal saline at 4°C, and then sectioned into two pieces. One fragment was fixed in 10% formalin and processed in paraffin for subsequent immunohistochemical studies, while the other was stored at -80°C until they were assayed. ## Evaluation of the Intestinal Kinetic Power The intestinal transit rate was calculated as follows: intestinal transit rate (%) = *A*/*B*×100% (*A*: length among the musculus sphincter pylori and the end of charcoal stained intestine; *B*: whole length of the intestine (distance from pylorus to rectum).) ## Assays of Blood Sample The levels of NO and NOS were determined by colorimetric method using commercial reagents. The VIP and SP concentrations in the serum were estimated by ELISA using commercially available kits. ## Immunohistochemical Analysis Unstained 5μm sections were cut from paraffin blocks for immunohistochemical (IHC) analysis. The sections were stained with rabbit anti-c-kit (1:100) at 4°C overnight. The secondary antibody and avidin-biotin peroxidase complex method were used according to the manufacture’s instructions. An immunoglobulin- negative control was used to eliminate nonspecific binding. Expressions of c-kit were determined by the semi-quantitative method. Using Carl Zeiss imaging system (Carl Zeiss Imaging Systems, Germany), the area of stained cells were measured under 40×objective lens of each slice for semi-quantitative analysis. ## RT-PCR Total RNA was extracted according to the manufacture’s protocol using TRIzol Reagent (Invitrogen, USA). The total RNA (1μg) was reverse-transcripted into cDNA using SuperScript II reverse transcriptase. The operations were done strictly in accordance with the manufacturer’s protocol. The expression level of c-kit was analyzed by semi-quantitative RT-PCR with β-actin as internal control. The primer sequences were designed as follows: c-kit forward 5´-AGATGACGAGCTGGCTCTGGA-3´and reverse 5´-CTGGCTGCCAA ATCTCTGTGAA-3´;β-actin forward 5´-ACACTGTGCCCATCTACG-3´ and reverse 5´-CAGGATTCCATACCCAAG-3´. After 30 cycles of amplification, 5μl of amplification product was used for gel electrophoresis. The software program Gel-Pro Analyzer (BIO-RAD, USA) was used for quantitative analysis of the stained gels. The integrated optical density (IOD) of the bands on digitized images was measured. All RT-PCR reactions were done 3 times for each sample. C-kit gene expression was expressed as the ratio of c-kit overβ- actin. The ratio between the PCR amplified gene and the amplified standard was obtained for each sample. ## Western Blot Analysis Colon samples were homogenized and lysed in SDS-PAGE sample buffer, boiled, centrifuged and supernatant recovered. Samples were run on 10% SDS polyacrylamide gels, electroblotted onto nitrocellulose membranes, incubated with blocking buffer for 1h. Immunoblotting was assayed using anti-SCF (1:200) antibodies. Anti-β-actin (1:200) was used as loading control. The detection and analysis were done by Odyssey Infrared Imaging System (Gene Company Ltd., Hongkong, China). ## Statistical Analysis Measurement data were expressed as mean±SEM and analyzed by SPSS 16.0 software. Significant differences between the experimental groups were analyzed by the Student’s *t* test or Mann-Whitney *U* test when appropriate. *P* \<0.05 was accepted as statistically significant. # Results ## The Intervention with TGP Partially Recovered Intestinal Function Compared with normal faeces, the faeces defecated by rats in the MC group were dry, small, hard and without burnish. This condition ameliorated after treatment by TGP. The number and moisture content of faeces decreased after oral gavage of atropine- diphenoxylate, whereas they increased when treated by TGP. showed the intestinal transit rate from 3 groups, respectively. The intestinal transit rate in MC group was statistically significantly lower than that in NC group (*P*\<0.01), whereas the rate in the treatment group was higher than that in MC group (*P*\<0.01). ## TGP Treatment Reduced the Inhibitory Neurotransmitters The values of NO and NOS of the MC group rats were increased when compared with the normal rats (*P*\<0.01, respectively). Elevated NO and NOS was significantly attenuated by TGP (*P*\<0.01, respectively). The serum levels of VIP obtained from MC group rats were increased significantly compared with those from NC group (*P*\<0.01). Administration of TGP attenuated the increased VIP (*P*\<0.01). Compared with normal rats, constipation rats attenuated SP levels(*P*\<0.01), and administration of TGP could not change the levels of SP (*P*\>0.05). ## TGP Ameliorated the Function of ICC To assess the change of ICC in intestinal mucosa, immunohistochemical analysis and RT-PCR were done to investigate expression levels of protein and mRNA of c-kit. The results revealed that fewer c-kit positive cells were observed in the intestinal mucosa of the STC animals than those in the normal rats. As shown in, c-kit positive signs were widely distributed in intestinal mucosa of STC animals with treatment of TGP. The positive cells areas also showed that c-kit expression levels in intestinal mucosa of constipation rats were reduced(*P*\<0.01). The decreased c-kit in constipation rats was significantly increased by TGP treatment (*P*\<0.01). The results showed the mRNA levels of c-kit in MC group was statistically significantly decreased compared with that in the NC group (*P*\<0.05), whereas the mRNA levels in TGP group was higher than that in MC group(*P*\<0.05). We also examined the expression of SCF, a ligand of c-kit. The protein expression of SCF in the MC group rats was statistically significantly lower than that in the NC group, whereas treatment with TGP increased SCF protein expression. # Discussion STC is a common health problem which usually attribute to abnormal motor function of the large bowel such as decreased colonic propulsive function, overall reduced electrical or motor activity of the large bowel, alterations of rectosigmoid contractile activity, and abnormal response to food ingestion.Studies have shown that the intestinal transit rate of atropine- diphenoxylate induced rats was significantly reduced compared with that of normal animals, which indicated the successful establishment of the animal model of STC. Our study also proved significant improvement of colonic transit after TGP treatment in STC rats. This result suggested that TGP showed promotion function on intestinal motility. The motor activity of the GI tract is a complex process involving multiple cell types such as enteric neurons, ICC and smooth muscle cells. Although ICC that transduces inputs from enteric motor neurons and generates intrinsic electrical rhythmicity is a minor component of the tunica muscularis of the GI tract, it plays a very significant physiological role in intestinal motility. Several neurotransmitter receptors have been identified in the ICC. Some of them are excitatory (SP) and others inhibitory (VIP and NO) transmitters. ICC accepts neurotransmitters via these receptors and then transfers them to neighboring smooth muscle cells via gap junctions. Previous studies showed that neurotransmitters such as NO, VIP and SP were crucial in the STC. Excessive production of NO and VIP may inhibit colonic motility in STC colon. Low numbers of substance P-immunoreactive (SP-IR) nerve fibers have been found in colonic circularmuscle from STC patients. This study found that the serum levels of NO, NOS and VIP in the model group were increased and SP was decreased. In the TGP group, the serum levels of NO, NOS and VIP were reduced compared with those in the model group. These data suggested that TGP played a part in intestinal motility, possibly by decreasing the inhibitory neurotransmitters. Many ICC express c-kit, which is a membrane receptor with tyrosine kinase activity. C-kit acts as a receptor for stem cell factor (SCF), which serves as its ligand. In its normal state, c-kit is present as a monomer in the cell membrane. This will activate intracellular signaling pathways that are pivotal for normal cellular growth and development. Neither antibodies to kit, nor SCF, nor tyrosine kinase inhibitors could inhibit the proliferation of these cells, while mature ICC were greatly affected. The c-kit signal pathway may be essential in ICC reduction in STC because the expression of c-kit mRNA and c-kit protein was significantly decreased in the colon of STC. In our experiment, the expression of c-kit mRNA and c-kit protein and SCF protein were down-regulated in the STC rats. As these decreased expressions were up-regulated by TGP, TGP had obvious protective effect on ICC. # Conclusions In conclusion, TGP promoted the intestinal motility by improving the function of ICC and regulating neurotransmitters. The present study provided scientific evidence that TGP is a possible agent for the treatment of constipation. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: FZ SX GX. Performed the experiments: FZ FC JJ GX. Analyzed the data: FZ SX YZ. Contributed reagents/materials/analysis tools: FC JJ. Wrote the paper: FZ GX.

# Introduction There has been a recent push in the literature towards rigorous impact evaluations of interventions, coupled with a strong emphasis on how performance varies by context, specifically the underlying tenure regimes that create different incentives for conservation and land use clearing. However, still very little is known about the heterogeneity of interventions based on the context. We address this gap by examining how a common conservation intervention—protected areas, has varying impacts depending on the underlying property rights and the profitability of forests. We focus on the Yucatan peninsula in Mexico, an area with high biodiversity and importance for climate change mitigation ([www](http://www/). <http://theredddesk.org/countries/mexico>. Accessed July 19, 2017), but also high threats of forest conversion (e.g.,). Despite Mexico’s importance for conservation and climate change mitigation, not much is known about the effectiveness of conservation and climate change mitigation efforts in the area. A handful of studies examine the effectiveness of Mexico’s protected areas using rigorous statistical methods that account for the non-random placement of the interventions. Honey-Roses et al use quasi- experimental techniques to assess the effectiveness of a butterfly reserve in central Mexico. In a country-wide analysis, Blackman et al. find no statistically significant overall impact of protection between 1993 to 2000, but significant heterogeneity within regions. Pfaff et al. examine the effectiveness of Mexico’s protected areas between 2000 and 2005 and find an overall 3.2% decrease in deforestation. The authors ascribe the change in effectiveness from 1990s to a shift in the conservation politics and funding. Sims & Alix-Garcia find that Mexico’s protected areas and payments for ecosystem services reduced forest loss on average by 20–25% between 2000 and 2012, despite differences in the impacts on livelihoods. However, no previous study controls for the type of property rights. Previous studies have suggested that the performance of protected areas may be significantly hindered by the lack of funds and effective institutions especially in developing countries. Where formal protection may not be possible or easily enforced, establishing community land tenure has been considered an alternative as a way to incentivize and empower local communities to manage forest resources sustainably. The reason is the increasing returns to scale when forestry is practiced on a larger area; with limited other employment opportunities, forestry can generate employment opportunities and income that can benefit the whole community. In these cases, theory suggests, communities have a strong incentive to maintain the forest stocks. Theory has suggested that collective management of natural resources can be successful as a conservation strategy in cases where communities have existed for a long time and have clear rules and membership, conflict resolution mechanisms, recognized rights to organize; where standing forests generate significant benefits; where the forest resource is clearly delineated and stationary; and where there is little migration, (). Conversely, in areas where these conditions do not hold, the creation of private property may be a way to conserve resources by incentivizing investment in the land and facilitating monitoring and enforcement. Large-scale empirical evidence on the role of tenure in forest conservation in Mexico is still largely lacking. To address the role of community land tenure in preventing deforestation, previous work on Mexico has relied on case studies, focused on only a particular type of common property resource—ejidos (), or does not sufficiently account for endogenous placement of interventions like protected areas. Using rigorous statistical techniques and large-scale data, we examine the role of property rights, specifically, their interactions with protected areas, in preserving forests and the carbon stored within. We consider formal protected areas, community forest tenure (specifically, ejidos), parceled ejidos—a form of private property established from dividing formal ejido land, and private property in reducing forest loss in the Yucatan peninsula in Mexico. Specifically, while controlling for differences in the location of different tenure regimes, which proxy for the deforestation threats and the ease of enforcement, we examine what tenure regimes are effective in preventing forest loss. In addition, differentiating between forests biomes and combining biomass data with deforestation data, we examine the impact of the different tenure regimes across different types of forest. To our knowledge, ours is the first study that uses statistically rigorous techniques and large-scale data to (1) examine the interactions between protected areas, communally held forests, and private forests and (2) quantify the causal impact of community land tenure (ejidos) relative to private property and parceled ejidos. It is also one of few studies to address the role of property rights in conservation using large-scale data and a rigorous empirical design (). Our results indicate that protected areas in the Yucatan are effective on average, but, unsurprisingly, their impact varies by tenure regime. On average, we do not find evidence that community tenure in the absence of protected areas reduced forest loss relative to private properties. However, controlling for differences in the location, we find that protected areas in ejidos may be more effective than protection in other types of tenure. Consistent with previous studies (e.g.,), we do find evidence that in the absence of protected areas, parceling ejidos may increase the probability of forest loss. Further, we provide evidence how the impacts vary by the type of forest (dry vs. moist broadleaf forests; high vs. low-biomass forests). Given the Yucatan’s importance for biodiversity conservation and climate change mitigation, our results underscore the need to understand the mechanisms through which interventions interact on the ground, in order to design effective conservation and development policies in the region. # Methods & data ## Study area Our study area covers the three states in the Yucatan Peninsula: Campeche, Quintana Roo, and Yucatan. The area is considered a biodiversity hotspot; as part of the Maya Forest region, it contains the largest remnant of natural vegetation in Mesoamerica, high levels of species diversity and endemism, but also experiences high threats from human disturbance. We focus on the dry and moist broadleaf forests, which are the two forest biomes that span most of the peninsula and store on average 311-533tC/ha and 455-755tC/ha, respectively. They differ in their commercial importance: the dry broadleaf forests are converted to agricultural land, and harvested for fuelwood, marketable timber and charcoal, whereas the moist broadleaf forests are used more for commercial timber harvesting, but, to a smaller extent, also are cleared for agriculture and harvested for fuelwood and charcoal production. Because of the variability of mangroves ecosystems comprised from shrub-like vegetation to trees, we exclude them from the analysis. Forest loss on the peninsula is primarily driven by (1) conversion to pasture for commercial cattle ranching, maize cultivation and, more recently, large- scale mechanized agriculture for crops like soybeans, sugar cane, and sorghum, (2) urban development especially in areas that are important for tourism or can accommodate waterfront properties, (3) parcelization of ejidos, which can follow increases in the demand for urban, tourism, or agricultural land and/or can be the result of informal land markets and local arrangements (e.g.,) (4) and the practice of the traditional for the area slash-and-burn subsistence agriculture (called *milpa*) that results in small temporary openings in the forest canopy that are planted with maize, beans, squash and other subsistence crops, harvested for up to three years, and then left to regenerate into secondary forest. In the Yucatán Península, greater forest cover loss from agricultural conversion was recently reported in Campeche and Yucatán compared to Quintana Roo, although forest cover loss from disturbances such as fires and from urbanization in tourist areas was greater in Quintana Roo. While timber harvesting is a very important source of revenue for some local communities, forestry practices in the region usually do not directly result in long-term conversion. However, small-scale, usually temporary, tree canopy loss from logging operations does occur: the creation of gaps when individual trees are felled, the clearing of forests to create log landings and in some cases patch cuts of up to 4000m² that allow for the natural regeneration of shade-intolerant species like mahogany. The main tenure regimes in the area consist of private properties; ejidos, which are a form of collective land ownership and management; and former ejidos whose land has been parceled and allocated to individuals. Ejidos comprise most of the study area: 5.5 Mha distributed among 3,613 properties. Private properties encompass 1.8 Mha in 11,532 polygons; parceled ejidos span 1.5 Mha, distributed across 8,889 properties. A large fraction of the properties was covered in forest in 2000. Established in 2 waves (1934–1940 and 1960s-1990s), ejidos traditionally have forests under communal use and agricultural land parceled and allocated to individual households. Only a portion of the people living within an ejido have rights over the use and benefits of ejido forest resources and are responsible for forest management (these individuals are known as *ejidatarios*); the remaining households do not have rights to the ejido forest and do not directly benefit from it, although they may receive indirect benefits in the form of bribes not to deforest or improved community infrastructure (as in). In ejidos, communal forests are managed separately from agricultural and urban areas; each ejidatario has access and decision rights to their portion of the agricultural and urban area, while management decisions for communal forest areas are undertaken by the ejido general assembly, made up of all ejidatarios. Although land use conversion inside communal forest areas is not permitted by the ejido and national law, illegal deforestation can take place under weak monitoring and enforcement. Private property has been present before the establishment of ejidos; private property owners are primarily engaged in agriculture, residential or commercial real estate, tourism, and to a lesser extent, commercial forestry, mostly for charcoal production. Some ejidos have recently been divided into individual landholdings (we refer to these as parceled ejidos), although these are mostly informal and often not legally recognized. Because parceled ejidos may be in a state of flux, with forest stocks transitioning from collectively to individually owned and managed, we consider as a them as a separate category. Our sample spans ejidos parceled between 1993 and 2012. Only a small portion of existing ejidos (11%), parceled ejidos (1%), and private lands (3%) fall within protected areas, which place additional constraints on forest conversion and land use practices. For example, the Agrarian Law (*Ley Agraria*) restricts ejidos in establishing settlements or urban areas within protected areas; agriculture and cattle raising activities are allowed if they comply with the applicable Secretariat of Environment and Natural Resources (SEMARNAT) regulations. However, protected area management plans and conservation strategies by the federal Commission for Natural Protected Areas (CONANP) and state environmental institutions intend to discourage deforestation in overlapping and contiguous properties. We use pixels as the unit of analysis. For the sake of brevity, we refer to sample units under protected areas within existing ejidos, private properties, and parceled ejidos as protected ejidos, protected private property, and protected parceled ejidos, respectively. Similarly, the sample units located outside protected areas as well as outside the boundaries of properties spanning protected areas within existing ejidos, private properties, and parceled ejidos are referred to as unprotected ejidos, unprotected private property, and unprotected parceled ejidos throughout the text. The details for the empirical analysis as well as the data used are described in the Methods and Data sections below. ## Methods Our analysis answers the following questions: 1. What was the average impact of protection within our study area between 2000 and 2014? 2. What was the average impact of protection in private properties, existing ejidos, and parceled ejidos between 2000 and 2014? 3. Did ejidos enhance the impact of formal protected areas more than private properties and parceled ejidos? Question A allows us to quantify the average causal impact of protected areas in reducing deforestation in the Yucatan, whereas Question B allows us to test for heterogeneity in the impacts of protection based on the underlying property rights. The latter comparisons involve protected and unprotected pixels falling *under the same* tenure regime. While these comparisons allow us to establish the average causal impact of formal protected areas within a tenure regime, they do not allow us to compare one tenure regime to another. The reason is that we cannot rule out that the differences of the different tenure regimes in effectiveness may be driven by differences in their location. Controlling for differences in location, Question C allows us to further examine how tenure regimes impact the effectiveness of protected areas. In that analysis we first compare observationally similar *protected* pixels under one tenure regime to *protected* pixels under another tenure regime and then observationally similar *unprotected* pixels under one tenure regime to *unprotected* pixels under another. We also examine how the emerging patterns vary by the type of forest. ### The need for matching techniques Characteristics of the land like the biophysical attributes and proximity to markets and infrastructure determine its suitability and profitability for agriculture and development. Because forests experience different conversion threats as a function of their characteristics (e.g., how suitable the soil is for agriculture, how accessible timber markets are etc), not controlling for the differences in those is likely to invalidate causal inference. That is, simple comparisons of deforestation in the different tenure regimes will not allow us to attribute how much of that is due to the tenure regime itself and how much-to the location of the sample unit (e.g.,). Depending on the question, we use the presence of a formal protected area or a tenure regime as the treatment and control for its endogenous placement. In order to quantify the *causal* impact of the treatment on deforestation, we use matching techniques, which are a quasi-experimental statistical approach that allows us to establish what would have happened to a sample unit, had it not been placed under a given tenure regime. The relevant statistical measure is the Average Treatment Effect on the Treated (ATT), with negative values indicating a reduction in the probability of forest loss due to the treatment. The choice of matching estimator for each specification is driven by the data: we use the matching estimator that results in the smallest covariate difference (as indicated by a normalized difference between the treatment and control distributions for that covariate) between the treatment and control groups. For the most part, we calculate the statistic using nearest neighbor matching with replacement, a Mahalanobis distance metric and trimming based on a propensity score. The propensity score is the linearized predicted probability of an observation being under a particular treatment and is derived from a logit model that has the binary treatment as the outcome and biophysical and physiogeographic covariates related to the placement of the treatment and drivers of deforestation as independent variables. In some of the specifications, we use propensity-score augmented matching. The type of matching for each specification used is reported in the Results summary tables. In all specifications, matching was done within a forest biome. Post-matching, we performed bias and variance adjustments. To examine the average impact of protection on deforestation (Question A above), we use as treatment pixels falling under a protected area, regardless of the tenure regime, and match them to observationally similar unprotected pixels, regardless of the tenure regime. In order to examine the average impact of protection in private properties, existing ejidos, and parceled ejidos (Question B), we perform matching within each tenure type, such that pixels falling under a protected area were considered as treatment and matched to observationally similar unprotected pixels from the same tenure regime, but not from the same property. Using analogous matching methods, we separately test for the possibility of spillovers resulting from displacing deforestation from the protected to the unprotected portions of a property. That is, for these specifications we consider the pixels from the unprotected pieces of properties partially spanned by a protected area as treatment and compare them to observationally similar pixels from properties not spanned by a protected area. We perform these tests for each tenure regime separately. Lastly, by controlling for the differences in the location, we compare each type of regime in terms of reducing deforestation (Question C). Specifically, in order to control for differences in the location of private properties and ejidos, we directly compare pixels in existing ejidos to observationally similar pixels under private property; we repeat the matching analysis for the pixels under parceled ejidos and private property, respectively. We perform the analysis for the protected and unprotected pixels, separately. That is, we compare forest loss in protected ejido pixels to observationally similar protected pixels in private property and forest loss in unprotected ejido pixels to observationally similar unprotected pixels in private property. We perform analogous tests comparing private properties and parceled ejidos. In all tests, we treat the two forest types separately. In order to examine how the impacts of formal protected areas and the three tenure regimes vary with the type of forest (old-growth vs. not), we run partial linear models (PLM) using on each of the matched samples difference weights. Specifically, we run the PLM on an equation of the form: $${tt}_{i} = \alpha + x_{i}\beta + f(z) + \epsilon_{i},$$ where *tt* is the treatment effect calculated for individual pair *i* (calculated as the difference between the outcome for the treated observation and its matched control), α is a constant, *ϵ_i* is an idiosyncratic error term, *x_i*-a vector of control variables, and *β*-a vector of the regression coefficients for each linear covariate. Motivated by the context in the Yucatan, the linear control variables include the distances to inland water, urban areas, roads (paved, unpaved, and highways), temperature, elevation, slope, precipitation, forest cover in 2000, and population density. z is the non-linear covariate—the amount of woody biomass in 2000, which proxies for old-growth vs. new forests. We generated 95% confidence intervals for the non-linear covariate via wild t bootstraps on 50 runs. The number of iterations was limited by the sample size and the resulting computational burdens. Apart from allowing us to examine the impact on forests that differ in their importance for climate change mitigation and biodiversity, the PLM analysis addresses concerns that our estimates may be picking up temporary forest conversion due to shifting agriculture (e.g., *milpa*). Thus, we are able to examine whether forest loss is driven by clearing of lower biomass forests as part of traditional *milpa* agriculture. ## Data Our source data include information on the boundaries of existing ejido lands, private property, parceled ejidos, national and federal terrains , formal protected areas, tree cover loss, and biophysical, socioeconomic, and physiographic covariates. In cases of overlap, property boundaries were manually adjusted using an established topological rule-set. Because of data unavailability, we do not differentiate between different types of ejidos (e.g., forestry, urban, or agricultural) and instead consider them a single category; we address this limitation of our analysis in the Discussion section. The formal protected area boundaries are based on the World Database on Protected Areas boundaries (Available at <http://www.protectedplanet.net/>. Accessed August 2018). Polygons listed as designated or inscribed and established before 2000 were included in the study. There are 63 designated protected areas on the peninsula under four different governance schemes; from those, we excluded lands within protected areas established after 2000, as newer protected areas may require more time to effect change. Our analysis does not distinguish between the different types of protected areas. ### Sampling Using a 30-meter pixel from the forest cover data as our unit of analysis, we randomly sample one percent of all areas forested in 2000; this resulted in \~1.6 million sample pixels; the sample size is constrained by the computational burden of very large datasets. We exclude all pixels that fell within urban locations and water bodies. To minimize the possibility of spatial dependence, we exclude spatially adjacent pixels (i.e., within 30 meters of another observation). To control for the possibility of spillovers within the same property, we exclude pixels that fall within the unprotected parts of protected tenure polygons; we use them to test whether protection displaces the forest clearing activities to the unprotected pieces of the properties (leakage or spillovers). An underlying assumption of the analysis is that spillovers from the protection of a property do not occur outside the property. Because of the small number polygons (n = 1) in the study area, we exclude all observations that are under indigenous lands or *comunidades*. While these also have common forest tenure regimes, they were formed by local, homogenous, indigenous groups. This is in contrast to ejidos that were created from often non-indigenous households from other regions and potentially different ethnic backgrounds. We further exclude observations from (a) two ejidos in Quintana Roo with current sustainable forest management certification (Noh Bec and Caoba, both of which obtained their Forest Sustainability Council (FSC) management certification in 2005), because of the changes in their governance, forestry practices, and access to financing that came with certification; and (b) four Quintana Roo ejidos that had their forest sustainability management certification revoked during the study period. The reason is that these areas were very badly affected by Hurricane Dean in 2007, which damaged huge swaths of land and necessitated clear cutting of the damaged forests. ### Covariates Our statistical approach aims to establish what would have happened to a sample unit, had it not been placed under a treatment. To this end, we need to account for the characteristics of the sample units that would make them suitable for alternative units (e.g., agriculture, tourism). The covariates we use reflect the ease of deforestation and the suitability of the sample units for alternative land uses. We focus on available data representing the biophysical, physiogeographic, and socio-economic characteristics that affect profitability of wood products and charcoal (e.g., type of forest, proximity to markets proxied by distances to urban areas and ports), the suitability for agriculture (slope, elevation, precipitation, temperature, the proximity to water bodies for irrigation, and accessibility, proxied by the distance to roads, ports, and urban areas), and the suitability for tourism (accessibility, proxied by proximity to roads and urban cities as well as the proximity to water bodies for the potential for waterfront properties). In addition, we use the woody biomass density in 2000, to proxy for the type of forest—old growth vs. not, which, in turn, correlates with ease of conversion to agriculture and the profitability of timber. Further, we include population density in 2000, in order to proxy for the degree of urbanization. To capture differences in travel cost, we distinguish between three road types: paved, unpaved, and large federal roads connecting airports. We also differentiate large urban areas (Merida, San Francisco de Campeche, Cozumel, Chetumal, Cancun, and Playa de Carmen) from smaller towns, as large tourist centers likely influence outcomes differently (e.g., by creating more incentives for forest conversion to urban and tourist areas) than smaller locales that are more dependent on local agricultural labor markets. The choice of covariates, determined by data availability, is consistent with previous studies. The covariates are calculated at the pixel level, which is our unit of analysis. The summary statistics are given in – Tables. ### Outcome We focus on the aggregate forest loss between 2000 and 2014 as our metric for conservation performance. The data come from a global dataset at 30 meter resolution. Several reasons underlie our choice of the Hansen data: First, they seem to be the dataset with finest resolution currently available for our study area. To the best of our knowledge, all other currently available datasets are at much coarser resolution (e.g., roughly 250-meters). Second, the methods behind Hansen dataset, including empirical uncertainty analysis and validation steps, are published and have been subjected to independent peer review, as part of the publication in high impact journal; the transparency sets the dataset apart from national datasets, for which methods and uncertainty assessments are often less accessible. Third, the Hansen dataset is currently one of the most widely used global dataset monitoring tree cover loss; its accessibility and widespread allows for comparisons across locations. Following Mexico’s national forest definition, we consider pixels as forested, if the percent tree cover in 2000 is greater than 10%. A pixel is counted towards forest loss if it was forested in 2000 and the forest was cleared between 2000 and 2014; pixels that were forested in 2000 and- remained forested in 2014 were considered no forest loss; pixels with no forest in 2000 are excluded from the analysis. The choice of cutoff to define tropical forest is still subject to debate in the literature. The concern is that a low cutoff may be capturing tree loss in areas that are not forests (e.g., a mixture of herbaceous and tree crops). To address this issue, we supplement the deforestation data with the spatially explicit woody biomass data for 2000 that are consistent with the Hansen dataset. Areas with trees that are not part of a forest have low values for woody biomass in 2000. Thus, by using partial linear regressions to examine the impacts of the interventions along a biomass gradient, we assess the sensitivity of our results to the choice of cutoff. We also repeated the analysis using 25% cutoff to define forest in 2000; the results are consistent with Tables – and available upon request. The ATT estimate gives us what would have happened to a treatment unit, had it not been placed under a treatment. In order to convert changes in the probability of deforestation to avoided deforestation and carbon dioxide emissions, we use back-of-the-envelope calculations based on the definition of the ATT estimate. Specifically, for each calculation we use the bias-adjusted ATT estimates and the mean probability of deforestation for the matched control group (Tables –) as well as its mean biomass levels (– Tables). That is, we multiply the ATT estimates by the mean deforestation rates within the matched control group and the area (in ha) of the matched control pixels. We then multiply the avoided deforestation area by the mean biomass levels corresponding to a forest and tenure types. Given the random sampling of the study area (roughly one percent sampled), we extrapolate the estimates to the whole study area by multiplying by 100. # Results ## Protected areas were effective in protecting broadleaf forests between 2000 and 2014 On average, protected areas reduced the probability of forest loss for dry and moist broadleaf forests by 4% and 7%, respectively. Using back-of-the-envelope calculations for the whole study area, these correspond to over 400ha and 4,100 ha of avoided forest loss and 20.9 MtC and 282.3 MtC of avoided carbon emissions over 14 years, respectively. However, the impact of protection varied by the amount of biomass in 2000. Protected areas were effective at protecting moist broadleaf forests except at very high biomass levels. For dry broadleaf forests, protection had a statistically significant impact only until intermediate biomass levels. These results suggest that across all tenure types formal protection may not be a deterrent for clearing high-biomass forests with high timber value, but can be effective when the forests may not have a very large commercial value. ## The protected area effectiveness between 2000 and 2014 varied by tenure regime In private properties with dry broadleaf forests and existing ejidos with moist broadleaf forests, formal protection reduced the probability of forest loss by 9% between 2000 and 2014. Using extrapolations of the whole study area, these correspond to about 231.7 ha (12.36 MtC of avoided emissions) and 3,148 ha of avoided forest loss (218.6 MtC of avoided emissions), respectively. In parceled ejidos with moist broadleaf formal protection reduced the probability of deforestation by 5%, which translates into about 32.9 ha and 1.66 MtC over the study period. Formal protected areas had no statistically significant average impacts elsewhere. Because of the small number of protected pixels under parceled ejidos in dry broadleaf forests and the very different characteristics of the protected and unprotected pixels for existing ejidos in dry broadleaf forests, we are omitting the results for the two specifications. The post-matching heterogeneity analysis suggests that the impact of formal protected areas varied with the amount of biomass: in dry broadleaf forests, the protection of private property was generally effective except at very low biomass values. In moist broadleaf forests, the impacts of protection was significant at low and intermediate values regardless of the tenure regime. However, at high biomass levels, only the formal protection of ejidos remained a significant deterrent for deforestation. The impact of protecting ejidos in moist broadleaf forests is statistically significant regardless of the biomass levels, but is distinguishable from the other tenure regimes only at very low and very high levels of biomass. Depending on the tenure regime, protection had different impacts on the unprotected forests in the same property. In parceled ejidos in moist and dry broadleaf forests as well as existing ejidos and private properties in dry broadleaf forests, formal protection resulted in deforestation being reduced in the unprotected parts of the same properties. Based on back-of-the-envelope calculations, these translate into 2,332.3 ha (135.9 MtC), 901.3 ha (39.9 MtC), 5703.5ha (297.2 MtC), and 253.8ha (12.22MtC) of avoided forest loss (avoided carbon emissions) for the moist forest parceled ejidos, dry forest parceled ejidos, dry forest existing ejidos, and dry forest private properties, respectively. However, in private properties with moist broadleaf forests, protection seems to have displaced the deforestation activities to the unprotected parts of the protected properties, resulting in the loss of 1139ha (59.73 MtC). When the spillover effects are added to the direct impacts, the protection of private properties across the two forest types contributed to 653.5 ha being deforested (35.1 MtC of additional carbon emissions) due to leakage. ## Existing ejidos were more responsive to protection than private properties and parceled ejidos Existing ejidos, parceled ejidos, and private properties are located in different areas and may experience different deforestation threats (– Tables). After controlling for the differences in location, the probability of dry broadleaf forest loss in unprotected ejidos was 0.13%, which is consistent with that for the observationally similar unprotected private properties; in most broadleaf forests, the probability of forest loss for unprotected ejidos exceeded that of observationally similar private properties, with the difference being statistically significant. However, protection reversed the results: dry broadleaf forests in ejidos had 3% lower probability of being lost than in observationally similar private properties; moist broadleaf forests in existing ejidos had the same probability of deforestation as observationally similar private properties. The results from the post-matching heterogeneity analysis indicate that protected ejidos benefit dry and moist broadleaf forests at intermediate biomass levels relative to observationally similar pixels in private properties (Figs &). Parceled ejidos seem to have a mixed impact on forest conservation during the study period. Our results comparing pixels under parceled ejidos to observationally similar pixels under existing ejidos, indicate that, on average there were no statistically significant differences between the two tenure types for both dry and moist broadleaf forests. In dry and moist broadleaf forests, unprotected ejidos had a lower probability of deforestation at medium to high levels of biomass relative to unprotected parceled ejidos (Figs &). Relative to parceled ejidos, protection in existing ejidos reduced the probability of dry broadleaf forest loss only at intermediate biomass levels and increased it in moist broadleaf forests. A potential explanation for the observed pattern is that the protected areas in our sample may allow multiple use and controlled timber harvesting, which may be practiced by the existing ejidos only. Unprotected parceled ejido pixels exhibited 2% lower and 4% higher probability of forest loss relative to observationally similar private property pixels in dry and moist broadleaf forests, respectively. The post-matching PLM analysis indicates that much of the value in dry broadleaf forests is driven by impacts at low and intermediate biomass levels ( Panel C); for moist broadleaf forests the effect did not vary with the biomass levels ( Panel C). In contrast, moist broadleaf pixels in protected parceled ejidos had on average 3% lower probability of deforestation relative to observationally similar private property pixels, with the effect most pronounced at intermediate biomass levels ( Panel C). The very small sample size does not allow us to estimate the impact of protected private property pixels relative to observationally similar protected parceled ejido pixels in dry broadleaf forests. The ambiguous impact of parceled ejidos may very likely be due to differences in the timing of ejido parcelization as well as potential livelihood strategies, with recently parceled ejidos deforesting more to clear land for other activities as they transition to private properties and/or different livelihoods. Our dataset indicating the boundaries of the parceled ejidos does not allow us to distinguish when the ejidos were divided. # Discussion Our study is the first to provide large-scale rigorous evidence of the impact of common and private forest tenure regimes and their interactions with protected areas in a biodiversity hotspot like the Yucatan peninsula. Our results indicate that protected areas in the Yucatan peninsula are generally an effective conservation strategy that can also contribute to climate change mitigation. Our estimates of conservation effectiveness of protected areas are smaller than those of previous studies. There are two possible explanations for the observed patterns. First, we use a different forest loss data source than Pfaff et al.; our cutoff to define forests is lower than the one used in Sims & Alix-Garcia, who also use a different sample unit (locality) instead of pixels, although robustness checks with higher cutoffs indicated similar patterns. Given the largely insignificant impacts of protection at low levels of biomass in our study area, a lower cutoff to define forests suggests that we are likely underestimating the impact of formal protected areas. Second, our analysis does not differentiate between the different types of protected areas. Most protected areas on the Yucatan peninsula are biosphere reserves (48,380km²), followed by mixed use (10,637km²) and strict protected areas (9,826.1km²). Of the private properties spanned by a protected area, 55% are under a mixed use, followed by biosphere reserves (31%) and strict protected areas (13%). Of the ejidos spanned by a protected area, 70% are under a biosphere reserve, 18% are under a mixed use, and 12% are under a strict protected area. While previous studies have suggested that the impact of protection might depend on the category of the protected area, we do not explicitly test for heterogeneity based on protected area category. The reasons are (1) the relatively small sample sizes in this study precluded us from explicitly testing this hypothesis and (2) digression from the scope of the article that aiming to establish the average impacts of protection interacted with tenure regimes and type of forest. Examining the interaction of the different types of protected areas and tenure regimes is an interesting venue for future research. We find that ejidos may be a successful tool in promoting forest conservation and climate change mitigation in the Yucatan. Our estimates regarding ejidos relative to protected areas are conservative, i.e., biased towards finding no impacts. The reason is that ejidos may have voluntary conservation areas and community forestry outside the formal protected areas; the delineation of such conservation zones is often not publicly available and not accounted for in the analysis. For this reason, if conservation zones that help maintain forest cover are more likely to exist outside formal protected areas (thereby effecting the control group), any additional protected area effect they create would diminish the effect of formal protected areas. That is, we would observe a smaller impact of the formal protected areas on average. Further, unprotected ejidos receiving payments to protect forests and provide ecosystem services as well as those receiving technical assistance for monitoring and enforcement from non- governmental organizations are also likely to have lower deforestation rates, although the effect may be short-lived. Similarly, the exclusion of the ejidos under Forest Sustainability Council (FSC) certification implies that the ejidos that are likely to perform best in terms of forest management have been excluded from the analysis and is, therefore, biasing the estimates of ejido effectiveness towards 0. However, as of now, these are only a small number. In contrast to previous studies that find that ejidos and protected areas have the same impact on forests or those that find that community forest tenure may reduce forest loss relative to private property in Mexico and elsewhere, we find that broadleaf forests in ejidos *in the absence of conservation interventions*, may have a higher probability of being cleared than those in private properties. There are two potential explanations for the observed patterns. First, we consider the impact of ejidos *in aggregate*. That is, our analysis does not distinguish between forestry and non-forestry ejidos as well as those receiving additional support and assistance (e.g., payments for ecosystem services, support for silvicultural practices etc). Because the non-forestry ejidos do not depend on the forest as a primary source of livelihoods, we expect that forest loss there may be greater there. Second, our data do not allow us to distinguish whether forestry is practiced in private properties, and if so, what species are harvested and how. Thus, if the latter harvest species different from mahogany, for example, the logging may not register as deforestation in the satellite data used to quantify the deforestation outcome. Because of data limitations, our analysis presents an average impact of ejidos on forests in the Yucatan. Other studies have shown that larger ejidos with greater numbers of ejidatarios are also associated with higher probabilities of forest loss, although these ejidos may also be effective in maintaining forest cover when permanent forest areas and silviculture are present. Furthermore, we do not account for the differences in the ejido size, area of remaining forest within ejidos, or tree species within the forest holdings, which impact the profitability of forests and, hence, the incentives to convert the land to other uses. Moreover, because large-scale data are not available, we do not account of the socio-economic characteristics of ejidos: factors such as low socio-economic heterogeneity within the ejido member group, lower numbers of ejidatarios, low poverty levels, high dependence on forests, longer history of forest use, history of cooperation, and indivisible well-delineated stationary resources have been associated with higher potential for collective action and better ability to enforce regulations and monitor forests.. Finally, domestic and international policies can also undermine the effectiveness of forestry ejidos. For example, recent policy changes such as an increase in agricultural subsidies and more liberal trade that allow for surges in timber imports from China–these policy interventions can undermine the profitability of Mexico’s domestic forestry enterprises and create incentives for the conversion of the land to other uses. The effects of these policies remain understudied. The protected areas in ejidos appeared to have a net positive effect on moist broadleaf forests. Furthermore, the existing ejidos appeared more responsive to protection than private properties. Previous studies suggest this may be due to protection increasing tenure security and contributing to greater political and economic equality in the ejido, thus, incentivizing and facilitating ejidos to protect forests. Because 60% of Mexico’s forests are currently managed by ejidos, it appears that strategic conservation efforts should target ejidos. In this context, three important venues for applied research emerge: (1) understanding how and why the forest ecosystems are changing in ejidos, (2) developing and testing mechanisms through which interventions in ejidos effect change on the ground, and (3) understanding the mechanisms to preserve ejidal forests as community managed and not as individual property. Similar to other studies relying on geospatial forest cover data, we do not account for changes in the composition of the forest ecosystems (such as the loss of mahogany), which may be changing in different ways due to differences in the forest management practices. Because ejidos are dynamic systems, it is important to identify the most strategic incentives to minimize environmental impacts and conserve forest ecosystems, while maintaining rural livelihoods and improving human well-being. Future research is needed to inform the development of policies to achieve desirable reductions in forest loss and promote the sustainable development of the region. Consistent with previous studies, we find that protected areas can be effective, with the impact varying based on location. These complex contingencies require well-designed conservation and development plans based on a thorough understanding of the land use decisions and socio-economic processes operating in different tenure regimes. Only then can decision makers adopt the combination of policies that are likely to be most effective in a given area. # Supporting information We thank staff of the Mexico REDD+ Alliance including Sebastien Proust, Yves Paiz, Rane Cortez, Maria Martinez Murillo Cuervo, Jose Canto Vergara, Juan Francisco Torres Origel, and Maria Elena Martínez Murillo Cuervo, for synthesis of publicly available geospatial datasets and feedback on study design and write-up. We thank Peter Veit, Laura Villalobos, and Allen Klaiber for helpful suggestions, and Rosa Goodman, and Trisha Gopalakrishna for help with GIS issues and Colette Naples for excellent research assistance. The authors are solely responsible for decisions about study design, data analysis, results, and conclusions. [^1]: The authors have declared that no competing interests exist.

# Introduction Human herpesvirus 6 (HHV-6) comprises two ubiquitous human beta-herpesviruses: HHV-6A and HHV-6B. HHV-6A and HHV-6B share about 95% nucleotide identity, but some regions have up to 25% divergence, which likely underlies the differences in tropism, drug susceptibilities and disease associations between these two viruses. These differences have led to their recent reclassification as separate viral species. Primary infection with HHV-6 occurs around the age of two, and as with other herpesviruses, latency persists for the life of the host. HHV-6B is the etiologic agent of roseola, a self-limiting febrile illness of early childhood. Primary infections with HHV-6 vary geographically; HHV-6B predominates in childhood infections in the US and Japan, while a recent report suggests that HHV-6A may predominate in asymptomatic childhood infections in Zambia. HHV-6A and HHV-6B have each been associated with diseases of the central nervous system (CNS) (reviewed) including multiple sclerosis (MS), , encephalitis and epilepsy (reviewed). These viruses are also known to reactivate following bone marrow, and solid organ transplant, causing complications that range from febrile illness to fatal encephalitides. The association of the HHV-6 viruses with these disorders has been partly established by the detection of elevated viral DNA levels in patients compared with healthy controls. Many studies report PCR amplification of HHV-6A and HHV-6B from blood, saliva, and urine, of healthy adults. However, the reported frequencies of detection are wildly discrepant. For example in PBMC, the reported detection of HHV-6 in healthy adult donors ranges from 9% to 90%, and in saliva from 9% to over 80%. These differences may be population-based, but they may also reflect the lack of a validated, standardized method for detecting HHV-6 viral DNA. Since in most healthy individuals the concentration of HHV-6 in the peripheral blood is low, its detection is dependent on PCR sensitivity and subject to high inter-replicate variance. Nested PCR has proven to be a valuable tool for detecting low frequency events; however, it is not quantitative and is highly prone to false positive signals. Though real time PCR is quantitative, its accuracy is limited by the need for an external calibrator. A new, non-nested digital PCR technology has recently emerged that enables absolute quantification and offers an extraordinarily high degree of precision. In digital PCR, DNA is distributed across multiple replicate reactions. These replicates enable the use of Poisson statistics for absolute quantification; the dynamic range increases proportionally with the number of replicates. The method employed in this paper, digital droplet PCR (ddPCR) uses water-in-oil droplets to increase the number of replicates up to 20,000 nanoliter-sized droplets that each support PCR amplification. As the droplet volume is known, the fraction of positive droplets is used to calculate the absolute concentration of the target. This system has recently been utilized to quantify bacteria and viruses including HIV, and CMV in various sample types. In light of increasing evidence regarding the biological differences between HHV-6A and HHV-6B, it is important that current research on this subject distinguish between the two. Therefore, we designed a multiplex ddPCR reaction for the simultaneous detection of HHV-6A and HHV-6B, with one primer set to amplify both viruses, and fluorescent probes that are specific for each. This SNP-like assay design enables comparable sensitivity and kinetics between amplification reactions, while differentiating HHV-6A and HHV-6B with high specificity and sensitivity. To our knowledge, this present study is the first report on the use of ddPCR for the detection of HHV-6, and the first report on its use for multiplexing two viruses. We report our findings on the quantitative detection of the HHV-6 viruses from human biological material including PBMC, serum and saliva. Notably, we demonstrate a high frequency of HHV-6A and HHV-6B coinfection in healthy controls, and this frequency is further elevated in the saliva of patients with MS, a disorder in which HHV-6 has been reported to play a role. # Materials and Methods ## Ethics Statement Prior to study inclusion, written informed consent was obtained from each subject (adult healthy volunteers and MS patients) in accordance with the Declaration of Helsinki. The study was reviewed and approved by the National Institutes of Health Institutional Review Board ## Clinical samples The 59 MS patients were composed of 32 males and 27 females, 31 with relapsing- remitting (RRMS), 18 with primary progressive (PPMS) and ten with secondary progressive (SPMS). Forty-seven of the MS patients had been untreated for at least 30 days prior to sample collection. The remaining 12 were on various disease-modifying treatments at the time of sample collection, including interferon-beta, glatiramer acetate and dimethyl fumarate. For the subset of the 110 unique healthy volunteers with known demographic information, there were 61 males and 42 females; 57 whites, 35 African Americans, six Asians, one Hispanic, and one Native American. Ten healthy donors donated both saliva and blood, sixteen healthy donors donated blood, and the remainder of healthy donors we analyzed either PBMC or serum. ## Primers and probes The NCBI reference genomes used to design the primer and probe sequences are NC_001664 for HHV-6A, strain U1102, and NC_000898 for HHV-6B, strain Z29. Ribonuclease P protein subunit P30 (*RPP30*) was used as a cellular housekeeping gene (Gene ID: 10556). The *RPP30* primers amplify a 62 base pair region, and the HHV-6 primers amplify an 89 base pair region *U57*, which encodes the major capsid protein. There is approximately 95–97% nucleotide homology between HHV-6A *U57* (Gene ID: 1487939) and HHV-6B *U57* (Gene ID: 1497059). The primers amplify both HHV-6A and HHV-6B, while the differentially labeled fluorescent probes are specific for HHV-6A or HHV-6B. The HHV-6A probe is FAM- MGBNFQ-labeled, while the HHV-6B probe is VIC-MGBNFQ-labeled. The housekeeping gene, RPP30, is VIC-MGBNFQ labeled. For each DNA sample, the HHV-6A and HHV-6B primers and probes were duplexed, while the *RPP30* primers and probe were separately singleplexed. When the primers and probes were duplexed (HHV-6A and HHV-6B) or singleplexed (*RPP30*), the final concentrations in each reaction were 900 nM per primer and 250 nM per probe. When the reaction was triplexed, the final concentrations of the HHV-6A and HHV-6B primers and probes remained as above, while the concentrations of *RPP30* were 450nM per primer and 125 nM for the probe. ## Sample collection/isolation To isolate PBMC, whole blood was diluted with PBS, layered over a Ficoll gradient (Lympocyte Separation Medium, Lonza, Walkersville, MD) and spun at 12000 rpm for 30 minutes at room temperature (RT). The cells were washed, cryopreserved and stored in liquid nitrogen until DNA extraction. To isolate serum, whole blood was collected in a red top tube and spun at 2000 rpm for 10 minutes at RT; the serum was aliquoted and stored at –80°C until DNA extraction. Salivettes (Sarstedt, Newton NC) were used for saliva collection. Patients and healthy volunteers chewed on a cotton swab for two minutes, which was then placed in a capped plastic container and spun at 1100 rpm for 5 minutes. The saliva was stored at –20°C until DNA extraction. ## DNA extraction DNA was extracted from saliva (200 μl) and PBMC or virus-infected cells (3×10⁶) using a DNeasy Blood and Tissue Kit (Qiagen, Valencia, CA) according to the manufacturer’s instructions. Buffer AVL was used for saliva extraction, while Buffer AL was used for PBMC extraction. DNA was extracted from serum (1 ml) using the QiaAmp Ultrasensitive Virus Kit (Qiagen, Valencia, CA) according to the manufacturer’s instructions. Elution volumes of purified DNA in Buffer AE were 150 μl for serum and saliva and 200 μl for PBMC. ## Viral Infections HHV-6A (U1102) and HHV-6B (Z29) were propagated in the T-lymphoblastoid cell lines HSB-2 and SupT1, respectively, as previously described. ## Digital Droplet PCR We used a modified version of the digital droplet PCR workflow described by Hindson, *et al.*. Briefly, for each replicate, 5 μl DNA was digested with the restriction enzyme HindIII in NEB Buffer 2 (New England Biolabs, Ipswich, MA) for 30 minutes at 37°C, and then diluted 1:5 with molecular biology grade water. Dilution reduces salts in the digestion buffer that may interfere with PCR amplification. A 20 μl mixture of primers and probes, Bio-Rad 2X Supermix and the digested diluted DNA was emulsified with droplet generator oil (Bio-Rad, Hercules, CA) using a QX-100 droplet generator according to the manufacturer’s instructions. The droplets were then transferred to a 96 well reaction plate (Eppendorf, Hauppauge, NY) and heat-sealed with pierceable sealing foil sheets (Thermo Fisher Scientific, West Palm Beach, FL). PCR amplification was performed in this sealed 96 well plate using a GeneAmp 9700 thermocycler (Applied Biosystems, Grand Island, NY) with the following cycling parameters: 10 minutes at 95°C, 40 cycles consisting of a 30 second denaturation at 94°C and a 60 second extension at 59°C, followed by 10 minutes at 98°C and a hold at 12°C. Immediately following PCR amplification, droplets were analyzed using a QX100 droplet reader (Bio-Rad, Hercules, CA), in which droplets from each well are aspirated, streamed toward a detector and aligned for single-file two-color detection. ## DdPCR data analysis Fluorescence data for each well were analyzed using QuantaSoft software, version 1.3.2.0 (Bio-Rad, Hercules, CA). Thresholds were determined manually for each experiment, according to the negative controls, which included a no template control and a negative sample. Droplet positivity was determined by fluorescence intensity; only droplets above a minimum amplitude threshold were counted as positive. Negative control DNA and a no template control were included in each experiment and resulted in zero positive droplets. For a given sample, target copies per μl were calculated by averaging over all replicate wells. Cellular DNA input was calculated by halving the number of *RPP30* copies, as there are two copies of *RPP30* per diploid cell. PBMC data are represented as viral copies per 10⁶ cells, while serum and saliva data are represented as viral copies per ml. # Results ## Characterization of ddPCR for the detection of HHV-6A and HHV-6B This assay was designed to multiplex HHV-6A and HHV-6B, such that coinfection within a given sample could be clearly visualized. The primers amplify the same region of *U57* from both viruses, while the fluorescently labeled probes distinguish between the two viruses by a three base pair mismatch. This design minimizes differences in primer sensitivity and amplification kinetics that can lead to differences in the dynamic range of viral DNA quantitation. To determine the optimal annealing temperature of these primers and probes, a temperature gradient experiment was performed using positive control DNA from HHV-6A and HHV-6B-infected cells. The optimal annealing temperature was determined to be 59°C (data not shown), which was used for all subsequent experiments. Primer/probe sequences are listed in. shows several examples of one-dimensional (1D) ddPCR plots. The x-axis corresponds to the number of analyzed droplets (event number) and the y-axis corresponds to the fluorescence amplitude of the droplets. The purple line (amplitude 2000) is the manually set threshold, and droplets below this threshold are negative; that is, they do not contain amplifiable DNA. The maximum fluorescence amplitude of the probes differs for each fluorophore (FAM and VIC), and is further influenced by thermocycling conditions and the avidity of each probe for its target sequence. The three base pair mismatch between the two probes confer a high level of specificity to each. depicts the results when HHV-6A DNA (HSB-2-infected cells, strain U1102) is tested with each probe. There is a positive droplet population (blue) in the left plot with the HHV-6A FAM probe, and the absence of positive droplets in the right plot with the HHV-6B VIC probe, indicating the specificity of the HHV-6A FAM probe for HHV-6A viral DNA. Likewise, when HHV-6B DNA (SupT1-infected cells, strain Z29) is tested with each probe, there is only a positive droplet population (green) in the right plot, with the HHV-6B VIC probe, indicating the specificity of the HHV-6B VIC probe for HHV-6B viral DNA. ## HHV-6A and HHV-6B coinfection detected in the peripheral blood and serum of healthy donors To extend previously published observations on the frequency of detection and quantitative range of viral DNA in healthy blood donors, 46 PBMC samples were analyzed by an HHV-6 multiplex ddPCR assay. display representative two- dimensional (2D) plots, with HHV-6A FAM on the y-axis and HHV-6B VIC on the x-axis, and the corresponding 1D housekeeping gene (*RPP30* VIC) plots as insets. These data are representative of three outcomes: no HHV-6 positivity, HHV-6 positivity with one virus, or HHV-6 positivity with both viruses. The presence of both HHV-6A and HHV-6B positive droplets is defined as coinfection. The 1D *RPP30* inset for all three donors displays a strongly positive droplet population, indicating the abundance of cellular material as expected from PBMC DNA. The *RPP30* concentration was used to calculate the number of viral copies per cell for each sample. is a group analysis of the total HHV-6 viral copies per 10⁶ cells. Open circles represent the detection of only HHV-6B, while closed circles represent the detection of both viruses. The frequency of HHV-6 DNA detection in 46 healthy PBMC samples was 57% (26/46) with a range of 0-3,000 total HHV-6 copies per 10⁶ cells, consistent with other studies. Of these 26 HHV-6 positive samples, HHV-6A and HHV-6B coinfection was detected in 13 (50%). The amounts of HHV-6A and HHV-6B within each coinfected sample are shown in. The 6A/6B ratio of copies/10⁶ PBMC ranged from 0.01–0.25, demonstrating that all healthy donor PBMC samples contained more HHV-6B compared to HHV-6A. Additionally, serum samples from 43 healthy blood donors were studied, and the results are displayed in. are representative 2D ddPCR plots, with the corresponding 1D *RPP30* plots as insets. The *RPP30* inset for all three donors displays a variably positive droplet population, with a reduced concentration compared to the PBMC DNA shown in. Though *RPP30* was detected in most serum samples to varying extents, the data in are represented as total viral copies per ml. The frequency of viral DNA detection from 43 healthy serum samples was 30% (13/43) with a range of 0–42,000 copies/ml. These results are similar to a previous study that employed a highly sensitive PCR technique. Of these 13 HHV-6 positive samples, HHV-6A and HHV-6B coinfection was detected in 8 (62%). The amounts of HHV-6A and HHV-6B within each coinfected sample are shown in. The 6A/6B ratio of copies/ml serum ranged from 0.1 to 0.66, demonstrating that all healthy donor serum samples contained more HHV-6B compared to HHV-6A. We did not observe correlations between HHV-6A and HHV-6B coinfection and gender or ethnicity (data not shown). These PBMC and serum samples were from the same healthy donor in 26 cases- the results of 16 are shown in, and the results of ten (who also donated saliva) are shown in. Of the 16 shown in, 11 were positive in at least one compartment, and nine were positive in both. To examine the possibility of HHV-6 detection in the serum reflecting viral DNA released from circulating blood cells, we analyzed the ratio of viral copies in serum to viral copies in PBMC (these data are calculated before normalization to copies/cell or copies/ml). As shown in, eight of the nine samples cluster around a ratio of approximately one, suggesting approximately equal amounts of virus in the PBMC and serum. One normal donor (25611 circled in red) has a serum/PBMC ratio of approximately 11, demonstrating an abundance of viral DNA in the serum compared to the PBMC. In, the viral copies per ml of serum are plotted against the copies of cellular housekeeping gene, demonstrating that in the majority of samples, serum viral copies do not correlate with cellular housekeeping gene levels. These data support that viral DNA detected in the serum may not be cell associated (that is, independent of PBMC-associated virus). ## Identification of blood donors with probable chromosomally integrated HHV-6A or HHV-6B and applicability of ddPCR for screening In the assessment of healthy donor blood samples, we identified two individuals with unusually high levels of HHV-6 in their PBMC. displays PBMC DNA from donor 27867. Normalizing to the housekeeping gene *RPP30*, this individual was calculated to have approximately one copy of HHV-6A per cell (0.98 copies/cell). is a ddPCR plot of PBMC DNA from another donor, 28319. Normalizing to *RPP30*, this individual was calculated to have approximately one copy of HHV-6B per cell (1.02 copies/cell). This high concentration of HHV-6 suggests that these two individuals display a phenomenon known as chromosomally integrated (CI) HHV-6, which has been described in approximately 1% of the general population (reviewed in). The identification of these two individuals with probable CI HHV-6 led us to design a triplex reaction for HHV-6A, HHV-6B and *RPP30*. Triplexing enables a reduction in DNA input and increases system throughput, which are important for clinical settings in which CI HHV-6 may be significant. In this triplex condition, the concentrations of the *RPP30* VIC primers and probe were halved, allowing for the two VIC-labeled probes (HHV-6B and *RPP30*) to be distinguished by their fluorescence amplitudes. The calculated number of viral copies per cell was comparable whether the samples were duplexed or triplexed. Donor 27867 was calculated to have 1.02 HHV-6A copies per cell when duplexed and 1.09 copies HHV-6A per cell when triplexed. Donor 28319 was calculated to have 0.98 copies HHV-6B per cell when duplexed and 1.04 copies HHV-6B per cell when triplexed. ## Elevated levels of HHV-6A and HHV-6B coinfection in the saliva of MS patients compared to healthy controls HHV-6 viral DNA was also investigated in the saliva of healthy donors, as salivary glands are an important reservoir for HHV-6. are representative 2D ddPCR plots of three MS patient saliva samples. The *RPP30* insets for these three donors display variably positive green droplet populations compared to the PBMC DNA, but greater than what is observed for serum DNA. As with HHV-6 viral DNA in serum, the saliva data are expressed as viral copies/ml because similar to Cone *et al.*, in our study, salivary cellular DNA levels did not correlate with viral DNA levels (data not shown). Collectively, these results demonstrate HHV-6 in the saliva of 67% (26/39) healthy donors. Of these HHV-6 positive saliva samples, only 12% (3/26) were found to be coinfected with HHV-6A and HHV- 6B (summarized in Table 3), lower than what was observed in the PBMC or serum. We did not observe correlations between coinfection and gender or ethnicity (data not shown). This high frequency of detection, and range of viral copies/ml saliva are both consistent with other published reports. Of the HHV-6 viruses, HHV-6A has been suggested to be more neurotropic than HHV-6B, and is associated with MS. Therefore, we asked if the level of coinfection (which in this study is defined as the presence of HHV-6A in addition to HHV-6B) is increased among MS patients compared to healthy donors. The rates of viral detection in saliva were similar between MS patients (63%) and healthy donors (67%), with a mean total viral load of 1.82×10⁴ copies/ml for MS patients and 1.38×10⁴ copies/ml for healthy donors (summarized in Table 3). However, there was a significantly elevated prevalence of coinfection in MS patient saliva (30% (11/37)) compared to the healthy controls (12% (3/26)) (Χ² test, p = 0.04), suggesting that the difference between MS patients and controls may be the increased presence of HHV-6A in addition to HHV-6B. The amounts of HHV-6A and HHV-6B within each coinfected sample are shown in. The 6A/6B ratio of copies/ml saliva ranged from 0.005-0.6 in healthy donors and from 0.04 to 4.0 in MS patients, demonstrating that in healthy donors, all coinfected samples contained more HHV-6B compared to HHV-6A, while in MS patients, a subset of the coinfected samples contained more HHV-6A than HHV-6B (2/11). In this particular MS cohort, there was no correlation between coinfection and expanded disability status scale (EDSS) score, brain lesion load or contrast enhancing lesions around the time of saliva collection (data not shown). # Discussion The HHV-6 viruses are detected at low levels in the majority of healthy individuals. This is supported by the widespread—but controversial—use of nested PCR to establish the frequency of HHV-6 in the blood of healthy donors, as well as data demonstrating that viral detection in blood increases with increasing cellular input. Such low levels of virus render the detection of viral DNA highly subject to both the sample integrity and assay sensitivity, and may underlie discrepancies in the literature regarding the frequency of HHV-6 detection among healthy individuals. In this paper, we report a novel HHV-6 detection method using a third generation quantitative PCR technology, ddPCR. DdPCR is ideal for detecting low frequency events, such as HHV-6 viral DNA in healthy donor samples, because it partitions a sample into discrete droplets, thereby minimizing random sampling error. Moreover, increasing the number of replicates proportionally increases the quantitative dynamic range; this ability to extend the lower limits of detection is ideal for assessing low-level targets, and important for detecting two viral species that may be present in different quantities. We detected HHV-6 viral DNA in the PBMC of 57% (26/46) of healthy donors, consistent with a number of previous studies. When considering prevalence data in a cross-sectional study such as this one, it is important to note that detectable viral DNA levels within an individual may fluctuate over time. This was demonstrated in a longitudinal study of healthy children, which reported that over several years, viral DNA was detected intermittently in the PBMC of 76% children, but consistently detected in the PBMC of only 12% of children. In our study of healthy donor PBMC, we fortuitously identified two individuals with probable CI HHV-6, one with HHV-6A and the other with HHV-6B, though analysis by fluorescent *in situ* hybridization (FISH) is necessary to confirm chromosomal integration. The precise mechanisms and clinical implications of chromosomally integrated HHV-6A and HHV-6B are active areas of study. From studies of healthy blood donors in the US and UK, the incidence of HHV-6 chromosomal integration in the general population is thought to be approximately 1%. We believe that the percentage of potentially chromosomally integrated individuals as calculated from this study (2/48, or 4%) reflects low numbers of surveyed PBMC. In the healthy blood donor with probable integrated HHV-6A (1.02×10⁶ copies HHV-6A per 10⁶ cells) we also detected HHV-6B viral DNA (306 copies HHV-6B per 10⁶ cells). These data are consistent with a recent report of HHV-6B detection in two individuals determined to be CI HHV-6A. Notably, we detected these HHV-6B positive droplets both when the sample was duplexed (306 copies HHV-6B/10⁶ cells) and triplexed (401 copies HHV-6B/10⁶ cells). These data highlight the suitability of ddPCR for assessing coinfection, as low copy numbers of one viral species can be detected and reproducibly quantified in the presence of excess copies of the other viral species. The detection of viral nucleic acid in blood may reflect active viral replication, chronic persistence or latency. We detected HHV-6 DNA in 30% (13/43) serum samples from healthy adult donors. These results are higher than many previous studies, – that employed conventional PCR approaches and reported undetectable levels of HHV-6 DNA in healthy donor serum. However, a frequency of 30% agrees with several other studies, including one that used a highly sensitive PCR ELISA assay and reported HHV-6 DNA in the serum of 21% (5/24) healthy Canadian donors and another that reported 20% detection among healthy Jordanian controls. It is currently thought that HHV-6 DNA detected in the serum indicates an active infection, but recent work cautions that viral DNA in the serum may in fact reflect DNA released from circulating blood cells. While cellular material in serum can indicate the presence of genomic DNA contaminants, we found no correlation between viral copies/ml serum and the cellular housekeeping gene (a subset shown), suggesting that viral DNA we detected in serum may not be cell-associated. An active infection can also be hypothesized based on the ratio of virus in the serum to virus in circulating peripheral blood cells. For example, most individuals in our study had a ratio of approximately one, but one individual had a ratio of 11. This 11-fold increase of virus in the serum compared to PBMC supports further studies on this individual to determine whether there is active or infectious virus in this compartment. A major finding of this report is the high frequency of HHV-6A and HHV-6B coinfection in healthy adult donors. There are few published studies on coinfection of the HHV-6 viruses, likely because many early reports did not distinguish between HHV-6A and HHV-6B, though the literature is shifting towards this distinction. Even in studies that report coinfection, there may be an underestimation of the frequency. For example, in one study of 44 adult PBMC samples, coinfection was detected in 18% of the HHV-6 positive samples. This is comparatively lower than our data, as we detected coinfection in 50% (13/26) of HHV-6 positive PBMC samples. We also report coinfection in 62% (8/13) of HHV-6 positive serum samples, and are not aware of other studies that observed this in healthy donor serum. This high percentage likely reflects the low denominator, and should be interpreted with caution. In addition to PBMC and serum, we also analyzed the HHV-6A and HHV-6B viral loads in saliva. The salivary glands are a known reservoir for HHV-6 and saliva is an easily collected clinical sample. Among healthy donors, we observed the lowest incidence of HHV-6A and HHV-6B coinfection in this compartment (12%, relative to 50% in PBMC and 62% in serum). However, this observation may be population-specific, as a recent study of HHV-6 in the saliva of healthy Brazilian donors reported twice the frequency of HHV-6A compared to HHV-6B. As we observed higher viral loads in the saliva compared to the blood of healthy donors, we chose this sample type to compare HHV-6 levels between healthy donors and patients with MS, a neurologic disease associated with HHV-6 (reviewed). Despite comparable detection frequencies and total viral loads between healthy donor and MS saliva, we detected significantly increased HHV-6A and HHV-6B coinfection in the MS samples. These data extend observations from a previous study in which we detected coinfection in a subset of HHV-6 positive saliva samples from MS patients (6%), but not healthy controls. The observation of increased detection of HHV-6A in MS patients is also consistent with other studies suggesting that this viral species is more prevalent among MS patients compared to healthy donors. In this report, coinfection is defined by the presence of HHV-6A in addition to HHV-6B, as singly infected samples were determined to be HHV-6B. This may reflect that HHV-6B is the predominant species of childhood infection among our donor population. The implications of coinfection should be considered in light of clinical reports suggesting that the abundance of detectable HHV-6A may be pathophysiologic. For example, in a study of renal transplant patients, the frequency of HHV-6A was significantly higher in patients with active infections, while the frequency of HHV-6B was significantly higher in patients and healthy donors with latent infections. In a separate study of critically ill patients, 54 of 101 had detectable HHV-6 viral DNA, and 53 of the 54 were positive for HHV-6A. The authors suggest that the selective reactivation of HHV-6A reflects the dominance of this viral species under illness-related stress. As HHV-6A may be more active than HHV-6B, and studies demonstrate different susceptibilities of each virus to therapeutics, quantitating the extent of coinfection may prove clinically relevant. It may be insufficient to assess only the presence or absence of each species in a given sample, as we determined great variability in the viral copy ratio of HHV-6A to HHV-6B among our coinfected healthy and MS saliva samples (0.005 to 4.0). It remains to be seen in a larger study whether there are clinical correlates to this ratio. In summary, we aimed to address the frequency of HHV-6A and HHV-6B coinfection by designing a multiplex digital droplet PCR assay to comparably amplify both viruses. We investigated blood and saliva samples from healthy adult donors and adult MS patients and demonstrate a heretofore-underappreciated high frequency of coinfection of these two viruses. Larger studies are required to learn to what extent the variability of frequency positive and coinfection are related to the site (PBMC, serum or saliva) versus the population, though our cohort in (ten donors from which we collected PBMC, serum and saliva) suggests that viral positivity is related to the site as opposed to the population. Moreover, larger studies are required to determine if there are clinical correlates of HHV-6A and HHV-6B coinfection (for example, markers of inflammation) and if the frequency of either virus is affected by treatment. As we understand more about each virus in health and disease, this observation of coinfection may have profound implications. We thank Sandy Chiang, John Chuckalovcak, and Dr. Bo Song from Bio-Rad Laboratories for technical support, Dr. Kory Johnson of the NINDS Bioinformatics Section for helpful discussions, and the Neuroimmunology Branch clinical staff for collecting patient specimens. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: ECL SJ. Performed the experiments: ECL GSB BC. Analyzed the data: ECL SJ. Contributed reagents/materials/analysis tools: KF JO DSR. Wrote the paper: ECL SJ.

# Introduction Chloroplasts have to respond quickly to changes in light quality and quantity in order to balance the excitation pressure between the two photosystems. Important aspects of this fast responding system are controlled by posttranslational modifications of resident proteins, e.g. phosphorylation (reviewed in –). Several chloroplast protein kinases are known that catalyze light-triggered acclimation responses, but their complete set of targets and their cross talk in a phosphorylation network are only partially understood. In Arabidopsis, the STN7 and STN8 kinases catalyze the phosphorylation of thylakoid membrane proteins. In addition to its role in controlling state-transitions, STN7 is also involved in the regulation of long-term acclimation processes that entail changes in gene expression of both nuclear and chloroplast encoded genes. Although the STN7-triggered long-term response has been characterized in detail, the exact mechanisms of signal transduction and the mediators involved are still largely unknown. Comparative phosphoproteomics with *stn7* mutants provided no indication for the STN7-dependent phosphorylation of proteins with a function in gene expression. It therefore appears likely that a direct control of plastid gene expression is exerted by other kinases such as the chloroplast sensor kinase CSK or casein kinase II. Plastid casein kinase II (pCKII, formerly also termed CKA4) activity was originally identified in the 1990's when phosphorylation of an RNA binding protein and the plastid RNA polymerase were reported. The classification of this phosphorylation activity into the GMGC-group of kinases was based on biochemical properties such as the phosphorylation of serine residues, the ability to utilize GTP as phosphate donor and its specific inhibition by heparin. pCKII activity is associated with the plastid RNA polymerase suggesting a function in the regulation of gene expression by phosphorylation of RNA polymerase subunits and sigma factors. Indeed, sigma factor phosphorylation influences the efficiency of RNA polymerase promoter binding, supporting a function of pCKII in transcriptional regulation. Henceforth, pCKII was named plastid transcription kinase (PTK). More recently, the gene for pCKII was identified. The gene model AT2g23070 encodes for a CKII alpha subunit that harbors the catalytic activity, possesses a canonical kinase domain and a functional plastid transit peptide. Recent large-scale phosphoproteomics screens identified many proteins that are phosphorylated at characteristic acidic CKII motifs and a statistical evaluation of motif utilization suggested pCKII as the major phosphorylation activity in chloroplasts. However, despite its many putative substrates, pCKII is of very low abundance in photosynthetic chloroplasts, while it accumulates to higher levels in non-photosynthetic organs such as roots. The identification of pCKII substrates is an important first step to understand the function of this kinase. Several distinct methods are available and were already used for the identification of kinase targets. For example, *in vivo* targets for the STN7 and STN8 kinases were identified by quantitative comparative phosphoproteomics analyses between kinase mutant and wildtype. The *in vitro* identification of kinase targets is based on the analysis of phosphorylation activity of a purified kinase or an extract on peptides or proteins. When performed at large-scale, such assays are mostly performed on protein or peptide arrays. Protein arrays are similar to a standard kinase assay in which γ-^32/33P-ATP is used as a substrate to monitor the phosphate transfer from ATP onto a protein *in vitro* with several protein substrates in parallel. Peptide Arrays are similar to the above assays but instead of proteins, synthetic peptides with a length of around 12–18 amino acids are used as kinase substrates. Such peptides are offered to the kinase either in solution as in the “Kinase Client (KIC)-assay” in which phosphorylated peptides are identified and quantified by mass spectrometry, or immobilized on a glass slide. In both instances, *in vitro* phosphorylation reactions are performed with a purified kinase or a kinase-enriched protein extract. Here we report the creation of the first chloroplast phosphopeptide microarray ChloroPhos1.0. The peptide microarray was specifically designed for the analysis of chloroplast protein kinases, using plastid phosphoproteomics data for peptide selection and phosphorylation site positioning (<http://phosphat.uni- hohenheim.de/>). By selecting peptides that were previously identified in phosphoproteomics experiments, we ascertain that all peptides on the microarrays are true *in vivo* kinase targets. We used the microarray to identify the substrates of pCKII. Our analysis shows that ChloroPhos1.0 is a suitable screening tool to establish currently unknown kinase/substrate relationships in the chloroplast phosphoproteome network. It allows extracting preferred motifs of chloroplast kinases thus facilitating the prediction of additional targets that may have escaped mass spectrometric detection. # Materials and Methods ## Preparation of the peptide microarray ChloroPhos1.0 We extracted plastid phosphoproteins and phosphopeptides from previously published phosphoproteomics data (Table S1) available in PhosphAT 3.0 (<http://phosphat.uni-hohenheim.de/>) using a chloroplast protein reference table (for details see). Peptides were synthesized with a linker at their N-terminus (N-(3-(2-(2-(3- amino-propoxy)-ethoxy)-ethoxy)-propyl)-succinamic acid). Peptide derivatives were synthesized via glycine ester linkage on cellulose membranes in a parallel manner using SPOT synthesis technology. After each coupling, remaining amino functions were acetylated using acetic anhydride in dimethylformamide in the presence of diisopropylethylamine to prevent deletion sequences. Subsequent to removal of the last Fmoc-protecting group with 20% piperidine in dimethylformamide, anthranilic acid was attached to the N-terminus. This coupling reaction transformed desired full-length peptides only, but not acetylated side products resulting from incomplete couplings during peptide synthesis, into 2-aminobenzoyl-derivatives. Following trifluoroacetic acid-mediated side chain deprotection the cellulose bound peptide esters were transferred into 96 well microtiter filtration plates (Millipore, Bedford, Massachusetts, USA) and treated with 200 µL of aqueous triethylamine (2.5% by vol) in order to cleave the peptides from the cellulose. Peptide-containing triethylamine solution was filtered off, and quality controlled by LC-MS, and solvent was removed by evaporation under reduced pressure. Resulting peptide derivatives (50 nmol) were re-dissolved in 25 µL of printing solution (70% DMSO, 25% 0.2 M sodium acetate pH 4.5, 5% glycerol (v/v).) and transferred into 384-well microtiterplates. Peptide derivatives were deposited onto epoxy-functionalized glass slides (PolyAn GmbH, Berlin, 3D-Epoxy, 25×75×1 mm) using the contact printer OmniGrid 300 equipped with 16 SMP2 pins (Telechem/ArrayIt). Our buffer conditions resulted in selective covalent bond formation between epoxy-functions and amino groups of 2-amino-benzoyl- derivatives. Printed peptide microarrays were kept at room temperature for 5 hours. Each microarray with dried deposited spots was analysed using an Axon 4000B microarray scanner. Microarrays which passed this quality control were quenched for 1 hour with sodium citrate buffered 1% BSA solution at 42°C, washed extensively with water followed by ethanol, resulting in ‘purified peptide’ spots, essentially free of deletion sequences (due to acetylation steps during synthesis) and truncated sequences (due to chemoselective immobilization). Printed microarrays were dried using a microarray centrifuge and stored at 4°C. ## Cloning and expression of MBP-tagged pCKII For bacterial overexpression of *Arabidopsis* cpCK2, pMALc5× (Amp⁺) expression vector was transformed into competent BL 21 DE3 *E.coli* cells. The advantage of this system is the N-terminal MBP (maltose binding protein)-tag on the protein, which enables fast purification with commercially available kits. The cloning of CK2 without transit peptide was done with the following primers NotI_FW, 5′ GTA TAG CGG CCG CGC TTC TCT TTA CCG TCA AC 3′ and BamHI_Rv, 5′ GTA TAG GAT CCT CAC TGG CTG CGC GGC G 3′. At an optical density of 0.5–0.8 at 595 nm, the protein expression was induced by the addition of 1 mM IPTG. The cells were shaken for 3 h at 250 rpm at room temperature and collected. Cell lysis and purification was done according to the NEB pMALc5× purification protocol. After the elution of the purified protein, some contaminating proteins were visible, which we deleted in the second purification step with XK 16/40 Sephacryl S100. A ∼85 kDa band (recombinant pCKII with MBP-tag) was obtained. The fractions were collected and concentrated. Kinase activity was tested with \[γ-³³P\] ATP as described below. ## Heparin Sepharose chromatography with *Sinapis alba* and *Arabidopsis thaliana* chloroplast extracts *Sinapis alba* seedlings were grown on soil under steady light conditions in a controlled environment chamber (24 h light, 150 µE·m⁻²·s⁻¹). *Arabidopsis thaliana* Col 0 seedlings were grown on soil under long-day conditions in a controlled environment chamber (16 h light/8 h dark, 150 µE·m−2·s−1). Plants were harvested after 4 days *(S. alba)* or after 17 days *(A. thaliana)*. Chloroplasts were isolated according to established protocols. The isolated chloroplasts were solubilized on ice in a hypotonic buffer (50 mM TRIS-HCl pH 7.6, 4 mM EDTA, 20 mM DTT, 25% glycerol, 0.1% protease inhibitor cocktail (Sigma P9599), 1% Triton ×100). The lysate was adjusted to 200 mM (NH₄)₂SO₄ and applied on a 2-ml Heparin-Sepharose CL-6B column (GE healthcare) equilibrated in buffer E (50 mM TRIS-HCl pH 7.6, 0.1 mM EDTA, 5 mM DTT, 10% glycerol, 0.1% protease inhibitor cocktail (Sigma P9599), 0.1% Triton ×100 and 200 mM (NH₄)₂SO₄). After 2 washing steps with equilibration buffer, the bound proteins were eluted in a single step with 1 M (NH₄)₂SO₄ at a flow rate of 0.4 ml/min. Dialysis was performed overnight at 4°C in buffer E with 20% glycerol and without (NH₄)₂SO₄. ## Kinase activity assays All fractions were tested for kinase activity as described in with minor modifications. 10 µg proteins were incubated with 5 µg dephosphorylated and heat inactivated stromal proteins as a complex substrate mixture at 30°C for 30 min in kinase activity buffer A (50 mM Tris-HCl pH 7.5, 50 mM NaCl, 10 mM MgCl₂, 1× PhosStop (Roche) and 0.1% protease inhibitor cocktail (Sigma P9599), 5 µM ATP, 60 nM \[γ-³³P\] ATP). In case of recombinant pCKII we used 0.4 µg enzyme for the activity assays. Several controls were performed, in which the indicated ingredient was added or omitted from the reactions. These included omission of Mg²⁺, addition of EDTA, addition of non-radioactive GTP and addition of the CKII inhibitor heparin at a concentration of 15 µg/ml. Following incubation, the reactions were stopped by the addition of SDS sample buffer and the proteins therein were separated by SDS-PAGE and exposed overnight on an autoradiography screen. ## MS analysis of the Heparin Sepharose eluates Eluted proteins were precipitated in 90% acetone at −80°C and sedimented at 22.000xg for 15 minutes. The protein pellet was dissolved in 25 mM (NH₄) HCO₃ and 0.1% RapiGest (Waters) to a final protein concentration of 1.6 µg/µl. In the presence of 10 mM DTT the protein mixture was reduced for 10 min at 60°C and alkylated with 30 mM iodacetamide for 30 min at room temperature in the dark. The tryptic digest was performed overnight at 37°C with a trypsin/protein ratio of 1/100. The digest reaction was acidified by HCl (pH\<2) in order to stop tryptic activity and centrifuged for 10 min at 13000 g. The supernatant was analyzed by LC-MS using an ACQUITY UPLC System coupled to a Synapt G2-S mass spectrometer (Waters, Eschborn, Germany). ## Nano-LC separation, HD-MS^E data acquisition and protein identification/quantification LC separation (140 min gradient) and HD-MS^E data acquisition was performed using 1 µl of the digested sample on a ACQUITY UPLC System coupled to a Synapt G2-S mass spectrometer (Waters, Eschborn, Germany). MS acquisition was set to 50–2000 Da. Data analysis was carried out by ProteinLynx Global Server (PLGS 3.0.1, Apex3D algorithm v. 2.128.5.0, 64 bit, Waters, Eschborn, Germany) with automated determination of chromatographic peak width as well as MS TOF resolution. Lock mass value for charge state 2 was defined as 785.8426 Da/e and the lock mass window was set to 0.25 Da. Low/high energy threshold was set to 180/15 counts, respectively. Elution start time was 5 min, intensity threshold was set to 750 counts. Databank search query (PLGS workflow) was carried out as follows: Peptide and fragment tolerances was set to automatic, two fragment ion matches per peptide, five fragment ions for protein identification, and two peptides per protein. Maximum protein mass was set to 250 kDa. Primary digest reagent was trypsin with one missed cleavage allowed. According to the digestion protocol fixed (carbamidomethyl on Cys) as well as variable (oxidation on Met) modifications were set. The false discovery rate (FDR) was set to 4% at the protein level. MS^E data were searched against the modified *A. thaliana* database (TAIR10, <ftp://ftp.arabidopsis.org>) containing common contaminants such as keratin and rabbit glycogen phosphorylase B (P00489) was used as internal quantification standard. Quantification was performed based on the intensity of the three most abundant proteotypic peptides. The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium (<http://proteomecentral.proteomexchange.org>) via the PRIDE partner repository with the dataset identifier PXD000981. ## Kinase activity assays on the microarray ChloroPhos1.0 The active kinase preparations were incubated on the microarrays for 2 h at 25°C in kinase activity buffer B (50 mM Tris-HCl pH 7.5, 50 mM NaCl, 10 mM MgCl₂, 1× PhosStop (Roche) and 0.1% protease inhibitor cocktail (Sigma P9599), 2 µM ATP, 240 nM \[γ-³³P\] ATP). These incubations included 25 µg of the recombinant MBP-tagged pCKII, 90 µg of the Heparin Sepharose eluates or 20 units of commercially available bovine heart protein kinase A (PKA, Sigma-Aldrich). The microarrays were washed in TBS pH7.5 to remove remaining substances of the incubation mixture, subsequently by H₃PO₄ (pH 2), to get rid of ATP bound to peptides and finally in H₂O, to neutralize the microarray surface. The dried microarrays were exposed on a Phosphorimager Screen for 1 week and scanned by a BAS-1800 reader (Fujifilm) with 50 µm pixel size. Analysis was performed with the GenePixPro 6.1 software (Molecular devices). The grayscale intensity median of all spot pixels were determined after background subtraction and averaged over all 9 spots/peptide. These signals were classified into 4 categories, which were defined as 3 – strong (\>50% over background intensity \[BI\]); 2 – medium (\<50%–\>20% BI); 1 – weak (\<20%–\>10% BI); 0 – none (\<10% BI). Due to the possible occurrence of signal artefacts such as dust particles, we included an optical control to exclude false positives. ## Expression of recombinant Alb3 constructs and phosphorylation assays His-tagged Alb3 wildtype and mutant constructs were expressed and purified as described previously. Site-directed mutagenesis constructs were generated using the QuikChange XL site-directed mutagenesis kit (Agilent Technologies) according to the manufacturer's protocol. Plastid CKII was expressed as His-tagged protein in *E. coli*. The mature sequence of pCKII corresponding to amino acids 56–433 was cloned into the pETDuet1 vector using the BamHI/SalI restriction sites. The plasmid encoding the N-terminally his-tagged pCKII was transformed into *E. coli* Rosetta2 (DE3) (Novagen) and shaken at 37°C until an optical density (600 nm) of 0.6–0.7 was reached. Protein expression was induced by the addition of 1 mM IPTG and continued for 1 h at 30°C. Afterwards the cells were harvested by centrifugation and immediately frozen to −20°C until use. His-pCKII was purified under native conditions using Ni-NTA resin (Qiagen). Protein purification was carried out as the manufacturer suggested using the following buffers: lysis/washing buffer 20 mM Na₂HPO₄/NaH₂PO₄ pH 8.0, 300 mM NaCl, 5 mM MgCl₂ and 20 mM imidazole; elution buffer 20 mM Na₂HPO₄/NaH₂PO₄ pH 8.0, 300 mM NaCl, 5 mM MgCl₂, 1 mM ATP, 5% (v/v) glycerol and 250 mM imidazole. Alb3 phosphorylation assays for the determination of the phosphorylation site contained 10 µg of His-tagged Alb3 WT or mutant protein and 10 µg of His-pCKII in 20 mM Tris-HCl pH 8.0, 300 mM NaCl, 5 mM MgCl₂ and 100 µM \[γ-³²P\] ATP (1 mM stock solution of non-radioactive ATP spiked with \[γ-³²P\]ATP in a final concentration of 0.6 µCi/µl). After incubation for 30 minutes at 30°C the phosphorylation reactions were stopped by the addition of SDS sample buffer and heating. Samples equal to 3 µg of each protein were separated by SDS-PAGE and the dried gels were subjected to autoradiography. ## *In vitro* Phosphorylation kinetics of Alb3 phosphorylation by pCKII For kinetic studies, we phosphorylated increasing amounts of wildtype Alb3 in two replicates for 2 and 4 minutes at 25°C using 0.3 µM recombinant pCKII. The Alb3 concentration was varied in a range between 50 nM and 5 µM. The reaction was performed in kinase activity buffer (20 mM Tris-HCl pH8; 300 mM NaCl, 10 mM MgCl₂, 1× PhosStop (Roche) and 0.1% protease inhibitor cocktail (Sigma P9599), 5 µM ATP, 25 nM \[γ-³³P\] ATP). All reaction mixtures were separated by SDS-PAGE and analyzed via an autoradiography screen. In parallel a \[γ-³³P\] ATP dilution series was spotted onto a dried SDS-gel and exposed on an autoradiography screen in the same way as the phosphorylation reactions. The autoradiograms were analyzed and quantified by the BAS Reader V2.26 and the TINA 2.09 software. The grayscale intensities of the pixels were quantified after background subtraction. Pixel intensities of the ATP-dilution were compared with the Alb3 phosphorylation signal. The resulting PSL values were correlated to the concentration of incorporated \[³³P\] thus resulting in quantitative values for phosphate incorporation into Alb3. # Results ## Design of ChloroPhos1.0: Selection of peptides and positioning of the phosphorylation site We extracted information on *Arabidopsis thaliana* chloroplast phosphoproteins from different published phosphoproteomics experiments at the status of January 2012, most of them being represented in the PhosPhAT 3.0 database (<http://phosphat.uni-hohenheim.de/>). Table S1 in lists all studies that were included for the peptide library generation. The majority of these studies were conducted at the level of the complete cell, only some early MS studies used organelles or isolated proteins for phosphopeptide detection. By means of a chloroplast protein reference table we extracted chloroplast phosphoproteins from these studies. Altogether, we identified 376 phosphoproteins without splice variants. In order to ensure accessibility of phosphorylation sites, we centered the peptides on the phosphorylated amino acid, i.e. we added 7 amino acids upstream and downstream to the phosphorylation site as determined by mass spectrometry. In cases where the phosphorylation site was not unambiguously localized, we allowed all hydroxyl-group carrying amino acids in a phosphopeptide to be present in the center position. We included two combinatory cases for phosphorylation sites in close proximity in order to decrease the number of different peptides on the microarray. In case of two neighboring phosphorylation sites, only the N-terminal amino acid was centered. In case of two phosphorylation sites separated by one amino acid, the amino acid between the two phosphorylation sites was centered. Phosphorylation sites at the N- or C-terminus of proteins (closer than 8 characters to the terminus) were not centered; instead the suitable 15 mer starting from the N- or C- terminus was built. Redundancies at the peptide level were eliminated. After applying these constraints, our peptide library contained 905 different 15 mers. Some peptide spots as process- or incubation- controls were added to better assess kinase activity on the microarray. Table S2 in lists all peptides spotted on the microarray. All peptides were synthesized with a linker at their N-terminus (N-(3-(2-(2-(3- amino-propoxy)-ethoxy)-ethoxy)-propyl)-succinamic acid). ## Phosphorylation activity of recombinant pCKII on the peptide microarray ChloroPhos1.0 Although casein kinase II activity is the major kinase activity in chloroplast extracts, we were unable to detect phosphorylation activity on the microarray with chloroplast protein preparations. In these experiments, we assayed up to 6 mg chloroplast protein, either from solubilized chloroplasts, where Triton X-100, or ß- DDM were utilized as nonionic detergents, or from detergent-free stroma extracts (data not shown). This is most likely due to low kinase concentrations and difficulties with the phosphorylation of peptide substrates because of competition from suitable protein substrates in close proximity to the active kinase. We therefore decided to enrich native chloroplast CKII by Heparin-Sepharose chromatography and expressed recombinant *Arabidopsis thaliana* pCKII in *E. coli* as a control. Phosphorylation activity of these two kinase preparations on the microarray was determined to identify pCKII phosphorylation targets. MBP-tagged pCKII was overexpressed in *E. coli* and purified in two steps by utilizing the maltose-binding protein tag on an Amylose column in the first step and by size-exclusion chromatography on Sephacryl S100 in the second step (Fig. S1). Activity tests using 2 µg casein as substrate show kinase activity of the recombinant protein and experiments with GTP as a phosphate donor and heparin as inhibitor reveal the specific CKII characteristics of this phosphorylation activity (Fig. S1). We performed a first microarray experiment and compared pCKII activity with that of bovine heart protein kinase A (PKA) to assess phosphorylation activity and specificity ( A). PKA phosphorylated 43 peptides and pCKII phosphorylated 33 peptides (Table S3). There was no overlap in peptide phosphorylation between these two activities suggesting high specificity of these two kinases with peptides on the microarray. We extracted the phosphorylation motifs from the set of phosphorylated peptides by WebLogo (<http://weblogo.berkeley.edu/>) and found that PKA prefers a motif with basic residues at positions −3 to −1 relative to the phosphorylation site, while pCKII prefers motifs with acidic residues at positions +1 to +3 relative to the phosphorylation site ( B). These motifs are in good agreement with alignments for these two kinases obtained from a collection of their phosphorylation targets in other systems. In conclusion, recombinant pCKII recognizes its bona fide substrates on the microarray and does so with a clear substrate preference that differs from that of PKA (Table S3). ## Preparation of native pCKII from *Arabidopsis thaliana* and *Sinapis alba* We next analyzed whether native pCKII isolated from chloroplast extracts phosphorylates the same set of peptides as the recombinant enzyme. To this end, we isolated pCKII from *Sinapis alba* and *Arabidopsis thaliana* chloroplasts by Heparin-Sepharose chromatography. *Sinapis alba* was introduced as a preparation control because the pCKII (PTK) purification protocol was originally established for this plant. The isolation of pCKII from Arabidopsis used the same method as described for *Sinapis alba* with minor modifications. The enrichment of pCKII in the eluates was analyzed *in vitro* using dephosphorylated and heat inactivated stroma extracts as substrate mixture. The specificity of the kinase preparation was assessed in control experiments using heparin as inhibitor and an excess of non-radioactive GTP as phosphate donor. These data indicate that pCKII was successfully enriched in the eluate fractions. To assess the efficiency of pCKII enrichment we identified and quantified proteins in the eluate fraction by mass spectrometry as previously described. Altogether, 518 proteins were identified. Chloroplast ribosomal proteins and CSP41 are the most abundant proteins in these fractions followed by components of the transcription machinery (Table S4). Despite successful enrichment, the pCKII amount is relatively low with 1.2 fmol measured in 1.5 µg chloroplast protein on column. In 1.5 µg chloroplast eluate protein, we also identified the thylakoid kinases STN7 (3.3 fmol on column) and STN8 (0.5 fmol on column) and the plastoglobuli associated kinase ABC1K8 (10.4 fmol on column) ( B). ## Phosphorylation activity of different pCKII preparation on the peptide microarray ChloroPhos1.0 We tested the phosphorylation activity of the three different pCKII preparations on the peptide microarray and found a robust set of peptides that are phosphorylated by all kinase preparations ( A, Table S5 in). Most importantly, the native pCKII preparation phosphorylated the same peptides as the recombinant pCKII. The common substrates of the recombinant pCKII and the native pCKII preparations are e.g. peptides of translation elongation factor EF1B, Alb3, Toc159, Tac10 and RH3. Unique targets of recombinant pCKII comprise furthermore unknown proteins, RNP31, several metabolic enzymes and Mg²⁺-chelatase. Several peptides are exclusively phosphorylated by the recombinant pCKII, which is most likely a consequence of its higher specific kinase activity. In this case we expect that the peptides phosphorylated by all three preparations are better pCKII substrates or contain more pCKII phosphorylation sites, because they apparently require less pCKII activity to produce a detectable phosphorylation signal. The peptides phosphorylated by all preparations do not contain more S, T or Y compared to those phosphorylated exclusively by the recombinant enzyme but they are more acidic and comply better with the canonical CKII consensus motif with acidic amino acids in position +1 to +3 relative to the phosphorylation site ( B). The peptides phosphorylated only by the recombinant enzyme are therefore most likely less favorable CKII substrates. In only two instances we observe a stronger peptide phosphorylation signal with the native Arabidopsis pCKII compared to the recombinant protein, i.e. with peptide KAEKDEVSDDEATIE of MSCS-LIKE 3 (AT1G58200) and peptide KYEGKKLSELSDDED of TAC10 (AT3G48500) ( A). Both peptides fulfil the requirements for acidic amino acids in a CKII phosphorylation motif and it is therefore not clear why the highly active recombinant pCKII is less active on them compared to the native enzyme. At this stage we cannot rule out the possibility that one of the kinases in the Heparin-Sepharose eluates are involved in the phosphorylation of the second serine or threonine in each peptide (see peptides above). This could make the peptide more acidic and thus elevate the phosphorylation activity by native pCKII. Such a “priming” phosphorylation would be missing in the recombinant enzyme preparation. It is also possible that pCKII interacts with proteins in the chloroplast extract and that this interaction alters its substrate preference for some of the substrates. Although at present speculative, both explanations make an excellent starting point for further investigations. ## *In vitro* phosphorylation of recombinant Alb3 validates peptide phosphorylation data The set of pCKII targets comprises proteins of the gene expression apparatus and proteins involved in chloroplast metabolism, suggesting that pCKII may catalyze crosstalk between different chloroplast functions. We selected Alb3 for further characterization because its phosphorylation by pCKII suggests a so far unknown type of crosstalk between the regulation of gene expression and thylakoid membrane assembly *in vivo*. We expressed Alb3 in *E. coli* with its wildtype sequence and as the phosphorylation site mutants S416A, S418A and S424A (Fig S2), and used affinity purified and desalted Alb3 protein for *in vitro* phosphorylation assays. Using recombinant MBP-tagged pCKII, we phosphorylated increasing concentrations of wildtype Alb3 for 2 and 4 minutes in two replicates each ( A). Under these conditions, the maximum velocity of the reaction (Vmax) was not reached at substrate concentrations up to 5 µM. Plastid CKII phosphorylates Alb3 efficiently as illustrated by the high velocity of phosphate incorporation into recombinant wildtype Alb3 at incubation times between 2 and 4 minutes ( A). The phosphorylation of Alb3 by the recombinant pCKII shows the specific properties of CKII phosphorylation activity because it can be competed with unlabeled GTP and is inhibited by low concentrations of heparin (Figure S3). To determine the exact site of pCKII phosphorylation we exchanged the serines S416, S418 and S424 in the Alb3 wildtype sequence with alanine (Fig. S2). The S414A and S418A mutant proteins are phosphorylated by recombinant His-tagged pCKII up to wildtype levels, suggesting that these sites are not targeted by pCKII ( B). In contrast, phosphorylation assays with the S424A mutant resulted in a complete loss (His-tagged pCKII) or a greatly diminished phosphorylation signal (MBP-tagged pCKII) ( B and Figure S3) depending on the specific activity of the pCKII preparation. This establishes S424 as the major pCKII phosphorylation site in Alb3, which is in line with the phosphopeptide mapping data for Alb3 reported by Reiland and colleagues. The residual phosphorylation observed with the MBP-tagged pCKII preparation is most likely due to phosphorylation of S420, which is placed in a putative CKII phosphorylation motif with acidic amino acids at positions −1 and +1 to +2 relative to the serine residue. The phosphorylation site S424 shows a clear preference of pCKII for the motif E-X-S-D-D-E (Fig. S2) for Alb3 phosphorylation. Such a motif in the C-terminal, stroma exposed region of Alb3 is present in *Arabidopsis thaliana*, *Zea mays*, *Hordeum vulgare* and one *Oryza sativa* homologue, but is missing from *Pisum sativum* and *Solanum tuberosum* and from the two *Chlamydomonas reinhardtii* homologues (Fig. S4). # Discussion ## Using peptide microarrays as screening tools for kinase targets We show here that chloroplast phosphoproteomics data are a useful resource for the design of an organelle-specific peptide microarray. The phosphoproteomics results were collected from published large-scale experiments with different Arabidopsis samples, most of them deposited in the PhosPhAT database (<http://phosphat.uni-hohenheim.de/>). By exclusively using peptides that were found phosphorylated *in vivo*, we increase the specificity of a positive *in vitro* phosphorylation result on the microarray with a chloroplast protein kinase. By combining prior knowledge on the few established chloroplast kinases with knowledge about their *in vivo* targets, we introduce a meaningful constraint in the design of a large-scale experiment. Such constraints reduce false assignments of kinase/substrate relationships, as previously shown for kinase motif prediction. We are thus confident that the peptides phosphorylated by pCKII in the experiments reported here are true *in vivo* substrates of this kinase, provided that the constraints are fulfilled (this is not the case for the phosphorylation of peptides by PKA, because the kinase does not co-localize with the substrates *in vivo*). The specificity of this assignment is supported by the fact that we do not find any overlap in the set of phosphorylated peptides between pCKII and PKA. However, there are a few issues that need attention in the interpretation of the *in vitro* results and their transfer to the *in vivo* situation. The specificity and efficiency of a kinase reaction with its substrate is determined at different levels *in vivo*, some of which cannot be assessed with a peptide microarray. These levels of regulation entail interaction of the kinase with its substrate via docking sites or scaffold proteins. Docking sites are often spatially separated from the kinase active site and the actual phosphorylation sites and they increase the local concentration of substrate in the vicinity of the protein kinase. Thus, docking sites may be required to facilitate the phosphorylation of a substrate *in vivo*. Since they are not represented in the 15mer peptide on the microarray, we expect phosphorylation to occur at lower rate. In fact, the lack of peptide phosphorylations with crude extracts and the necessity to enrich kinases for their use on the microarray support this view. The same holds true for scaffold proteins that are required to establish a productive kinase/substrate interaction. In both cases, “false negative” results are the consequence. With the peptide microarray we furthermore neglect the concentration ratio between kinase and its substrate that exists *in vivo*. This may result in “false positive” assignments because protein concentrations are a potential regulator of kinase activity *in vivo*, that may operate for example by substrate competition. The primary determinant of kinase substrate specificity is the catalytic cleft, e.g. its depth, hydrophobicity and charge distribution. Based on the depth of the catalytic cleft, kinases are roughly grouped into two different families, serine/threonine kinases on the one hand and tyrosine kinases on the other hand. Although the catalytic domains are similar, tyrosine kinases have a deeper catalytic cleft that allows a tyrosine residue to span the distance between the substrate peptide backbone and the y-phosphate of the phosphate donor. Apart from the phosphorylated residue, the amino acids that are directly neighboring the phosphorylation site are the most important specificity determinants for the recognition of a substrate by its kinase. It was shown, that the active site of a kinase interacts roughly with four amino acids up- and downstream of the phosphorylation site. In the case of CKII, basic residues in subdomains VIII, I, II and III are predicted to make contact with the acidic amino acids of the substrate, such determining the high specificity of CKII for acidic residues proximal to the phosphorylation site (subdomain numbering system according to. This substrate specificity is reflected in the consensus motif of CKII phosphorylation in general, and it was correctly reproduced on the peptide microarray data reported here ( B and 4 B). ## The set of chloroplast pCKII targets The set of phosphorylated peptides on the microarray is relatively small, i.e. much smaller than the set of predicted CKII targets based on phosphorylation motif recognition. For reasons detailed above we conclude that we have assembled a highly stringent dataset with a high number of false negative observations, i.e. peptides that were not phosphorylated *in vitro* although they are targets for pCKII *in vivo*. Exceptions are the phosphorylations of Toc159 and SNAP33, both of which are strongly phosphorylated by all pCKII preparations on the microarray. These proteins are normally not in contact with pCKII because the acidic A-domain of Toc159 is reaching into the cytosol and SNAP33 (AT5G61210) is an endosomal contaminant in the chloroplast reference proteome. Nonetheless, these proteins are excellent targets for CKII and it is very likely that they are phosphorylated by one of the other CKII alpha subunits in the nucleo- cytoplasmic compartment. The set of pCKII targets comprises several proteins involved in plastid gene expression. TAC10 is an essential component of the plastid RNA polymerase complex containing an S1 RNA-binding domain. Its exact function is currently unknown. The phosphorylation site is located C-terminal to the S1 domain at Ser431 and/or Ser434. The function of phosphorylation is not known, but early analyses revealed that phosphorylation of the plastid RNA polymerase complex by pCKII results in a decrease of *in vitro* transcription activity. The hypothesis that pCKII phosphorylation of the RNA polymerase has an inhibitory effect aligns with the higher concentration of pCKII in non- photosynthetic organs. Here, phosphorylation of the RNA polymerase could prevent the transcription of genes for photosynthetic proteins, that are not needed e.g. in roots. A function of pCKII in the regulation of posttranscriptional processes of plastid gene expression is supported by the phosphorylation of the DEAD box RNA helicase RH3. RH3 has a predicted transit peptide length of 60 amino acids and the phosphorylation site is located at Ser80. Thus, this phosphorylation site is N-terminal to the helicase domain and close to the N-terminus of the mature protein. Because of this location it seems unlikely that phosphorylation alters the catalytic properties of RH3. It is more likely that phosphorylation alters its interactions with other proteins in RNA degrading complexes. The RNA-binding protein RNP31 is already a known pCKII target. This protein belongs to a family of RNA-binding proteins that have an acidic domain and two tandem RNA recognition motifs (RRM). They were among the first identified pCKII targets in chloroplasts. Phosphorylation by pCKII occurs in the acidic domain at Ser128 and is known to alter the RNA binding properties by this RNP family. In addition to targets in the gene expression system, pCKII appears to be involved in the regulation of the plants central energy metabolism. This entails the carbohydrate metabolism via phosphorylation of the starch branching enzyme (SBE2.1), the small subunit of RubisCO, and RubisCO activase, and photosynthesis via phosphorylation of Alb3. A regulatory role of CKII in carbohydrate metabolism was reported earlier for mammalian systems. Here, CKII phosphorylates glycogen synthase, phosphoglucose isomerase and glycerol-3-phosphate acyltransferase (GAT). A new and previously unreported function of pCKII in the regulation of photosynthesis may be mediated by phosphorylation of Alb3. Alb3 is essential for the integration of light-harvesting complex proteins into the thylakoid membrane and lack of Alb3 results in an albino phenotype. This protein is well conserved in land plants and algae and structurally related to bacterial YidC proteins. Using recombinant Alb3, we could establish Ser424 as the major site for phosphorylation. This phosphorylation site is not strictly conserved (Fig. S4) suggesting specialization in the regulation of LHC complex insertion into thylakoid membranes. # Conclusions We show here that the peptide microarray ChloroPhos1.0 is a suitable screening tool to identify novel kinase substrates and to characterize the preferred phosphorylation motif of currently uncharacterized plastid kinases. By using *in vivo* phosphorylation sites of proteins that co-localize with chloroplast kinases, we aim for stringent substrate recognition and a low false positive rate in the assignment of *in vivo* kinase substrates. This requires that the constraints are correct, i.e. that the substrates truly co-localize with the kinase, which is not the case for PKA. At the same time, we expect a relatively high false negative rate for structural reasons such as the lack of interaction domains and docking sites that facilitate substrate recognition by its cognate kinase *in vivo*. The fact that we were unable to measure phosphorylation activity with complex extracts supports this hypothesis. Because of its design, the microarray is restricted to assay targets for Arabidopsis chloroplast protein kinases and may be extended to the analysis of closely related species such as mustard. We will constantly update the microarray and the number of peptides will increase as more information about chloroplast protein phosphorylation becomes available. # Supporting Information We thank Stefan Schlömer and Lukas Sydow for initial experimental support. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: AS BA DS MS SB. Performed the experiments: AS EB SH BD. Analyzed the data: AS EB BA BD DS MS SB. Contributed reagents/materials/analysis tools: MS SH. Wrote the paper: AS EB SH BD DS MS SB.

# Introduction Circulating tumor cells (CTCs), which are tumor cells circulating in the peripheral blood, have been attracting attention as a new cancer biomarker. Previous studies reported that CTCs were present in the blood of cancer patients from the early stage of cancer development and that the numbers of CTCs were correlated with patient prognosis and treatment efficacies. In addition, because CTCs are living cancer cells, their detection provides definitive proof of the presence of tumors. The CellSearch^TM System has been approved by the U.S. Food and Drug Administration (FDA) as a CTC detection system. In this system, CTCs are detected as epithelial cell adhesion molecule (EpCAM)/cytokeratin-positive cells in the blood; however, it is difficult to detect CTCs undergoing epithelial-mesenchymal transition (EMT) by the CellSearch^TM System, because EpCAM expression declines on tumor cells undergoing EMT. It has become clear that EMT-induced CTCs (EMT-CTCs) have increased metastatic potential and that EMT-CTCs are a biomarker of patient prognosis and chemotherapy resistance. Detection of EMT-CTCs is thus important to predict patient prognosis and the efficacy of cancer treatment. In order to efficiently detect EMT-CTCs, several novel CTC detection methods based on differences in the physical properties between CTCs and blood cells, including cell size, cellular surface charge, and density, have been developed. However, these methods are often inefficient and lacking in specificity, although these methods are relatively easy to operate and cost-effective. In order to address the above-mentioned problem of a reduction in EpCAM expression on tumor cells undergoing EMT, a CTC detection system using a green fluorescence protein (GFP)-expressing conditionally replicating adenovirus (Ad) has been developed. The conditionally replicating Ads contained the tumor- specific promoter-driven E1 gene expression cassette and the cytomegalovirus (CMV) promoter-driven GFP expression cassette. The conditionally replicating Ads showed tumor cell-specific replication, resulting in tumor cell-specific GFP expression. CTCs were detected as GFP-positive cells. Furthermore, we added the following modifications to the conditionally replicating Ads in order to detect CTCs more efficiently. First, the fiber protein was substituted with that of Ad serotype 35 (Ad35), which recognizes human CD46 as an infection receptor. The conventional conditionally replicating Ad was based on species C Ad serotype 5, which recognizes coxsackievirus and adenovirus receptor (CAR). CAR expression is often downregulated on malignant tumor cells and tumor cells undergoing EMT, whereas CD46 is ubiquitously expressed on almost all cells except for erythrocytes. In addition, CD46 expression is up-regulated on malignant tumor cells. Substitution with the fiber protein of Ad35 leads to efficient detection of CAR-negative CTCs. Second, miR-142T-3p-targeted sequences were inserted into the 3’-untranslated region (UTR) of the E1 gene and GFP gene. miR-142T-3p is expressed in a blood cell-specific manner, resulting in significant suppression of GFP expression in normal blood cells. The conditionally replicating Ad containing the modifications described above, rAdF35-142T-GFP, efficiently detected the CTCs in the blood samples of lung cancer patients without significant production of false-positive cells however, higher levels of GFP expression are preferable for efficient detection of CTCs. In this study, we have developed four types of novel conditionally replicating Ads expressing GFP and examined their CTC detection efficiencies in the blood samples of lung cancer patients. Among the four types of recombinant Ads, rAdF35-E1-2A-GFP-ADP, which contained the E1A gene fused with the 2A peptide- coding sequence, the GFP gene in the E1 region and the ADP gene in the E3 region, mediated the highest GFP expression in the human cultured tumor cells, and detected CTCs in the blood samples at levels comparable to rAdF35-142T-GFP developed in the previous study. # Materials and methods ## Cell lines H1299 (a human non-small cell lung cancer cell line), MCF-7 (a human breast cancer cell line), THP-1 (a human monocyte-derived cell line), and human peripheral blood mononuclear cells (PBMCs) (Precision for Medicine, Frederick, MD) were cultured in RPMI 1640 containing 10% fetal bovine serum (FBS), 100U/mL penicillin, and 100μg/mL streptomycin. HepG2 (a human liver carcinoma-derived cell line, JCRB1054, JCRB Cell Bank, Tokyo, Japan), T24 (a human bladder carcinoma-derived cell line), and PANC-1 cells (a human pancreatic carcinoma- derived cell line) were grown in Dulbecco’s modified eagle medium (DMEM) containing 10% FBS, 100U/mL penicillin, and 100μg/mL streptomycin. All cells were cultured at 37°C in a humidified atmosphere of 5% CO₂-95% air. ## Plasmid The Ad plasmid for rAdF35-E1-CMV-GFP, pAdHM34-hAIB-miR142T-miR142T-EGFP-CMV, was constructed as follows. First, a human telomerase reverse transcriptase (hTERT) promoter-driven E1 gene expression cassette, in which the E1A and E1B genes were linked by an internal ribosome entry site (IRES), in pHM5-hAIB-142-3pT and a cytomegalovirus (CMV) promoter-driven GFP expression cassette containing the four copies of miR-142-3p-targeted sequences were inserted into pHM5, producing pHM5-hAIB-miR142T-miR142T-EGFP-CMV. The hTERT promoter-driven E1 gene expression cassette and the CMV promoter-driven GFP expression cassette were inserted into pAdHM34 to obtain pAdHM34-hAIB-miR142T-miR142T-EGFP-CMV. The Ad plasmid for rAdF35-E1-2A-GFP, pAdHM34-hAGB-miR142T, was constructed as follows. The hTERT promoter-driven E1 and GFP gene expression cassette in which the P2A peptide-coding sequence and the GFP gene were located downstream of the E1A gene was chemically synthesized (GENEWIZ Japan Corp, Tokyo, Japan) and was inserted into pAdHM34 to obtain pAdHM34-hAGB-miR142T. The Ad plasmid of rAdF35-pIX-2A-GFP, pAdHM34-hAIB-miR142T-pIX-P2A-EGFP-miR142T, was constructed as follows. The pIX gene fused with the 2A and GFP genes and four copies of miR-142-3p-targeted sequences was chemically synthesized (GENEWIZ Japan Corp) and was inserted into pAdHM34 by homologous recombination, producing pAdHM34-pIX-2A-GFP-miR142T. An hTERT promoter-driven E1 gene expression cassette was inserted into the E1 deletion region of pAdHM34-pIX-2A-GFP-miR142T, producing pAdHM34-hAIB-miR142T-pIX-P2A-EGFP-miR142T. The Ad plasmid of rAdF35-E1-2A-GFP-ADP, pAdHM309-E1-F35-hAGB3-miR142T, was constructed as follows. pAdHM309-E1-F35 was created by substituting the fiber gene of pAdHM309-E1 with that of Ad35. The hTERT promoter-driven E1 and GFP gene expression cassette in pHM5-E1-2A-GFP-miR142T was inserted into pAdHM309-E1, producing pAdHM309-E1-F35-hAGB3-miR142T. Details of the plasmid construction are available upon request. ## Production of conditionally replication-competent Ads Recombinant Ads were produced by transfection of the *Pac*I-digested Ad plasmids into HEK293 cells using Lipofectamine2000 (Invitrogen, Carlsbad, CA). All recombinant Ads were grown in H1299 cells, purified by two rounds of cesium chloride-gradient ultracentrifugation, dialyzed, and stored at -80°C. The virus particle (VP) titers were measured using a spectrophotometric method, and biological titers were determined using an Adeno-X-rapid titer kit (Clontech, Mountain View, CA). The ratio of the particle-to-biological titer was between 4.3 and 15 for each Ad used in this study. ## Flow cytometry analysis of GFP expression levels in human cultured cells Human cultured tumor cells were seeded with 2-4x10⁴ cells in a low- binding 24-well plate and incubated with the recombinant Ads at 30 and 300 virus particles (VP)/cell. After a 24-h incubation, cells were collected, washed twice with cold PBS, and incubated with trypsin for 2 minutes at 37°C. GFP expression levels in the cells were analyzed using a flow cytometer (MACS Quant Analyzer; Miltenyi Biotec, Bergisch Gladbach, Germany). Data were analyzed by FCS Multicolor Data Analysis Software (Flowjo; BD Biosciences, San Jose, CA). ## Real-time PCR analysis of Ad genome copy numbers in human cultured cells Human tumor cells were seeded at 4x10⁴ cells/well in a low-binding 24-well plate and were incubated with the recombinant Ads at 30 and 300 VP/cell. After a 24-h incubation, total DNA containing the Ad genome was recovered from the cells using DNAzol (Molecular Research Center, Cincinnati, OH). Ad genome copy numbers were determined by quantitative PCR using THUNDERBIRD Next SYBR qPCR Mix (TOYOBO, Osaka, Japan) as previously described. ## Detection of CTCs in the blood samples of lung cancer patients CTCs in the peripheral blood samples of lung cancer patients were detected as previously described with slight modifications. Briefly, cells were recovered from the 3 ml blood samples of patients with non-small cell lung cancer (NSCLC). Following enrichment of CD45-negative cells, cells were incubated with 1.0x10⁸ VP of GFP-expressing conditionally replicating Ads at 37°C for 24 h. The cells were washed and stained with phycoerythrin (PE)-labeled anti-human CD45 antibody (clone HI30; BioLegend, San Diego, CA) and then observed under a fluorescence microscope. CTCs were defined as GFP+/CD45- cells. The procedures for obtaining peripheral blood from patients with lung cancer were approved by the Institutional Review Board at the Juntendo University School of Medicine (M16-0154). All patients provided written informed consent. Information about cancer stages and histological cancer types of lung cancer patients is shown in. ## Statistical analyses Statistical analysis was performed using Student’s *t*-test or one-way ANOVA followed by a Tukey post hoc test. Data are presented as the means ± S.D. # Results ## Development of novel types of GFP-expressing conditionally replicating Ads for efficient detection of CTCs In order to improve the GFP expression levels in CTCs and titers of GFP- expressing conditionally replicating Ads, we first developed three types of novel conditionally replicating Ads. In the previous study, we used rAdF35-142T-GFP, which contained the CMV promoter-driven GFP expression cassette in the E3 region and miR-142-3p-targeted sequences in the 3’-UTR of the E1B gene and the GFP gene. rAdF35-E1-CMV-GFP contained the CMV promoter-driven GFP expression cassette in the region downstream of the hTERT promoter-driven E1 gene expression cassette in the reverse orientation. In rAdF35-E1-2A-GFP, the 2A peptide-coding sequence and the GFP gene were inserted downstream of the E1A gene in order to transcribe the GFP gene by a tumor cell-specific promoter. The E1B gene was driven by the native E1B gene promoter in rAdF35-E1-2A-GFP. rAdF35-pIX-2A-GFP had the 2A peptide-coding sequence and the GFP gene fused with the pIX gene. Our group previously reported that the pIX gene was efficiently expressed 24 h after infection. These four types of conditionally replicating Ads, including rAdF35-142T-GFP, contained the Ad35 fiber proteins, which recognize CD46 as an infection receptor, and 4 copies of sequences perfectly complementary to miR-142-3p, which is expressed in a blood cell-specific manner. ## GFP expression levels in human cultured tumor cell lines following treatment with GFP-expressing conditionally replicating Ads In order to examine the GFP expression levels following treatment with conditionally replicating Ads, rAdF35-E1-CMV-GFP, rAdF35-E1-2A-GFP, and rAdF35-pIX-2A-GFP were added to five types of human tumor cells cultured in suspension. rAdF35-E1-2A-GFP treatment yielded the highest numbers of GFP- positive cells in all the human tumor cell lines. The percentages of GFP- positive tumor cells following infection with rAdF35-E1-2A-GFP at 300 VP/cell were more than 80% in all the human tumor cell lines, with the exception of T24 cells. Next, in order to examine the GFP expression levels in blood cells following treatment with conditionally replicating Ads, cells of THP-1, a human monocyte cell line, were incubated with the recombinant Ads. Statistically significant numbers of GFP-positive cells were not found following treatment with rAdF35-E1-CMV-GFP, rAdF35-E1-2A-GFP, or rAdF35-pIX-2A-GFP. These results suggested that none of the three conditionally replicating Ads produced GFP- positive blood cells. Collectively, the above results indicated that rAdF35-E1-2A-GFP realized the most efficient mediation of GFP expression in human tumor cells among the three types of conditionally replicating Ads, without detectable levels of GFP expression in blood cells. ## GFP expression levels by the conditionally replicating Ad possessing the ADP gene in human tumor cells In order to further improve the GFP expression levels following infection with rAdF35-E1-2A-GFP in human tumor cells, the E3-deleted region in rAdF35-E1-2A-GFP was modified to leave the ADP gene in the conditionally replicating Ad genome, producing rAdF35-E1-2A-GFP-ADP. The ADP gene plays a crucial role in promotion of progeny virus production. Titers of rAdF35-E1-2A-GFP-ADP were slightly but significantly higher (about 4-fold) than those of rAdF35-142T-GFP. VP/infectious unit (IFU) titer ratios were comparable between rAdF35-E1-2A-GFP-ADP and rAdF35-142T-GFP. These data suggested that inclusion of the ADP gene in the Ad genome improved the production of conditionally replicating Ads. Next, we examined the GFP expression levels in human tumor cells following treatment with rAdF35-142T-GFP, rAdF35-E1-2A-GFP, and rAdF35-E1-2A-GFP-ADP. rAdF35-E1-2A-GFP-ADP mediated slightly but significantly larger numbers of GFP- positive PANC-1 and HepG2 cells than the other types of the recombinant Ads, although the percentage of GFP-positive cells following infection with rAdF35-E1-2A-GFP-ADP was slightly lower than those following infection with rAdF35-142T-GFP in H1299 cells. Statistically significant differences in the virus genome copy numbers were not found in the tumor cells for the three types of the recombinant Ads, except when H1299 cells were infected at 300 VP/cell. Although flowcytometric analysis showed that rAdF35-E1-2A-GFP-ADP produced significantly higher levels of GFP-positive than mock-infected THP-1 cells at 300 VP/cell, GFP expression levels were below the detection limit by fluorescence microscopy. In addition, detectable levels of GFP-positive PBMCs were not found following infection with the three types of the recombinant Ads. These data indicated that rAdF35-E1-2A-GFP-ADP produced comparable or slightly higher levels of GFP-positive human tumor cells, compared with rAdF35-142T-GFP, without significant levels of GFP expression in blood cells. ## Conditionally replicating Ad-mediated detection of CTCs in the blood of lung cancer patients In order to compare the CTC detection efficiencies of rAdF35-142T-GFP and rAdF35-E1-2A-GFP-ADP in the blood of cancer patients, non-blood cells in the blood of lung cancer patients were concentrated, followed by infection with the recombinant Ads. GFP+/CD45- cells that were considered to be CTCs were detected for both types of the recombinant Ads ( arrow head). More than 10-fold larger numbers of DAPI- cells were found, compared with DAPI+ cells (arrows). The numbers of GFP+/CD45- cells following incubation with rAdF35-E1-2A-GFP-ADP and rAdF35-142T-GFP were almost comparable. GFP+/CD45- cells were found in 10 of 17 patients (58.8%) for both types of recombinant Ads. No clusters of viable circulating tumor microemboli (CTM) were found in the samples in this study. The mean longitudinal diameters of GFP+/CD45- and GFP-/CD45+ cells were 7.81 μm and 4.84 μm, respectively. Previous studies reported that the diameters of CTCs were larger than those of normal blood cells. No GFP+/CD45- cells were detected in the blood samples of healthy individuals following treatment with rAdF35-E1-2A-GFP-ADP or rAdF35-142T-GFP. These results indicate that rAdF35-E1-2A-GFP-ADP and rAdF35-142T-GFP can detect CTCs at comparable efficiencies in lung cancer patients without production of false-positive cells. # Discussion In this study, we generated novel conditionally replicating Ads by modifying the GFP expression cassette and leaving the ADP gene in the Ad genome in order to detect CTCs more efficiently than by the recombinant Ad used for detection of CTCs in the previous study. The novel recombinant Ad, rAdF35-E1-2A-GFP-ADP, that was developed in this study, and rAdF35-142T-GFP, that was developed in the previous study, exhibited comparable levels of CTC detection efficiencies in our present study. In this study, 1x10⁸ VP of the recombinant Ads were applied to the non-blood cells isolated from 3 ml of blood samples after enrichment of non- blood cells, although blood cells, including leukocytes, were included. When the CTCs were observed under a florescence microscope, the total number of DAPI+ cells, including CTCs, was approximately 10⁴ cells/sample. In addition to DAPI+ cells, DAPI- cells, including platelets and erythrocytes, were also found in the samples. The numbers of DAPI- cells were significantly higher (approximately 10–50-fold) than those of DAPI+ cells. All blood cells except for erythrocytes are CD46-positive. Taking these findings into consideration, approximately 10⁵ CD46-positive cells were incubated with 10⁸ VP of the conditionally replicating Ads in this study. Hence, the conditionally replicating Ads were added to the blood cells at approximately 10³ VP/cell. When human tumor cell lines were incubated with the conditionally replicating Ads at 300 VP/cell, more than 90% of the tumor cells were GFP-positive, suggesting that sufficient amounts of the conditionally replicating Ads were added to the blood cells of cancer patients for detection of CTCs under this experimental condition. Previous studies using rAdF35-142T-GFP, also sometimes referred to as OBP-1101, reported that CTCs were detected using rAdF35-142T-GFP in 42.5% and 69.1% of lung cancer patients, although the methods of treatment with the conditionally replicating Ads and observation of CTCs were slightly different between the studies. In this study, 58.8% of NSCLC patients were found to be CTC-positive by using rAdF35-142T-GFP and rAdF35-E1-2A-GFP-ADP. These data indicated that the sensitivity of CTC detection in this study was comparable to those in the previous studies. In previous studies using a CellSearch^TM system, CTCs were detected in 22.5% and 43.4% of patients with NSCLC. Previous studies using the GFP-expressing conditionally replicating Ads containing the hTERT promoter-driven E1gene expression cassette reported that CTCs were efficiently detected in the other types of cancer patients (50.0% of obstructive colorectal cancer patients, 26.0% of cervical cancer patients, 39.6% of gynaecological cancer patients), although precise experimental conditions were different between the studies. These findings suggest that a CTC detection method using a GFP-expressing conditionally replicating Ad can be utilized for detection of CTCs derived from various types of tumors. Comparison between rAdF35-E1-2A-GFP and rAdF35-E1-2A-GFP-ADP showed that the percentages of GFP-positive tumor cells at 24 h after infection were significantly elevated by inclusion of the ADP gene. The virus genome copy numbers of rAdF35-E1-2A-GFP-ADP tended to be higher than those of rAdF35-E1-2A-GFP, although statistically significant differences were not found. The ADP gene was mainly expressed at the late phase of infection and contributed to the release of progeny virus from infected cells, leading to an improvement of progeny virus infection of neighboring cells. On the other hand, it remained unclear whether the ADP contributed to virus genome replication. The above- described findings of our present experiments suggested that the ADP gene is involved in virus genome replication in an unknown manner and/or that progeny virus would be released from the infected cells within 24 h, followed by infection of neighboring cells. Further examination is necessary to elucidate the role of the ADP in virus genome replication. On the other hand, no apparent differences were found in the numbers of GFP-positive tumor cells and CTC detection efficiencies between rAdF35-142T-GFP and rAdF35-E1-2A-GFP-ADP (Figs). Other differences observed between the Ad genomes of these two types of Ads, including the structure of the Ad genome, the promoters driving the GFP gene expression, and the presence or absence of the ADP gene affected the GFP expression levels in the tumor cells. Recently, nanoparticle-based CTC detection systems have been developed. Nanoparticles containing CTC-targeting ligands and antibodies on the surface efficiently captured CTCs, resulting in efficient enrichment and detection of CTCs, however, expression of CTC markers on cellular surface is essential for efficient capture of CTCs. In addition, CTC marker proteins are different between tumor cell types. Tumor cell-specific marker proteins, including hTERT, are often intracellularly expressed. Conditionally replicating Ads mediate tumor cell-specific replication *via* utilizing tumor cell-specific intracellular proteins, resulting in tumor cell-specific GFP expression. The conditionally replicating Ads used in this study contained the hTERT promoter for the E1 gene expression. The conditionally replicating Ads efficiently replicated and killed hTERT-positive tumor cells, on the contrary, hTERT-negative normal cells were resistant to the conditionally replicating Ads. hTERT activity or mRNA expression was not found in all of clinical tumor samples. Although it is now unclear whether hTERT was efficiently expressed in CTCs, hTERT-negative CTCs might not be efficiently detected by the conditionally replicating Ads used in this study. Conditionally replicating Ads containing another type of tumor-specific promoter should be used for detection of hTERT- negative tumor cells. In summary, in order to detect CTCs more efficiently than by the recombinant Ad used in the previous study. One of the novel recombinant Ads, rAdF35-E1-2A-GFP- ADP, exhibited CTC detection efficiency comparable to that of rAdF35-142T-GFP developed in the previous study. This study provides important clues for the future development of recombinant Ad-based systems for the evaluation of biomarkers. # Supporting information The authors are grateful to all participating patients and their families. We also thank Eiko Sakai, Yukiko Toba, Kosuke Shibata, Victoria Nicholson (Graduate School of Pharmaceutical Sciences, Osaka University, Osaka, Japan) and Toshiyoshi Fujiwara (Okayama University Graduate School of Medicine, Dentistry and Pharmaceutical Sciences, Okayama, Japan) for their support. [^1]: The authors have declared that no competing interests exist.

# Introduction In eyeblink conditioning an originally neutral conditional stimulus (CS), such as a tone, is followed by a reflex eliciting unconditional stimulus (US), such as a puff of air to the cornea. Initially, the subject will produce an unconditional blink response (UR) to the air puff, but if the tone and air puff are paired repeatedly the subject will eventually acquire a conditioned blink response (CR), which begins before the onset of the US and peaks near the expected US. The CR is adapted in time to the specific CS-US interval used. Several lines of evidence demonstrate that the cerebellum plays a critical role in the acquisition and expression of adaptively timed conditioned eyeblink responses. This opens up for the possibility that eyeblink conditioning can be used as a measure of cerebellar function and, by extension, of cerebellar dysfunction. An increasing number of studies indicate that the cerebellum is involved in cognition and language. Cerebellar abnormalities have been detected in patients with dyslexia, specific language impairment (SLI), and attention deficit hyperactivity disorder (ADHD). Autism spectrum disorders (ASD) have also been linked to cerebellar abnormalities, although this link has later been contested. Consistent with the pivotal role of the cerebellum in eyeblink conditioning, children with ADHD, ASD, and dyslexia have displayed atypical timing and learning patterns during eyeblink conditioning, although no such effect was found in children with SLI. However, the results are inconsistent concerning studies of individuals with ADHD. Other conditions, such as fetal alcohol syndrome (FAS), fragile X syndrome, Down’s syndrome, and schizophrenia, have also been linked to cerebellar deficits and poor performance in eyeblink conditioning. Rats exposed to alcohol as neonates suffer from loss of cerebellar neurons and show deficits in eyeblink conditioning as adults. Collectively, these studies constitute strong evidence that factors that influence the cerebellum also affect performance in eyeblink conditioning. Given that the cerebellum, like the rest of the brain, undergoes changes throughout life, it is perhaps not surprising that performance in eyeblink conditioning is influenced by age. Adults 20–50 years old, perform better than children and individuals older than 60 years. Yet, no age effects were observed in a study comparing adults, adolescents and typically developing children older than 9 years. Surprisingly, five month old infants reach similar levels of CRs as adults, although differences in CR timing remained post training. The aim of this study was to examine performance in eyeblink conditioning in school aged children (6–11 years). Our relatively large sample of 46 children and 30 adults allowed us to correlate age with various CR parameters, including the temporal profile of the CR. # Methods ## Ethics statement The regional ethics committee in Lund, Sweden, approved this study and all associated procedures. ## Participants A total of 76 subjects, 46 children and 30 adults, participated in the study. The children were recruited from two local elementary schools in lower middle class to higher middle class socioeconomic areas in southern Sweden. The age of the pupils ranged from 6 to 12 years old. The adults were recruited mainly from the student population at Lund University. Participants and legal guardians were informed about the purpose of the study and signed an informed consent form prior to testing. Three children and four adults were excluded from the analysis due to technical problems with the registration of eyelid movements. In addition, one child was excluded due to a non-verbal IQ score below 65 (percentile 1), tested by Raven’s colored progressive matrices. Thus, the participants included in the analysis were 42 children, 6–11 years old (mean = 8.8, SD = 1.3) and 26 adults (21 students and 5 former students), 20–55 years old (mean = 29.3, SD = 8.6). Among the children 22 were female (mean age = 9.0 years, SD = 1.4, range 6.8–11.1 years) and 20 were male (mean age = 8.6 years, SD = 1.3, range 6.9–10.6 years). Among the adults 18 were female (mean age = 30.3 years, SD = 9.3, range 21.6–55.9 years) and 8 were male (mean age = 26.8 years, SD = 6.8, range 20.6–40.8 years). All participants were screened for normal hearing with a modified Hughson-Westlake method (ISO 8253–1), at 20 dB hearing level. The average IQ score on Raven’s colored progressive matrices was 105, with a standard deviation of 16 points, for the children. The right hand was the dominant hand for 91.3 percent of the participants. The participants had no eye deficit or disease and all exhibited normal motor development. None of the children received extra support in school or used any medication. ## Procedure The test sessions took place in a calm and quiet room (Leq 60 seconds = 45 dB(A) measured with Brüel & Kjær 2225 sound level meter), away from school or university activity. The conditioning session lasted 35 minutes on average. The participants watched a movie both as distraction from the test situation and as motivation. It also helped to fixate their gaze and eye position. The children watched a cartoon while the adults watched a classic comedy movie. The movies were selected to vary as little as possible in sound intensity. The eyeblink conditioning was performed using a Shebot (Neurasmus, The Netherlands). This device consisted of a computer, air compressor, air-puff generator, sound generator, magnetic distance measurement technique controlling hardware, movie goggles with a circular air puff opening of 1.5 millimeters in diameter, magneto-sensitive sensor, noise excluding circumaural Sennheiser HD 201 headphones and a helmet that held the goggles and headphones in place. The eyelid movements were measured with the magnetic distance measurement technique. The sensor was placed on the cheek straight below a magnet (\~0.1 g, 5 x 3 x 1 mm) that was attached to the eyelid, close to the eyelash. The CS, a 1 kHz tone, was presented binaurally through the headphones at 68 dB SPL during 500 ms. The tone and movie sound were calibrated through the headphones and a coupler connected to an ISO-TECH SLM52N sound level meter. The tone was clearly audible to all of the participants through the soft background sound of the movie of \~41 dB(A), but not strong enough to trigger a blink reflex by itself. The US, a 15 ms air puff of 1 bar, was released through the movie goggles, placed 1–2 centimeters from the eyes, towards the left cornea. In a few cases the duration of the US was adjusted (+/- 5 ms at most), to ensure that it elicited a clear blink reflex, without being perceived as too aversive. A classical eyeblink conditioning delay paradigm, where the CS and US overlapped in time, was programmed in National Instruments LabVIEW 2011. Of the 42 children, 22 received a total of 70 trials (53 paired) while 20 received 100 trials (76 paired). The 26 adults received 80 trials (67 paired). The eyeblink conditioning protocols differed in number of trials since the children at the two schools and the adults had different amounts of time to spend on the eyeblink conditioning sessions. Yet, given that the settings were similar and the protocols, up until a certain time, were nearly identical, we considered it justified to compare and combine data from these three groups. The inter-trial interval varied randomly between 15 and 25 seconds. In paired trials the CS preceded the US by 485 milliseconds and the two stimuli co-terminated in time. If a spontaneous blink occurred during the 500 ms preceding the expected CS onset, the trial started over. As illustrated in, the majority of the trials in the three protocols were paired trials and the CS alone trials were concentrated in the last part of the conditioning session, after the first 50 trials. The participants were instructed to try to relax, concentrate on the movie and not pay attention to the stimuli or their own reactions during the experiment. They were informed that the air puffs could feel somewhat stronger in the beginning of the session. Before the conditioning session, the children were presented with 5–10 trials of paired tones and air puffs to make the test situation more familiar. ## Data analysis The magnetic distance measurement registrations, sampled at 1 kHz, were stored in a SQL database (MySQL Server 5.1, MySQL Workbench 5.2 CE). Eyelid movements were analyzed offline trial-by-trial with a semi-automatic SQL-based visual trial identifier program made in National Instruments LabVIEW 2011. A trial was considered invalid if the registration of the response was too noisy. A CR was defined as an eyeblink with an onset between 100 and 490 ms after the CS onset on paired trials, or between 100 and 700 ms after CS onset on CS alone trials. For every valid trial with a CR, the onset latency, peak latency, and peak amplitude were determined. The CR onset was defined as the earliest point in time where there was a change in the eyelid position of at least 3 SD, compared to the baseline. The CR peak was defined as the point in time the first eyelid closure appeared after the CR onset. CRs with an amplitude less than 15 percent of a participant’s mean UR amplitude were excluded. The UR was analyzed in terms of onset latency (minimum change of 3 SD compared to the baseline), peak latency (first maximum eyelid closure), and peak amplitude (\~100 percent eye closure), in paired and US alone trials. Participants blinked before the US on 2% of the US alone trials. ## Statistical analysis For the statistical analysis, the conditioning session was divided into blocks of ten trials. Since the protocols differed slightly between the groups after the fifth block, the analysis was mainly focused on the first five blocks. Average measures of CRs in CS alone trials after the fifth block were also used to describe the level of conditioning post training. When referring to CS alone trials without specifying actual blocks in the result and discussion sections, block 4 and the following blocks until the end of the session were included. CR percentages were used to investigate the learning during conditioning. While analysis of the CR onset was done in both paired and CS alone trials, analysis of the peak latency was done only in CS alone trials since URs often interfered with the peak latency of the CR in paired trials. All the participants were included in the analyses regardless of level of learning. The statistical analyses were made in SPSS Statistics 23 (IBM). With repeated measures ANOVAs the CR measures in the first five blocks as within-subjects factor, and sex and age as between-subjects factors, were analyzed. Greenhouse-Geisser correction were made whenever the assumption of sphericity was violated according to Mauchly’s test of sphericity. Post-hoc pairwise comparisons were made with Bonferroni confidence interval adjustment, with the significance level 0.05. Between-subjects effects were also analyzed with standard linear regression models in different parts of the session (paired trials in block 1, blocks 2–5, and in CS alone trials post training). # Results ## CR performance While the rate of learning varied substantially among participants, repeated measures ANOVAs with percent CRs on successive blocks as within-subjects factors showed that for both the children and the adults, the rate of CRs increases during training (Children: F(4, 152) = 6.601, p = 0.000, η² = 0.148. Adults: F(2.984, 71.625) = 5.656, p = 0.002, η² = 0.191). Post-hoc pairwise comparisons show that the initial increase in CRs levels off between the first and the second block for the children (block 1 vs. 2 p = 0.032, 1 vs. 3 p = 0.037, 1 vs. 4 p = 0.006, 1 vs. 5 p = 0.008), and between the first block and the third block for the adults (block 1 vs. 2 p = 0.538, 1 vs. 3 p = 0.000, 1 vs. 4 p = 0.017, 1 vs. 5 p = 0.138). There are no differences between the rest of the blocks compared to each other (p = 1.000). When comparing the first block to blocks 2–5 combined, the CRs increased with 12 percentage points among the children (t₄₁ = 4.269, p = 0.000, 95% CI\[1, 17\]) and 15 percentage points among the adults (t₂₅ = 5.400, p = 0.000, 95% CI\[1, 21\]). The CRs increase further, though only among the children, with 8 percentage points (t₄₁ = 3.642, p = 0.001, 95% CI\[3, 12\]) from blocks 2–5 to CS alone trials during later parts of the session. ### Effects of age and sex on the CR performance As illustrated in, the children produced on average 20 percentage points fewer CRs than the adults (t_48.52 = -3.027, p = 0.003, 95% CI\[-34, -7\]). The difference is only significant in blocks 2–5. This age effect is also present when examining only the children. Pearson product-moment correlation coefficient shows positive correlations between age and percentage of CRs in the first block (r = 0.328, p = 0.034, n = 42), in blocks 2–5 (r = 0.476, p = 0.001, n = 42), and in the CS alone trials (r = 0.490, p = 0.001, n = 42). No such correlation is present among the adults. The females produced more CRs than the males. On average, the girls produced 16.6 percentage points more CRs than the boys in blocks 2–5 (t₄₀ = 2.243, p = 0.031, 95% CI\[1.6, 31.5\]), and 15.5 percentage points more CRs than the boys in CS alone trials (t₄₀ = 2.031, p = 0.049, 95% CI\[7.8, 31.0\]). The women produced 26 percentage points more CRs than the men in block 1 (t_2.536 = 2.502, p = 0.020, 95% CI\[4.3, 46.1\]), and 29 percentage points more CRs than the men in blocks 2–5 (t_21.403 = 3.302, p = 0.003, 95% CI\[10.8, 47.5\]). Girls 9 years (median age) or older reached a similar level of CRs in blocks 2–5 as the women, and boys 9 years or older performed similar to the men. The CR production change from block 1 to blocks 2–5 did not differ significantly between the females and males, neither among children nor among adults. A repeated measures ANOVA, with the test session divided into the first five blocks and age split by median age (children = 9 years, adults = 28 years), demonstrates an effect of age (F(1,38) = 10.484, p = 0.003, η² = 0.216), and a marginal effect of sex (F(1,38) = 3.596, p = 0.066, η² = 0.086) on CR percentage among the children, without any significant interaction effect of age and sex (F(1, 38) = 0.051, p = 0.822). Similarly, there are a between-subjects effect of sex on the CR percentages among the adults (F(1, 24) = 6.730, p = 0.016, η² = 0.219). The influence of the age and sex of the children is also evident in a standard linear regression model with age and sex as predictors of average CR percentages. Together, age and sex account for 29.5% of the variance in the CR percentage in blocks 2–5 (F(2, 39) = 8.144, p = 0.001), and 29.3% of the variance in the CR percentage in CS alone trials (F(2, 39) = 8.074, p = 0.001). In block 1 there is only an age effect, which explains 10,8% (F(1, 40) = 4.836, p = 0.034) of the variance. Among the adults, there are some evidence of an effect of sex both in block 1 and in blocks 2–5, but not in CS alone trials. The regression model accounts for 14,5% (ß = -0.381, F(1, 24) = 4.075, p = 0.055) of the variance in CR percentages in block 1 and 23.8% (ß = -0.488, F(1, 24) = 7.492, p = 0.011) of the variance in CR percentages in blocks 2–5. ## CR timing Among the children there was no change in the CR onset latency, or in the CR onset variability within the first five blocks. By contrast, among the adults the CR onset latency changed during training (F(2.147, 32.203) = 4.555, p = 0.016, η² = 0.233). Post-hoc pairwise comparisons with Bonferroni adjustment show that onset latency increased between block 1 and block 2 (p = 0.022), and between block 1 and block 3 (p = 0.026) but not between the later blocks. There was also an onset latency increase of 38 ms between block 1 and blocks 2–5 combined (t₁₇ = 2.173, p = 0.044, 95% CI\[1,74\]), while there was no increase between blocks 2–5 and CS alone trials. Statistically significant CR latency differences between children and adults are not found in any part of the session. ### Effects of age and sex on the CR timing Older children and adults produced CRs that were more closely-timed to the US than younger children. Among the children, age correlates positively with CR onset latency (r = 0.347, p = 0.028, n = 40), and negatively with the variability of the CR onset latency (r = -0.398, p = 0.020, n = 34), in blocks 2–5 only. Standard linear regression analyses, while not repeated measures ANOVA on the first five blocks, show effects of age on the CR onset latency (ß = 0.347, R² = 0.121, F(1, 38) = 5.216, p = 0.028), and on the CR onset variability (ß = -0.398, R² = 0.159, F(1, 32) = 6.038, p = 0.020), in blocks 2–5. No correlations between age and any CR latency measure show among the adults. Standard linear regression analysis, while not repeated measures ANOVA on the first five blocks, shows an effect of sex on the CR onset latencies in blocks 2–5 (ß = 0.498, R² = 0.248, F(1, 24) = 7.923, p = 0.010), where the adult men produced CRs with later onsets than the women. There is also some effect of sex during CS alone trials on the CR peak latency among the children (ß = 0.400, R² = 0.160, F(1, 36) = 6.873, p = 0.013), with later peaks among boys than girls, and on the CR peak latency variability among the adults (ß = -0.405, R² = 0.164, F(1, 21) = 4.128, p = 0.055), with less variability among men than women. ## UR production and timing The US consistently elicited URs in both children and adults. While the UR latency was not affected by training, it was affected by the age of the subjects. Specifically, the children’s URs started on average 15 ms later (t_52.07, p = 0.000, 95% CI\[10, 18\]), and the children’s UR peaks were on average 28 ms later (t_64.228, p = 0.000, 95% CI\[21, 37\]), than the adults’ in blocks 2–5. Moreover, among the children, age correlates negatively with the UR onset (r = -0.303, p = 0.051, n = 42; ß = -0.303, R² = 0.092, F(1, 40) = 4.039, p = 0.051), and peak (r = -0.330, p = 0.033, n = 42; ß = -0.330, R² = 0.109, F(1, 40) = 4.878, p = 0.033). No effect of sex on the UR latency is found among the children. No correlations between age and UR onset or peak latencies, or effects of age and sex on the UR latencies are found among the adults. # Discussion This study demonstrates that as children get older their performance in eyeblink conditioning improves. Older children produced more CRs with better and less variable timing, and they also had earlier UR onsets and peaks. Adults showed earlier UR onsets and peaks than the children. Consistent with earlier results, adults reached higher rates of CRs than the children. Our school aged children never reached the same level as the adults, as infants earlier have been observed to do. We did not find any effects of age among the adult participants in our study. Yet, since the majority of the adults in this study were 20–30 years old, this is in line with results of other studies on adults that have reported a drop in performance in eyeblink conditioning only above 60 years of age. This declining performance in older adults has been attributed to age related degeneration of the cerebellum. In addition to the age effects, our study shows that females produced more CRs than males both among the children and adults. Sex differences are present in a variety of sensory and motor tasks. Animal studies show that adult females outperform age-matched males on a number of different learning tasks including classical conditioning. Sexual dimorphism of brain development, triggered by sex hormones may potentially contribute to these differences, although other factors, including genetic, social, and environmental factors, could also influence learning and consequently contribute to our results. In humans, cerebellar white and gray matter develop at different rates in girls and boys during childhood up to pre-puberty. Developmental processes, whichever their causes, could potentially explain the fact that the older (9 years or older) girls in our sample performed similar to the women and the older boys performed similar to the men, whereas the younger (below 9 years old) girls and boys produced fewer CRs. However, even though the females showed greater CR production, the increase during the session compared to the first block was not significantly greater among any of the sexes. Above this, the adult men had better timed CRs than the women during later parts of the acquisition. Post training, the boys showed later CR peaks than the girls, and the men showed less CR peak variability than the women. Sensitivity to the UR might be one factor that has an effect on the acquisition, and that perhaps differs between the sexes. ## Limitations It is legitimate to ask if the participants were properly conditioned. After training the children produced CRs on \~30% of the trials and the adults reached \~50% CRs. While this is low, these levels are above chance and they are similar to those reported in other studies on humans. A longer training period would probably have resulted in more CRs. However, to sit still for long sessions of eyeblink conditioning is generally hard, especially for young children. Another option would be to split the training into several shorter sessions, which has been observed to result in more CRs, for adults and for infants. On the other hand, in this and other studies, much of the learning occurs in the first block. Extra training might have caused the age and sex effects to level out. Indeed, male rats, despite their initial disadvantage, perform on par with females after a few days of training. Some caution about the sex effects among the adults is also warranted since we only tested eight adult men. The learning curves and percentage of CRs reached by different participants were highly variable. While some individuals produced close to 100% CRs, others barely produced any CRs at all. Similarly, some individuals showed incremental numbers of CRs, while others had flat, or even negative learning curves. In other words, the relatively smooth learning curves, in this and other studies, is often not representative of the individuals in different groups. The background sound of the movie, although soft, may have affected individuals differently, and could easily have been omitted to better control the test situation. Though, perhaps not without any impact on the concentration or motivation for some of the participants. A disadvantage in this study is that not all children were trained with identical protocols. However, despite protocol variations we did not find any statistical differences in percentage of CRs in either paired or CS alone trials between children trained with different protocols. We therefore considered grouping of the children justified although we cannot rule out potential impact on the results. # Conclusions Our results demonstrate that performance in eyeblink conditioning improves during development in childhood. This suggests that it may be possible to use performance in eyeblink conditioning as a measure of cerebellar maturity, or at least brain maturity. Moreover, our results indicate that sex is an important variable when it comes to eyeblink conditioning in humans. If our goal is to use eyeblink conditioning to explore cerebellar dysfunction in at-risk groups such as ADHD, ASD, FAS and SLI, disorders that are often more prevalent in males, it is important to consider the age and sex effects demonstrated here. Above that, we must be careful not to misinterpret immaturity as a disorder. # Supporting information We wish to thank Jan Willem Potters at Erasmus MC for technical assistance, Joost van de Weijer at Lund University for statistical consultation and Birgitta Sahlén and Magnus Lindgren for support and valuable contributions. Finally, our gratitude goes to the children and staff at the participating schools, as well as the adult participants and colleagues at the Humanities Laboratory at Lund University, for helping us conduct the study. [^1]: The authors have declared that no competing interests exist. [^2]: **Conceptualization:** KL AR GH. **Data curation:** KL RB AR HJB SK CDZ. **Formal analysis:** KL RB AR HJB SK. **Funding acquisition:** AR CDZ GH. **Investigation:** KL. **Methodology:** KL RB AR HJB SK CDZ GH. **Project administration:** KL GH. **Resources:** KL RB AR HJB SK CDZ GH. **Software:** HJB SK CDZ. **Supervision:** GH. **Validation:** KL RB AR HJB SK CDZ GH. **Visualization:** KL RB AR. **Writing – original draft:** KL. **Writing – review & editing:** KL RB AR HJB SK CDZ GH.

# Introduction Influenza virus mutates quickly and unpredictably creating emerging pathogenic strains that are difficult to detect, diagnosis, and characterize. Each year there are millions of flu cases and tens of thousands of deaths in the United States alone. The wide variety of circulating strains of Influenza virus at any given time contributes to the difficulty of selecting candidate strains for yearly seasonal flu vaccine development as well as making the study of Influenza virus as a species challenging. Strains may suddenly emerge through antigenic shift (reassortment) and become pandemic, such as the 1918 H1N1 “Spanish flu”, 1957 H2N2 “Asian flu”, or 2009 H1N1 “swine flu”. These pandemic strains often remain circulating in the population for decades, undergoing gradual antigenic drift to cause milder, yearly epidemics until they completely disappear from circulation. Influenza virus relies on two glycoproteins on its enveloped surface to successfully bind to and later release from epithelial cells of the host respiratory tract—hemagglutinin (HA) and neuraminidase (NA). Influenza HA’s are known to multivalently bind to sialic acid (SA) residues of the host cell’s glycocalyx during the first step of the viral infection cycle. This SA binding triggers viral entry through endocytosis and mediates subsequent endosomal membrane fusion resulting in the release of viral ribonucleoprotein into the host cell. One of the major differences among HA subtypes of Influenza virus is the preferential binding to a specific conformation of the sialic acid glycosidic linkage to glycolipids and glycoproteins of the host cell glycocalyx. For instance, HA subtypes of human Influenza A virus preferentially bind to sialic acid with an α2,6 glycosidic linkage which are abundant on the human pulmonary epithelium. Whereas HA subtypes of avian Influenza A strains prefer an α2,3 glycosidic linkage. This preferential receptor binding is likely one of the major barriers preventing the spread of highly pathogenic avian influenza A virus in humans. Conversely, Influenza B is an exclusively human circulating class of Influenza virus that has a much lower mutation rate in the HA encoded region of its genetic material. The Influenza B virus HA contains a conserved, narrower sialic acid binding site that preferentially binds to the α2,6 glycosidic conformation by discriminating against the avian α2,3 glycosidic linkage. However, there is evidence that the sialic acid residue preference of HA is not absolute. Species crossover has been proven to begin before changes in the HA binding domain occurs, such as during the 1997 Hong Kong bird flu outbreak caused by an avian H5N1 influenza A virus. All eight gene segments of the virus were of avian origin and maintained preference for binding the avian α2,3 glycosidic linkage, yet were able to cause significant pathogenicity in the human population. Better tools are needed to elucidate the mechanism of infection for these crossover events. Conventional tools to study and characterize virus, such as next generation sequencing, genome amplification (RT-PCR), and serological antibody testing, are not adequately suited to rapidly mutating pathogens like Influenza virus where the success of infection heavily depends on the phenotypic expression of surface glycoproteins. Bridging the gap between genome and pathogenic expression remains a challenge. Mutations in PCR primer or probe binding regions significantly impact diagnostic sensitivity and often result in false-negative results. And producing and harvesting strain specific antibodies is a labor and time intensive process plagued by low sensitivities. However, these techniques remain the gold standard for Influenza virus diagnosis, providing guidance for current epidemiological tracking and vaccine strain selection. Recent incremental progress on more complex techniques, such as digital electrochemical enzyme-linked immunoassay (ELISA), for Influenza virus which has a low detection limit. In addition, glycan microarrays are employed to determine binding specificities on a diverse set on glycan configurations, but the synthesis and/or purification of well characterized oligosaccharides may be time consuming and prohibitively complex for many diagnostic device labs. There remains a great need for a paradigm shift in approach to influenza virus diagnostic and characterization techniques. The future of single-molecule biosensors depends on high spatial resolution and immobilization of viable virus to achieve enhanced molecular information, dynamic interactions, and detection sensitivity. The development of open platform technologies is sorely needed to allow for the rapid testing of a variety of unconventional tools, from novel antigenic testing to spectroscopy techniques. Some groups have demonstrated the merit of such techniques by immobilizing isolated Influenza virus sourced antigens to study binding affinity and dynamic binding of Influenza HA and NA to sialic acid. Others have used an impedimetric-based detector to differentiate between species of Influenza A virus. Yet, to the best of our knowledge, universal Influenza binding for glycoprotein characterization studies on viable, whole virus has not been realized without the use of serotype specific antibodies or species-specific galactose residues. These studies would require the capture and immobilization of all Influenza virus present, agnostic to the specific strain or serotype. This has motivated the development of the novel capture coating in this study, allowing for universal Influenza virus immobilization. In addition, the evolution of the capture coating design was particularly motivated by the advancement of vibrational spectroscopy techniques, such as infrared spectroscopy, Raman spectroscopy, surface enhanced Raman spectroscopy, and tip enhanced Raman spectroscopy. These techniques are capable of detecting molecular-level phenotypic changes such as those that occur during a viral envelope protein mutation. For example, Sun et al. used surface enhanced Raman spectroscopy (SERS) based immunosensing techniques to detect clinically isolated Influenza A down to 10 pfu/mL. For the cleanest spectra, consideration must be given to the potential background signal created by the immobilization technique used. This motivated the pursuit of an avidin-biotin complex-based technique that can be layered onto optical spectroscopy compatible substrates, such as glass and sapphire, for greater flexibility in data acquisition techniques. Furthermore, application directly to glass and similar substrates would facilitate a variety of microfluidic integrations that befit from virus immobilization, including interrogation for virus characterization and diagnosis or high throughput drug screening. Towards this end, this study demonstrates the successful development of a novel universal Influenza virus capture coating that harnesses sialic acid binding to capture viable whole Influenza virus. A strong base layer of avidin biotin complex (ABC) binding combined with a biomimicking pegylated sialic acid tether is utilized to capture and immobilize the virus relative to the substrate in an aqueous environment without compromising the structural or functional integrity of the virus itself. In this study we confirm virus capture using fluorescent probes and atomic force microscopy (AFM). # Materials and methods ## Reagents and biologics Biotinylated bovine serum albumin (bBSA), avidin, blocker BSA, and serotype specific anti-Influenza virus FITC conjugated antibody probes were sourced from Thermo Fisher Scientific (PA1-73044, PA1-73047, PA1-73036). The custom assay probes are biotinylated polyethylene glycol (MW 2000) conjugated with sialic acid (bPEG_2kSA), produced by Nanocs, Inc. The non-binding assay probe control is biotinylated polyethylene glycol (MW 2000) conjugated with thiol (bPEG_2kSH), also produced by Nanocs, Inc. Virus strains were acquired from ATCC (IAV H1N1 A/Virginia/ATCC1/2009, IAV H3N2 A/Victoria/3/75, B Yamagata B/Wisconsin/1/2010, and B Victoria B/Florida/78/2015), rehydrated and diluted in PBS 1x and stored at -80°C with glycerol until thawed for experimental use. ## Avidin-biotin sialic acid capture coating fabrication 1000μg/mL bBSA in PBS was adsorbed onto hydrophobic substrate (black-walled non- treated polystyrene microplate or sapphire slide) via incubation at 37°C for 2 hours. Substrate was then rinsed with PBS and incubated with 40mg/mL blocker BSA in PBS for 1 hour at 37°C. Substrate was again rinsed and incubated with 100 μg/mL avidin in PBS for 1 hour at 4°C followed by another rinse and incubation with 10 μM bPEG_2kSA (or 10 μM bPEG_2kSH) in PBS for 1 hour at 4°C. Substrate underwent a final rinse with PBS, sealed, and stored at 4°C for up to four weeks without noticeable capture efficiency loss. ## Virus preparation On testing days, virus stock was removed from -80°C storage and thawed in a 37°C water bath. All stocks consisted of PBS-diluted active virus originating from either pooled allantoic fluid or MDCK propagation supernatant sourced from the supplier, ATCC. Prepared substrates were removed from 4°C storage at this time and brought to room temperature. Once thawed, virus stocks were further diluted in PBS as required to achieve final concentration and incubated on capture coated substrate at 37°C for 1 hour followed by a PBS wash. ## Capture coating optimization Concentrations of bBSA and avidin were systematically varied from 0 μg/mL to 1000μg/mL and bPEG_2kSA varied from 0 μM to 100 μM to optimize capture coating efficiency to the final concentrations described under Avidin-Biotin Sialic Acid Capture Coating Fabrication ( and). Serotype specific (H1N1, H3N2, B) anti-Influenza virus FITC conjugated antibody probes were used to detect Influenza virus immobilized on the capture coating. Fluorescent images were captured for each well of a black walled microplate during exposure with an X-CITE 120 fluorescent illuminator fitted with a 480 nm excitation filter with a focal point power of 8.1 mW. Emission was imaged with a SPOT Insight camera through a 40x Nikon Plan Fluor objective and SPOTAdvanced software set to a 519 nm monochrome colorizing palette. Relative fluorescent unit measurements were made using ImageJ opensource software on the captured images. Statistical p values shown in and were calculated using a Welch’s t-test with a threshold of α = 0.05 on the log₁₀ transformation of the fluorescence data. Fluorophore-conjugated antibodies have been determined to have intensity measurements that follow a lognormal distribution. The geometric mean was then calculated by taking the antilog of the mean log₁₀ transformed data and reported with the geometric standard deviation in and **.** ## Atomic force microscopy As an additional confirmation of capture coating efficacy, Atomic Force Microscopy (AFM) was used to image Influenza virus immobilized on the capture coating. A double side polished 460μm thick c-plane sapphire wafer (MSE Supplies) was cut into approximately 3/4” square pieces and layered with capture coating according to Avidin-Biotin Sialic Acid Capture Coating Fabrication protocol and incubated with Influenza virus according to Virus Preparation. Following incubation, slides were gently washed with sterile DI water and allowed to dry under the hood. Sapphire slides with immobilized virus were then placed on the stage of a Park Systems XE series AFM and imaged in soft non- contact mode using an approximately 30nm diameter cantilever probe (Applied NanoStructures, Inc.) with Park Systems XE software. AFM images were processed, and influenza virus particles counted and characterized using Gwyddion open source software for SPM data analysis. Each AFM image was leveled using Gwyddion mean plane subtraction, scars corrected, and background subtracted such that the minimum value was set at zero microns. Using the mark grains feature, a threshold was set to 100 nm height and a 70 nm equivalent radius (r_eq) filter applied to account for the 30nm radius AFM probe tip causing broadening edge artefacts of the typically 40–100 nm radius Influenza virus particles. Particles were counted by their r_eq properties. Those with a r_eq larger than 150 nm were double counted as a clustered pair of viruses. An example of this process is depicted in. In addition to the r_eq, height above capture coating was recorded for each virus. # Results and discussion ## Capture coating development and optimization The virus capture coating of the immobilization platform is comprised of bPEG_2kSA linked to bBSA with avidin. Concentration combinations of the three coating components (bBSA, avidin, and bPEG_2kSA) were investigated to find the optimal concentration based on viral binding performance. This was accomplished by first determining effective combinations of bBSA and avidin. Twenty-five concentration combinations of immobilized bBSA and FITC-conjugated avidin were evaluated by fluorescence imaging. Black-walled non-treated polystyrene microplates were used to prevent fluorescence emission from leaking into adjacent wells during imaging. Fluorescence readings for the FITC-biotin probe provided many promising combinations, as shown by high levels of fluorescence (green) in the upper left triangle of the test matrix. Higher fluorescence readings indicate a higher quantity of potential binding sites for the next layer of bPEG_2kSA tether. This agrees with other similar ABC based protocols in literature commonly using avidin to biotin ratios ranging from 1:1 to 1:10. Avidin-biotin binding was chosen as the base of our capture coating because of its many advantageous properties. It is one of the strongest known non-covalent bonds between a protein and ligand with a high degree of affinity (K_D ≈ 10⁻¹⁵ M) and specificity. Avidin and biotin are widely available reagents with an interaction that is stable over a wide range of temperatures and pH, providing a robust base that would allow for a variety of downstream testing on captured virus. Next, substrates were prepared with candidate bBSA/avidin combinations and incubated with different concentrations of bPEG_2kSA. The binding performance of each coating was assessed for three influenza virus strains (A H1N1, A H3N2, and B Yamagata) using serotype specific FITC conjugated anti- influenza probes. The FITC conjugated anti-influenza probes were raised against similar strains as each of the three used in this work (anti-H1N1, anti-H3N2, and anti-B Yamagata). shows the relative fluorescence intensity values for different concentrations of bPEG_2kSA with 10⁴ CEID₅₀/mL (IAV H3N2 and IBV Yamagata) or 10⁴ PFU/mL (IAV H1N1) of virus. These experimental strains were chosen as representatives of major Influenza serotypes currently endemic in the human population–IAV H1N1, IAV H3N2, and IBV Yamagata, respectively. Results indicate that each of the experimental strains of Influenza virus were best immobilized using 10 μM of bPEG_2kSA receptor with 1000 μg/mL bBSA and 100 μg/mL avidin, as indicated by having significantly higher fluorescent reading compared to the 0 μM bPEG_2kSA controls (IAV H1N1 p = 0.041, IAV H3N2 p = 0.032, IBV Yamagata p \< 0.001). In addition, IAV H3N2 showed significant capture via fluorescence detection at 100 μM bPEG_2kSA (p = 0.031), however this was not the case for the other two strains at the same receptor concentration. The high standard deviation of these fluorescence tests, especially at the higher 100 μM bPEG_2kSA concentration, may be due to steric hindrance effects caused by a high concentration of receptors. Steric hinderance may be why 10 μM of bPEG_2kSA receptor significantly outperformed a higher concentration of 100 μM bPEG_2kSA receptor. Other less successful avidin/bBSA combinations with bPEG_2kSA are not shown. Successful immobilization of the representative strains demonstrates proof of concept for this novel capture coating technique to be used universally with Influenza viruses type A and B. Furthermore, detection of Influenza in the concentration demonstrated, 10⁴ PFU/mL or CEID₅₀/mL depending on strain used, points toward potential clinical use of this novel immobilization platform. The limit of detection for point of care Influenza virus detection tests has been found to range from 5.4 to 8.9 log copies/mL and 4.8 to 7.3 log copies/mL for Influenza H3N2v and H7N9 viruses strains, respectively. These findings track well with 10⁵ to 10⁷ RNA copies per mL of mean viral load found in human nasopharyngeal isolates. Using an established conversion rate of 20–60 viral genome copies needed per PFU, this amounts to a typical clinical range of 10⁴ to 10⁶ PFU/mL in a clinical isolate. A control experiment using an alternative thiol-functionalized bPEG_2kSH tether was used to ensure that the Influenza virions were binding to the sialic acid tether and not simply being caught in the tendril-like structures of the capture coating. Results, show consistently lower fluorescence values in this non-binding control coating for all Influenza strains tested compared to the bPEG_2kSA tether. Using FITC probes as a quantitative analog to virus concentration was not possible because the different serotype specific anti-Influenza FITC probes used may vary in binding affinity and fluorescence levels across strains. To our knowledge, this is the first known successful use of the bPEG_2kSA molecule in an ABC-based capture coating designed specifically for universal immobilizing infectious virus. Past studies have often used fetuin and/or mucin as a viral receptor, due to an abundance of endogenous sialic acid end chains, in order to study Influenza virus binding characteristics. In these cases, the endogenous receptors do not permanently immobilize the virus due to the release mechanism of the vial NA protein. While useful to study binding kinetics, this release mechanism would hinder interrogation by any technique that requires a highly spatially resolved location of the virus, such as super resolution microscopies and spectroscopies. This effect may notably be incompatible in a high laminar flow environment such as a microfluidic. Specifically, in the case of molecular spectroscopies such as Raman spectroscopy and infrared spectroscopy, molecules like fetuin and mucin add an uncontrolled layer of variable molecular makeup that is much more difficult to deconvolve from spectra than well-regulated proteins and molecules like biotin, avidin, and PEG. A related pegylated sialic acid was recently used to study dynamic binding of HA and NA to sialic acid, including a demonstration of the irreversibility of HA binding in the absence of NA cleaving activity, also previously described by Guo et al.. Similarly, our biotinylated polyethylene glycol (MW 2000) conjugated with sialic acid (bPEG_2kSA) for virus capture was configured without the addition of a galactose-sialic acid linkage to prevent NA cleaving the virus from the pegylated sialic acid tether. This cleaving allows for the severance of progeny virus from the infected host cell *in vivo*. However, it has been shown that this action causes unstable binding of virus *in vitro* as HA mediated binding competes with NA mediated cleaving. Future iterations of this platform may explore incorporating a galactose-sialic acid functionalized end chain with either an α2,6 or α2,3 glycosidic linkage. This may allow for studying the HA mediated binding and NA mediated cleaving of Influenza strains originating from different host species. However, for the scope of this research, only human strains preferential to α2,6 were used, and the glycosidic linkage was not included in the bPEG_2kSA probe to prevent NA mediated cleaving. ## AFM visualization and analysis The developed capture coating was characterized using Atomic Force Microscopy (AFM) imaging and topography analysis. While AFM is not practical in a clinical diagnostic setting, it provides valuable validation of Influenza virus capture and allows for characterization of the capture coating itself which lead to the calculation of the coating’s *capture efficiency*. displays topographical images and profile analysis results comparing images of the sapphire substrate, capture coating on sapphire substrate without virus, a cluster of Influenza A H3N2 virions immobilized by capture coating on sapphire substrate, and a single Influenza A H3N2 virion immobilized by capture coating on sapphire substrate. The measured height of the individual virion, 36 nm, and diameter, 160 nm, agree well with the known size of Influenza virus given the relatively large diameter (\~30 nm) of the AFM probe tip causing a broadening edge artefact in the diameter measurement. Throughout these studies multiple 50 x 50 μm, 10 x 10 μm, and 1 x 1 μm AFM images were taken. Using Gwyddion software, virus height and radius were taken for each virus captured. These distributions are plotted in. Further, roughness parameters were extracted from these profiles using Gwyddion software and are summarized in. Differences in sapphire substrate and capture coating profiles support successful deposition of the capture coating onto the substrate. The average roughness (Ra) increased by approximately 24 nm which agrees well with the estimated 20–30 nm thickness of the bBSA-avidin- bPEG_2kSA coating combination. The regularly spaced and spiked profile of the capture coating also indicates successful adsorption of the bBSA and subsequent orientation of the avidin and bPEG_2kSA binding to create tendril-like protrusions from the substrate surface. These protrusions are better visualized surrounding the AFM image of a cluster of influenza virions which support the capture coating design and execution first depicted in. The apparent embedding of the virus into the capture coating is likely due to the brief drying out of the slides before AFM interrogation, causing the PEG tethers to crumple under the weight of the virion. It should be noted that features in the AFM topography images and subsequent roughness parameters may be smaller than they appear due to the relatively large diameter (\~30 nm) of the AFM probe tip causing broadening edge artefacts. ## Capture efficiency The capture coating of the virus immobilization platform utilized a unique bPEG_2kSA linker that provides a biomimicking sialic acid receptor for Influenza HA to bind to. Capture efficiency was calculated for each of the four virus strains. When the total number of virus particles applied to the substrate is less than that which can fully occupy the substrate surface, capture efficiency (C_Eff) is defined as the ratio of virus particles captured over virus particles applied. <img src="info:doi/10.1371/journal.pone.0247429.e001" id="pone.0247429.e001g" /> C E f f = P b o u n d P t o t a l = S \* C n \* I Where the number of virus particles bound to the substrate, P_bound = S\*C with S being the substrate size factor (total substrate area/size of the AFM scan area) and C the virus particle count from AFM image. To determine the virus particle count from an AFM image, a 10 x 10 μm² AFM scan was analyzed for each of the four Influenza virus strains. Virus particles present were counted using Gwyddion’s mark grain feature with minimum thresholds set at 120 nm height and 70 nm equivalent radius to account for the 30nm radius AFM probe tip causing broadening edge artefacts. A demonstration of particle counting via AFM image with Gwyddion mark grain feature is provided as. For each Influenza strain, n = 3 AFM images were analyzed for particle count and the mean taken. The total number of virus particles applied to the substrate P_total = n\*I, where n is the number density of virus particles in the stock solution and I is the volume of sample applied to the substrate. The number density is calculated from the infectious virus concentration of the virus stock \[V\] multiplied by a conversion factor. For virus concentration in PFU/ml, n = \[V\]\*CTPR, where CTPR is the virus particle “count to PFU ratio” found in previous studies. When the virus concentration is CEID₅₀/ml, n = \[V\]\*PtC\*CTPR, where PtC is the PFU to CEID₅₀ ratio conversion factor. Results and inputs parameters used to calculate the *capture coating efficiency* are detailed in. For this work, it is estimated that the total number of virus particles incubated for 1 hour on a functionalized slide (P_total = 2.29x10⁷, 5.14x10⁶, 2.00x10⁸, 1.39x10⁸) was less than the total number of available binding sites (I_total = 1.03x10¹⁰, 9.02x10⁸, 1.03x10¹⁰, 1.03x10¹⁰) for IAV H3N2, IAV H1N1, IBV Yamagata, and IBV Victoria, respectively. Capture efficiency (C_Eff) ranged from 90.0% to 99.7%. These promising results are likely due to the irreversibility of HA binding to sialic acid in absence of active NA cleaving or due to strong avidity from bivalent or multivalent binding of sialic acid by individual Influenza virions. Multivalent binding of Influenza virus to our capture coating may be caused by multiple bPEG_2kSA binding to a single avidin molecule in the ABC complex. Alternatively, the 333 nm peak spacing found via the AFM topography data had a standard deviation of 90 nm. The influenza virus may also be showing preference to binding locations where multiple bPEG_2kSA binding locations are within reach, providing access for multivalent binding. These results demonstrate the strong immobilization capability of this novel bPEG_2kSA capture coating. However, this strong binding capability was generated under ideal, static incubation conditions and should be further vetted through flow cell observations beyond the scope of this study. Sialic acid is a common binding receptor used by a variety of viruses to initiate host cell entry. Since the goal of this research was to create a universal infectious virus Influenza immobilization platform, the sialic acid functionalized bPEG probe was synthesized without galactose and, therefore, without the NA glycosidic linkage cleaving site. While this work was limited to just a single conformation of bPEG_2kSA, advanced binding studies may be done by altering the glycan chemistry, such as functionalizing the PEG with a α2,6 glycosidic conformation favored by human pathogenic Influenza virus or, alternatively, an avian Influenza favored α2,3 glycosidic linkage. In addition, further studies may open the possibility to use this capture coating with other SA binding virus families such as adenovirus and coronavirus, amongst others. Like Influenza virus (K_d = μM-pM), adenovirus (K_d = 19 μM), and some coronaviruses (K_d = 49.1 μM) *multivalently* bind to SA with K_d in a high affinity range. Together, these fluorescence and AFM studies show the first successful use of a novel ABC based sialic acid capture coating for universal Influenza virus immobilization and its potential as a diagnostic tool platform. An avidin-biotin complex base with a biotin-PEG-sialic acid functionalized surface was utilized to create an Influenza virus capture coating capable of immobilizing whole virus while keeping the virion intact. Atomic Force Microscopy studies confirmed and described the profile of the capture coating as well as demonstrated the ability of using the capture coating to help characterize a single Influenza virus by AFM topography. From AFM topography data we were able to determine the capture efficiency of the capture coating to be above 90% of virus particles at a concentration of 10⁵ CEID₅₀/mL. We hope this capture coating technique creates an adaptable platform for further characterization of virus and development of novel phenotype diagnostic techniques. # Supporting information The authors would like to acknowledge the contributions of Dr. Gina Shreve and Chris Thrush. 10.1371/journal.pone.0247429.r001 Decision Letter 0 Dague Etienne Academic Editor 2021 Etienne Dague 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 Dec 2020 PONE-D-20-36336 Avidin-Biotin Complex Based Capture Coating Platform for Universal Influenza Virus Immobilization and Characterization PLOS ONE Dear Dr. Trexler, 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 one questions the novelty. Please emphasize on the originality. Reviewer 2 is concerned about the AFM images and wonder if the particules are really virus. Please take care to consider this comment carefully. Please submit your revised manuscript by Feb 07 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, Etienne Dague, 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\. Our staff editors have determined that your manuscript is likely within the scope of our Call for Papers on Influenza. This editorial initiative is headed by PLOS ONE Guest Editors Dr. Meagan Deming and Dr. Deshayne Fell. The Collection encompasses research on influenza prevention on every level, including in vitro, translational, behavioral, and clinical studies; disease and immunity modelling; as well as new approaches to influenza prevention. Additional information can be found on our announcement page: <https://collections.plos.org/call-for-papers/influenza/>. Currently, your manuscript is included in the group of papers being considered for this call. Please note that being considered for the Collection does not require additional peer review beyond the journal’s standard process and will not delay the publication of your manuscript if it is accepted by PLOS ONE. We would greatly appreciate your confirmation that you would like your manuscript to be considered for this Collection by indicating this in your next cover letter. If you would prefer to remove your manuscript from collection consideration, please specify this in your cover letter. 3\. To comply with PLOS ONE submission guidelines, in your Methods section, please provide additional information regarding your statistical analyses. In addition, please report your p-values to support your claims. For more information on PLOS ONE's expectations for statistical reporting, please see <https://journals.plos.org/plosone/s/submission-guidelines.#loc-statistical- reporting>. \[Note: HTML markup is below. Please do not edit.\] 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 \*\*\*\*\*\*\*\*\*\* 2\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: No Reviewer \#2: No \*\*\*\*\*\*\*\*\*\* 3\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 4\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 5\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: This manuscript reports influenza virus capturing system using avidin-biotin complex and sialic acid (SA). The authors optimized the surface functionalization protocol and suggested the evidences (AFM and fluorescence) of virus capturing. SA is a well-known universal receptor for influenza viruses and avidin-biotin interaction has been widely used in biological experiments. Although the authors claimed that this is the first known successful use of the bPEG2kSA molecule in an ABC-based capture coating designed specifically for immobilizing infectious virus, I think the novelty of this work is insufficient to publish in PLOS one. 1\. What is the main purpose of this system? For the diagnosis of influenza viruses, there are several methods. I should be compare the current method with pervious approaches. If the development of surface functionalization method is the key achievement, it will be nice to apply the method to various sensing surfaces. 2\. In Table 1, why the fluorescence signals are so irreproducible? Error data is even higher than the average values. 3\. For the calculation of capturing efficiency, the related AFM data should be provided. Reviewer \#2: The authors provide an interesting work about capturing flu virus. The topic is appealing an interesting, but I found the AFM part a little weak. In particular, figure 2c pretends to be an individual virus particle. However, the height provided by the topographical profile reaches almost 500 nm. It is known that influenza virus is a pleomorphic virus whose diameter is around 100 nm (Biophysical Journal (2014) 106(7) 1447–1456), but the AFM topographical height ranges between 60nm and 120 nm. In fact, viruses height on the surface can be more or less collapsed due to the virus-surface interaction (Current Opinion in Virology Volume 18, June 2016, Pages 82-88, Seminars in Cell & Developmental Biology Volume 73, January 2018, Pages 199-208). I wonder if what they show in this figure is a virus. The authors should provide more statistics about the AFM data, such as size and height distributions. Supplementary figure S3 shows a region with several viruses. How is the statistics of virus heights on the surface? How many individual viruses are like this shown in figure 2? \*\*\*\*\*\*\*\*\*\* 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.0247429.r002 Author response to Decision Letter 0 20 Jan 2021 The following is copied from the enclosed form "Response to Reviewers". Thank you again for your time, consideration, and insightful comments and recommendations that have led to the improvement of our manuscript. Dear Dr. Dague and Reviewers, Thank you for giving us the opportunity to submit a revised draft of the manuscript “Avidin-Biotin Complex Based Capture Coating Platform for Universal Influenza Virus Immobilization and Characterization” for publication in PLOS ONE. We are grateful of the time and effort that you and the reviewers have dedicated to providing feedback on our manuscript and are appreciative of the insightful comments which have contributed to valuable improvements to our paper. We have incorporated many of the suggestions made by the reviewers and have supplemented our analysis and discussion with emphasis on originality and AFM evidence for the presence of virus. Please see below for a point-by-point response to the reviewers’ comments and concerns. All line numbers refer to the revised manuscript with tracked changes. Sincerely, Micaela Trexler, M.Eng Doctoral Candidate Biomedical Engineering Wayne State University Detroit, MI 48202 Reviewers’ Comments to the Authors: Reviewer 1 1\. Comment: What is the main purpose of this system? For the diagnosis of influenza viruses, there are several methods. Compare the current method with pervious approaches. Author Response: Thank you for pointing out the weakness in our claims on originality and for giving us the opportunity to emphasize the novelty and purpose of this work, especially as it compares to current and previous approaches. We have addressed this concern on multiple fronts. On the comparison to current diagnostic techniques and how our capture coating may integrate into a novel diagnostic set up, we have added clarifying language and strategic emphasis on the adaptable potential of this work to second half of the Introduction section, lines 70 to 117. These paragraphs detail the pros and cons of existing diagnostic techniques (PCR, ELISA, and glycan microarrays) and the progress, potential, and, up to this point, difficulty of molecular spectroscopy techniques that require universal Influenza virus immobilization. Studies using our novel bPEG2kSA capture coating in conjunction with nano- resolution vibrational spectroscopy to differentiate between Influenza virus strains are ongoing, but we felt are outside the scope of this stand-alone manuscript describing the capture coating itself. Regarding the novelty of the capture coating itself, we have addressed this in a new paragraph in the Results and Discussion section, lines 228 to 238. Here we reference previous studies by Lai et. al (2019), Onkhonova et. al (2019), and Sakai et. al (2017) that used fetuin or mucin as a common method of studying virus binding characteristics and why these endogenous receptors would not be ideal for our future diagnostic purposes: “Past studies have often used fetuin and/or mucin as a viral receptor, due to an abundance of endogenous sialic acid end chains, in order to study Influenza virus binding characteristics\[31-33\]. In these cases, the endogenous receptors do not permanently immobilize the virus due to the release mechanism of the vial NA protein. While useful to study binding kinetics, this release mechanism would hinder interrogation by any technique that requires a highly spatially resolved location of the virus, such as super resolution microscopies and spectroscopies. This effect may notably be incompatible in a high laminar flow environment such as a microfluidic. Specifically, in the case of molecular spectroscopies such as Raman spectroscopy and infrared spectroscopy, molecules like fetuin and mucin add an uncontrolled layer of variable molecular makeup that is much more difficult to deconvolve from spectra than well-regulated proteins and molecules like biotin, avidin, and PEG.” In addition, in the following paragraph lines 239 to 243, we reference the only previous study (Guo et. al 2018) we are aware of that uses a pegylated sialic aid conjugate for Influenza virus immobilization. We use this study as support for the receptor binding functionality of our bPEG2kSA end chain, though our study goals are different. 2\. Comment: In Table 1, why is the fluorescence signal error data so high? Author Response: We agree with your assessment that the standard deviation of our fluorescence experiments summarized in Table 1 is high and have addressed potential reasons with the addition of lines 206 to 209 in the revised manuscript: “The high standard deviation of these fluorescence tests, especially at the higher 100 µM bPEG2kSA concentration, may be due to steric hindrance effects caused by a high concentration of receptors. Steric hinderance may be why 10 µM of bPEG2kSA receptor significantly outperformed a higher concentration of 100 µM bPEG2kSA receptor.” Reflection on this also motivated us to solidify a claim of significance with new statistical analysis (Welch’s t-test, described in Methods section, lines 152 to 156), which required handling the statistic calculations with data from the lognormal distribution typical of fluorescence tagged data. This allowed us to claim the optimized bPEG2kSA concentration of 10 µM had “significantly higher fluorescent reading compared to the 0 µM bPEG2kSA controls (IAV H1N1 p =0.041, IAV H3N2 p = 0.032, IBV Yamagata p \< 0.001).” (revised manuscript lines 203-204) despite the high standard deviation of the data. Values that met the significance criteria are now marked with a “\*” in Table 1 and S1 Fig. To reflect these changes, we report the data as geometric mean and standard deviation. 3\. Comment: For the calculation of capturing efficiency, the related AFM data should be provided. Author Response: Thank you for this observation that made us realize our mistake in erroneously omitting the proper standard deviation data in the “Mean Particle Count” column of Table 3 which is used to calculate the capture efficiency. This data has been added. Further improvements have been made to the AFM section of this manuscript in regard to additional analysis, characterization, and visualization of individual viruses and the capture coating. In the interest of brevity, please refer to our responses to the specific requests of Reviewer 2. Reviewer 1, thank you again for your invaluable insight and questions pertaining to the originality and data behind this work. We hope we have satisfactorily addressed your concerns on the novelty of our bPEG2kSA and ABC-based capture coating and readily invite additional feedback as you see fit. Reviewer 2 1.Comment: The dimensions of the virus particle in Fig 2c does not match established reference work. Authors Response: We agree with your assessment, especially as it pertains the topographical profile of the sample in Fig 2C that we had erroneously claimed to be a single virion. In response to your insight, we thoroughly reviewed our many AFM images taken over the course of our experiments and determined this image to be a cluster of Influenza viruses, not a single virion. We have added Fig 2D to our revised manuscript which provides evidence of a single virion immobilized by the capture coating with a height of 36 nm and a diameter of 160 nm that better matches the reference studies you provided, given the probable collapsed height due to virus-surface interactions and the edge broadening artefacts of our 30 nm AFM probe. These corrections, with suggested references (de Pablo et. al 2017, Li et. al 2014, and Marchetti et. al 2016), have been added to the revised manuscript, both in a revised Fig 2 and in lines 255 to 261: “Fig 2 displays topographical images and profile analysis results comparing images of the sapphire substrate (Fig 2A), capture coating on sapphire substrate without virus (Fig 2B), and a cluster of Influenza A H3N2 virions immobilized by capture coating on sapphire substrate (Fig 2C), and a single Influenza A H3N2 virions immobilized by capture coating on sapphire substrate (Fig 2D). The measured height of the individual virion, 36 nm, and diameter, 160 nm, agree well with the known size of Influenza virus given the relatively large diameter (\~30 nm) of the AFM probe tip causing a broadening edge artefact in the diameter measurement. \[37-40\].” In addition, we added a new column to Table 2 with the corresponding characterizing surface parameters for the individual virion in Fig 2D. We feel it is important to keep Fig 2C as it provides a wonderful visual of the tendril- like structure of the capture coating. However, now it is properly and clearly described as capturing a cluster of viruses as to not mislead the reader. 2\. Comment: How many individual viruses are like this shown in figure 2? Author Response: The vast majority of the virions captured by our coating and imaged using AFM closely resembled the newly referenced profiles in conjunction with Fig 2D with heights in the 20 to 60 nm range and radii in the 35 to 70 nm range. We provided evidence for this in response to your next comment below. In all, we were able to evaluate the height and radius dimensions of 218 particles of Influenza A H1N1, 88 of Influenza A H3N2, 93 of Influenza B Victoria lineage, and 42 of Influenza B Yamagata lineage. 3\. Comment: The authors should provide more statistics about the AFM data, such as size and height distributions. Supplementary S3 Fig shows a region with several viruses. How is the statistics of virus heights on the surface? Author Response: Prompted by this suggestion, we took advantage of the mark grain feature in the Gwyddion software, as used in the capture coating efficiency related particle counts detailed in S3 Fig, and applied the same applicable thresholds to all AFM images captured over the course of our various experiments. This included all the 10 x 10 µm scans used in the Ceff calculations (including the one shown in S3 Fig), gross 50 x 50 µm scans, and detailed 1 x 1 µm scans. Radius and height for each marked grain (i.e. virus particle) were gathered using Gwyddion’s distributions of various grain characteristics feature and summarized in the histogram plots of the new S4 Fig for each of the four Influenza stains used throughout our experiments. Reviewer 2, thank you again for your insight and for challenging us to look closer at our AFM data. We believe these changes have improved our manuscript greatly and hope we have addressed your concerns aptly. We gladly invite additional feedback as needed. 10.1371/journal.pone.0247429.r003 Decision Letter 1 Dague Etienne Academic Editor 2021 Etienne Dague 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. 8 Feb 2021 Avidin-biotin complex-based capture coating platform for universal Influenza virus immobilization and characterization PONE-D-20-36336R1 Dear Dr. Trexler, 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, Etienne Dague, 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 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: Partly \*\*\*\*\*\*\*\*\*\* 3\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes Reviewer \#2: 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 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's responses solved the concerns of reviewers I think that the manuscript now suitable for PLOS ONE Reviewer \#2: The authors have replied to my questions. However, figure 2 is complicated. The font size of the charts is very narrow and difficult to read. I am still missing some statistical AFM data analysis in the main 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 Reviewer \#2: No 10.1371/journal.pone.0247429.r004 Acceptance letter Dague Etienne Academic Editor 2021 Etienne Dague 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 Feb 2021 PONE-D-20-36336R1 Avidin-biotin complex-based capture coating platform for universal *Influenza virus* immobilization and characterization Dear Dr. Trexler: 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. Etienne Dague Academic Editor PLOS ONE [^1]: The authors have declared that no competing interests exist.

# Introduction In mammals, the innate immune system is the first line of host defense against invading pathogens and is mediated by innate immune cells including macrophages. Macrophages are critical effector cells contributing to the innate immune response against infection, as they are the most efficient pathogen scavengers and the predominant source of pro-inflammatory cytokines such as IL-1β, IL-6 and TNF-α, which are pivotal to promote inflammation at the site of infection and fight against pathogens. However, excess or inappropriate production of pro- inflammatory cytokines has serious consequences including tissue damage and septic shock. Therefore, inhibiting the synthesis or release of these cytokines and other cellular mediators emerges as a potential therapeutic approach for septic shock-like diseases associated with inappropriately amplified inflammatory responses. The Toll-like receptor (TLR) family is one of the best-characterized pattern recognition receptor (PRR) families and is responsible for sensing invading pathogens. Recognition of bacterial components such as lipopolysaccharide (LPS) by TLRs results in the recruitment of multiple cytoplasmic signaling molecules involving MyD88, TRAF6, TIRAP and IRAK, which eventually activate downstream signaling components such as SAPK/JNK, p38 and NF-κB. As an E3 ubiquitin ligase, TRAF6 interacts with various protein kinases including IRAK, SRC, and apoptosis signal-regulating kinase 1 (ASK1) and provides a link between distinct signaling pathways. ASK1, a MAP kinase kinase kinase, plays an essential role in cytokine- and stress-induced apoptosis in mammalian cells by activating the MAP kinase kinase 4 (MKK4) and MKK3, which in turn lead to the activation of JNK and p38 pathways. Recent studies have shown that ASK1 formed a complex with TRAF6 in response to LPS, and this complex formation and subsequent activation of p38 pathway required LPS-induced production of reactive oxygen species (ROS). Chromones have been extensively studied as bioactive compounds. They possess remarkable biological activities, including potent anti-inflammatory actions. Our previous results revealed the diphenolic chromone derivatives as a new class of anti-inflammatory leads which showed potent inhibitory activity on nitric oxide (NO) production. In this study, we investigated anti-inflammatory property of (E)-5,7-dihydroxy-3-(3-oxo-3-phenylprop-1-en-1-yl)-4H-chromen-4-one (DCO-6), a novel diphenolic chromone derivative. The results indicate that DCO-6 significantly reduced LPS-induced production of NO, IL-1β and IL-6 and decreased the levels of iNOS, IL-1β and IL-6 mRNA expression in macrophages. Moreover, LPS-induced activation of p38 MAPK was remarkably impaired by DCO-6 due to its inhibitory effects on the production of ROS and formation of TRAF6-ASK1 complex. In addition, DCO-6 alleviated LPS-induced mortality in murine model of septic shock, suggesting that DCO-6 may have use as a treatment for inflammatory diseases. # Results ## DCO-6 inhibits LPS-induced NO, IL-1β and IL-6 Production in RAW264.7 and Peritoneal Macrophages The synthesis of (E)-5,7-dihydroxy-3-(3-oxo-3-phenylprop-1-en-1-yl)-4H-chromen-4- one (DCO-6) is outlined in, and the ¹H NMR spectrum of DCO-6 was shown in. DCO-6 at concentrations ranging from 1 to 30 µM failed to affect cell viability of both mouse RAW264.7 macrophages and primary mouse peritoneal macrophages. Thus, up to 30 µM of DCO-6 was used in the following *in vitro* experiments. To assess the effects of DCO-6 on LPS-induced production of cellular mediator in RAW264.7 and peritoneal macrophages, cell culture medium was harvested. Measuring nitrite as the index of NO production by the Griess method, we found that LPS treatment for 24 h resulted in a large amount of NO release in macrophages. Co-incubation of DCO-6 with LPS inhibited the formation of NO in a concentration-dependent manner. In addition, LPS-induced production of IL-1β and IL-6 was significantly reduced in macrophages treated with DCO-6. When RNA was isolated and quantitative real-time PCR was performed to examine the effects of DCO-6 on gene expression, Co-incubation of DCO-6 with LPS also decreased the levels of iNOS, IL-1β and IL-6 mRNA expression, suggesting that NO, IL-1β and IL-6 reduction by DCO-6 may be related to transcriptional inhibition. However, neither TNF-α release nor mRNA induction was altered by DCO-6 at concentration of up to 30 µM. ## DCO-6 Inhibits the Activation of p38 MAPK as well as NF-κB Signaling Induced by LPS in RAW264.7 Cells In LPS signaling, activation of mitogen-activated protein kinases (MAPKs) and transcription factor NF-κB play essential roles in transcriptional induction of those genes involved in inflammation, such as iNOS, COX-2, TNF-α, IL-1β and IL-6. Here, we assessed the effects of DCO-6 on activation of MAPKs in RAW264.7 upon response to LPS. As shown in, DCO-6 inhibited p38 MAPK phosphorylation induced by LPS in a concentration-dependent manner without any effect on total p38 MAPK expression. In contrast, DCO-6 did not affect phosphorylation of JNK, ERK induced by LPS. In spite of the failure in inhibition of phosphorylation of IKKα/β, DCO-6 slightly reduced phosphorylation of IκBα. Moreover, the nuclear localization of p65 subunit of NF-κB was inhibited by DCO-6, in line with the blockade of DNA binding activity of p65 in LPS-stimulated RAW264.7 cells. ## DCO-6 Inhibits p38 MAPK Activation Dependent on TLR Ligands in RAW264.7 Cells Next we used LPS (a ligand for TLR4), poly (I:C) (a synthetic double-stranded RNA and a ligand for TLR3) or unmethylated CpG (a ligand for TLR9) to stimulate RAW264.7 cells. As shown in, p38 MAPK phosphorylation was substantially induced upon any stimuli. However, DCO-6 failed to reduce p38 phosphorylation induced by poly (I:C) or CpG, suggesting that DCO-6 specifically inhibited TLR4-dependent p38 activation. In an *in vitro* kinase assay to determine the direct effect of DCO-6 on the p38 MAPK, we found that, unlike p38 MAPK inhibitor SB203580, which can inhibit p38 MAPK at 10 µM, DCO-6 did not affect the kinase activity of p38 MAPK. ## DCO-6 Inhibits the Production of Intracellular ROS and Formation of the TRAF6-ASK1 Complex in RAW264.7 Cells Given that the generation of ROS is required for TLR4-dependent activation of p38 but not JNK, we examined whether DCO-6 inhibited LPS-induced phosphorylation of p38 by reducing ROS production. After treatment for 6 h, LPS_, but not poly (I:C) or CpG, resulted in a large amount of ROS production in RAW264.7 cells. DCO-6 inhibited LPS-induced ROS production in a concentration-dependent manner and 30 µM of DCO-6 showed the inhibitory activity comparable to 1 mM of the antioxidant N-acetyl-L-cystein (NAC). In addition, 1 mM of H₂O₂ in RAW264.7 cells induced p38 phosphorylation and Co- incubation of DCO-6 with H₂O₂ inhibited the phosphorylation of p38 in a concentration-dependent manner. It is known that ROS mediates LPS-induced p38 activation by inducing the formation of the TRAF6-ASK1 complex. Then, we examined the effects of DCO-6 on interactions between TRAF6 and ASK. As shown in, TRAF6 did not coimmunoprecipitate ASK1 in the absence of LPS stimulation, whereas there was definite interaction between TRAF6 and ASK1 in RAW264.7 cells upon LPS stimulation. DCO-6 at 30 µM had no effect on the protein expression of TRAF6 and ASK1, but completely disrupted their interaction. ## DCO-6 Protects Mice from LPS-induced Septic Shock with a Significant Inhibition of p38 Activation To assess the protective effects of DCO-6 against LPS-induced lethal shock *in vivo*, BALB/c mice were intraperitoneally administered LPS (10 mg/kg) and DCO-6 (10 and 20 mg/kg). Cumulative proportions of mice surviving after lethal dose of LPS were shown in. Single administration of DCO-6 improved survival in a dose- dependent manner. At 48 h after LPS injection, survival rate of DCO-6 treated mice (20 mg/kg) was 50% while none of untreated mice were alive. Moreover, LPS- induced IL-6 and IL-1β were significantly reduced in serum of DCO-6-treated groups. In addition, inhibition of p38 activation and ROS production was also observed in peritoneal macrophages from septic mice treated with DCO-6. # Discussion This study examined the anti-inflammatory effect of the synthetic chromone derivative DCO-6 and its underlying mechanism. Our results suggest that the inhibition of TRAF6-ASK1-p38 signaling pathway by DCO-6 contributes to its anti- inflammatory action in reducing LPS-induced production of NO, IL-1β and IL-6 in macrophages and protecting against LPS-induced septic shock in mice. In contrast, DCO-6 did not affect the other two MAPK signaling cascades, ERK and JNK, elicited by LPS. LPS-induced activation of p38 MAPK in macrophages has been widely demonstrated to correspond to the *in vitro* and *in vivo* effects of LPS on positive regulation of a variety of genes involved in inflammation. p38 MAPK signaling is implicated in LPS-induced transcriptional activation of pro-inflammatory genes, such as iNOS, COX-2, IL-1β and IL-6. DCO-6 concentration-dependently inhibited LPS-induced p38 phosphorylation, an index of p38 MAPK activation. Accordingly, DCO-6 substantially decreased the levels of iNOS, IL-1β and IL-6 mRNA expression in macrophages. However, TNF-α production and mRNA induction was unaffected by DCO-6 *in vitro*. Multiple MAPK pathways activated by LPS are involved in the regulation of TNF-α expression. It is possible that the p38 inhibition alone can not lead to significant attenuation of TNF-α transcriptional activity. Consistent with this finding, the selective p38 inhibitor SB203580 shows ineffectiveness on TNF-α promoter activity. Despite the inhibition of p38 MAPK signaling induced by LPS, DCO-6 is not a direct inhibitor of p38 MAPK’s catalytic function as evidenced by the *in vitro* kinase assay, where DCO-6 did not affect the kinase activity of p38 MAPK. Upon stimulation by different TLR ligands, DCO-6 specifically inhibited TLR4-dependent p38 activation, suggesting that DCO-6 might block upstream events required for LPS-induced p38 MAPK activation. Indeed, we found that LPS_, but not poly (I:C) or CpG, resulted in a large amount of ROS production. And DCO-6 significantly inhibited ROS production in macrophages stimulated by LPS. Accumulating evidence has shown ROS are not only injurious by-products of cellular metabolism but also essential participants in cell signaling and regulation. In LPS signaling, ROS can selectively mediate the formation of a complex between TRAF6 and the redox-sensitive ASK1, which in turn triggers p38 activation. In current study, DCO-6 completely disrupted the interaction between TRAF6 and ASK1 in RAW264.7 cells upon LPS stimulation, although the expression of TRAF6 and ASK1 protein was unaffected. These results indicate that DCO-6 may inhibit ROS-dependent activation of TRAF6 -ASK1-p38 pathway which leads to the production of inflammatory factors. p38 MAPK’s role in the regulation of inflammatory cytokines and enzymes responsible for inflammation–like COX2, iNOS and MMPs–makes it an attractive drug target. Several small molecular inhibitors targeting p38 have been developed and evaluated in animal models of inflammatory diseases. In this study, treatment with DCO-6 alleviated LPS-induced septic shock in mice and inhibited p38 activation *in vivo*. The anti-inflammatory potential of an oral p38 MAPK inhibitor was also evaluated in human endotoxemia. Despite the strong rationale for MAPK inhibitors in human disease, direct proof of concept in the clinic has yet to be demonstrated, with most compounds demonstrating dose- limiting adverse effects. Identification of new classes of compounds with greater specificity or a deeper knowledge of the other targets in p38 MAPK pathway could overcome these side-effects. Selective inhibition of more upstream events involved in p38 MAPK signaling may be a feasible strategy. Although chromones and their structural analogues are known to play an important protective role against oxidation processes, many chromone-based compounds have demonstrated more fascinating properties. For example, it is reported that cromoglycate, a chromone complex, increases survival during an experimental model of sepsis by inhibiting HMGB1 release. Unlike DCO-6, cromoglycate showed no effect on the LPS-induced p38 phosphorylation. 30 µM of DCO-6 slightly, but significantly, reduced DNP-induced β-hexosaminidase release in Rat basophilic leukemia RBL-2H3 cells. The results indicate that DCO-6 might have inhibitory effect on anaphylactic response through a different pathway from mast cell stabilizer cromoglycate. Recent study has shown that some chromone derivatives were synthesized and evaluated as p38 MAP kinase inhibitors. However, the novel synthetic compound DCO-6 did not affect the kinase activity of p38 MAPK. DCO-6 showed low cytotoxicity and strong anti-inflammatory activity. Blocking the upstream events required for p38 MAPK action by DCO-6 may provide a new therapeutic option in the treatment of human inflammatory diseases. # Materials and Methods ## Cells and Reagents Murine RAW264.7 macrophages, obtained from the American Type Culture Collection (Rockville, MD), were cultured in DMEM (Invitrogen Corp., Carlsbad, CA) containing 5% fetal bovine serum (GIBCO, Grand Island, NY), 100 U/ml penicillin, and 100 µg/ml streptomycin in 5% CO₂ at 37°C. Peritoneal macrophages elicited by thioglycollate broth (Sigma, St Louis, MO) were harvested by lavage of the peritoneal cavity. LPS from *Escherichia coli* (0111:B4), poly (I:C) and unmethylated CpG were purchased from Sigma, Invitrogen and Invitrogen, respectively. All other chemicals were obtained from Sigma. ## Mice Male BALB/c, 6–8 weeks of age, were purchased from Experimental Animal Center of Jiangsu Province (Jiangsu, China). They were maintained with free access to pellet food and water in plastic cages at 21±2°C and kept on a 12 h light/dark cycle. Animal welfare and experimental procedures were carried out in accordance with the Guide for the Care and Use of Laboratory Animals (Ministry of Science and Technology of China, 2006) and the related ethical regulations of our university. All the animal experiments were approved by Nanjing University Animal Care and Use Committee (NJU-ACUC) and made to minimize suffering and to reduce the number of animals used. ## Measurement of Cytotoxicity, Nitrite and Cytokines Cells (5×10⁴ per well) were cultured in a 96-well plate, and treated with various concentrations of DCO-6. After 24 h of incubation, LDH in the culture supernatant and in cell lysate were tested by using the CytoTox 96® Non- Radioactive Cytotoxicity Assay (Promega Corp., Shanghai). The percentage of LDH released from the cells was determined using the formula: % release  =  LDH activity in supernatant/(LDH activity in supernatant + LDH activity in cell lysate). For nitrite measurement, cells were treated with various concentrations of DCO-6 in the absence or presence of LPS (500 ng/ml) for 24 h, and then nitrite in the culture supernatant was measured by Griess reagent (Promega). And the nitrite concentration was calculated by using standard solution of sodium nitrite in the culture medium. For cytokine assay, IL-1β and IL-6 production were measured using ELISA kits from R&D systems (Minneapolis, MN) according to the manufacturer’s instructions. ## Real-time PCR Real-time PCR was performed as described previously. Briefly, RNA samples were treated by DNase and subjected to quantitative PCR, which was performed with the ABI Prism 7000 sequence detection system (Applied Biosystems, Foster City, CA) using SYBR Green I dye (Biotium, Inc.), and threshold cycle numbers were obtained using ABI Prism 7000 SDS software version 1.0. Conditions for amplification were 1 cycle of 94°C for 5 min followed by 40 cycles of 94°C for 30 s, 58°C for 30 s, and 72°C for 45 s. The primer sequences used in this study were as follows: iNOS forward, 5′-CAACATCAGGTCGGCCATCACT-3′; iNOS reverse, 5′-ACCAGAGGCAGCACATCAA AGC-3′; IL-1β forward, 5′-CTTCAGGCAGGCAGTATC ACTC-3′; IL-1β reverse, 5′-TGCAGTTGTCTAATGGGAACGT-3′; IL-6 forward, 5′-ACAACCACGGCCTTCC CTAC-3′; IL-6 reverse, 5′-TCTCATTTCCACGATT- TCCCAG-3′; β-actin: forward, 5′-TGCTGTCCCTGTATGCCTCT-3′; β-actin reverse, 5′-TTTGATGTCACGCACGATTT-3′. ## Western Blotting Analysis Western blotting was performed as described previously. Cells were collected and lysed in the lysis buffer containing Triton X-100. After 10,000 g centrifugation for 10 min, the protein content of the supernatant was determined by a BCA™ protein assay Kit (Pierce, Rochford, IL). The protein lysates were separated by 10% SDS-PAGE and subsequently electrotransferred onto a polyvinylidene difluoride membrane (Millipore Corp., Bedford, MA). The membrane was blocked with 5% nonfat milk for 1 h at room temperature. The blocked membrane was incubated with the indicated antibodies. Protein bands were visualized using Western blotting detection system according to the manufacturer’s instructions. For extraction of nucleoprotein, cells were collected and lysed in the lysis buffer (10 mM Hepes pH 7.9, 1.5 mM MgCl₂, 10 mM KCl, 0.5 mM DTT, 2% NP-40, 1 mM PMSF) for 20 min, and the lysis buffer was centrifugated at 280 g for 10 minutes. the protein content of the supernatant was collected as cytoplasmic protein. Precipitation were washed twice and lysed in the lysis buffer containing Triton X-100 as nucleoprotein. ## p38 MAP Kinase Activity Assay Cells were stimulated by 500 ng/ml LPS for 6 h and the protein was collected. p38 MAP Kinase activity was detected by p38 MAP Kinase Assay Kit (Cell Signaling Technology). Briefly, endogenous kinases were immuno- precipitated from cell lysates using phospho-p38 (Thr180/Tyr182) antibody bound to protein-A agarose. The beads were washed twice with lysis buffer and twice with kinase buffer (25 mM Hepes pH 7.4, 25 mM MgCl₂, 25 mM β-glycerophosphate, 100 mM sodium orthovanadate, 2 mM DTT). The immunoprecipitates were incubated with DCO-6 (30 µM) or SB203580 (10 µM) for 10 min before the addition of kinase buffer containing ATP and ATF2 fusion protein as a substrate for p38. After further incubation for 30 min at 30°C, the phosphorylated ATF2 products were resolved by SDS-PAGE and analyzed by immunoblot according to the manufacturer’s instructions. ## Electrophoretic Mobility Shift assay (EMSA) Assay EMSA was performed as described previously. Briefly, nucleoprotein (5 µg) was incubated in a 20 µl reaction volume (10 mM Tris–HCl, pH 7.8, 1 mM EDTA, 100 mM KCl, 5 mM MgCl₂, 1 µg/ml poly dI.dC) for 20 min at 37°C and then loaded onto a 6% nondenaturing polyacrylamide gel. After cross-link transferred DNA to membrane, the bands were detected by chemiluminesecence. A 75-fold excess of unlabelled oligonucleotide probe and mutant probe were added in binding reaction to prove if the shifted bands binds specifically to NF-κB oligonucleotides. For supershift assays, 5 µg of rabbit anti-p65 polyclonal antibody (Cell Signaling Technology, Danvers, MA.) was incubated with the protein extract for 30 min prior to binding reaction. ## Measurement of Intracellular ROS Cells (2×10⁵ per well) were cultured in a 6-well plate, and treated with various concentrations of DCO-6 in the presence of different TLR ligands for 6 h. Then the cells were harvested and incubated with 2,7-dichlorofluorescein diacetate (DCFH-DA, Invitrogen) at 37°C for 20 min and washed twice with cold PBS. DCF fluorescence distribution was detected by flow cytrometry on a FACScan (Becton Dickinson) at an excitation wavelength of 488 nm and an emission wavelength of 525 nm. Data were analyzed by Cell Quest software (Molecular Devices Corporation). ## Coimmunoprecipitation Assay For coimmunoprecipitation, cells (1×10⁸) were lysed in lysis buffer containing Triton X-100 and cell lysates were immunoprecipitated with polyclonal mouse antibody to TRAF6 or control mouse immunoglobulin G with protein A/G–Sepharose. The beads were washed, separated by SDS-PAGE and analyzed by immunoblot with antibodies to ASK1 and TRAF6 (Santa Cruz). Protein bands were visualized using Western blotting detection system. ## LPS-induced Septic Shock in Mice BALB/c mice were administered LPS at 10 mg/kg intraperitoneally and survival was monitored for 48 h (n  = 10 per group). DCO-6 (10 or 20 mg/kg) or vehicle (olive oil) was administered intraperitoneally at the day of LPS injection. In some experiments, blood samples were collected 3 h post LPS, and serum levels of IL-1β and IL-6 were measured using ELISA kit from R&D systems according to the manufacturer’s instructions. ## Statistical Analysis All data represent as means ± S.D. of three independent experiments. Statistical analyses were performed using one-way analysis of variance (ANOVA), followed by Student two-tailed t test. Mortality differences between groups were evaluated by the Kaplan-Meier method. P \<0.05 was considered significant. # Supporting Information [^1]: Conceived and designed the experiments: HLL Y. Shen QX. Performed the experiments: HLL RX LLF PT. Analyzed the data: HLL Y. Shen WJG. Contributed reagents/materials/analysis tools: LW NC CQ XFW Y. Sun JXL. Wrote the paper: HLL Y. Shen QX. [^2]: The authors have declared that no competing interests exist.

# 1. Introduction The problematic use of alcohol and other drugs has been a worldwide concern for decades. Globally, national and international policies and interventions have been installed aiming to tackle the harmful consequences of alcohol and drug misuse. Action is especially needed with regard to alcohol misuse since it is the most prevalent psychoactive substance worldwide. According to the World Health Organization, the problematic use of alcohol remains one of the five most important causes of disease, disability and death across the globe. A staggering 5.9% of all deaths worldwide are caused by harmful alcohol use. Indeed, alcohol misuse has been indisputably identified as a public health problem. This is especially true for young people, such as university and college students, as the transition from high school to university or college is often accompanied by high levels of substance use and more problematic alcohol use. Alcohol use among students has been studied extensively in recent years and has received much media attention. A large-scale Flemish study indicated that 98% of university and college students have ever used alcohol and 93% of these students had drunk alcohol in the past 12 months. Half of all these students (49.7%) showed risk characteristics of problematic alcohol use. Moreover, excessive consumption patterns such as binge drinking, have been reported as a common practice among young people, increasing the risk of experiencing alcohol-related consequences. Several studies have addressed the problematic consequences of students’ drinking behaviour, such as academic problems, injuries, assaults, driving under the influence and sexual assault, not only harming the student, but also other people around the student and within society as a whole. Given the immense burden that alcohol puts on society in terms of health, social and economic outcomes, it is vital that alcohol research is based on valid and reliable instruments to measure the consequences of alcohol misuse. In recent decades, considerable effort has been put into developing scales to measure the consequences of alcohol (and drug) use among young people. However, reliability and validity testing of some of these instruments is lacking. The Core Alcohol and Drug Survey (CADS) was developed in 1990 as a self-report instrument to assess the nature, scope and consequences of alcohol and other drug use amongst college students. Although numerous studies have used the consequences scale, a subscale of the survey, little attention has been paid to its psychometric properties. The initial developers presented this consequences scale as a unidimensional construct, without extensively investigating its factor structure, while another research study found that this scale had a two- factor structure that identified personal consequences (such as having a hangover) and consequences with others (such as getting into an argument or fight). Moreover, these studies were all performed in the US, creating a dearth of knowledge of the factor structure of the CADS in other contexts. The primary aim of this research study is, therefore, to assess the psychometric properties of a Dutch version of the CADS in a large sample of 19,253 Flemish university and college students. As alcohol is currently the most prevalent psychoactive substance worldwide, we focus especially on assessing the scale with regard to alcohol consequences. We examined the factor structure of the Dutch CADS by comparing the one- and two-factor model as presented in the literature by using confirmatory factor analysis and adapting the models if necessary. In addition, we tested for composite reliability and both construct (i.e., convergent and discriminant) and criterion-related (i.e., concurrent) validity to verify the consistency as well as the accuracy of the factors. # 2. Materials and methods ## 2.1 Procedure and participants The analyses are based on data collected by the inter-university project ‘Head in the clouds’. A cross-sectional survey was sent to students of the eleven universities and colleges in Flanders (Belgium) who were willing to participate. Students were invited by email and other methods (e.g., student magazine) to participate anonymously to an online survey. They had four to six weeks to participate in the period February to April 2013 and no reminder was sent. Students could voluntarily decide whether or not to participate by actively clicking on the link in the email which would lead them to the online survey. The introduction clearly stated that the data would be anonymous. To increase response rate, some incentives (e.g., the chance to win a number of prizes, including an iPad) were offered to the participating students and only if they agreed to provide an email address. Five colleges were excluded from the sample because they had a very low response rate (\< 3.5%). This resulted in a final sample of 19,253 college and university students (22.1% response rate). The study was performed according to the ethical standards of the American Psychological Association and was approved by the Ethics Committee of Ghent University Hospital (EC UZG 2013/065). Of the 19,253 participants, 35.7% (n = 6,867) were male and 64.3% (n = 12,386) were female. Mean age was 21.12 years (SD = 3.251). provides an overview of sex and age distributions among participating institutions. We also performed bivariate analyses (ANOVAs) with age as the dependent variable and sex as well as institution as the group variable to verify any significant differences in participants’ age between men and women and between institutions. Results indicated that age significantly differs between institutions (*F*(5) = 49.733, *p* \< 0.000). With regard to sex, however, no significant difference was found between the age of male and female participants (*F*(1) = 0.117, *p* = 0.732). ## 2.2 Measures *Negative consequences of alcohol use* were measured using the CADS. Participants were asked how often they have experienced a list of 19 consequences (e.g., got into an argument or fight) as a consequence of their drinking or drug use during the last year. The internal consistency of the items was reported to be high with a Cronbach’s Alpha of 0.90. The CADS was translated into Dutch by two independent translators. Both translations were almost similar. Any differences that do existed were discussed in the working group responsible for the questionnaire. Moreover, five students pre-tested the usability and comprehensibility of the questionnaire. The answer categories of the CADS were ‘none’, ‘one’, ‘two’, ‘three to five’, ‘six to nine’ and ‘10 or more times’. Frequencies were coded using mid-points of the categories, respectively 0, 1, 2, 4, 7.5 and 11.25 times for the upper category (10 times plus half range to midpoint of adjacent category). The complete list of consequences is presented in. *The Alcohol Use Disorder Identification Test (AUDIT)* was developed by the World Health Organization (WHO) and measures problematic alcohol use with 10 items. The scale has proven to be useful and reliable in measuring problematic alcohol use among students. The AUDIT was officially translated into Dutch with the approval of the WHO and has proven to be a reliable screening instrument. In this study, we used the shortened version, the AUDIT-c, which has proven to be an equally good or even better indicator for measuring problematic alcohol use. The AUDIT-c consists of three questions, measured on a 5-point scale: ‘How often do you drink alcohol (in general)’; ‘if you drink, how many glasses do you usually drink per day’; ‘how often does it happen that you drink six glasses or more in one single occasion’. The reliability of the AUDIT-c in the present study was good (α = 0.795). *Binge drinking w*as measured by asking students to indicate how often they drank four glasses or more (for women) or six glasses or more (for men) during a time span of two hours. One glass refers to a standard glass of alcohol containing 10 g or 12.7 ml pure alcohol. This amount corresponds to approximately 1 glass of beer (25 cl), wine (10 cl), non-distilled beverage such as sherry (5 cl), or spirits (3.5 cl). Response options ranged from 1 = never, 2 = less than monthly, 3 = monthly, 4 = weekly, to 5 = daily or almost daily. The time-frame used to measure binge drinking was within the previous year. ## 2.3 Analytic strategy Data were analyzed using IBM SPSS Statistics 22 and IBM SPSS Amos 22. We only included those participants who reported drinking alcohol within the past 12 months (n = 17,756) in the analyses. Firstly, we performed descriptive analyses to describe drinking characteristics and the related alcohol consequences in our sample. Next, we examined the factor structure of the CADS by performing confirmatory factor analyses. The analyses are a mix of the alternative models approach and a model generating approach, as defined by Jöreskog, in which we compare two models as presented in the literature and modify them with the goal of finding a model that fits the data well and has a theoretically meaningful interpretation. We started with the one-factor model as described by Presley (i.e., Model 1a) and the two-factor model indicated by Martens et al. (i.e., Model 2a). These initial models were adapted and compared, based on model fit and their composite reliability. Martens et al. (2005) made several decisions in their analyses to improve model fit. First of all they excluded all the items experienced by 5% or less of the participants. In addition, they excluded items 11 (had a memory loss) and 12 (done something I later regretted) as they loaded high on both factors. We employed a similar strategy for our data. We used several goodness-of-fit indices to measure model fit. The classic goodness-of-fit index is χ². However, it is well known that χ² is almost always significant in the case of large sample sizes. We therefore also reported the Root Mean Square Error of Approximation (RMSEA), Standardized Root Mean Square Residual (SRMR), Comparative Fit Index (CFI) and Tucker-Lewis Index (TLI). We also reported the Akaike Information Criterion (AIC), as this index allows a comparison between non-nested models. The following (strict) cutoff criteria were used to evaluate model fit: SRMR \< 0.08; RMSEA \< 0.08 = adequate fit; \< 0.05 = good model fit; CFI and TLI \> 0.95; factor loading (FL) \> 0.50. Since the CADS is not normally distributed (0 is very frequently answered), we used the ADF estimator in AMOS. Item 1 is the reference item in the one-factor model. In the two-factor model, item 1 is the reference item for the ‘personal consequences’ factor and item 19 is the reference item for the factor ‘social consequences’. We used Jöreskog Rho = (Sum(FL))² / ((Sum (FL))² + Sum (1-FL²)) to evaluate the composite reliability of every model. We also tested the validity of the best fitting model. As indicated by the International Test Commission, we provided evidence on both construct validity as well as criterion-related validity. First of all, construct validity was measured by both convergent and discriminant validity. As Brown (2006) describes, “*convergent validity is indicated by evidence that different indicators of theoretically similar or overlapping constructs are strongly interrelated”*. In other words, all items of one construct need to be interrelated with factor loadings above 0.50 (or even better above 0.70). A more strict evaluating tool of convergent validity is measuring average variance extracted (AVE = (Sum of FL²)/(Sum of FL²+ Sum (1-FL²)). Strictly speaking the AVE needs to be higher than 0.50. Discriminant validity *“is indicated by results showing that indicators of theoretically distinct constructs are not highly intercorrelated”*. In other words, we do not want items of one construct loading onto another construct, or items of different constructs correlating with each other. The covariance of factors needs to be lower than 0.80–0.85. Secondly, we also addressed the concurrent validity by replicating a well-known correlation with two external variables (binge drinking and AUDIT-c). Missing items were deleted using listwise deletion. # 3. Results ## 3.1 Descriptive results provides the sample responses on binge drinking and on the AUDIT-c. gives an overview of the item score distribution of the CADS. ## 3.2 Fit of the one-factor models ### Model 1a We started by testing the one-factor structure of the CADS, containing all of the 19 items. As shown in Tables and, the fit of model 1a was bad, except for the RMSEA. 14 of the 19 factor loadings were below 0.50, and the factor loadings of items 13, 14 and 18 were not significant on a *p* \< 0.001 level. Composite reliability was good with rho = 0.710. ### Model 1b We excluded certain items as they were rarely endorsed (i.e., 5% or less) by the participants. This resulted in an exclusion of 8 items, namely items 3, 4, 13, 14, 15, 16, 17 and 18. The 11-item scale was tested as a one-factor model. As shown in Tables and, the model fit was not good. RMSEA indicated a good model fit, but the other fit indices clearly did not. Some factor loadings were still low \[loadings ranging from 0.204 (item 7) to 0.837 (item 1)\], although all loadings were significant. Composite reliability was adequate with rho = 0.784. ### Model 1c Because model 1b did not have an acceptable fit, we eliminated one by one all the items with a low factor loading (standardized loading \< 0.50) from our analyses. After each elimination, we evaluated the model fit, resulting in a one-factor structure containing 5 items (1, 6, 8, 11, 12). Standardized factor loadings were higher than 0.50 and all were highly significant (*p* \< 0.001). This model was seen to be the ‘best’ model of all the one-factor models. As shown in, the model had an acceptable model fit, although CFI, and especially TLI could be improved. Composite reliability was calculated to be 0.812. presents the one-factor models. ## 3.3 Fit of the two-factor models ### Model 2a We first tested the two-factor model as described in Martens et al. (2005), identifying personal consequences (items 1, 6, 7 and 8) and consequences with others (items 2, 3, 4, 5, 9, 10, 19) which we further refer to as social consequences. We used CFA with correlated factors (similar to an oblique rotation) to test this model. The results are presented in Tables and. All loadings were significant, but not all of them were higher than 0.50. Moreover, the model fit was not good (χ² = 407.528; RMSEA = 0.023; SRMR = 0.088; CFI = 0.916; TLI = 0.893; AIC = 453.528). Correlation of the two factors was 0.82. Composite reliability was good for factor 1 (0.711), but not for factor 2 (0.592). However, as our dataset is different from that of Martens et al. (2005), we extrapolated the decisions they made (cfr. 2.3 Analytic strategy) to our dataset and tested two additional models. As items 3 and 4 were experienced by less than 5% of the participants, these items were also excluded in our analyses (Model 2b). Since we did not know whether items 11 and 12 would load high on both factors, we included them in model testing (Model 2c). In the two models we eliminated items if necessary. ### Model 2b When testing the initial Model 2b, we concluded that the model fit was similar to Model 2a. Factor loadings were significant, but some were really low (\< 0.50). As a consequence, these items were deleted one by one and model fit was evaluated each time. This process of testing and evaluating the fit resulted in the following model: Personal consequences (items 1, 6, 8) and Social consequences (items 5, 9, 19). The results of this model are shown in. All factor loadings were significant and higher than 0.50. The model fit was good, as can be seen in (χ² = 174.137; RMSEA = 0.036; SRMR = 0.034; CFI = 0.960; TLI = 0.925; AIC = 200.137). Correlation of the two factors was 0.76. Composite reliability was good for factor 1 (0.773), but not for factor 2 (0.534). ### Model 2c Model 2c was based on Model 2b, but included items 11 and 12 as well. It was clear that item 11 ‘had a memory loss’ belonged to the factor of personal consequences. For item 12 ‘done something I later regretted’, however, it was somewhat unclear whether it is a consequence that only relates to the drinker or to other people as well. We therefore performed two CFA’s: one where item 12 was part of factor 1 and another where it belonged to factor 2. As the second CFA gave a better fit (AIC of 409.557 compared to 385.866), we included item 12 in the factor of social consequences. However, since item 11 had a high cross loading (similar to Martens et al. (2005)) with the factor social consequences, we still excluded item 11 from the model. This resulted in a major improvement of the model fit. This process of testing and evaluating fit resulted in the following model: Personal consequences (items 1, 6, 8) and Social consequences (items 5, 9, 12, 19). The results of this model are shown in Tables and. Factor loadings were all significant and model fit was good (χ² = 202.125; RMSEA = 0.030; SRMR = 0.033; CFI = 0.956; TLI = 0.929; AIC = 232.125). Correlation between the two factors was 0.78 and composite reliability for the two factors was 0.776 and 0.662. presents the two-factor models. ## 3.4 Conclusion ‘best’ model We performed CFA’s on both one- and two-factor structures of the CADS, starting from two models in the literature, and adapting them based on the (significance of) factor loadings, modification indices and goodness-of-fit indices. As shown in, Model 2b has the lowest AIC and thus the best model fit, closely followed by Model 2c. However, the factor ‘social consequences’ of Model 2b has a low composite reliability, which is much better in Model 2c. In deciding which model best represents the data, it is crucial that not only model fit is evaluated, but also composite reliability is taken into account. Based on this, it can be concluded that Model 2c is the best fitted model in understanding the consequences of alcohol misuse, as it has both a good model fit and an acceptable composite reliability. In the next step, we test the construct and concurrent validity of this model. ## 3.5 Construct and concurrent validity We evaluated the validity of Model 2c and we focused on both construct and concurrent validity. ### 3.5.1 Construct validity Construct validity was measured by both convergent and discriminant validity. As described in the analytic strategy (section 2.3), all items of a construct need to be highly interrelated (factor loadings \> 0.50) to measure convergent validity. Model 2c complies with this standard, and in particular the factor loadings of personal consequences are very high. Only item 5 has a slightly lower factor loading (0.49). The stricter evaluating tool of convergent validity (AVE), however, shows mixed results. Factor 1 with an AVE of 0.537 has a good convergent validity. Factor 2, with an AVE of 0.334, however, has a lower convergent validity. If the AVE is \< 0.50, this means that the variance of the measurement error is larger than the variance explained by the factor, which makes the validity of the factor and the individual indicators questionable. The validity of factor 2 is thus less strong than that of factor 1. Nevertheless, all factor loadings are significant and close to or larger than 0.50. Furthermore, the factors have a high discriminant validity, as there are no cross-loadings between the indicators of the two factors and the covariance of the two factors is lower than the threshold of 0.80–0.85. ### 3.5.2 Concurrent validity As heavy episodic drinking is linked to negative consequences which students experience, we tested whether two drinking variables (binge drinking and AUDIT-c) correlated with Model 2c. At first, we included AUDIT-c in the model (item 3 as reference category). It appears that the model fit is not as it should be. Although RMSEA and SRMR both have acceptable values (0.059 and 0.051, respectively), CFI and TLI are too low (0.866 and 0.812, respectively). However, since the response to the first question of the AUDIT indicates whether or not the respondents need to proceed with the rest of the AUDIT-questions, the bad model fit could be explained by a possible error term correlation for the first two questions. If the respondents indicated that they had never drunk alcohol before in question 1, they did not need to fill in the whole AUDIT. Consequently, we decided to freely estimate this error term correlation. As a result, the model fit improved substantially (χ² = 459.194; RMSEA = 0.029; SRMR = 0.0426; CFI = 0.968; TLI = 0.954; AIC = 507.194). In a final step we included the variable ‘binge drinking’ as a one-indicator construct. The error variance was equalized to 0.0865 based on the following formula: *Var (E) = (1-REL)\*VAR (indicator) → Var (E) = (1–0*.*9) \* 0*.*865*. An error variance of 0 is not preferred, since misinterpretation of the question is possible. The fit of this final model is very good (χ² = 531.926; RMSEA = 0.029; SRMR = 0.0390; CFI = 0.964; TLI = 0.948; AIC = 587.926). The covariance between AUDIT-c and both personal and social consequences is 0.854 and 0.626, respectively. The covariance between binge drinking and both personal and social consequences is 0.764 and 0.594, respectively. The covariances are significant. # 4. Discussion Alcohol research should rely on valid and reliable instruments to measure consequences of alcohol misuse. Although considerable research has used the negative consequences scale of the Core Alcohol and Drug Survey, little is known about its psychometric properties, especially when not used in English. Therefore, the primary aim of this research was to address the research gap regarding the psychometric properties of a Dutch version of the CADS in a sample of 19,253 Flemish university and college students. We focused especially on alcohol consequences and examined the factor structure of the Dutch CADS by comparing two models from the literature, using confirmatory factor analysis and adapting the models if necessary. Reliability and validity issues were also addressed. Based on the literature, we started with a one-factor structure containing the 19 items as developed by Presley et al. (1993) and a two-factor structure as suggested by Martens et al. (2005). These initial models were adapted based on the factor loadings, modification indices and goodness-of-fit indices. As a result, CFA was performed on 6 models and fit indices were compared. In addition, composite reliability was measured for every model. The best model (CADS_D) was a two-factor structure, identifying personal consequences (had a hangover; got nauseated or vomited; missed a class) and social consequences (got into an argument or fight; been criticized by someone I know; done something I later regretted; been hurt or injured) (Model 2c). This model was identified as the best based on both the model fit and composite reliability of the two factors. Although Model 2b had the lowest AIC, and thus the best model fit, the composite reliability of the second factor was not acceptable. Since Model 2c had a much better composite reliability and only a slightly higher AIC, Model 2c was preferred over Model 2b. Our findings confirm the fact that the negative consequences of alcohol misuse should be measured as a two-dimensional scale, focusing not only on consequences that affects the drinkers themselves, but also consequences harming other people around them. Finally, the validity of the CADS_D was assessed. Construct validity was evaluated as acceptable, with good discriminant validity, although the convergent validity of the factor ‘social consequences’ could be improved. Concurrent validity was measured by testing the known correlation of two drinking variables (binge drinking and AUDIT-c) with the negative consequences students experience. Concurrent validity was evaluated as good. We need to take some limitations of the study into account. Firstly, we excluded the consequences which were encountered by less than 5% of the participants. This does not mean that these consequences were of minor importance. On the contrary, these deleted items are often more severe than the ones included in the analyses and therefore remain important. Secondly, the CADS was measured as an interval variable using frequencies. In this way, a higher weight is given to a student who, for example, experienced a hangover six times last year compared to a student who had been arrested for driving while intoxicated (DWI)/driving under the influence (DUI) twice last year. Future studies should analyze the CADS in a dichotomous way and establish whether the same results are found. And finally, the assessment of the concurrent validity could be improved by measuring the correlation between the CADS and other consequences scales, such as the Young Adult Alcohol Consequences Questionnaire or the Rutgers Alcohol Problem Index. However, these scales were not available in the dataset and thus these analyses could not be performed. Despite these limitations, the current study aimed to enhance the knowledge of the psychometric properties of the CADS. We did this by addressing the factor structure, reliability and validity of a Dutch version of the CADS in a large sample of 19,253 Flemish university and college students. The study findings have both theoretical and practical implications. Theoretically, the results indicate that a two-factor structure, identifying personal and social consequences, had the best model fit. This current study will help future researchers working with this scale to address alcohol-related consequences correctly. From a practical point of view, the CFA results indicate that the shortened Dutch version of the CADS (CADS_D) is a valid and reliable instrument to screen for alcohol-related consequences among college students, with the ultimate aim of preventing these consequences. Moreover, we expanded the debate about evaluating models and encourage not blindly evaluating model fit, but also taking reliability and validity issues into account. University and college students: The use of the concepts ‘college’ and ‘university’ differs between countries worldwide. In Belgium, colleges offer professional bachelor degrees, whereas universities offer academic bachelor and master degrees as well as doctoral degrees. # Supporting information Membership of the Task Force substance use in Flemish universities and colleges: Johan Rosiers (Association for Alcohol and other Drug problems, Brussels, Belgium, <johan.rosiers@vad.be>); Anne Hublet (Department of Public Health, Ghent University, Ghent, Belgium); Maura Sisk (Student Health Center, Catholic University of Leuven, Leuven, Belgium); Yassira Si Mhand Benali (Student counselor, Catholic University of Limburg, Belgium); Lea Maes (Department of Public Health, Ghent University, Ghent, Belgium) The authors wish to thank Dr Jude Murison and Tessa James for proof-reading the manuscript. ADF Asymptotic Distribution Free AIC Akaike Information Criterion CADS Core Alcohol and Drug Survey–in this research the abbreviation refers specifically to the question battery of negative consequences CFA Confirmatory factor analysis CFI Comparative Fit Index df Degrees of Freedom DUI Driving under the influence DWI Driving while intoxicated FL Factor loading MI Modification Indices RMSEA Root Mean Square Error of Approximation SRMR Standardized Root Mean Square Residual TLI Tucker-Lewis Index χ² Chi-square [^1]: The authors have declared that no competing interests exist. [^2]: ¶ The membership of the Task Force can be found in the Acknowledgments section.

# Introduction Adenoviruses (Ads) may cause invasive, life-threatening infections in hematopoietic stem cell transplant (SCT) recipients, especially in the pediatric population, such as hepatitis, pneumonitis, colitis, nephritis, and encephalitis. The profound immune suppression following allogeneic bone marrow transplantation, due to ablation of host T cell immunity and administration of immunosuppressive drugs to prevent or treat graft vs. host disease makes patients susceptible to both reactivation of latent Ad infections and primary Ad infections. Group C serotypes (Ad types 1, 2, and 5) common in the general population and the uncommon group B serotypes (Ad types 11 and 35) have most frequently been associated with disease post-SCT, but infections with most serotypes have been described. Although treatment with the toxic antiviral cidofovir may inhibit Ad replication, patients need to develop cell-mediated immunity to recover from invasive Ad disease. Given the limited therapeutic options, donor lymphocyte infusions have been under investigation as a treatment for Ad disease in allogeneic SCT recipients. As a precedent, infusions of donor lymphocytes or virus-specific T cells have been effective in the treatment of Epstein Barr virus-associated lymphoproliferative disease and cytomegalovirus infections in SCT recipients. Case reports of donor lymphocyte infusions have suggested some benefit against invasive Ad disease. More recently, Ad-specific T cells, isolated by collecting IFN-γ-secreting T cells after short *in vitro* stimulation with Ad lysate, were infused into 8 pediatric SCT pediatric recipients with Ad infections. Patients were also treated with cidofovir. Four of 5 patients with diarrhea alone or no symptoms resolved their infections, whereas all 3 patients with disseminated Ad disease died. In another study, 11 SCT recipients were treated with donor lymphocytes stimulated *in vitro* with autologous lymphoblastoid cell lines infected with an E1-deleted Ad. Ad-specific T cell responses were detected only in patients (5) with positive Ad cultures. Reductions in Ad viral loads were documented in 3 pediatric patients, and clinical improvement noted in one patient with Ad pneumonia. A detailed understanding of the T cell response to Ad is required to identify antigen-specific donor T cells that have rapid responses, high proliferation potential, and longevity in order to optimize immunotherapy strategies. Recent advances in T cell biology have shown that two main cell types are involved in memory CD8+ T cell responses – effector memory cells (Tem) (CD62L-/CCR7-) and central memory cells (Tcm) (CD62L+/CCR7+). Tcm are concentrated in secondary lymphoid tissue and have limited effector functions, but they are highly efficient in self renewal and divide in response to homeostatic cytokines including IL-7 and IL-15. Tem, on the other hand, migrate to peripheral tissues and exhibit pronounced immediate cytolytic (effector) activity but only modest proliferation upon antigenic stimulation. Together, Tem and Tcm contribute to protective immunity depending on the nature and route of infection. Additional levels of complexity in memory CD8+ T cell phenotypes exist between distinct peripheral tissues and in different infectious models. Such functional, anatomic, and phenotypic heterogeneity in the CD8+ T cell memory pool has significant consequences for immunotherapy strategies. Therefore, it is important to determine both the T cell subset(s) and the viral antigen targets able to generate the most efficacious recall responses to Ads. Additionally, it is crucial to take into consideration the fact that Ads express unique immune evasion proteins that help the virus evade T cell recognition. Notably, the Ad early region 3 E3-19K glycoprotein specifically downregulates cell surface expression of MHC class I molecules, which inhibits recognition of infected cells by CD8+ T cells E3-19K binds to both class I molecules and TAP proteins in the endoplasmic reticulum, which efficiently prevents class I antigen transport to the cell surface. Thus, elimination of infected cells prior to E3-19K mediated downregulation of MHC class I molecules may be more efficient in controlling Ad replication and dissemination. Targeting cells early after infection may also help avoid the effects of other early Ad proteins such as the E3-10.4K/14.5K RID complex (receptor internalization and degradation) and E3-14.7K that confer resistance to apoptosis. We have previously shown that both early regulatory proteins and late structural proteins from Ad can be recognized by human CD8+ CTLs. Additionally, our lab has identified several HLA-A2 restricted epitopes from the early region 2 protein DNA polymerase (e.g P-977) and late capsid protein hexon (e.g. H-892) that are well conserved among different Ad serotypes. Hexon represents the most abundant viral protein and is synthesized late after infection following DNA replication. Rooney and colleagues have also identified other CD8+ T cell epitopes from hexon. Most groups are now focused on expanding hexon-specific T cells *ex vivo* for immunotherapy of Ad disease in SCT recipients. The goal of this study was to compare the frequency, memory and activation phenotypes, and functional aspects of early protein DNA polymerase- and late protein hexon-specific CD8+ T cells in both peripheral blood and tonsils directly *ex vivo*. CD8+ T cells targeted to HLA-A2 restricted epitopes P-977 from the DNA polymerase and H-892 from hexon were measured *ex vivo* using epitope-specific pentamers. Activation and lymphocyte homing markers and cytokine expression were compared between early and late protein-specific CD8+ T cells in both PB and tonsils. Additionally, P-977 and H-892 epitope specific CTL lines were generated from PBMC by *in vitro* stimulation with Ad5-infected dendritic cells (DC), followed by stimulation with peptide, to compare the kinetics of cytotoxicity. # Methods ## Study participants Buffy coat collections were obtained from healthy adult volunteers through the Thomas Jefferson University Hospital Blood Donor Center. Additionally, 1–2 blood specimens (50 ml each) were obtained from SCT recipients with documented invasive Ad infections. Tonsil samples were obtained from patients undergoing tonsillectomies at Thomas Jefferson University Hospital. The research protocols were approved by the Institutional Review Board. All donors were screened by HLA typing or flow cytometry with HLA A2-specific antibody W6/32 to identify HLA A2-positive donors. ## Purification of lymphocytes Lymphocytes were purified from PBMC by density gradient centrifugation using Ficoll-Hypaque (Sigma Aldrich, St. Louis, MO). For purification of lymphocytes from tonsils, tonsil tissue was cut into small fragments and cells were dispersed by vigorously agitating sample with the flat end of a 5 ml syringe against bottom of culture dish on ice. The fractionated tonsil specimen and cellular suspension were passed through a 100 µm sieve and rinsed well with cold HBSS, until the tissue was clear. Cells were resuspended by repeated pipetting until all cell clumps dissolved and then underlayed with Ficoll-Hypaque. The cell suspension was spun at 900× g for 30 min without brakes, and mononuclear cells at the interphase were harvested. Lymphocytes were aliquoted in media containing 10% DMSO and stored in liquid nitrogen until use. ## Viruses Group C Ad5 prototype strain was originally obtained from M. Horowitz, Albert Einstein College of Medicine (Bronx, NY). The Ad5 E3-19K deletion mutant dl704 was obtained from W. Wold, St. Louis University (St. Louis, MO). Ad virions were purified from virus-infected A549 cell lysates by an affinity chromatography method (Clonetech, Mountain View, CA) as per the manufacturer's instructions. The Ad5 virion preparation was titered by ELISA (Cell Biolabs, San Diego, CA). ## Synthetic peptides The HLA-A2-restricted epitopes H-892 (LLYANSAHAL) and P-977 (VLAWTRAFV) were previously identified from the Ad5 hexon and DNA polymerase protein sequences, respectively. Crude peptides were synthesized by Research Genetics (Huntsville, AL) or Sigma-Genosys (The Woodlands, Texas). Stock solution (10 mM) for each peptide was prepared in 90% DMSO and stored in small aliquots at −80°C. ## IFN-γ ELISPOT assay This assay was performed as previously described. ## Pentamer staining and flow cytometry assays (CFC) APC-labeled MHC class I HLA A2 Pro5 ^Tm Pentamers with H-892 and P-977 peptides were custom synthesized at ProImmune (Oxford, UK). PE-labeled mAbs to human CCR7, CD25, TNF-α, IL-2, Perforin, INF-γ and FITC-labeled human anti-CD8 were purchased from BD Pharmingen (San Diego, CA). For immunostaining, 1–2×10⁶ lymphocytes were stained with a pre-titered amount of pentamer reagent (HLA A2:H-892 or HLA A2:P-977) at room temperature, shielded from light, for 20 min. HLA A2 negative control Pro5 ^Tm Pentamers were included to assess nonspecific binding in flow assays. Cells were washed and then stained with PE-labeled CD8, CD25, or CCR7 mAbs on ice for 20 min. Thereafter cells were washed and fixed in 0.4% formaldehyde solution. Cell surface phenotype analysis was done by gating on the CD8+ pentamer+ population. ## Cytokine analysis CD8+ T cells were isolated from PBMC or tonsils by negative immunomagnetic selection (Miltenyi Biotec, Auburn, CA), and analyzed by intracellular cytokine staining and flow cytometry. Briefly, 1- 2×10⁶ CD8+ T lymphocytes were incubated with or without peptide for 2 h. Cells were treated with brefeldin A (Golgi Stop, BD Pharmingen) for an additional 4 h and then washed, permeabilized, and fixed using the Cytofix/Cytoperm reagent (BD Pharmingen). PE- conjugated anti-human IFN-γ, TNF-α, perforin and IL-2 mAbs were used to measure cytokine secretion. PE-conjugated isotype control cocktail as well as purified blocking antibody cocktail were used to measure background. Samples were collected on a FACSCalibur flow cytometer (BD Pharmingen), and analysis was performed using CellQuest software. ## Statistical analysis Differences in values between experimental groups were examined for significance with Graph Pad Prism software using the Student *t* test. A probability (*P*) values \<0.05 was considered significant. ## Reverse transcriptase PCR (RT PCR) The kinetics of mRNA expression from the Ad DNA polymerase and hexon ORFs was measured by infecting fibroblasts with wild type (WT) Ad5 (200 pfu/cell) and harvesting cells between 6–24 h post infection. RNA was extracted from infected cells via affinity chromatography (Qiagen, Valencia, CA) and RT PCR was performed using the Quantiscript™ RT (Qiagen). The following primers were used for amplification of DNA polymerase - (forward): 5′ GACTTCACTCTAGAAAGATGATACTCAGAGAGCCCCTC 3′ and (reverse): 5′ GACTTCACGGATCCTAACTGGAGGAGGCGCCGCA 3′ - and hexon - (forward): 5′ GACTTCACTCTAGAAGGTCAGGAAAATGGATGGG 3′ and (reverse): 5′ GACTTCACTCTAGAGATTCCCCAGTCCTTTTCCTCCC 3′ - to produce 1 Kb and 0.3 Kb products, respectively. A control reaction without RT was run for each time point to ensure that the RT PCR products were RNA specific. ## Preparation of Ad-infected dendritic cells CD14+ cells were isolated from PBMC by positive immunomagnetic selection (Miltenyi Biotec) per the manufacturer's directions. CD14+ cells (1×10⁶/ml) were placed in a 6-well plate in RPMI 1640 supplemented with 5% human AB sera, HEPES, glutamine, and P/S containing GM-CSF (800 U/ml) (Berlex, Wayne, NJ) and IL-4 (1000 U/ml) (BD Pharmingen). Fresh cytokines were added on day 3. Non-adherent cells (immature DC) were harvested on day 5 and infected with Ad (200 pfu/cell) in 200 µl media for 1.5 h or left untreated. DC were re-plated in 6-well plates, and maturation was induced by incubation with 10 ng/ml IL-1β, 1000 U/ml IL-6, 10 ng/ml tumor necrosis factor-α (TNF-α) (BD Pharmingen), and 1 µg/ml prostaglandin E2 (PGE2) (Sigma Aldrich) for 48 h. ## Generation of peptide-specific CTL lines Bulk Ad-specific T cells were amplified *in vitro* by co-culture of 10×10⁶ lymphocytes with 2×10⁶ Ad5-infected DCs for one week. Then, cells were re-stimulated with peptide (H892 or P977)-loaded, irradiated PBMC on day 7. IL-2 (20 U/ml) was added to each line on day 8, and every 3 to 4 days thereafter. Specificity was confirmed by IFN-γ ELISPOT and the CTL lines were assayed for cytotoxicity around day 14. ## Cytotoxic T cell assay Cytotoxicity was measured by a calcein release assay as previously described. HLA-A2-positive MRC-5 fibroblasts (targets) were infected for 6, 8, 12, 16 or 24 h with Ad5 (200 pfu/cell). In some cases, fibroblast targets were pre-treated for 24 h with 150 U/ml of human IFN-γ prior to infection. Targets were labeled with 5 µg/ml calcein (Molecular Probes, Eugene, OR) for 30 min at 37°C, washed extensively, and incubated with effectors in microtiter wells for 3 h. Assays were performed in triplicate using a range of effector/target ratios. Calcein release was measured from supernatants on a Victor 2 1420 multilabel counter (Wallace, Gaithersburg, MD). ## Proliferation assay Proliferation of lymphocytes *in vitro* was measured using CFSE (Molecular Probes) labeling. Briefly, 20×10⁶ purified CD8+ T cells were labeled with 1.5 µM CFSE in PBS for about 8 min. Cells were then incubated for 10 min at 37°C with pre-warmed FBS to allow for efflux of excess CFSE. Cells were washed vigorously three times in PBS containing 1% FBS, and a day 0 sample was obtained to determine initial labeling. For peptide stimulation experiments, 2×10⁶ CFSE-labeled cells were placed in a 24-well plate and cultured in RPMI with 10% FBS with or without P-977 or H-892 peptides for 4 days. # Results ## Early and late protein-specific CTLs can be detected in peripheral blood and lymphoid tissue Nine HLA A2-positive donors were tested for CD8+ T cell responses to both early and late protein HLA A2-restricted epitopes P-977 and H-892 in PB by *ex vivo* pentamer staining. These epitopes are both highly conserved among different Ad serotypes. PBMC samples were obtained from three different groups of individuals: 1) six healthy adults; 2) one recently infected immunocompetent adult; and 3) two adult SCT recipients who recently recovered from documented invasive Ad infections (2 and 12 months post-infection). All donors had positive T cell responses to Ad lysate by IFN-γ ELISPOT assay (data not shown). Overall, as shown in, P-977+CD8+ T cells were detected at lower frequencies than H-892+CD8+ T cells in PB (mean 0.38% vs 0.57%, p = 0.03). Recently infected donors (immunocompromised and immunocompetent) exhibited high, comparable frequencies of P-977- and H-892-specific CD8+ T cells. In contrast, the frequencies of P-977+CD8+ T cells were relatively lower than H-892+CD8+ T cells in PB samples from healthy adults. Two of 6 healthy donors did not have detectable H-892 or P-977 responses in PB, while one donor had a low H-892 response but no P-977 response using this method. Tonsils from 5 of 5 HLA A2+ donors tested exhibited positive responses to one or both epitopes by *ex vivo* pentamer staining. Notably, in contrast to PB responses, P-977+CD8+ T cells were detected at significantly higher frequencies than H892+CD8+ T cells in tonsils (mean 1.1% vs. 0.48%, p = 0.01). One tonsil sample exhibited a high frequency of P-977+CD8+ T cells (1.1%) but did not have a detectable H-892 response. ## Early and late CTLs from peripheral blood and tonsils have distinct phenotypes Cell surface expression of the lymph node homing receptor CCR7 and the activation receptor CD25 (IL-2R) was analyzed in PB and tonsil mononuclear cell samples. CD8+ T cells from both early protein P-977 pentamer+ and late protein H-892 pentamer+ populations had negligible cell surface expression of CCR7 in all PB samples. In contrast, CCR7 was highly expressed in tonsil samples by both early and late protein pentamer+ CD8+ T cells. The signature of the T cell activation receptor CD25 was more diversified. PB samples from remotely infected adults showed a distinction between early and late protein epitope-specific T cell populations. P-977+CD8+ T cells from remotely infected adult PB showed low expression of CD25, whereas most H-892+CD8+ T cells were positive for CD25. In comparison, the vast majority of both early and late protein epitope+ CD8+ T cells from recently infected donor PB (both immunocompetent and immunocompromised) exhibited cell surface expression of CD25. Tonsil samples had negligible expression of CD25 from both early and late protein epitope-specific CD8+ T cells consistent with a central memory phenotype, except for 1 tonsil that exhibited moderate expression of CD25 from P-977+CD8+ T cells only. ## Functional profile of peripheral blood and tonsil P-977- and H-892-specific CD8+ T cells Cytokine secretion and proliferative potential of early and late protein epitope-specific CD8+ T cells from lymphoid tissue and PB were compared. Within each compartment, early P-977- and late H-892-specific CD8+ T cells had similar cytokine secretion profiles by intracellular cytokine flow analysis. IFN-γ, TNF-α, and perforin were expressed by PB CD8+ T cells stimulated 6 h with either the P-977 or H-892 peptide, but there was relatively little IL-2 secretion **(**). The cytokine release pattern in tonsils was distinct from that of PB samples. Lymphoid CD8+ T cells had significant release of both IL-2 (which was negligible in PB CTLs) and IFN-γ but expressed little or no perforin and TNF-α when stimulated *in vitro* with either peptide). The proliferative potential of lymphoid and PB epitope-specific CD8+ T cells was also compared by CFSE staining following *in vitro* antigen exposure for 4 days from single donors. Lymphoid CD8+ T cells exhibited a higher proliferative potential than PB CD8+ T cells when stimulated with either P-977 or H-892 peptides. Additionally, within both compartments, early P-977-specific CD8+ T cells had a higher proliferative potential than late H-892 peptide-specific T cells. Thus, early protein P-977 CD8+ T cells from the lymphoid compartment had the highest proliferation potential of all. ## P-977-specific and H-892-specific CTLs have distinct cytotoxicity profiles PB lymphocytes from one of the healthy donors were stimulated *in vitro* with Ad-infected DCs for 1 week, followed by stimulation with individual peptides, to generate early protein P-977 and late protein H-892 peptide-specific CTLs (as described). These early P-977 or late H -892 CTLs were incubated with fibroblast targets infected with either WT Ad5 or the Ad5 E3-19K deletion mutant dl704. Targets were infected for 6, 8, 12, 16, or 24 h, and cytotoxicity was measured by calcein release assay (as described). Early P-977 CTLs lysed WT Ad5-infected targets as soon as 8 h post infection (**(i)**); cytotoxicity peaked at 12 h and was absent at later time points. In the absence of E3-19K expression, early P-977 CTLs were able to lyse Ad dl704-infected cells efficiently up to 16 h post infection. In contrast, late H-892 CTLs did not efficiently kill WT Ad5-infected targets (**(i)**). Our lab has previously shown that IFN-γ can help overcome the E3-19K-mediated inhibition of class I cell surface expression. Therefore, fibroblast targets were treated with IFN-γ prior to infection, and the cytotoxicity assays were repeated. As shown in **(ii)**, the early P-977 CTLs lysed WT Ad5-infected targets pre-treated with IFN-γ 8 h post infection, and cytotoxicity peaked at 16 h, similar to the kinetics using dl704-infected fibroblasts. Moreover, as shown in **(ii)**, late H-892 CTLs efficiently killed WT Ad5-infected targets pre-treated with IFN-γ, but cytotoxicity did not occur until 16–24 h post infection. Similarly, the late H-892-specific CTLs killed Ad dl704-infected fibroblasts without IFN-γ pre-treatment at 16–24 h. These data confirm the role of E3-19K in inhibiting CTL recognition of hexon but not DNA polymerase. Finally, the kinetics of DNA polymerase and hexon expression was studied by RT- PCR analysis of Ad5-infected fibroblasts. Robust expression of DNA polymerase mRNA was observed as early as 8 h post infection and started to decline by 16 h. In contrast, hexon mRNA could not be detected until 12 h following infection and expression peaked at 16–24 h post infection. The kinetics of mRNA expression paralleled the above cytotoxicity data in that early protein P-977 CTLs started killing infected cells at 8 h, whereas late protein H-892 CTL cytotoxicity was delayed until 16 h post infection. # Discussion A detailed understanding of Ad-specific T cell responses is needed in order to optimize immunotherapy strategies for SCT recipients infected with Ads, pathogens which may be associated with significant morbidity and mortality. Although most healthy adults exhibit memory T cell responses to Ads, little is known about the kinetics or memory compartmentalization of Ad-specific T cells. This is the first study to identify the phenotype and functional status of Ad- specific CD8+ T cells in tonsils and to compare the responses of secondary lymphoid and PB compartments following natural Ad infection. Additionally, this study compares the characteristics of CTLs targeted to highly conserved epitopes from the early regulatory protein DNA polymerase and late capsid protein hexon in both PB and lymphoid compartments. This study documents that DNA polymerase- specific CTLs kill fibroblasts earlier after infection than hexon-specific CTLs, prior to Ad E3-19K-mediated down regulation of HLA class I expression. Additionally, central memory type CTLs targeted to DNA polymerase dominate the lymphoid compartment and have a higher proliferative potential compared to hexon-specific CTLs. CD8+ T cell responses to HLA A2-restricted epitopes from DNA polymerase (P-977) and hexon (H-892) were detected in both PB and tonsil samples using epitope- specific pentamers. Recently infected subjects exhibited high, comparable frequencies of P-977+CD8+ and H-892+CD8+ T cells in PB. Although, as expected, remotely infected donors exhibited significantly lower responses to Ad-specific epitopes in PB, H-892+CD8+ T cells were detected at significantly higher levels than P-977+CD8+ T cells. In contrast, there were significantly higher frequencies of P-977+CD8+ T cells compared to H-892+CD8+ T cells in tonsil samples. Thus, early protein DNA polymerase-specific CD8+ T cells dominate the lymphoid compartment, while late protein hexon-specific CD8+ T cells are predominant in PB of remotely infected adults. Phenotypic analysis of Ad-specific CTLs from PB and tonsils revealed significant differences between compartments, as well as between epitopes. Several reports have described two subsets of memory CD8+ T cells based upon expression of lymph node homing receptors: CD62L+CCR7+CD8+ central memory T cells -Tcm and CD62L-CCR7-CD8+ effector memory T cells -Tem. In PB, both P-977+CD8+ and H-892+CD8+ T cells from recently infected subjects exhibited high level expression of the T cell activation receptor CD25 and were negative for the lymphoid homing marker CCR7. In tonsils, on the other hand, both P-977+CD8+ and H-892+CD8+ T cells exhibited negligible expression of CD25 and high expression of CCR7. Thus, P-977-specific CD25-CCR7+ CD8+ Tcm-like cells predominate in the lymphoid organs while H-892-specific CD25+CCR7-CD8+ Tem-like cells are found mostly in PB. Interestingly, there was a mixed phenotype of Ad-specific CD8+ T cells in PB from remotely infected adults. Although Ad-specific T cells from PB in remotely infected subjects were all CCR7 negative and most P-977+CD8+ T cells did not express CD25, the majority of the higher frequency H-892+CD8+ T cells were CD25+, similar to the phenotype in recently infected adults. These data suggest that over time following infection, DNA polymerase-specific CTLs preferentially home to secondary lymphoid tissue, while hexon-specific CTLs remain in PB. This is consistent with the fact that DNA polymerase is a regulatory protein that is expressed transiently in active infection, as reflected by the cytotoxicity and RT-PCR kinetics data presented, and thus there is little if any residual antigen in vivo. In contrast, hexon is the major capsid protein in the virion and is over-expressed in Ad-infected cells. Therefore, hexon-specific CTLs likely remain in an activated state in PB due to persistent antigen exposure. Studies have documented the persistence of viral antigens from other lytic viruses such as influenza and vesicular stomatitis virus. More recently, respiratory DCs that migrate to draining lymph nodes were identified as the major source of persistent antigen following acute influenza virus infection. The presence of activated hexon-specific CTLs in PB in remotely infected adults may also reflect persistent low level Ad replication and/or boosting of hexon-specific T cells by re-exposure to exogenous Ads. The central memory type CTLs (Tcm) are considered the “true” memory subset of CD8+ T cells, and several reports have documented that viral replication *in vivo* is more effectively controlled by CD8+ Tcm. Wherry *et al* also showed that CD8+ Tcm are able to persist long term *in vivo* by self renewal, are able to secrete IFN-γ, and rapidly expand upon re-encounter with pathogen. CD8+ Tcm were also shown to elicit greater protection against experimental tumors after adoptive transfer followed by vaccination. Analysis of cytokine expression and proliferation demonstrated functional differences in Ad-specific T cells primarily based upon the compartment, not the epitope. In PB, most CD8+ T cells targeted to either P-977 or H-892 secreted IFN-γ, TNF-α, and perforin but not IL-2, consistent with a Tem phenotype. These data are consistent with a recent study in which Ad E1-deleted vector-stimulated CD8+ T cells were found to secrete both IFN-γ and perforin; however, TNF-α secretion was not detected. Tonsil T cells targeted to either P-977 or H-892 secreted both IL-2 and IFN-γ but exhibited little or no expression of perforin and TNF-α consistent with a CD8+ Tcm phenotype. Additionally, Tcm-type tonsil CTLs showed higher proliferative potential when stimulated with cognate antigen in vitro than Tem- type PB derived CTLs. In both compartments, however, P-977-specific CTLs exhibited a higher proliferative ability than H-892-specific CTLs. The fact that lymphoid derived Tcm and DNA polymerase-specific CTLs have a proliferative advantage over hexon-specific CTLs has implications for protective immunity, as more rapid proliferation will eventually lead to greater expansion of this subset *in vivo*. Additionally, only Tcm-type tonsil CTLs produced IL-2, consistent with their greater potential for *in vivo* expansion. Ad-specific Tcm may also be exposed to more efficient antigen presentation *in vivo* as these CTLs home to primary and secondary lymphoid organs. Infusion of large number of Tem-type cells may provide immediate protection but fail to reconstitute long term memory. Thus, for immunotherapy applications, the superior ability of Tcm to reconstitute the memory T cell pool may be highly relevant. Based on other studies, it may also be important to infuse both CD4+ and CD8+ Ad-specific T cells in order to help amplify and maintain CD8+ T cell responses, Notably, P-977 CTLs lysed fibroblasts *in vitro* significantly earlier following infection compared to H-892 CTLs and evaded E3-19K-mediated inhibition of cytotoxicity. In parallel with the kinetics of mRNA expression, P-977 CTLs killed fibroblasts as early as 8 h post infection. In contrast, H-892 CTLs were unable to kill infected fibroblasts unless targets were treated with IFN-γ to up regulate HLA class I antigens prior to infection, and cytotoxicity was delayed until 16–24 hours. This result confirms our prior study in which bulk Ad- specific CTLs did not kill infected fibroblast targets without IFN-γ pre- treatment as a result of E3-19K-mediated down regulation of class I antigens. Thus, CTLs targeted to DNA polymerase may be able to more rapidly and efficiently eliminate Ad-infected cells *in vivo* compared to hexon-specific CTLs. Targeting Ad-infected cells prior to viral DNA replication will also prevent new virus production. We acknowledge that the small sample size is a major limitation of this study. Unfortunately, acute Ad infections are infrequently identified in adults and the number of adult SCT patients who develop and recover from invasive Ad infections is very limited. It took several years to identify the 3 HLA-A2 positive recently infected donors used in this study. Additionally, we had access to a limited number of PBMC from each donor, especially the recently infected patients, so analyses had to be carefully selected. Although there were statistically significant differences in the precursor frequencies of H-892- and P-977-specific CD8+ T cells in PB and tonsil samples, the differences in cytokine expression and proliferation need to be confirmed in a larger study. Also, these analyses were performed using a single epitope for each protein, which may not necessarily reflect the total CTL response to each protein. Additionally, the intracellular cytokine analysis was done with a single stain at a time, so this study did not characterize the expression of multiple cytokines from individual cells. Finally, although it would be ideal to evaluate the protective potential of these CTLs after adoptive transfer in an animal model, a suitable animal model that mimics human Ad infection is not available, and animal Ads, such as mouse Ads, have significantly different properties. In summary, these data show that in contrast to hexon-specific CTLs, central memory type CTLs targeted to DNA polymerase dominate the lymphoid compartment and have a higher proliferative potential. Additionally, DNA polymerase-specific CTLs kill fibroblasts earlier after infection compared to hexon-specific CTLs, prior to both E3-19K-mediated down regulation of HLA class I expression and Ad replication. In contrast, hexon CTLs cannot recognize Ad-infected cells unless exposed to exogenous IFN-γ to maintain HLA class I expression. Thus, use of CTLs targeted to both early protein DNA polymerase and late protein hexon may improve the efficacy of immunotherapy for life-threatening Ad disease in SCT recipients. # Supporting Information We wish to thank all of the donors who volunteered for this study. We also thank Matthew Farabaugh for excellent technical assistance with flow cytometry assays. [^1]: Conceived and designed the experiments: AJ PF. Performed the experiments: AJ BZ CR. Analyzed the data: AJ PF. Contributed reagents/materials/analysis tools: AJ PF DR. Wrote the paper: AJ PF. [^2]: The authors have declared that no competing interests exist.

# Introduction Bacteria are able to colonize a wide variety of habitats, including the most extreme environments and even living organisms. This remarkable feature likely reflects a high degree of adaptability and the presence of specific genetic programs devoted to the exploitation of nutrients present in these diverse habitats. The bacterial ability to use defined carbohydrates to support cell survival or growth implies the availability of these substrates in their habitat. Moreover, it requires the recognition of these molecules and the coordinated induction of particular uptake systems and of metabolic enzymes. Characterization of the repertoire of genes involved in signal perception or transduction and in carbohydrate utilization, in conjunction with analysis of the regulation of their expression, is likely to provide key information about the interaction and adaptation of bacteria with their environment. The analysis and comparison of complete genomic sequences of numerous bacteria from diverse phylogenies and habitats have shown a relationship between the ecological niches that a bacterium occupies and the proportion of genes involved in signal perception and transduction. Bacteria that inhabit stable environments (extremophiles, obligate parasites or symbionts), which generally have small genomes, possess fewer sensory and regulatory genes than free-living bacteria found in complex and changing environments such as those living in soil or in association with plants. Therefore, it was proposed that sensors and regulators can be used as “descriptors of bacterial lifestyle”. With the aim to study the molecular mechanisms controlling adaptation of phytopathogenic bacteria to their host-plants, we undertook a global analysis of receptors and regulators of *Xanthomonas campestris* pv*. campestris* (*Xcc*), the causal agent of black rot of crucifers. This pathogen infects a wide range of *Brassicaceae* plants of economic interest, including cabbage, cauliflower and radish as well as the model plant *Arabidopsis thaliana*. This epiphytic bacterium naturally infects host plants *via* wounds in the leaves or hydathodes which are specialized pores on the leaf margins of higher plants that connect to the vascular system. Then bacteria multiply and progress in vascular tissues. During the past two decades, classical molecular and genetic studies led to the characterization of several determinants controlling pathogenicity of *Xcc*, such as secretion of extracellular plant cell wall degrading enzymes, cell-cell signaling, biofilm formation and *hrp* genes coding for a type III secretion system. The characterization of new virulence factors should now be greatly facilitated by the availability of complete genome sequences of two *Xcc* strains (strain ATCC33913 and strain 8004). Moreover, comparative genomics would improve the analysis of virulence and host adaptation in *Xanthomonas*, since the genomic sequences of four other *Xanthomonas* strains displaying different host specificities and representing three other species are also available, i.e. *Xanthomonas axonopodis* pv. *citri* (*Xac*), the causal agent of citrus canker, *Xanthomonas campestris* pv. *vesicatoria* (*Xcv*), the causative agent of bacterial spot disease on pepper and tomato plants, and *Xanthomonas oryzae* pv. *oryzae* (*Xoo*) (strain Kacc10331 and strain MAFF311018), the causal agent of bacterial blight of rice. Thus, the *Xanthomonas* genus, which affects two major model plants (*Arabidopsis* and rice), constitutes a very attractive model to study plant-pathogen interactions. Our analysis of the *Xcc* (ATCC33913) genome revealed an overrepresentation of a particular family of receptors, named TonB-dependent receptors (TBDRs). These proteins are located in the outer membrane of Gram-negative bacteria and are mainly known to transport iron-siderophore complexes and vitamin B12 into the periplasm. In most cases, the expression of the genes encoding these receptors is under the control of the Fur (<u>F</u>erric <u>u</u>ptake <u>r</u>egulator) repressor and activated under conditions of iron starvation. In contrast to diffusion through porins, transport *via* TBDRs requires energy which is provided by the inner membrane energy-coupling TonB-ExbB-ExbD protein complex. The exploration of complete genome sequences of 226 Gram negative bacteria showed that the overrepresentation of TBDRs is restricted to a small proportion of these bacteria, but is a common trait of all sequenced *Xanthomonas* species. Interestingly, most of the bacteria displaying this particularity have diverse lifestyles and belong to different taxonomical lineages, but they all share the ability to exploit complex carbohydrates. Therefore, we postulated that some *Xcc* TBDRs might be involved in the transport of plant-derived molecules. This hypothesis was recently reinforced with the characterization of a TBDR, named MalA, in the oligotrophic aquatic bacterium *Caulobacter crescentus*, which transports maltodextrins. A systematic study of *Xcc* TBDRs, based on mutagenesis, expression analyses and transport assays identified one *Xcc* TBDR involved in the transport of sucrose, a major plant sugar. This TBDR gene is required for full virulence on *Arabidopsis* and is associated with genes required for sucrose metabolism, thus forming a “sucrose utilization locus”. Our study also suggests the existence of other TBDR-containing loci involved in the utilization of plant cell wall compounds which are conserved in a wide range of bacteria displaying TBDR overrepresentation. # Results ## Identification of *Xcc* TBDRs TBDR proteins consist of two domains, with a C-terminal membrane embedded β-barrel domain that is sealed by the N-terminal plug domain. Hidden Markov models (HMMs), PF00593 and PF07715, corresponding to these two domains, are available in the Pfam database. Seventy-six proteins carrying one or both domains were detected in the proteome of the *Xcc* ATCC33913 strain. Among these proteins, 64 possess both domains. The remaining 12 proteins, which possess only one of the two domains, can be divided into three groups: 3 proteins (*XCC2208, XCC4131* and *XCC4132*) which do not seem to have a canonical plug domain, one protein (*XCC2497*) which has no domain detected by the PF00593 HMM, and finally 8 proteins (*XCC0304-0305, XCC1750-1751, XCC3215-3216* and *XCC3270-3271*) which are truncated and correspond to 4 pseudogenes having a frame-shift in their coding region. Each of these 4 pseudogenes was thereafter studied as a unique entity. Other conserved features of TBDRs were then used to further characterize these identified proteins. In the N-terminal part of the plug domain, TBDRs display a conserved sequence, the TonB-box, that interacts with the TonB protein. A TonB- box, with the *Xcc* consensus sequence tLDXVXV (lower case indicates less highly conserved amino acid, X indicates any amino acid) was detected in all studied proteins. The localization of TBDRs in the outer membrane implies their transport across the inner membrane and thus the presence of a signal peptide at their N-terminal end. The proteins were therefore analyzed for the presence of such a motif. TBDRs which were not predicted to possess a signal peptide in the annotation proposed by da Silva and colleagues were manually re-annotated, thus revealing putative signal peptides. Out of 72 proteins, 1 TBDR (XCC2385) does not have a signal peptide. Finally, the 12 C-terminal amino acids of outer membrane proteins (OMPs) form a membrane anchoring β-sheet with the last residue being aromatic (F in the vast majority of OMPs). The 12 C-terminal amino acid of 55 of the proteins studied here are predicted to form a potential β-sheet, ending with an aromatic residue. Among the TBDRs which did not display a typical C-terminal sequence, 8 have a β-sheet ending with an aromatic amino acid followed by a short extension of 1 to 4 amino acids. Taking all this information together, 48 proteins possess all typical features of TBDRs in *Xcc* ATCC33913 and were thus considered as putative functional TBDRs. The other proteins lacking or having degenerated TBDR domains were also included in our study and then named Ps-TBDRs (for Pseudo-TBDRs). Finally, some TBDRs contain a N-terminal extension, located between the signal peptide and the plug domain. Two types of N-terminal extension have been identified: the transducer and the Oar-like extensions. In *Xcc*, among the 9 TBDRs having an N-terminal extension, one is a member of the TBDR-transducer subclass, seven are in the Oar-like subclass and one displays a N-terminal extension that does not correspond to the two types already identified. Among these TBDRs, only two have all canonical features of TBDRs. ## TBDRs are overrepresented in *Xanthomonas* spp. and in bacteria scavenging complex carbohydrates A survey of TBDRs was performed in 226 eubacterial completely sequenced genomes, by identifying proteins detected by both PF07715 and PF00593 HMMs. This analysis showed that most bacteria (71%) hold less than 14 TBDRs per proteome, whereas the remaining bacteria have a very broad variation in TBDR number, ranging from 14 to 120. In fact, 27% of the bacteria studied here have no TBDR proteins and 43% possess between 1 and 13 TBDRs. Only a very small proportion of bacteria (15.5%) possesses more than 30 TBDRs, thus forming a particular class in which TBDRs seem to be over represented. It is worth noting that most bacteria in this class either belong to the α or γ-Proteobacteria classes or to the *Bacteroides* genus. The ratios between the number of TBDRs found in each proteome and the genome size or the number of annotated coding sequences (CDS) displayed a very similar distribution pattern, thus showing that there was no major bias due to annotation. Therefore, the TBDR/Mbp ratio was used to define several bacterial classes: bacteria displaying a ratio higher than 5 were considered as members of a class showing TBDR overrepresentation, while those having a ratio ranging from 3 to 5 belong to an intermediary class. The 4 *Xanthomonas* species whose genomes have been sequenced belong to the class displaying TBDR overrepresentation. It is worth noting that phytopathogenic bacteria, such as *Pseudomonas syringae* pathovars or *Erwinia carotovora* subsp. *Atroseptica* belong to the intermediary class. ## Seven *Xcc* TBDR and two Ps-TBDR genes are Fur-regulated As TBDRs are mainly known to be involved in iron uptake, it was important to determine how many of these receptors are assigned to this function in *Xcc*. In most cases, the expression of genes encoding TBDRs involved in iron transport is regulated by the iron status in the medium. These genes are activated under iron depletion conditions and repressed under iron repletion conditions *via* the Fur repressor (<u>F</u>erric-<u>u</u>ptake <u>r</u>egulator), that binds to a specific DNA sequence element called the “Fur-box”, which is found in the target promoters of iron-regulated genes. These features led us to analyze the regulation by the iron status for all identified *Xcc* TBDR and Ps-TBDR genes. The 76 *Xcc* TBDR and Ps-TBDR genes were mutated by insertion of the suicide plasmid pVO155, leading to transcriptional fusion with the promoterless *uidA* reporter gene. Two insertions (in *XCC1990* and *XCC3209*) were found unstable. Thus, we constructed deletion-mutations in these genes using the *cre-lox* system. In order to analyze the expression of these 2 deleted TBDR genes, their promoter regions were cloned upstream of a promoterless *lacZ* gene in a reporter plasmid (see), and these constructions were introduced into the wild- type strain. Using β-glucuronidase or β-galactosidase expression assays, we monitored the expression of all *Xcc* TBDRs and Ps-TBDRs in iron-repleted and -depleted media. Seven genes (*XCC0158, XCC0768, XCC1391, XCC2772, XCC3050, XCC3518*, and *XCC4162*) and 2 Ps-TBDR genes (*XCC3216*-*3215 and XCC3595*) are induced by iron starvation, compared to iron-replete conditions. With the exception of Ps-TBDR *XCC3216-3215*, this regulation pattern was confirmed by quantitative real-time reverse transcriptase polymerase chain reaction (qRT-PCR) in a wild-type background (data not shown). We then checked whether these genes are under the control of the Fur regulator. XCC1470, the orthologue of the Fur regulator identified in *Xoo* and in *Xanthomonas campestris* pv. *phaseoli* (*Xap*), was mutated using the manganese method. A point mutant in this latter gene was obtained (*fur*1). As expected for a *fur* mutation, this mutant produced more siderophores than the wild-type strain and this production remained unaffected in response to increased iron levels (from 1 to 50 µM), as determined by using the chrome azurol S assay (data not shown). It is worth noting that the *fur* mutant is unable to induce any disease symptoms (data not shown), as already reported in *Xoo*. We then compared by qRT-PCR analysis the relative expression of the TBDR and Ps- TBDR genes induced by iron starvation, in the wild-type strain *versus* a *fur* mutant, in an iron-containing medium. As presented in, the expression of all these genes was repressed by the Fur protein. However, the repression level of the 2 Ps-TBDR genes was low (less than twofold). We confirmed that the deregulation phenotype of the *fur* mutant was the result of mutation in the *fur* gene by complementation experiments: the repression by iron of the 7 TBDR and the 2 Ps-TBDR genes was restored by providing the wild-type Fur protein on the pL-*XCC1470* plasmid in the *fur* mutant. ## The 9 Fur-regulated TBDR/Ps-TBDR genes, but not the other *Xcc* TBDR genes, possess a putative Fur-box The DNA regions located upstream of the 9 Fur-repressed TBDR and Ps-TBDR genes were analyzed using the MotifSampler program, to identify putative Fur-boxes. For genes arranged in putative operons (*XCC3050* and *XCC0768*), the DNA region located upstream of the first gene of the putative operon was also studied. *XCC3595*, which belongs to the TBDR transducer subclass, is preceded by two genes displaying significant similarities with the *fecI* (*XCC3593*) and *fecR* (*XCC3594*) genes, coding for a sigma factor of the ECF subfamily and its anti- sigma factor, respectively. In most cases, the FecI/FecR system is associated with a TBDR of the transducer subclass and is involved in iron signaling and transport by regulating the expression of the TBDR gene and other “iron- associated” genes (for review see). We thus also analyzed the DNA regions upstream of these two genes. This analysis allowed the identification of a single significant 19-bp palindromic motif (9-1-9 inverted repeat) in the upstream region of most genes or operons studied here. For *XCC3595*, the palindromic motif was detected in the upstream region of *XCC3593* (*fecI*). For *XCC0768*, two motifs were identified: one in the upstream region of *XCC0767* and one in the coding region of this gene. This motif (AATGAgAATcATTctCATT; lower case indicates less highly conserved) only showed a weak similarity with the *E. coli* 19 bp Fur-box consensus sequence (GATAATGATAATCATTATC) : it matched this sequence in 10 bp positions out of 19 (52% identity). However, the conservation was significantly higher with Fur-box consensus sequences identified in *Shewanella oneidensis* (AATGATAATAATTATCATT; 84% identity) and *Bacillus subtilis* (aaTGAtAATnATTaTCAtt; 84% identity). Interestingly, a multiple alignment of these 4 Fur-box consensus sequences showed that although the *E. coli* consensus sequence seems more divergent, it perfectly matches the 3 other sequences after a 3 bp shift. These results suggest that the palindromic motif identified in this study might correspond to a possible *Xcc* Fur-box. We then explored DNA sequences located upstream of the other *Xcc* TBDR or Ps-TBDR genes as well as the entire *Xcc* genome sequence to detect putative Fur-boxes, using the MotifScanner program or the PatScan pattern matcher software. No putative Fur-box was detected in the promoter region of any other TBDR gene. These analyses identified more than 20 genes having candidate Fur-boxes in the region from −300 to +20 relative to the start of translation (data not shown). Several of these genes are orthologs or homologs of genes regulated by Fur in other bacteria, such as *feoA* (*XCC1834*) and *bfd* (*XCC0481*), or involved in iron storage such as *piuB* (*XCC3955*), thus reinforcing the validity of our Fur-box consensus sequence. Altogether, this analysis suggests that only 7 TBDR and 2 Ps-TBDR are directly associated with iron uptake in *Xcc*. The role of the other TBDRs remained to be deciphered. ## Two TBDR genes belong to the *hrp* regulon of *Xcc* During the study of TBDR gene promoters, we identified a pip/hrp_II box in the promoter of 2 TBDR genes: *XCC1041* and *XCC1719*. In *Xanthomonas, hrp* genes coding for a TTSS, as well as other functions related to pathogenicity are positively regulated by *hrpG/hrpX* regulatory genes. HrpG regulates the expression of *hrpX*, which controls the expression of genes possessing a pip-box (TTCGCN₁₅TTCGC) or a hrp_II-box (TTCGN₁₆TTCG). By qRT-PCR experiments, we showed that *XCC1041* and *XCC1719* TBDR genes are activated by HrpG and HrpX. Interestingly, a mutant in the *XCC1719* TBDR gene is weakly affected in pathogenicity (data not shown, see below). This gene seems to be specific to *Xcc* as it is absent in the genomes of *Xcv, Xac, Xoo* and *Xf*, although its surrounding genes are conserved in syntenic regions in all these strains. On the other hand, *XCC1041* is present in all *Xanthomonas* species (but absent in *Xf*) and encodes a TBDR with an atypical N-terminal extension. ## *Xcc* TBDRs and plant interactions: *XCC3358* plays a major role in pathogenicity In order to assess whether *Xcc* TBDRs control the interaction with plants, we tested the 76 TBDR mutants constructed in this study for pathogenicity on the *Arabidopsis thaliana Sf-2* ecotype. Among these mutants, 17 were altered in symptom development (data not shown). However, the alteration was weak and was not reproducible in all the tests we performed, suggesting either that their involvement is not crucial or that functional redundancy mask their individual contribution. However, the *XCC3358* insertion mutant was clearly and reproducibly altered in pathogenicity, showing a clear delay in symptom development in comparison to the wild-type strain. The *XCC3358* gene is likely to form an operon with the downstream gene, *XCC3359*. The ortholog of the *XCC3359* gene, named *suh*, has been characterized in *Xanthomonas axonopodis* pv. *glycines* (*Xag*), the causal agent of bacterial pustule disease on soybean. This gene codes for a sucrose hydrolase, SUH, which plays a major role in sucrose metabolism in *Xag*. A mutant in this gene is moderately affected in pathogenicity on soybean. This prompted us to construct a non polar deletion- mutation in *XCC3358* (*XCC3358*Δ1, see). A deletion-mutant was also generated into *XCC3359* (*XCC3359*Δ1, see). These two mutants showed an altered phenotype, i.e. delayed symptom development, similar to that observed with the pVO155 insertion mutant in *XCC3358*. The non polar mutation in *XCC3358* was complemented by introducing the plasmid pL-*XCC3358*, which carries a functional *XCC3358* gene, into *XCC3358*Δ1 mutant. This result confirms that the *XCC3358* TBDR plays a role in virulence. Similarly, the deleted mutant in *XCC3359* was complemented by a plasmid carrying this gene. ## *XCC3358* TBDR belongs to a locus required for full pathogenicity In *Xcc* genome, *XCC3358* and *XCC3359* are preceded by two other genes that might be related to sugar metabolism: *XCC3356* codes for a putative transcriptional repressor of the LacI/GalR-family and *XCC3357* encodes a putative sugar transporter of the major facilitator family, allowing the transport of substrate molecules through the inner membrane. We constructed insertion mutants in these two genes, presuming that they are monocistronic. These insertions lead to transcriptional fusions with the promoterless *uidA* reporter gene. The insertion mutant in *XCC3357* showed an altered phenotype on the *A. thaliana* Sf-2 ecotype, resembling that using *XCC3358* or *XCC3359* mutants, whereas the insertion mutant in *XCC3356* was not affected in pathogenicity. These results suggest that *XCC3357*, *XCC3358* and *XCC3359* are required in the same process related to sucrose metabolism and/or transport. The phenotype of a mutant in *XCC3356* was not surprising since this gene encodes a putative repressor, which might negatively control the expression of the other genes of the locus. A mutation in this gene should induce the constitutive expression of the other genes without affecting their function (see below). ## The *XCC3356*-*3359* locus is required for sucrose utilization Insertion mutants in *XCC3357, XCC3358* and *XCC3359* and the deletion mutant *XCC3359*Δ1 grew like the wild-type strain on minimal medium containing glucose or fructose (data not shown), but were all affected in growth on sucrose (20 mM). The growth of the insertion mutant in *XCC3356* and the *XCC3358*Δ1 mutant was not impaired in minimal medium containing sucrose. The fact that the insertion mutant in *XCC3358* was impaired in growth on sucrose whereas the deletion mutant in the same gene was not, suggested that the phenotype of the insertion mutant was due to a polar effect of the insertion on *XCC3359* gene expression. This hypothesis was confirmed by complementation experiments. The growth of insertion mutants in *XCC3358* and *XCC3359* on sucrose media was restored when XCC3359 was supplied *in trans* on the expression plasmid pL-*XCC3359*, which allows constitutive expression of the *XCC3359* gene. No complementation was observed when the *XCC3358* gene was supplied *in trans* in the *XCC3358* insertion mutant. These results confirmed that *XCC3358* and *XCC3359* form an operon. They also pointed out that this putative operon and the entire locus is important for sucrose utilization in *Xcc*. Thus, we renamed this locus *sux*, for <u>s</u>ucrose <u>u</u>tilization in *<u>X</u>anthomonas*. For the purpose of this study, the TBDR-amylosucrase operon was renamed *suxAB*, the putative sugar transporter gene, *suxC*, and the putative regulatory gene, *suxR*. However, the role of the TBDR SuxA in sucrose utilization remained elusive: this transporter gene was not required for growth on sucrose, but a non polar mutant in this gene was altered in pathogenicity, like *suxB* and *suxC* insertion mutants. This observation suggested that this *suxA* gene might have a particular function. ## SuxA and SuxC allow sucrose uptake To clarify the role of the SuxA TBDR, we investigated its involvement in sucrose transport, by performing sucrose uptake experiments using \[¹⁴C\]sucrose. First, the uptake rates into the *Xcc* wild-type strain were compared after an overnight preculture in the presence (induced) or absence (uninduced) of sucrose. As shown in, induced cells took-up \[¹⁴C\]sucrose quicker than uninduced cells, but the values obtained after 2 hours of incubation were lower for induced cells. For further experiments, we worked with uninduced cells, in order to study the effect of different mutations on the transport induction. Competition experiments were performed to test transport specificity. A 10-fold excess of unlabelled sucrose reduced the transport rate to 52% of the level obtained without unlabelled sucrose, and with a 100-fold or 200-fold excess, the transport rates were reduced to 10.5 and 7% respectively. Moreover, the addition of unlabelled fructose or glucose (sucrose degradation sub-products) slightly affected sucrose transport; a 200-fold excess of these carbon sources reduced the sucrose transport to about 80% of the control level. The involvement of the SuxA TBDR in sucrose transport was then checked by comparing the uptake rate of \[¹⁴C\]sucrose into the wild-type strain and into a *suxA* non polar mutant (Δ*suxA*). Sucrose transport was much lower in the Δ*suxA* strain, compared to the wild-type strain: after 8 minutes, it reached only 3.5% of the transport rate into the wild-type strain. When *suxA* was supplied *in trans* on a constitutive expression plasmid, sucrose transport was more efficient and rapid, with a transport rate value at 8 minutes corresponding to 484% of that of the wild-type strain. This result confirms that SuxA is required for sucrose entry into *Xcc*. We then studied sucrose transport in *suxR* and *suxC* insertion mutants (*suxR*::pVO and *suxC*::pVO, respectively), and in a *suxB* deletion mutant (Δ*suxB*). When the inner membrane transporter encoded by *suxC* was absent, the sucrose transport rate reached only 2.1% of the value obtained for the wild-type strain. Thus, this protein is necessary for sucrose entry into *Xcc*. On the contrary, sucrose transport was greatly enhanced in the *suxR* repressor mutant and in the Δ*suxB* amylosucrase mutant (488% and 241% respectively of the value obtained for the wild-type strain after 8 minutes). These results confirm the repressor function of SuxR (see below) and suggest that SuxB also has a repressor activity on *suxA* expression. When *suxB* was supplied *in trans* in the Δ*suxB* mutant, the sucrose transport was reduced to about the same level as the wild-type strain (76.4% of the transport rate level of the wild-type strain). Concentration-dependent sucrose transport experiments showed a biphasic kinetics, with a fast rate between 0.01 and 0.1 µM sucrose (*K_d* = 0.033 µM), and a slow rate between 0.25 and 5 µM sucrose (*K_d* = 0.59 µM). This biphasic pattern is very similar to those observed in energy coupled transports of vitamin B12 into *E. coli* through the BtuB TBDR, and more recently of maltose transport into *C. crescentus* through the MalA TBDR. In both systems, it was concluded that the first phase reflects binding of the transported molecule to the OM TBDR and that the second slower phase reflects binding to a cytoplasmic membrane transporter. Thus, we presume that the low *K_d* value (0.033 µM) mainly reflects binding to SuxA and that the higher *K_d* value (0.59 µM) binding to SuxC. Altogether, these data indicate that both SuxA and SuxC are required for sucrose transport. It is worth noting that if the Δ*suxA* mutant and the *suxC* insertion mutant have a different phenotype regarding their growth ability in presence of sucrose, both mutants show a very similar alteration in sucrose transport rate after 8 minutes. However, after 240 minutes, sucrose transport rate in the *suxA*-deleted mutant reached almost 23% of the value obtained for the wild-type strain, whereas in the *suxC* insertion mutant, this rate was less than 7% of that of the wild-type strain. This result suggested the existence of two sucrose uptake pathways, an active and a passive pathway, both requiring transport through SuxC. The active transport pathway depends on SuxA for translocation through the outer membrane, whereas the passive pathway does not. It is assumed that the sucrose uptake rate observed into the *suxA*-deleted mutant is determined by slow and passive diffusion of sucrose across the outer membrane, which is sufficient to support growth in the presence of 20 mM sucrose in the medium. ## Sucrose uptake is energy-dependent Active iron transport through TBDRs depends on the proton motive force (PMF). Thus, we performed sucrose transport experiments in the presence of a PMF inhibitor, carbonyl cyanide 3-chlorophenyl-hydrazone (CCCP). Addition of 20 µM CCCP inhibited \[¹⁴C\]sucrose transport rate to 1.13% of the transport rate in the absence of CCCP, demonstrating that sucrose transport through SuxA and/or SuxC is dependent on the proton motive force and does not occur by facilitated diffusion. These experiments did not permit us to conclude whether SuxA-mediated transport depended on the PMF, since SuxC belongs to the Na⁺-melibiose cotransporter and H⁺ cotransporters family, which require the transmembrane potential for transport across the cytoplasmic membrane. We then tested whether TonB was required for sucrose uptake. Eight genes coding for proteins displaying significant similarities to TonB are predicted in the genome of *Xcc*. These 8 genes were mutated by insertion of the pVO155 plasmid. As the insertion in *XCC2927* was found to be unstable, we constructed a deletion mutant in this gene. Sucrose transport was not significantly impaired in the 8 putative *tonB* mutants. This result suggests a complete or partial functional redundancy between at least two *Xcc* TonB proteins. Such a redundancy has already been observed in various other bacteria, e.g. *Serratia marcescens*, *Vibrio cholerae* and *Pseudomonas aeruginosa*,. ## The expression of *sux* genes is induced by sucrose and repressed by SuxR Using qRT-PCR analyses, we observed that the expression of *suxR*, *suxC* and *suxAB* is specifically induced by the presence of sucrose in the medium. As *SuxR* has a high degree of similarity with members of the LacI/GalR family of transcriptional repressors, we tested whether this protein regulates the expression of *sux* genes by comparing their expression by qRT-PCR in the *suxR* insertion mutant and in the wild-type strain, cultivated in minimal medium with or without sucrose. *suxAB, suxC* and *suxR* were over-expressed in a *suxR* mutant as compared to the wild-type strain in the absence of sucrose, suggesting a negative and effector-dependent control of *sux* genes by SuxR. We also studied the expression of the *suxAB* operon in the presence of different sucrose concentrations. For this purpose, the promoter region of this operon was cloned upstream of the promoterless *LacZ* gene in a reporter plasmid. This plasmid, named pPr-*suxA*, was introduced into the *Xcc* wild-type strain, the *suxR* and *suxC* insertion mutants (*suxR*::pVO and *suxC*::pVO, respectively), and the *suxA* and *suxB* deletion mutants (Δ*suxA* and Δ*suxB*, respectively). These strains were used to perform β-galactosidase assays after 6 hours growth in minimal medium containing a range of sucrose concentrations. Induction of the expression of the reporter gene was detected for sucrose concentrations ranging from 20 µM to 20 mM. Maximal induction (3.5 to 4-fold induction) was observed for concentrations higher than 200 µM in the wild-type background. As expected, the expression of *suxA* was highly induced in the *suxR* mutant, confirming the repressor activity of this gene. However, we observed a significant diminution of the induction level at 2 and 20 mM sucrose in this mutant, suggesting the existence of a second level of control. A high and constitutive induction of *suxA* expression was also observed in the *suxB* deletion mutant, suggesting that the amylosucrase gene plays a role in the repression of the *suxAB* operon. At 2 mM sucrose concentration, a reduction in *suxA* expression level was also observed in this mutant, but it was smaller than that detected in the *suxR* mutant. The deregulation of *suxA* expression in the *suxB* mutant is in agreement with transport studies which showed a higher sucrose transport rate in this mutant. No induction of *suxA* expression was observed in the *suxC* mutant, whatever sucrose concentration was used. This lack of induction certainly reflects the absence of sucrose entry into the cytoplasm of this mutant, thus preventing the alleviation of SuxR repression. This result clearly shows that the induction of *suxA* expression by sucrose requires the entry of this molecule into the cell. The pattern and level of expression of *suxA* observed in the *suxA* mutant were very similar to those observed in the wild-type strain, except that this expression was significantly higher in the mutant in the presence of 2 mM sucrose. This difference was reproducibly observed and further investigations are needed to understand this phenomenon. Expression assays suggested that sucrose transport through SuxA is not high enough to influence the expression of the *suxAB* operon. Sucrose entry across the outer membrane by slow diffusion seems to be sufficient to promote induction of *sux* genes after 6 hours growth in the presence of sucrose. ## *suxA* is required for its rapid induction as revealed by using low sucrose concentrations We performed time-course experiments comparing *suxA* expression in the wild- type strain and the *suxA* deletion mutant grown in presence of different sucrose concentrations (0, 20 µM, 50 µM, 100 µM, 200 µM 500 µM, 1 mM, 2 mM, 5 mM and 20 mM). For clarity, only results obtained with 100 µM and 20 mM are shown in. In the presence of 20 mM sucrose (triangles on), *suxA* expression was not significantly different after 6 hours growth in the wild-type and the *suxA* mutant backgrounds. However, a slight difference was observed in the induction level of *suxA*, from 60 to 120 minutes after sucrose addition. The expression level of *suxA* was 1.5 to 1.6 fold higher in the wild type strain than in the *suxA* mutant. Later, this difference tended to decrease and the ratio reached a value of 1.1 after 6 hours. A reproducible and clear difference in curve profiles was noticed in the presence of 100 µM sucrose (squares on). A rapid induction was observed in the wild-type context, from 30 to 120 minutes after sucrose addition, followed by a plateauing, in which expression seems to remain constant. On the other hand, in the *suxA* deletion mutant background, the induction was linear and less rapid. Between 30 and 120 minutes after sucrose addition, the expression level of *suxA* was 2 fold higher in the wild-type strain background than in the *suxA* deletion mutant background, although after 360 minutes, the expression levels were identical in both strains. Similar results were obtained with sucrose concentrations ranging from 20 to 200 µM (data not shown). We propose that these results could reflect the existence of two pathways controlling *suxA* expression, which can be clearly differentiated with low sucrose concentrations. One of these pathways is controlled by SuxA. These results can be related to the transport analyses showing the existence of two types of sucrose transport across the outer membrane, one being SuxA- dependent and the other one, slower, being dependent on passive diffusion. We thus postulate that the induction of *suxA* indirectly reflects these two means of transport. Moreover, this induction is specific to sucrose and is not due to its degradation products glucose and fructose, as no induction was observed following addition of 100 µM of these molecules in the wild-type background. Therefore, these experiments suggest that SuxA plays a role in sucrose transport which might be crucial at low concentrations of this sugar. ## Identification of other *Xcc* TBDRs in *loci* probably involved in plant carbohydrates utilization The genome of *Xcc* was then explored to see whether other TBDRs could belong to loci involved in the utilization of other plant compounds. *Xcc* has an extensive repertoire of plant cell-wall degrading enzymes, with cellulolytic, pectinolytic and hemicellulolytic activities. The analysis of carbohydrate active enzymes identified in the predicted proteomes of 209 Gram negative bacteria and referenced in the CAZy database (<http://afmb.cnrs-mrs.fr/CAZY/>) showed that, after *Bacteroides* sp. and *S. degradans*, which are well known specialists for polysaccharide degradation, *Xcc* has one of the highest number of genes involved in polysaccharide metabolism per megabase (29.9, total 152). These genes encode 82 predicted glycosyl hydrolases, 45 glycosyl transferases, 5 polysaccharide lyases, 18 carbohydrate esterases and 2 carbohydrate binding proteins. Interestingly, 46 of these proteins are encoded in the vicinity of 24 TBDR/Ps-TBDR genes, thus suggesting the existence of 19 new loci putatively involved in carbohydrate utilization. Among those loci, 7 also contain an inner membrane transporter coding gene and a regulatory gene. For 3 loci identified in this analysis, the substrate probably utilized and transported could be easily deduced from the nature of the degradative enzymes: the *XCC0120* TBDR gene is localized upstream of genes coding for a pectin methyl esterase and a pectate lyase and might belong to a locus involved in pectin utilization; the *XCC4120* TBDR gene is found in a cluster of genes probably involved in xylan metabolism; and the *XCC2469* TBDR gene might be related to maltose/maltodextrin utilization, since it is associated with genes coding for a cyclomaltodextrin glucanotransferase, two α-glucosidases, a sugar transporter and a maltose transport gene repressor. To verify whether these TBDRs are really associated with carbohydrate utilization, we studied their regulation by plant compounds. For this purpose, we performed β-glucuronidase expression assays using all *Xcc* TBDR pVO155 insertion mutants. Cells were grown in rich medium or in minimal medium supplemented or not with polygalacturonic acid (PGA), arabinose, glucose, maltose, sucrose, xylose or xylan. Most TBDR (or Ps-TBDR) genes (44 out of 74) were repressed in rich medium, compared to minimal medium. We also observed that in minimal medium, 48 *Xcc* TBDR genes are repressed in the presence of sucrose, xylose, arabinose and/or glucose, suggesting that these genes are submitted to catabolic repression. However, the repression patterns were variable, suggesting the existence of several repression pathways. It is worth noting that most Fur- regulated TBDR genes were submitted to catabolic repression. Three TBDR/Ps-TBDR genes were induced by PGA (*XCC0120, XCC1749* and *XCC1750-1751*), one by maltose (*XCC2469*), seven by arabinose (*XCC0050, XCC1749, XCC1750, XCC1892, XCC2828, XCC4120* and *XCC4222*), three by xylan (*XCC2828, XCC4120* and *XCC4237*) and finally two by xylose (*XCC2828* and *XCC4120*). Interestingly, the 2 TBDRs induced by xylose were also induced by xylan and arabinose, and *XCC1749* and *XCC1750* were both induced by PGA and arabinose. *suxA* is the only TBDR gene for which expression is specifically induced by sucrose, and *XCC2469*, the orthologue of *C. crescentus malA* TBDR gene, is the only one induced by maltose. It is worth noting that the expression of the *XCC0120, XCC4120* and *XCC2469* TBDR genes was specifically induced by the postulated substrate of the associated CAZy referenced enzymes. Altogether, these data suggest that several TBDRs are part of loci which seem to be involved in carbohydrate utilization. We thus propose the existence of 6 putative loci named <u>c</u>arbohydrate <u>u</u>tilization containing <u>T</u>BDRs (CUT) loci, which were defined by the presence of genes coding for carbohydrate degradative enzymes, inner membrane transporters and sugar related regulators beside TBDR genes. We also identified 15 putative partial CUT loci in the *Xcc* genome. ## Conservation and distribution of TBDRs and CUT *loci* in Xanthomonads When compared with proteins in the databases using BlastP and the “distance tree of results option” displayed on the BlastP result page, the best homologous genes of *Xcc* TBDRs/Ps-TBDRs were putative TBDRs of other *Xanthomonas* species and in some cases of *Xyllela* strains. These conserved genes might be considered as orthologs. We compared the distribution of *Xcc* TBDRs among Xanthomonads strains for which genome sequences are available, i.e. *Xcc* strain 8004, *Xcv* (strain 85–10), *Xac* (strain 306), *Xoo* (strains KACC10331 and MAFF311018) and *Xf* (strains 9a5c and PD). A search similar to that used for *Xcc* allowed us to identify 72 TBDR/Ps-TBDR in the proteome of *Xcc* strain 8004, 74 in that of *Xac*, 61 in *Xcv*, 41 and 42 in *Xoo* strains KACC10331 and MAFF311018 respectively, and finally 10 in each *Xf* strain. To further study the relationship between these TBDRs and *Xcc* TBDRs, we performed a comparative study using alignments generated by ClustalW. To carry out this comparative analysis, we only used TBDRs which seemed to be complete and Ps-TBDRs over 600 amino acids, in order to avoid bias in alignments. Examination of the phylogenetic tree, coupled with Blast results analysis, clearly confirmed that most *Xcc* TBDRs are well conserved in other *Xanthomonas* strains. All TBDR or Ps-TBDR genes detected in *Xcc* strain ATCC33913 are present in strain 8004; 55 seem to have orthologs in *Xac*; 49 are conserved in *Xcv* and 34 only are conserved in *Xoo* strains which possess significantly less TBDRs. In most instances, the grouping and distribution pattern of orthologous TBDRs matched the phylogenetic classification of *Xanthomonas* species –. This observation suggests that this protein family is ancient in the *Xanthomonas* genus. It is worth noting that branch lengths were variable, suggesting differential evolution rates. Moreover, we noticed that some truncated *Xcc* Ps-TBDRs could be functional in other strains: the Ps-TBDR XCC1750-1751 doublet corresponds to unique and complete TBDR proteins in *Xcv, Xac* and *Xcc* strain 8004. Similarly, the 2 Ps-TBDRs XCC3215-3216 and XCC3270-3271, which are also truncated in *Xcc* strain 8004, correspond to unique proteins in both *Xac* and *Xcv*. Interestingly, *Xac, Xcv* or *Xoo* TBDR orthologs of *Xcc* Ps-TBDR having non-canonical C-terminal regions (β-sheet ended by an aromatic amino acid followed by a short extension of 1 to 4 amino acids), also displayed this feature (data not shown). In most cases, these non canonical C-terminal domains were identical in all strains. This conservation might reflect a specific functional feature of *Xanthomonas* TBDRs. The analysis of the phylogenetic tree also showed that Xanthomonads TBDRs can be divided into three main groups, with most TBDRs belonging to group 1. This group can be divided into 8 subgroups (1A to 1H). We noticed that (i) *Xcc* Oar-TBDRs are clustered in subgroup 1D, (ii) most plant carbohydrate induced TBDRs (8 out of 11) are grouped in subgroup 1C, and (iii) most *Xcc* Fur-regulated TBDRs (6 out of 8) are grouped in subgroup 1F. Among these Fur-regulated TBDR genes, *XCC3518*, *XCC3595* and *XCC4162*, which belong to subgroup 1F, seem specific to *Xcc* strains. The comparison of genes located adjacent to orthologous TBDR genes in the different *Xanthomonas* strains showed that in most cases, these regions are syntenic. Thus, all putative CUT loci identified in *Xcc* genome are well conserved in *Xac* and *Xcv* but only 2 out of 6 were found in *Xoo*. Similarly, almost all putative partial CUT loci are present in both *Xcv* and *Xac*, but only 9 out of 15 are conserved in *Xoo* strains. The putative polygalacturonate utilization locus (*XCC0120-XCC0122*) seems unique to *Xcc*. The genes bordering this locus are conserved and contiguous in *Xcv* and *Xac*, thus suggesting insertion or deletion events. Finally, among the 10 TBDRs identified in *Xf* strains, 6 are conserved in *Xcc, Xac, Xcv* and *Xoo*, 1 is conserved in *Xcc, Xac* and *Xcv*, whereas only 1 is present in *Xcc*. The 2 remaining *Xf* TBDRs (XF0339/PD1711; XF0599/PD1552) seem more divergent and thus specific to *Xf* strains. However, XF0599 displays weak similarities with XCC2772, a Fur-regulated TBDR of *Xcc*. None of the other *Xf* TBDRs are related to *Xcc* Fur-regulated TBDRs nor to *Xcc* TBDRs belonging to putative CUT loci identified in this study. Only 2 *Xf* TBDR genes (*XF2713* and *XF1036*) correspond to *Xcc* TBDRs present in partial CUT loci. ## *Xcc* TBDRs are conserved in aquatic bacteria and/or phytopathogenic bacteria For each *Xcc* TBDR included in the phylogenetic study, we analyzed the BlastP results obtained on the nr databank, to characterize the next best homologous genes after Xanthomonads orthologs. This analysis allowed us to cluster *Xcc* TBDRs on the basis of the bacterial origin of the homologous TBDR. Thus, *Xcc* TBDRs could be divided into 4 main groups. The first group corresponds to TBDRs showing their next best homologies (after Xanthomonads orthologs) with other *Xanthomonas* TBDRs rather than with TBDRs from other genera. This observation mainly concerned TBDRs found in subgroups 1B, 1C and 1D of the phylogenetic tree. Their positions in the tree suggested that they could be generated by successive duplication events that occurred before *Xanthomonas* speciation. The second group corresponds to TBDRs which display high similarities with TBDRs of β-Proteobacteria and/or *Pseudomonas* species. Interestingly most of these TBDRs are clustered in subgroup 3 of the phylogenetic tree. Moreover, some of these TBDRs showed high similarities with TBDRs of phytopathogenic bacteria such as *Acidovorax avenae* subsp. *citrulli, Pseudomonas syringae* pathovars or *Ralstonia solanacearum*. The third group contains two *Xcc* TBDRs proposed to be involved in iron uptake and showing significant similarities with putative TBDRs from *Cyanobacteria*. The fourth group, which is the largest one (with 35 TBDRs out of 70), corresponds to TBDRs which displayed very significant homologies with putative TBDRs of a wide range of aquatic bacteria belonging to the α or γ classes of Proteobacteria. It is worth noting that in most cases, *Xcc* TBDRs were not affiliated to TBDRs of a specific bacterial class, but there was rather a variety of origins. Thus, for 16 *Xcc* TBDRs the best similarities (after Xanthomonads similarities) were obtained with putative TBDRs of *C. crescentus* CB15 and *Caulobacter* sp. K31 strains, which are aquatic oligotrophs belonging to the Caulobacterales order of the α-Proteobacteria. Homologies were also observed with TBDRs of other α-Proteobacteria living in aquatic habitats, such as *Oceanicaulis alexandrii, Maricaulis maris* or *Parvilarcula bermudensis* HTCC2503, or found in multiple environments like *Sphingomonas* sp. SKA58, *Sphingopyxis alaskensis* or *Novosphingobium aromacitovorans*. Significant similarities were also obtained with putative TBDRs of aquatic γ-Proteobacteria classified in the *Alteromonadales*, including *Alteromonas macleodii, Saccharophagus degradans* 2–40, *Pseudoalteromonas* and *Shewanella* species. One common trait of most of these bacteria is that they show TBDRs overrepresentation with values of TBDR number per megabase ranging from 7.4 to 15.7. This raised the question of whether homologies between these TBDRs were fortuitous and a consequence of their large number or whether they reflect common biological functions. ## Conservation of TBDR regions and CUT loci in Gram-negative bacteria Some regions surrounding TBDR genes were found to be conserved in closely related bacteria such as *Vibrio parahaemolyticus* or *P. aeruginosa*. Thus, the XCC3050 Fur-regulated TBDR belongs to a cluster of 6 genes showing similarities with the *pvuA*-*pvsABCDE* gene cluster of *V. parahaemolyticus*, involved in the uptake and biosynthesis of the siderophore vibrioferrin. Similarly, a group of proteins encompassing XCC3067 TBDR and putatively involved in cobalamin uptake and biosynthesis in *Xcc* was conserved in *Pseudomonas* species. Preliminary experiments showed that the expression of this TBDR gene is repressed by the presence of vitamin B12 in *Xcc*, thus suggesting that this cluster is functional (data not shown). Several CUT loci identified in *Xcc* were not conserved in taxonomically related bacteria but rather in the group of aquatic bacteria showing TBDR conservation and in particular in *C. crescentus*. The *sux* locus, 3 of the putative CUT loci and 4 of the putative partial CUT loci identified in this study were entirely or partially conserved in this group of bacteria. Thus, the putative maltose CUT locus was partially conserved with the *mal* locus recently identified in *C. crescentus*, which contains the MalA TBDR proposed to be involved in the uptake of maltodextrins. Interestingly, MalA shows significant similarities with the *XCC2469* TBDR whose expression is induced by maltose and maps in the maltose CUT locus. Similarly, the putative xylose locus, containing the *XCC2828* TBDR gene, is also very well conserved in *C. crescentus*. The corresponding *CC0999* TBDR gene was shown to be induced by xylose in this bacterium. Moreover, *CC2832*, which was also shown to be xylose-induced in *C. crescentus*, displays significant homologies with the xylose-induced TBDR gene *XCC4120* belonging to a xylose CUT locus. Furthermore, the 20-bp palindromic motif conserved upstream of genes of the *C. crescentus* xylose regulon, was also found upstream of genes putatively involved in xylose metabolism in *Xcc*, as well as in the promoter region of *XCC2828* and *XCC4120* TBDR genes. The *sux* CUT locus seems less well conserved and showed some degree of variation. The amylosucrase gene (*XCC3359*) is only well conserved in the corresponding *C. crescentus* locus. In the loci conserved in other bacteria, the degradation of sucrose seems to involve more classical pathways (for review see). Moreover, the MFS transporter of the *Xcc* locus is different from that of the *C. crescentus* sucrose locus encoded by *CC1133*, although it is well conserved in both *Sphingomonas* SK58 and *Erythrobacter litoralis*. These differences show the existence of some degree of plasticity in the evolution of the putative CUT loci. Further investigations are needed to see whether all these partially conserved loci are involved in the utilization of the same molecule. However, this wide conservation of CUT loci is in favor of their existence. These results also confirm that the similarities observed between TBDRs are probably not fortuitous. # Discussion ## Overview The outer membrane (OM) of Gram negative bacteria serves as a selective permeation barrier, excluding hydrophilic solutes, including most nutrients. However, OMs contain embedded integral proteins, named outer membrane proteins (OMPs), which allow sensing and entry of nutrients into the cell. A major class of OMP with a certain substrate specificity, called porins, allows the translocation of hydrophilic solutes through the OM. Another class of OMPs, the TonB-dependent receptors (TBDRs), is mainly known to be involved in iron or vitamin B12 uptake. Recently, the MalA TBDR of *C. crescentus* was shown to be involved in the ExbBD-dependent uptake of maltodextrins. Here, we show that several Gram negative bacteria, belonging to different lineages and having diverse habitats, display an overrepresentation of TBDRs, which might be related to the uptake of plant-derived carbohydrates. ## *Xcc* TBDRs belonging to the Fur regulon Our global study of *Xcc* TBDRs showed that only a small fraction of them seems to be involved in iron uptake. Among the 72 TBDR/Ps-TBDRs identified in the *Xcc* genome, only 9 are up-regulated under iron-limiting conditions. We established that these 9 genes are repressed by the Fur repressor and that they are the only TBDR genes having a Fur-box in their promoter region. Recently, a proteomic approach carried out in *P. aeruginausa* PAO1, which contains 34 putative TBDR genes, identified a very similar number of TBDR genes regulated by iron-stavation. In fact, in this bacterium, 7 TBDRs are produced under iron- starvation conditions and 4 others are specifically induced by the presence of heterologous siderophores, under iron-restricted conditions. Therefore, although it is possible that some *Xcc* TBDR genes need specific heterologous siderophores for their expression, as observed in *P. aeruginosa*, the number of TBDR genes involved in iron uptake seems very comparable in both bacteria. This suggests that the other putative TBDRs/Ps-TBDRs might be required for different biological functions. ## The identification of CUT loci/systems in *Xcc* suggests a relationships between TBDRs and carbohydrate utilization Our data reveal that several *Xcc* TBDRs are related to plant carbohydrate utilization. A large proportion of *Xcc* TBDR genes are coupled with carbohydrate active enzymes. For 6 of them, an inner membrane transporter gene and a regulatory gene are also present in the same region, thus defining the existence of putative CUT loci involved in the uptake and utilization of plant carbohydrates. Moreover, 15 *Xcc* TBDR genes belong to partial CUT loci, thus suggesting the existence of CUT systems composed of different parts scattered in the genome. The existence of such multipartite CUT systems was supported by the observation that 11 TBDR/Ps-TBDR genes are specifically induced by plant carbohydrates such as sucrose or plant cell wall derived compounds including arabinose, xylan/xylose, pectin/polygalacturonate or maltose. Moreover, in 5 cases, there is a correlation between the inducing carbohydrate and the degradative enzyme(s) present in the CUT locus. This allows us to propose the existence of a sucrose CUT locus as well as loci involved in the utilization of complex carbohydrates such as pectin, xylan or starch. ## The *sux* CUT locus is functional The existence of functional CUT loci was confirmed by the detailed study of the sucrose CUT locus, which showed its involvement in the entry and utilization of sucrose in *Xcc*. This locus comprises four genes, *suxA, suxB, suxC* and *suxR*, coding for a TBDR, an amylosucrase, a sugar inner membrane transporter and a regulatory protein, respectively. \[¹⁴C\]sucrose uptake experiments showed that SuxA and SuxC are both required for sucrose entry into the cell. Concentration-dependent sucrose transport experiments showed a biphasic kinetic, with a similar pattern observed for vitamin B12 uptake into *E. coli* through the BtuB TBDR, and more recently for maltose transport into *C. crescentus* through the MalA TBDR. In both of these systems, it has been concluded that the first phase reflects the binding of the transported molecule to the OM TBDR, and that the second slower phase reflects the binding to a cytoplasmic membrane transporter. Thus, we presume in our experiments that the low *K_d* value (0.033 µM) mainly reflects binding to SuxA and that the higher *K_d* value (0.59 µM) binding to SuxC. Sucrose transport was significantly lower in the *suxC* mutant than in the *suxA* non polar mutant. This difference might explain the differential phenotype of these two mutants observed for growth on sucrose: the *suxA* non polar mutant was not impaired in growth on media containing sucrose, whereas the *suxC* mutant was unable to grow under these conditions. These observations suggest the existence of an alternative pathway that supports facilitated diffusion of sucrose across the OM, whereas it seems that there is a unique route for crossing the inner membrane, depending on SuxC. The existence of this alternative pathway was also revealed by expression analyses of the *sux* locus. The regulation of *sux* genes seems to follow a classical inducer/repressor control, mediated by the SuxR repressor, sucrose being the inducer. The expression of *sux* genes is induced by sucrose but not by fructose or glucose. This induction was detected with sucrose concentrations ranging from 20 µM to 20 mM. These experiments showed that SuxA transport influenced sucrose induction of *suxA* at low sucrose concentrations (20 to 200 µM), whereas this effect is masked at higher concentrations, probably by interference by passive diffusion. Our data also indicated that sucrose transport through the *sux* system is active and depends on the proton motive force. However, we could not conclude whether it depends on the TonB-ExbBD energy coupling system. In addition to TonB, the *Xcc* genome harbours 7 TonB-like proteins, which might substitute one for another. Another *Xcc* feature is the presence of a second *exbD* gene, named *exbD2*, mapping downstream of the *tonB-exbBD1* locus. Interestingly, *exbD2* is not required for iron uptake but is essential for HR induction, whereas *tonB*, *exbB* and *exbD1* genes are necessary for both processes. This differential behavior might be related to the existence of at least two classes of TBDRs in *Xcc*, one involved in iron uptake and the other one in transport of plant compounds. Further work is needed to address this hypothesis. ## The *sux* locus represents a new sucrose utilization system The *Xcc sux* locus is clearly different from other sucrose utilization loci already found in bacteria (for review see). In particular, it differs from the *scr* sucrose-utilization system from enteric bacteria, which is also present in the plant pathogenic bacterium *Erwinia amylovora*, where it plays a major role in plant colonization. The differences concern regulation, sucrose utilization and transport. The *scr* genes are regulated by a LacI/GalR family repressor, but fructose is the inducer. In this system, the degradation of sucrose uses a sucrose-6-phosphate hydrolase/fructokinase degradation pathway. On the contrary, the *Xcc sux* locus contains an amylosucrase orthologous to SUH, a unique sucrose hydrolase previously characterized in *Xag*. This enzyme is responsible for intracellular sucrose hydrolysis. It is active on sucrose but not on sucrose-6-phosphate. In the s*cr* system, sucrose metabolism and uptake are coupled by the phosphoeneolpyruvate-dependent carbohydrate:phosphotransferase system (PTS), which controls crossing of the inner membrane and sucrose phosphorylation. The transport across the outer membrane is mediated *via* a porin, named ScrY. Thus, the transport of sucrose is very different in *sux* and *scr* systems. Interestingly, the *K_d* value of sucrose binding to SuxA is 1500- to 3000-fold lower than that of the *E. coli* ScrY sucrose porin, which varies from 13 mM to 50 mM. Similarly, *C. crescentus* MalA TBDR transports maltodextrins with *K_d* values 1000-fold lower than those of the LamB porin, which facilitates the passive diffusion of maltodextrin. It is worth noting that the *K_d* values of sucrose binding to SuxA and maltodextrins binding to MalA are comparable. Thus, SuxA and MalA represents a new class of outer membrane carbohydrate transporters showing a much higher affinity for their substrate than porins. The importance of the *sux* locus in *Xcc* is highlighted by the fact that it is required for full virulence on *Arabidopsis thaliana*. The phenotype of *suxB* and *suxC* mutants on plants can be related to their inability to grow on medium containing sucrose (20 mM). However, the non polar *suxA* mutant, which grows as well as the wild type strain on medium containing sucrose, was also affected in pathogenicity. Thus, it appears that the ability to scavenge sucrose with a very high affinity plays a key role during the interaction with host plants. In conclusion, data obtained on this sucrose CUT locus strongly support the existence of the other CUT loci identified in *Xcc*, which might have a similar mode of action (see model presented). This suggests the presence of several systems which seem to partially overlap and which are involved in the scavenging of plant molecules. They might form a complex network required for the exploitation of plant resources but which might also participate in signaling. ## TBDRs and CUT loci are ancient in the *Xanthomonas* genus The importance of TBDRs and CUT loci is not restricted to *Xcc* since they are well conserved in *Xac* and *Xcv*. A phylogenetic study of Xanthomonads TBDRs suggests that these proteins are an ancient class of proteins in the genus. Moreover, the grouping pattern observed in our phylogenetic study seems to correlate with functional features of *Xcc* TBDRs, thus suggesting the existence of structure/function relationships. However, it seems that there is some degree of variation in the repertoire of TBDR genes and CUT loci among *Xanthomonads*, as the number of genes and loci was lower in *Xoo* strains and also in *Xylella*. This latter species is considered as a minimal pathogen in the Xanthomonads with a restricted habitat and a reduced genome. *Xylella* possesses a significantly lower number of TBDRs and only one partially conserved CUT locus. ## Conservation of *Xcc* TBDRs and CUT loci reveals carbohydrate scavenging abilities in other bacteria One of the main surprises of this work was the observation that the large majority of *Xcc* TBDRs (35 out of 72) display significant similarities with TBDRs of bacteria which are mostly found in aquatic habitats and which are not closely related to Xanthomonads. Phylogenetic studies place Xanthomonads as a deep branch in the γ-Proteobacteria class, close to that of the β-Proteobacteria. The bacteria possessing similar TBDRs belong either to the α-Proteobacteria class or to the Alteromonadales order of the γ-Proteobacteria. Most bacteria in this latter order have been collected from diverse aquatic environments and belong mainly to 6 different genera: *Alteromonas, Colwellia, Idiomarina, Pseudoalteromonas, Saccharophagus* and *Shewanella*. The α-Proteobacteria mentioned in this study are also found mostly in aquatic habitats and are members of several orders, including Caulobacterales, Sphingomonadales, Rhodospirillales and Rhodobacterales. Similarities were not restricted to TBDRs genes, and genome context analyses showed that at least 8 putative (partial) CUT loci identified in *Xanthomonas* species were conserved among members of these aquatic bacteria. Several of them are found in association with abiotic or biotic surfaces, such as the surface of alga or shellfish. Some are adapted to live in oligotrophic environments, while others are abundant in nutrient-rich habitats or in association with particulate of organic matter. Some others are associated with decaying tissues of plant or algae. Although *Xanthomonas* seems to be very different from these bacteria in terms of habitat, lifestyle and taxonomy, we identified several common traits. All these bacteria belong to the class showing a TBDR overrepresentation (with TBDRs/Mbp ratios\>7.4). Most of them are able to degrade a wide variety of plant molecules or other complex carbohydrates such as chitin, alginate, as well as various aromatic compounds. This combination of characters and the similarities with *Xcc* suggest the existence of a shared biology which might be related to the ability to scavenge carbohydrates. If the situation in these bacteria is similar to that observed in *Xcc*, we can speculate that there is a significant proportion of TBDRs involved in the uptake of molecules other than iron- siderophores complexes. Moreover, as previously noticed, we observed that bacteria having a restricted habitat, such as obligate parasites or symbionts, have no or only a very small number of TBDRs. Thus, these proteins might be considered as good indicators of bacterial lifestyle. ## Association between TBDRs and carbohydrate utilization in other Proteobacteria A specific role for TBDRs has already been attributed to bacteria displaying an overrepresentation of this protein family. Sphingomonads are widely distributed in nature and are mostly found in soils and aquatic environments. Some strains are associated with plants. Strains like *Sphingomonas* sp. A1 are able to take up and metabolize macromolecules such as alginate produced by brown seaweed and certain bacteria. This uptake is mediated by large structures, so called superchannels, which form a pit in the outer membrane, and which function as a funnel or concentrator (for review, see). Four TBDRs seem to be part of this superstructure and are thus proposed to participate in alginate transport. Recently, the manipulation of these superchannels revealed the importance of these structures in bioremediation. In *C. crescentus*, beside the prediction of 67 TBDRs, a genome analysis revealed the presence of several genes for the breakdown of plant polysaccharides as well as transport systems, suggesting that plant polymers are a significant source of nutriment for this organism. Interestingly, several *C. crescentus* TBDR genes have been shown or proposed to be associated with the uptake of plant compounds. As described above, Neugebauer and colleagues showed that *C. crescentus* can grow on maltodextrins and that the transport of these molecules is mediated by the MalA TBDR. Proteomic and transcriptomic studies have shown that TBDR genes are expressed at higher levels in minimal medium than in rich medium,, as we observed in *Xcc*. Moreover, several TBDR genes are specifically induced by the presence of xylose. Interestingly, 2 TBDRs induced by xylose are conserved between *C. crescentus* and *Xcc*. Strikingly, putative regulatory *cis*-element motifs are conserved in the promoters of these homologous genes in both species. Moreover, one of these 2 *Xcc* TBDR genes (*XCC2828*) belongs to a putative partial CUT locus which is conserved in *C. crescentus*. This suggests a possible lineage between these loci. Moreover, 3 other *Xcc* CUT loci are entirely or partially conserved in the *C. crescentus* oligotrophic bacterium, including the maltose and sucrose CUT loci. TBDRs in oligotrophs like *C. crescentus* might play a very important role by allowing the foraging of carbohydrates in nutrient poor environments. ## Common themes between *Xanthomonas* and aquatic bacteria It is clear that the life cycle of *Xanthomonas* spp. presents common features with the different lifestyles of bacteria described above. The leaf surfaces encountered by *Xanthomonas* during their epiphytic development might correspond to an oligotrophic environment. Plant leaves release secondary metabolites which can be used by epiphytic microorganisms. However, carbon resources have been shown to be the most limiting resource on plants. Simple sugars, like glucose, fructose and sucrose, are the most dominant carbon sources on plants that have been examined. Studies with bacterial biosensors *in situ* on plants revealed a high heterogeneity of sucrose availability, with an average accessibility of only about 20 µM on moist bean leaves. Interestingly, this sucrose concentration allowed the induction of *suxA* gene in our expression assays, thus suggesting that the *sux* uptake system might function on plant surfaces. The involvement of *sux* locus in epiphytic life is now being investigated. Furthermore, TBDRs might facilitate the exploitation of plant debris generated during disease development, thus resembling bacteria living on dead tissues. This phenomenon might have an impact on *Xanthomonas* life cycle by increasing *Xanthomonas* population size and thus facilitating new infections. ## TBDRs and virulence on plants It appears that in *Xanthomonas*, some TBDRs such as SuxA play an additional and specific role by controlling virulence. Therefore, it is tempting to propose that a strategy shared with non-pathogenic bacteria and based on the active uptake of plant-derived nutrients, could have been diverted by *Xanthomonas* for the control of pathogenicity. It is worth noting that several phytopathogenic bacteria such as *P. syringae* or *E. carotovora* subsp. *Atroseptica*, as well as *Pseudomonas* species associated with plants, have an intermediate overrepresentation of TBDRs, suggesting that this feature could be shared by other bacteria interacting with plants. In *R. solanacearum*, the expression of *hrp* genes coding for a type III secretion system (TTSS) which controls disease and HR development, is specifically induced upon contact with plant cells. This signaling is mediated by PrhA, a TBDR belonging to the transducer subclass. This receptor senses contact with plant cells and transduces this signal into the cytoplasm via PrhI and PrhR, which are FecI/FecR homologs. Apparently, this regulatory system does not require the transport of plant molecules. Although *Xac, Xcc, Xoo* or *Xcv* do not carry any TBDR showing a high homology with PrhA, recently a TBDR controlling both HR and pathogenicity was described in *X. oryzae* pv. *oryzicola*, the causal agent of bacterial leaf streak of rice. This TBDR does not belong to the transducer subclass and only shows a weak similarity with PrhA. This TBDR gene is highly conserved in *Xoo* strain MAFF 311018 (*XOO0785*), *Xcv* (*XCV3654*) and *Xcc* (*XCC0674*). Interestingly, in *Xoo* it is located close to *trh* (*XOO0783*), a regulatory gene controlling the expression of *hrp* genes. The orthologue of this regulatory gene is also conserved in *Xcc* (*XCC0672*), in the vicinity of the *XCC0674* TBDR gene. Thus, it is possible to speculate that both genes might be involved in a common circuit controlling the expression of *hrp* genes in *Xanthomonas* spp. In our study, a mutation in *XCC0674* did not affect the pathogenicity of *Xcc*. However, as there are more TBDRs in *Xcc* than in *Xoo* strains, we can speculate that functional redundancy might have masked the effect of the mutation. Further work is needed to see whether these genes regulate *hrp* genes in *Xcc*. Nevertheless, we established a link between *hrp* genes and TBDRs in *Xcc*. Indeed, we observed that 2 *Xcc* TBDR genes are regulated by the *hrpG* and *hrpX* regulatory genes. Therefore, it seems that there is an overlap between *hrp* genes and at least 2 TBDR genes. What is the function of these two TBDRs? Are they specifically required during the infection of plants, through the action of the *hrp* regulon, to exploit specific released plant molecules? ## What other functions for *Xcc* TBDRs? We have been able to define a putative role for 21 TBDRs out of 72 identified in *Xcc*. The nature of the molecules putatively transported by the other TBDRs remains to be discovered. It is probable that TBDRs are not restricted to the transport of carbohydrates and that they can take up various other molecules produced by plants. *Azospirillum irakensis*, a plant associated bacterium, is able to metabolize salicin, a phenolic glycoside produced by plants. Interestingly, the *sal* operon, which controls the degradation of salicin, contains a TBDR gene which was proposed to be involved in salicin uptake. Thus, secondary metabolites including phenolic compounds might be assimilated through TBDRs. We are now trying to characterize which new molecules may be transported by *Xcc* TBDRs, to better understand the adaptation of this pathogen to host plants. ## TBDRs, CUT loci and evolution in Gram-negative bacteria This work has identified the existence of an ensemble of bacteria that have an overrepresentation of TBDR genes, and that share specific loci for the scavenging and utilization of carbohydrates. They belong to very different lineages in Proteobacteria and this raises the question of the origin of these TBDRs and CUT loci. Did they arise by convergent evolution or were they transferred from species to species by lateral gene transfer? Our data suggest that this latter hypothesis is most likely. Recently, a genomic comparative study established that *Xcc* and *Xac* have close to 40% of their genes showing highest similarities to genes from non γ-Proteobacteria, especially from α-Proteobacteria (20%). These genes seem to belong to genomic islands, denominated “unusual best-match islands” (UBIs). Interestingly, among 35 UBIs thus identified in *Xcc* genome, 14 contain TBDRs and/or CUT loci. However, there was no significant difference between the GC content of each of these loci and the overall content of the genome. It is therefore possible that these loci were acquired very early in the evolution of Xanthomonads and thus played a key role in their adaptation to plants. Most bacteria showing a TBDR overrepresentation possess TBDRs or CUT loci conserved with *Xcc*. However, there is a main exception with members of the Bacteroidetes phylum, as none of the TBDRs characterized in this phylum showed very significant similarities with those identified in *Xcc*. Bacteroidetes can be encountered in two very different niches, the marine environment and the human intestine. Marine Bacteroidetes such as *Gramella forsetii* are associated with particulate of organic matter, whereas those found in the intestine are assembled on partially digested food particles. In both cases, these bacteria are able to consume biopolymers. *Bacteroides thetaiotaomicron*, which is a prominent mutualist in the distal intestine of adult humans, has the largest ensemble of TBDR genes and glycobiome yet reported. Studies have shown that this bacterium has a carbohydrate foraging behavior. It is well known to bind starch through a protein complex of the outer membrane, which comprises SusD and the SusC TBDR protein. One hundred and six paralogs of SusC and fifty three paralogs of SusD were predicted in the *B. thetaiotaomicron* genome. In our study, we did not detect in *Xanthomonas* genomes any protein displaying similarities with SusD. Moreover, none of the TBDRs identified in *Xcc* showed strong similarities with SusC. A Blast analysis suggested that the SusC TBDR and the SusD OMP are specifically conserved in Bacteroidetes (data not shown). These results suggest that TBDRs involved in carbohydrate uptake evolved independently in Proteobacteria and Bacteroidetes. The analysis of the function and evolution of TBDRs in these phyla will certainly help us to better understand the adaptation of bacteria to their environment. This knowledge, which concerns the utilization of plant molecules that are widespread in the environment, will have a major impact not only in plant pathology, but also in human health as well as in the cycling of carbon and geobiology in marine environments. # Materials and Methods ## Bacterial strains, plasmids and growth conditions The *Xanthomonas campestris* pv. *campestris* (*Xcc*) strains and plasmids used in this study are listed in. *Xcc* cells were grown at 30°C in MOKA rich medium (Yeast Extract 4 g/l, Casamino acids 8 g/l, K₂HPO₄ 2 g/l, MgSO₄.7H₂O 0.3 g/l) or in MME minimal medium. *E. coli* cells were grown on LB medium. Antibiotics were used at the following concentrations for *Xcc*: rifampicin, 50 µg/ml; kanamycin: 50 µg/ml; tetracycline: 5 µg/ml. Antibiotics were used at the following concentrations for *E. coli*: ampicillin, 50 µg/ml; kanamycin: 50 µg/ml; tetracycline: 10 µg/ml. Growth curves were generated using the Bioscreen C instrument (Labsystems, Helsinki, Finland) in three independent experiments. Growth measurements were realized in 200-well microtiter plates on 350 µl volumes of a minimal medium containing 20 mM sucrose, inoculated at an OD₆₀₀ = 0.15 from a washed starter culture. Non-inoculated wells were used as asepsis controls. Optical densities at 600 nm values were measured every 30 min over a period of 2 to 3 days at 28°C. The microplates were shaken for 5 sec before each measurement. ## Construction of *Xanthomonas campestris* pv. *campestris* mutants Insertion mutants were constructed using the suicide plasmid pVO155. Oligonucleotide primers used for PCR amplification will be provided upon request. Amplicons were 300 bp in average. Location of insertions are indicated in. Deletion mutants in *XCC3358, XCC3359, XCC1990, XCC3209* and *XCC2927* were constructed using the *cre-lox* system adapted from Marx and colleagues. Deleted regions are indicated in. A *fur* mutant strain was obtained using the manganese mutagenesis method in LB medium containing 5 mM MnCl₂. After incubation for 48 h at 30°C, surviving colonies were harvested for siderophores over-expression on CAS agar plates adapted for *Xanthomonas* (K₂HPO₄ 1 g/l; MgSO₄ 1 mM; Casamino acids 0.15 g/l; (NH₄)₂SO₄ 1 g/l; sucrose 20 mM; CAS 60.5 mg/l; HDTMA 72.9 mg/l; FeCl₃ 10 µM). A resistant strain over-expressing siderophores was selected; the *fur* gene was sequenced and contains a point mutation (T212A leading to L71Q). ## Plasmid constructions The *XCC3358, XCC3359* and *XCC1470* genes (*suxA, suxB* and *fur* respectively, see) were amplified by PCR using appropriately designed primers (Oligonucleotide primers used for PCR amplification will be provided upon request). PCR products corresponding to *suxA* and *suxB* genes were cloned into pCZ525, a derivative of pSC154, without *cyaA*' coding sequence. The obtained plasmids were cloned into pLAFR6 to give pL-*XCC3358* and pL-*XCC3359* respectively. The PCR product corresponding to the *fur* gene with its promoter and terminator sequences was cloned into pFAJ1700 to give pL-*XCC1470*. The *XCC3358, XCC1990* and *XCC3209* promoter regions were PCR amplified with appropriately designed primers. These promoter regions were cloned as *Hind*III-*Xba*I fragments, into the pCZ750 plasmid, a pFAJ1700 derivative containing the *Kpn*I-*Asc*I *lacZ* gene from the pCZ367 plasmid. ## Expression studies β-galactosidase and β-glucuronidase assays: bacterial cultures in the appropriate medium were harvested at different time points and β-galactosidase and β-glucuronidase assays were performed as previously described. Quantitative RT-PCR (qRT-PCR): a 6 hour bacterial culture in the appropriate medium was harvested at an OD₆₀₀ = 0.4 to 0.6. RNA were extracted using the Rneasy Mini Kit (QIAGEN). One µg of RNA was treated with RNase-free DNase I (GE-Healthcare) for 15 min at 37°C. After DNase inactivation (10 min at 75°C), RNA were reverse-transcribed by Superscript II (Invitrogen) using random hexamers (Biolabs), for 2 min at 25°C followed by 1 h at 42°C. Quantitative-PCR amplification was performed on Light Cycler (Roche): 10 min 95°C, 1 cycle; 10 sec 95°C, 10 sec 65°C, 20 sec 72°C, 40 to 50 cycles). Experiments were carried out in three independent biological experiments. Oligonucleotide primers used for quantitative-PCR amplification will be provided upon request. As a control for real-time PCR, we used the 16S rRNA as described. The 16S rRNA forward primer (5′-TGACGGTACCCAAAGAATAAGCA-3′) and 16S rRNA reverse primer (5′-ACGCTTGCACCCTTCGTATTA-3′) amplicon was 72 bp in length. ## Pathogenicity tests Pathogenicity tests were conducted on *Arabidopsis thaliana* Sf-2 ecotype as previously described. Each strain was tested on sets of 4 plants with 4 leaves per plant. Disease development was scored at days 5, 7 and 9 post-inoculation using a disease index ranging from 0 (no symptom), to 4 (leaf death). ## \[¹⁴C\]sucrose transport experiments Overnight cultures in minimal medium (MME) without carbon source (uninduced) or with 20 mM sucrose (induced) were centrifuged. Pellets were resuspended in MME and the OD₆₀₀ was adjusted to 1. \[¹⁴C\]sucrose (PerkinElmer, specific activity of 21.8 GBq/mmol) was added to a final concentration of 0.5 µM. For competition experiments, sucrose, fructose or glucose was added to \[¹⁴C\]sucrose at a final concentration ranging from 0.5 to 100 µM. After different times, from 20 sec to 2 hours, samples of 0.2 ml were collected on cellulose nitrate filters, washed with 10 ml water, dried, and finally, the radioactivity was determined in a liquid scintillation counter. The concentration-dependent initial sucrose transport was determined using the rapid dilution method as described. Cells were precultured in minimal medium without sugar. After centrifugation and adjustment to an OD₆₀₀ of 1, cells were incubated for 15 sec in presence of 0.01, 0.025, 0.05, 0.1, 0.25, 0.5, 1, 2.5 and 5 µM \[¹⁴C\]sucrose and 0.2 ml samples were diluted into 5 ml MME supplemented with 0.1 mM sucrose. Cells were collected by filtration, washed with 10 ml MME supplemented with 10 mM sucrose, dried and the radioactivity was determined. For inhibition of the proton motive force (PMF) with carbonyl cyanide 3-chlorophenyl-hydrazone (CCCP), cells were incubated for 10 minutes at 30°C with 20 µM CCCP prior to the addition of \[¹⁴C\]sucrose. ## *In silico* analysis Location of the signal sequence responsible for the outer membrane localization was determined using the SignalP 3.0 server (<http://www.cbs.dtu.dk/services/SignalP/>) with default parameters for Gram- negative bacteria. Comparison alignments used to identify the TonB-box were realized using ClustalW (<http://www.ebi.ac.uk/clustalw/>) or Multalin (<http://prodes.toulouse.inra.fr/multalin/multalin.html>) softwares. β-sheets in the last 50 amino-acids of *Xcc* TBDRs were located using the secondary structure prediction method on the PSIPRED Protein Structure Prediction Server, (<http://bioinf.cs.ucl.ac.uk/psipred/>). The MotifSampler program, was used to identify a motif corresponding to the *Xcc* Fux-box upstream of the 9 Fur-repressed TBDR and Ps-TBDR genes. MotifScanner program or the PatScan pattern matcher software were used with the identified motif to locate all the Fur-boxes in the *Xcc* genome. Pip boxes (TTCGCN₁₅TTCGC) and hrp_II boxes (TTCGN₁₆TTCG) were identified in the *Xcc* genome using the PatScan software <http://www-unix.mcs.anl.gov/compbio/PatScan/HTML/patscan.html>). For phylogenetic analysis, amino acid sequences were aligned and phylogenetic trees were reconstructed by the neighbor-joining method as implemented in ClustalX. # Supporting Information We would like to thank J. Cullimore for critical comments on the manuscript, J. Gouzy and S. Carrere for help in sequence analysis. [^1]: Conceived and designed the experiments: EL SB DM MA. Performed the experiments: EL SB DM AB ML CG ND JV. Analyzed the data: EL SB DM MA. Contributed reagents/materials/analysis tools: EL SB DM MA. Wrote the paper: EL MA. [^2]: Current address: UMR PaVe, Centre INRA, Beaucouze, France [^3]: Current address: Laboratoire de Microbiologie et Génétique Moléculaires, Centre National de la Recherche Scientifique (CNRS) UMR5100, Toulouse, France [^4]: Current address: Signalisation chez les Végétaux, Pole de Biotechnologie Végétale, Centre National de la Recherche Scientifique (CNRS) UMR5546, Castanet-Tolosan, France [^5]: The authors have declared that no competing interests exist.

# Introduction Stillbirth, defined as the expulsion of a fetus with no sign of life remain a global issue of public health concern. In practice, gestational age of 20 to 28 weeks or a birth weight of 350 to 1000g is usually required to determine stillbirth. According to the World Health Organization (WHO) stillbirth is a neglected tragedy which poses economic, social and psychological effects to families, especially, the affected mother often leading to social withdrawal, loneliness and depression. Globally, 1 in 72 babies are stillborn and this translate to nearly 2 million stillbirths annually. Although the current estimates represent a global reduction of 35% overall since 2000, the rate of reduction is thought to be inadequate, and hence, the World would not achieve its target of 12/1000 live births by 2030. Of the global estimates, 84% is accounted for by lower and middle-income countries particularly, sub-Saharan Africa and southern Asia where three of four global stillbirth estimates are reported. In sub-Saharan Africa, stillbirth increased from 0.77 million in 2000 to 0.82 million in 2019, representing a 42 percent global stillbirths and, suggesting that the Africa region can only attain the 12 per 1000 lives birth target in 2050. Averagely, 21.7 per 1000 babies were stillborn in Ghana in 2019 and although this is an improvement over previous years, it is yet high compared to the regional average. Stillbirths remain a critical indicator of maternal and child health care performance and reflect negatively on weak health systems, particularly in lower and middle-income countries. Recently, the WHO launched an action plan campaign for newborn care, Every Newborn Child Action Plan (ENAP) to re-vitalize efforts toward reducing global stillbirths with emphasis on access to quality maternal health care as a necessary means to achieving country-level targets. Although the ENAP is yet to receive the required attention, the country-level programme of ’free’ maternal health care policy was introduced with the object of driving access to maternal health care, and ultimately improving newborn care outcomes. In line with the access to care policy, Ghana declared the ‘free’ registration of pregnant women as an exemption package of its national health insurance scheme (NHIS) in 2008 aimed at bridging the access gap to care to improve utilization of maternal health care services, thereby mitigating inequalities effect to enhance newborn care survival. While the ‘free’ policy targeted access to maternal health care, in particular, its broader intentions included comprehensive caregiving of newborns up to 90 days post-delivery. The ‘free’ maternal health care policy (FMHCP) initiative received £42.5 in funding support from the then UK and has since served over 3 million beneficiaries since its inception. Although the ‘free’ policy lacked an implementation framework at its inception, it nonetheless gained popularity, as pregnant women received cost ‘free’ maternal health care at a cost to the NHIS. In this paper, we compare stillbirth outcomes between mothers who registered and benefitted from the ‘free’ maternal health care policy since its inception and mothers who did not. Given the quintessence of health system factors effect on stillbirth, we further explored the views of service providers and pregnant women from selected health facilities to add context to the quantitative results to inform our discussion and conclusions. ## Conceptual framework We hypothesized that certain factors undermine the successful operationalization of the ‘free’ maternal health care policy in Ghana and hinder the intention of the policy and its ability to bring about a decline in not just maternal mortality, but also stillbirth and perinatal mortality in the medium to long term. The ‘free’ maternal health care policy is administered via the NHIS which in itself is bedeviled with funding constraints culminating in delays in payment of claims over the years. Consequently, it appears the existing challenges affect the effective management of accredited service provider facilities thereby threatening the credibility of the purchaser-provider split concept. On another level, the Ghanaian society presents itself as a keeper of pregnant women with cultural demands that upset the effective implementation of the ‘free’ policy. These unintended practices seem to derail the efforts of health care professionals and policymakers and hence, frustrate their efforts of achieving reduced mortality outcomes of newborn care. Studies have shown that maternal age, rural/urban area of residence, twin pregnancy, negative pregnancy outcome, income level, education and marital status play roles to moderate the outcome of stillbirth and perinatal mortality. In this current study, we ask to what extent does the FMHCP intervention affect the outcome of stillbirth and perinatal mortality against the background of the policy implementation bottleneck while accounting for the moderating factors? ## Contextual definition **Impact:** Reduction in stillbirth and perinatal mortality over time between 2008 and 2014. **Stillbirth:** The birth of a fetus after 28 weeks gestation with no signs of life. **Perinatal mortality:** The death of newborns within 6 days of life. **‘Free’ maternal health care policy:** Pregnant women registrants of the national health insurance scheme (NHIS). # Materials and methods ## Study design The current study employed a sequential mixed method design: first, analyzing repeated cross-section data of Ghana Demographic and Health Survey (GDHS), 2008 and 2014 as baseline and end line, respectively. The two-rounds of DHS data sets were merged, and two groups created: the ’free’ maternal health care policy group and the ‘no-free’ maternal health care policy group. We then used the NHIS registration status of women as a proxy for our exposure variable (the ‘free’ maternal health care policy) and constructed stillbirth and perinatal death as our outcome variables using the under 5 mortalities variable from the merged data sets. The first author conducted one-on-one in-depth interviews with 5 medical doctors and 18 midwives at post to the antenatal clinics and labor units of the selected health facilities and conducted 8 focus group discussions (FGD) for pregnant women who accessed care using the ‘free’ policy in the selected hospitals, to obtain user perspective for a contextual understanding of health system factors that affect stillbirth outcomes. Given the time disparity of the secondary data and qualitative interviews, facility-level data from two regions of Ghana, obtained from the District Health Information Management Systems (DHIMS) was analyzed and triangulated with those of the regression output from the GDHS to provide a current undertone to the outcomes of stillbirth and perinatal death for a meaningful inference. ## Qualitative study setting The Upper East and Northern regions are among the poorest regions of Ghana. The Upper East region has a total land area of about 8,842sq km with Bolgatanga as its capital and an estimated population of approximately 1.5 million. The region has a national health insurance enrollment rate of 6.3% of its total population with a considerably good number of midwives compared to other regions of Ghana. Between 2016 and 2020, stillbirth figures increased in trend as per the regional data from DHIMS. On the other hand, the Northern region shared a boundary with the Upper East Region at the time of the DHS data collection (now with the North East region) with a relatively low literacy rate. The Northern region has a land mass of about 70,765.2km² with an estimated population of about 1.8m representing 9.6% of the total population of Ghana and shows a considerably stable but inconsistent decline in stillbirth proportionate to facility delivery utilization per the DHIMS record. ## Study participants, sampling and variable construction ### Quantitative Stillbirth and perinatal death variables were generated from the under 5 mortality variables using STATA 15 to construct binary outcomes of ‘1’ representing ’stillbirth’, born dead or dying within day zero, and ’0’ representing ’no stillbirth’. Also, ’1’ was constructed to represent perinatal death (newborn death within 6 days of life), and ’0’ representing ’no perinatal death’. ### Qualitative The two regions (Upper East and Northern regions) were zoned into three and a hospital each selected purposively for the study. Two health centers per zone were also selected to add rural context to the data. Service providers (midwives and medical doctors) from the antenatal care clinics and labor units of the selected health facilities were then selected purposively and conveniently i.e., doctors or midwives on duty and not busy, who consented to participate and met the selection criteria, was recruited for one-on-one in-depth interview. Pregnant women in attendance to the antenatal care clinics were also recruited from the same facilities for the focus group discussions (FGDs). The use of multiple sources of data was to explore the idea of multiple realities and this added to data verification from multiple sources as argued by Creswell. ## Data collection and analysis ### Tools and pretesting Interview guides were developed to unearth health systems challenges that confront the ’free’ policy’s successful operation and how that may have affected stillbirth outcomes as per our conceptual framework which was guided by Mosley and Chen’s analytical framework for child survival. The qualitative tools were pre-tested among random midwives in a random hospital in one of the regions, which was not part of the selected study sites. All the authors then revised the tools based on our observations to include the eliciting of critical community and facility level factors relevant to our study. ### Inclusion criteria Only doctors and midwives with at least 3 years of working experience in the labor units and antenatal care clinics of the selected health facilities were included in the in-depth interviews, to ensure that they have adequate experience working with the ‘free’ maternal health care policy. Also, only pregnant women registrants of the NHIS were included in the focus group discussion. ### Exclusion criteria Pregnant women whose vital signs were outside the normal range using the standard of the American Psychology Association or pregnant women who were receiving treatment for a medical condition were excluded from the study. Pregnant women who were less than 16 years, considered minors under the 1992 Constitution of the Republic of Ghana were excluded from the focus group discussions. ### Sample weighting We generated a weighted variable by dividing v005 by 1000,000 to cater for 6 decimal places, usually not reported by DHS. We then applied the weighted variable to the GDHS survey data sets in STATA using \[pw = wgt\], psu (v021) strata (v022) to set the data to cater for clustering and stratification. Thereafter, all Stata command prefixes *‘svy’* to take the weighting into account across the secondary data analysis. We then checked for sensitivity and overdispersion using the negative binomial regression model. ### Confounding variables Maternal age, area of residence, employment status, abortion history, caesarean section, marital status, and educational status were adjusted for as confounding as these were statistically significant (p \<0.05) with the outcome variable of interest (stillbirth and perinatal) or the exposure variable (NHIS, proxy to the ‘free’ maternal health care policy). ### Quality control and trustworthiness The sample weighting catered for clustering and stratification across rural and urban areas of the complex DHS data design. The regression analysis also used the Taylor linearization technique to achieve reduced standard error. The use of purposive sampling for the qualitative data was deliberate to achieve trustworthiness through the acquisition of information from the right sources of service providers, doctors and midwives and pregnant women as policy users. This was critical, giving the ‘free’ policy is implemented by medical doctors and midwives. Also, the inclusion of an expert informant, a regional director of health services validated the field data which was useful and catered for the idea of multiple sources of information. ### Data analysis—Quantitative The overall stillbirth and perinatal mortality ratios were estimated between the two rounds of DHS for comparison. We then estimated the prevalence of stillbirth and perinatal mortality between the baseline (2008) and end line (2014) to compare the outcomes, pre and post the ‘free’ policy intervention. Finally, we merged the two rounds of data sets and analyzed for risk of stillbirth and perinatal mortality among the ‘free’ maternal health care policy using multiple logistic regression. We then tested for sensitivity using Poisson regression and checked for over-dispersion using negative binomial regression. All regression outputs are reported in Tables and. ### Data analysis—Qualitative One-on-one interviews and group discussions were transcribed verbatim into Microsoft office, double-checked for accuracy, and imported into INVIVO 10 for analysis by coding the data in line with the study objectives of stillbirths and context factors of maternal health care utilization. The transcripts were read multiple times and grouped into similar and dissimilar statements with annotations. Significant statements were then categorized under constructed themes, reviewing each theme carefully for relevance. Statements that did not align themselves to a particular theme were thoroughly examined for relevance and excluded altogether if they didn’t speak to the study objective. Constructed themes were based on common phrases and similar statement approaches. Relevant statements are quoted verbatim in reporting the qualitative results to convey participants’ impressions and aid explanatory power. ### Ethical consideration This study is part of the PhD research work of the first author and received ethical clearance from the Ghana Health Service Ethical Review Board reg. no. GHS-ERC: 002/04/19. The secondary data was obtained from Measure DHS after completing an online application process. All study participants for the primary data consented to participate in the study and completed a consent form. All interviews were conducted in private rooms, while focus group discussions were held in open spaces under chalets for aeration as part of the COVID-19 protocol. All the pregnant women had their vital signs checked by a registered nurse for normalcy prior to joining the focus group. Also, this study protocol received four double-blinded external reviewers and was published by BMC Reproductive Health Journal. # Results ## Quantitative findings ### Stillbirth and perinatal death in the Upper East and Northern regions Antepartum stillbirth is increasing in both regions since 2019, particularly in the Upper East region of Ghana and this is reflected in the overall rise in stillbirth in the region despite the increase in uptake of facility delivery. ### Distribution of maternal and population characteristics As shows in, more women accessed the ‘free’ policy in 2014 (68%) compared to 2008 (39%). Of antenatal care uptake, 62% of the pregnant women made 4 plus visits under the free maternal health care policy group. Also, more women delivered in health facilities (65%) among the ’free’ policy group, compared to the no- ‘free’ policy group. Of the maternal and population characteristics maternal group age (p \< 0.0001), area of residence (p \< 0.0001), history of abortion (p = 0.0104), employment status (p = 0.0271), maternal education (p \< 0.0001), wealth index (p = 0.0001), marital status (p \< 0.0001) and region (p \< 0.0001) were statistically significant between the two groups. ### Prevalence of stillbirth and perinatal mortality In total, 174 stillbirths were recorded between 2008 and 2014 rounds of DHS. Of this, 55 (28.7%) were reported in the 2008 DHS, compared to 119 (43.1%) in 2014, showing an increase in percentage points 14.4. Also, 243 perinatal deaths were reported between 2008 and 2014 of which 88 (45.6%) were reported in 2008, compared to 155 (56.4%) in 2014, representing a 10.8 percentage point increase. ### Stillbirth and perinatal mortality rate Stillbirth birth rate for 2008 was 19 per 1000 live births, compared to 21 per 1000 live births in 2014, while perinatal mortality rate was 31 per 1000 live births in 2008 and declined to 27 per 1000 lives birth in 2014. ### Risk of stillbirth in the ‘free’ maternal health care policy group Babies were 1.64 times more likely to be stillborn in the FMHCP group, compared to the no FMHCP group; aOR: 1.64; 95% CI: 1.02 to 2.65; p = 0.041. The results are similar across the Poisson and negative binomial regressions models, aPR: 1.34; 95% CI: 1.00 to 1.79; p = 0.045 respectively as shown in. ### Risk of perinatal mortality in the ‘free’ maternal health care policy group Babies were 2.04 times likely to die within 6 days of life in the FMHCP group compared to their counterparts in the no FMHCP, aOR: 2.04; 95% CI: 1.28–3.25; p = 0.003. for the results also compare similarly for the Poisson and Negative binomial regressions, aPR: 1.34 and these were statistically significant, p = 0.006, respectively. Women with a secondary level of education were more likely to register perinatal mortality compared to women with no formal education, aOR: 1.89; 95% CI: 0.98 to 3.63. However, this was not statistically significant, p = 0.056. Women with a history of abortion were also more likely to record perinatal mortality compared to women with no history of abortion, and this was statically significant, aOR: 1.91; 95% CI: 1.07 to 3.41; p = 0.028. ## Qualitative findings ### Study participants In all 67 service providers and pregnant women participated in the qualitative study. Of the service providers, midwives were 18, and doctors/directors were 5. Of the pregnant women participants, 43 (98%) were married, 38 (86%) were employed, while 40 (81%) gave birth previously. ## Common themes *Rising stillbirths and related causes*. Stillbirth was on the rise based on the facility records and the regional data, but the service providers attributed the rise to multiple reasons, late reporting to antenatal care clinics, delayed arrival to delivery centers by pregnant women in labor and the use of local herbs as oxytocin. Service also providers argued that improved record keeping associated with increased utilization make the numbers of stillbirths look worse than it appears. In other words, pregnant women benefiting from the ‘free’ policy are more likely to report to and deliver in a health facility and have their data captured as compared to the previous data capture rate under out-of- pocket payment. A medical doctor and charge midwives explain. > *"Actually, just this half–year, stillbirth numbers weren’t > encouraging. It was bad. We had 22, but 13 were macerated. Then we had > 9 fresh stillbirths. The numbers are going up \[increased\]” > **(Midwife 1, IDI, Bongo District)*** > > *“In the region, when you look at the picture, despite the so many > interventions, one will say SBs \[stillbirth\] are still high. But > when you look at it critically, it is the reporting which is also > going up, so it makes you think that the policy is not helping” > **(Doctor 1, IDI, UER).*** > > *“We have a high rate. This time we are getting mothers who are coming > with Intra–Uterine Fetal Death (IUFD). This year we had 15 for the > first 6 months. When you compare, I will say because we are taking > records, that is why the numbers are high…previously there was no > documentation" **(Midwife 1, IDI, Zebilla District).*** Antepartum stillbirths were commonly recorded among facility deliveries, and service providers deemed this as a prove that babies died in utero before arrival to a health facility and blamed this on community and individual level factors rather care giver related. Another phenomenon reported was previous maternal health care history which appear to negatively influence the outcomes of stillbirth. Reasons for delays bothered on two items as numerated by the service providers; some pregnant women want less of vaginal examination, explaining that the experience is uncomfortable and secondly, women with previous cesarean section (CS) avoided hospitals in order not to invite another CS. Vaginal delivery is a source of pride. Pregnant women will risk delivering per vagina as a sign of womanhood. Service providers explained further as follows. > *“Few cases also dodge the hospital … maybe she had two previous > cesarean sections (CS) and thinks that if she comes to the hospital, > there will be another CS, so they avoid the hospital and when there > are complications, then they quickly come. They want to deliver per > vagina at all costs” **(Doctor 1, IDI, UER)*** > > *“…and when they come, we manage them at postnatal care…. they say > they didn’t know that labor had started, some say they don’t want the > examination. One woman was frank, she said when they come, they put > fingers on her vagina and that one she doesn’t like it**” (Doctor 2, > IDI, UER).*** Additionally, the cultural practice of given herbal preparations to women during labor was reported from the study sites and this seem to play a role which the midwives inferred contributed to undesirable outcome of stillbirths. Pregnant women are served locally prepared mixtures known as *kaligutiem*, to speed up uterine contraction during labor at the blind side of midwives and doctors. Essentially, the herbs act as oxytocin and potentiates the effect of prescribed medicines when combined, with lethal consequences of risk of excessive uterine contraction. Usually, mothers-in-law administers the potion before coming to the hospital or may secretly give a dose to the woman in labor if they judged a labor-process of having prolonged. Some communities are notorious for the use of the local herbs and this impedes the pathway to accessing health care during labor. The midwives shared their experiences. > *“They also take ‘kaligutiem’ \[local oxytocin to aid uterine > contraction\] before coming to the hospital, and when they come the > contractions will be too high” **(Midwife 2, IDI, Sagnarigu > District).*** > > *“Some also come with excessive contraction because of the > ‘kalgutiem’, especially those coming from Dotoyille and Kunyevilla. > The mothers–in–law. They will give it to them and follow them to > hospital as well" **(Midwife 1, IDI, Sagnarigu District).*** During the focus group discussions, pregnant women claimed knowledge of *kalgutiem* but denied usage of same, unsurprisingly. > *“We have heard about ‘kuligutiem’, but we don’t use it. We don’t know > anybody who uses it. The midwives have complained and have advised us > against it. **(Pregnant Woman 3, FGD1, Sagnarigu).*** *Poor use of delivery partograph for labor monitoring*. Surprisingly, it merged that labor process and progress were poorly monitored as delivery partograph, a tool recommended by the World Health Organization was sparingly used across delivery suits in the selected hospitals. A senior medical officer and director of the regional health services noted during the expert informant interviews that, > *“Even though we use partograph to monitor labor we realize a > substantial number are not monitored. Using a partograph to monitor > will tell you the condition of that baby. So that if you realize that > the baby has difficulties, that baby can be delivered \[via caesarean > section\].” **(Regional Director of Health Services, KII, UER).*** The delivery partograph, according to experts served as an indicator for initiating advance action of cesarean operation, necessary to save the life of the unborn child. Although the use of delivery partograph is not for all pregnant women, its use during the labor management process on eligible mothers is not optional, but this was not the case at the study site. The director continued, > *“Yes! A significant proportion of labor are not monitored \[with > delivery partograph\]. Those women who are eligible, it should be > 100%…” **(RDHS, KII, UER).*** A senior midwife with over 10 years work experience in one of the regions also shared her experience when they carried out a monitoring exercise on behalf of UNICEF. Her observation of records of partograph use was a major concern. She claimed. > *“As for partograph dier! It is 0 out 100 in…hospital \[a particular > hospital\]. We went for monitoring on behalf of UNICEF and what we saw > was not good all at” **(Midwife 2, Tamale West)*** *Little or no interest in ‘matters’ of stillbirth*. During the one-on-one interview sessions, a medical doctor observed that somehow, little attention is given to stillbirth issues as compared to maternal mortality. Himself as a doctor, does not get to hear about stillbirth in his ward unless there was a review of a visiting team from the regional health directorate. Not even the media were interested in stillbirths as much as they were interested in maternal mortality. The doctor added thus, > *"We don’t pay attention to matters of stillbirth the way we do for > maternal deaths. One mother will die and the whole hospital will hear > about it. I don’t even know the stillbirths in the labor ward. They > don’t tell me…unless we are reporting. But when there’s maternal > mortality, eeeiii! even the media is interested” **(Doctor3, IDI, > UER)*** The medical doctor’s view appears to align with a seemingly common practice of midwives ignoring apparent calls of pregnant women in labor, which caught the attention of one of the pregnant women and she shared her views during the group discussions. > *“Sometimes you can be crying, and they won’t mind you. One time I was > suffering, and the midwives didn’t bother to check on me. I said my > baby is coming….by the time they came my baby was gone. They don’t > care about our babies" **(PW4, FG3, Bongo District).*** Even though the reasons for their non-response were beyond the scope of this study, the descriptions during the focus group discussion suggest that there are underlying challenges that perhaps explains the poor monitoring of the labor process and the eventual outcomes of stillbirths in the study sites. Pregnant women participants added, thus, > ***“**they \[midwives\] don’t pay attention to our babies. One woman > nearly gave birth on the bench. She was calling the midwives…the baby > is coming; the baby is coming. Oh! I felt very sad” **(PW2, FG1, > Zebilla District)*** Conversely, midwives disclosed that pregnant women had a laid-back attitude towards the survival of their unborn babies, sometimes refusing surgery as may be required and also providing inaccurate reproductive history which affects the caregiving process and hence, influencing stillbirths and perinatal deaths. Charge midwives had these to say in one of the district hospitals. > *“a woman came, and the liquor was small, so the best we could do was > conduct CS. When we told them, they told us that if the water is not > ok, can’t you fetch water and add it. They’re opposed to cesarean > section. They went home…came back some few days later and the baby was > dead…" **(Midwife 2, IDI, Zebilla District)*** > > *"Taking history is key… The woman misled the midwives concerning her > parity. We started inducing, and she raptured, then, we asked a > relative (her daughter) and she said her mother had 6 children and 1 > died. Such a person should be induced…we were misled.” **(Midwife 2, > IDI, Bongo District)*** *Intermittent shortage of medicine commodities*. Intermittent shortages of drugs were also reported during the in-depth interviews. Pregnant women are routinely asked to purchase some medicines outside of the health facility set up to augment their required supplementary intake. Not only did this affect the economic situation of mothers and their families, but it also frustrates quality-of-care processes of the health care professionals. The medical officers shared their experiences at the antenatal clinics. > “*Our environment is not good*. *Personal hygiene is poor*. *Unlike > other places where they think that labor is a sterile procedure*, > *here*, *we routinely put all our clients on the antibiotic cover*, > *whether you’re on episiotomy*, *assisted delivery*, *or not*.*"* > ***(Doctor 1*, *IDI*, *UER)*** > > *“When you visit the facility and certain medication is not available, > they are written for you in a prescription. So far as our facility is > a concern, if a medication is not available…we put it on a > prescription for you to find a pharmacy shop to procure…” **(Doctor 1, > IDI, UER)*** > > *“The issue has to do with drugs. The challenge here is that most of > the time the hospital runs out of stock. When they run out of stock, > the patient must buy… because of the poverty level, most of them > cannot afford the drugs…” **(Doctor 2, IDI, UER)*** > > *“We use antibiotics and pain killers for Cesarean Section. Then we > also have hematinic. The better once, usually we want them to buy > those outsides…. eenh! And IV fluids too. There are certain times we > go virtually down, they buy virtually everything” **(Doctor 3, IDI, > UER)*** Midwives also bemoaned the difficulty in getting drugs at the facility level, more so as some of these drugs are considered an emergency requirement yet not in supply, and this adversely affects the effective functions of caregiving with a direct consequence on the unborn/newborn child. > *"And after that, they pay for vitamin k, which we give to the > child…that is when it is a normal delivery. When it is a Cesarean > section, antibiotics like Cefuroxime, Amoxyclav, and Gentamycin are > ordered by the doctor. If it is not there, they go to buy…” **(Midwife > 2, KNNM, UER)*** > > *“When the dispensary does not have hematinic (iron III), you ask them > to go and buy, it is a problem…. what about if she comes for ANC and > you write for her and in the end, she goes and not buy? She will come > back with anemia…” **(Midwife 3, IDI, Bongo District)*** Folic acid, a dietary supplement giving during pregnancy as recommended by the World Health Organization as essential in minimizing the risk of stillbirth, is sometimes in short supply and pregnant women are told to purchase some from the open market. They shared experiences during the focus group discussion as follows. > *“Whenever we come, we have been buying the drugs. Most of the time > when we come, they do write for us to go and buy the drugs. The yellow > and the red drugs” **(PW1, FG1, Bongo District)***. > > *“The last time I delivered, my husband was made to buy water > \[intravenous fluids\] for infusion…the is a drug store outside the > hospital, that’s where we bought it.” **(PW5, FG2, KNNM, UER)*** # Discussion Generally, utilization outcomes improved over time between 2008 and 2014 showing statistically significant differences between the Ghana Demographic and Health Survey data. In a similar vein, there was a corresponding increase in stillbirth and perinatal mortality and although, population growth is one plausible explanation to this, the introduction of the ’free’ maternal health care policy was also key to increasing utilization and this may put pressure on the health system capacity to deal newborn care, as previously reported. Stillbirth was accounted for mainly in 2014, with a statistically significant difference, p \< 0.0344. Conversely, we found that the perinatal mortality rate declined in 2014 by 4 per 1000 live births, moving from 31 per 1000 live births in 2008 to 27 per 1000 live births in 2014. On the other hand, stillbirth rate was worse off, increasing over time by 2 per 1000 live births between 2008 and 2014. By implication, while perinatal mortality is declining, stillbirth is rising, and this supports the views espoused by the service providers during the in- depth interviews (IDIs). The inverse relationship between stillbirth and perinatal mortality is rather surprising because perinatal mortality feed directly on stillbirth, hence, the expectation would be that as one decreases, the other should also, but this is contrary to the current findings. Nevertheless, the current findings suggests that early neonatal deaths were declining at a factor rate, perhaps outpacing the rate of stillbirth, and hence reflecting in the overall decline in perinatal mortality rate. Secondly, home deliveries may be recording low stillbirths, compared to facility-level deliveries. This is not farfetched given the observations during the IDIs that pregnant women in labor turn to rash to the hospital after attempting and failing home delivery. Successful home deliveries will likely be those without complications and perhaps fewer mortality outcomes, yet, the rising number of stillbirths, which is consistent with the current increase in numbers of stillbirths in sub-Saharan Africa as reported by the WHO is a worrying development and a challenge for health systems and policy. Both stillbirth and perinatal deaths were prevalent among the ‘free’ maternal health policy group compared to the no ‘free’ maternal health care policy group. The qualitative exploration revealed that even though the ‘free’ policy may have led to increase in access to maternal health care, it was against a background of shortage of medical consumables, poor and/or inadequate monitoring of pregnancy and labor process and out-of-pocket purchasing of supplementary medicines. Iron tablets (folic acid, and ferrous sulphate) for example, are routine drugs served at antenatal clinics as supplements to prevent anemia in pregnancy and these also aid in combating stillbirth, yet, these medicines were consistently in short supply in the selected study sites. Indeed, findings in earlier studies show that folic acid intake during pregnancy is associated with reduced stillbirth, and therefore, intermittent shortages perhaps poses increased risk stillbirth among pregnant women. Under the current situation, quality of care is perhaps affected by the lack of medical commodities, a key component of the WHO building blocks of health systems framework and perhaps also affect technical quality. Despite the overall reduction in the perinatal mortality rate in 2014, perinatal mortality still increased proportionately high between 2008 and 2014 by 10.8 percentage points, moving from 45.6% to 56.4% within the DHS survey period. The results compare intriguingly with the proportion of stillbirths between 2008 and 2014 which showed a much higher increase in percentage points of 14.4 in 2014 despite the introduction of the ‘free’ maternal health care policy. The findings imply that stillbirth failed to show a decline in 2014 both in rate and in proportion to under 5 mortalities between 2008 and 2014, while perinatal mortality declined in overall rate but increased in proportion between 2008 and 2014. Although, this is unexpected, it represents some level of gain in the face of the ‘free’ policy. Arguably, the situation could have been worse without the ‘free’ maternal health care policy. The prevailing health systems factors of poor monitoring, delayed arrival, inadequate attention to stillbirth ’matters’ and intake of local herbs, although inconclusive probably throw some light as to why the numbers of stillbirth are high. It is imperative perhaps to consider the possibility of increased data capture and although this study did not independently explore the influence of data capture, the service providers insinuated during the IDIs that increased record keeping may have influenced the numbers of stillbirth among the ‘free’ maternal health care policy group. Even though the gains in overall perinatal mortality could be due to increase in access to care at the neonatal period, and consequently improved immunization, perinatal death and stillbirth are twin concepts that work together and reasonably, a decline in one was expected to show a similar pattern in the other, unless early neonatal mortality was significantly declining. Central to labor monitoring is the use of delivery partograph and although its implementation comes with striking challenges including form complication, midwifery staff shortages, and the lack of appreciation of its importance, perhaps it is imperative to state that in a situation where less partograph is used by midwives during labor, a rise in numbers of stillbirths may not be unexpected. The WHO recognized the challenges associated with using delivery partograph and approved its modification in Ethiopia, but the findings of the current study suggest yet another reason for further engagement of midwives on the need to use delivery partograph for labor monitoring. Partograph use was more of a problem in the Upper East region than the Northern region, and although the reasons are unclear in the current study, a closer look at the regional data from DHIMS and those of the quantitative output shows that facility delivery was high in the Upper East region compared to the Northern region and therefore, increased workload may have played a role in affecting midwives’ ability to use the delivery partograph. Perceived lack of care and attention also emerged from the FGDs among the pregnant women participants. The pregnant women perception somewhat lends credence to the IDI’s revelation that not much attention was paid to stillbirth ’matters’ as much as maternal mortality. This is consistent with the recent report by the WHO and UNICEF, which points out that stillbirth was receiving less attention from policy and resources, and thus, it was not surprising that stillbirth declined less in the last decade of the first century compared to maternal and under 5 mortality. The effect of this is that pregnant women may lose trust in facility-level delivery in the long run and turn up late and in a complicated state, thus affecting care outcomes. Stillbirth and perinatal mortality are sensitive indicators of health systems’ weaknesses and a test of the quality of care dimensions. The use of *kaligutiem* to speed up contraction in labor, unaware of its adverse effect on the unborn child also emerged from the IDIs. It seems a given that women in labor will want a fast-track process of labor, yet the practice of using local herbs against medical advice demonstrates some lack of confidence in the modern health care system. The account of midwives suggests that pregnant women who take the are at risk of excessive uterine contraction with increased risk of uterine rupture and therefore stillbirth. ## Strength and limitation The use of DHS data sets was appropriate to achieve external validity and generalizability as the data was large enough and representative. Suffice to say, the ‘free’ policy primary intent was to increase utilization and access to maternal health care. This study measured stillbirth and perinatal mortality outcomes in Ghana relative to the ‘free’ maternal health care policy using a mixed method design and this added context to the study findings and discourse. The study show limitation as well. The quantitative analysis was based on association using regression models. Although the analysis used multivariate regression models to test sensitivity, a quasi-experimental design would perhaps have measured the treatment impact with precision and give robust results. On the qualitative side, the selected regions were from the northern section Ghana, based on the quantitative findings of stillbirth outcomes, thus, excluding pregnant women’s perspective of the ‘free’ policy from the southern section of Ghana. ## Conclusion Although perinatal mortality rate declined overall in 2014, stillbirth rate increased within the period suggesting a significant decline in neonatal mortality. This is a gain in that, while the ‘free’ maternal health care policy is yet to translate to reduced stillbirths, early neonatal mortality is declining. Giving these findings are within the health systems context of poor monitoring of labor process, and intermittent shortage of drug consumables for pregnant women, the factors may have exerted negative influence on the outcomes of stillbirths. Delayed arrival during labor and the intake of local oxytocin may be compounding pregnancy outcomes and consequently, increasing stillbirth. It is recommended that health system thinking approach be adopted by the MoH and GHS to ensure regular supply of drug supplements and outright stoppage of local herb usage among pregnant women for better outcomes. There is the urgent need for leadership to monitor the use of the delivery partograph in managing eligible mother in labor. # Supporting information The DHS Programme funded by the USAID is hereby acknowledged for their role and ownership of the DHS data sets. We also acknowledge Ghana Health Service and DHIMS for allowing us access data for aspect of this study. We appreciate the cooperation of the directors who granted entrée to hospitals and health centres for the primary data. Finally, we acknowledge the voluntary participants who volunteered to participate in the study. 10.1371/journal.pone.0274573.r001 Decision Letter 0 Lupattelli Angela Academic Editor 2022 Angela Lupattelli This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 29 Nov 2021 PONE-D-21-35524Evaluating the impact of maternal health care policy on stillbirth and perinatal mortality in Ghana: a mix method approach using two rounds of Ghana demographic and health survey data sets and qualitative design techniquePLOS ONE Dear Dr. Azaare, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. Please submit your revised manuscript by Jan 13 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, Angela Lupattelli, PhD Academic Editor PLOS ONE Journal Requirements: 1\. When submitting your revision, we need you to address these additional requirements. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at  <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:  (We also acknowledge the cooperation and support of Ghana Health Service and the Directors who granted entrée to hospitals and health centers for this study.) 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 noted in your submission details that a portion of your manuscript may have been presented or published elsewhere. Please clarify whether this publication was peer-reviewed and formally published. If this work was previously peer-reviewed and published, in the cover letter please provide the reason that this work does not constitute dual publication and should be included in the current manuscript. 4\. Please include your full ethics statement in the ‘Methods’ section of your manuscript file. In your statement, please include the full name of the IRB or ethics committee who approved or waived your study, as well as whether or not you obtained informed written or verbal consent. If consent was waived for your study, please include this information in your statement as well.  5\. We note that Figure 2 in your submission contain \[map/satellite\] images which may be copyrighted. All PLOS content is published under the Creative Commons Attribution License (CC BY 4.0), which means that the manuscript, images, and Supporting Information files will be freely available online, and any third party is permitted to access, download, copy, distribute, and use these materials in any way, even commercially, with proper attribution. For these reasons, we cannot publish previously copyrighted maps or satellite images created using proprietary data, such as Google software (Google Maps, Street View, and Earth). For more information, see our copyright guidelines: <http://journals.plos.org/plosone/s/licenses-and-copyright>. We require you to either (1) present written permission from the copyright holder to publish these figures specifically under the CC BY 4.0 license, or (2) remove the figures from your submission: A. You may seek permission from the original copyright holder of Figure 2 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/> 6\. We note you have included a table to which you do not refer in the text of your manuscript. Please ensure that you refer to Table 7 in your text; if accepted, production will need this reference to link the reader to the Table. 7\. Please include captions for your Supporting Information files at the end of your manuscript, and update any in-text citations to match accordingly. Please see our Supporting Information guidelines for more information: [http://journals.plos.org/plosone/s/supporting- information. ](http://journals.plos.org/plosone/s/supporting-information. ) Additional Editor Comments: Dear author, I do have some request regarding the structure of your article, before it can be assessed further. 1\) Please format the article in a way that can be more easily assessed. The Introduction is extensively long; please shorten it and make it to the point. The "contextual definition" does not need to have a specific subsection; these are terms that you can define in the text. 2\) Please state your objectives clearly. 3\) Please remove the formulas for generating ratios, that is superfluous. 4\) Please summarize the amount of results presented in the article, in terms of number of tables, figures and text - consider moving not essential data to the supplementary material if possible. \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] 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.0274573.r002 Author response to Decision Letter 0 24 Feb 2022 The Academic Editor PLOS ONE Journal Dear Editor, RE: RESPONSE TO REVIEWERS: PONE-D-21-35524 - EVALUATING THE IMPACT OF MATERNAL HEALTH CARE POLICY ON STILLBIRTH AND PERINATAL MORTALITY IN GHANA: A MIX METHOD APPROACH USING TWO ROUNDS OF GHANA DEMOGRAPHIC AND HEALTH SURVEY DATA SETS AND QUALITATIVE DESIGN TECHNIQUE 1\. The manuscript formatting has been revised per PLOS journal requirements including Tables and Figures. We are willing to revise further if necessary, any aspect that does not meet the requirement of the journal. However, this may have to be pointed out clearly to us to guide us. 2\. The study objective has been restated to make it much clearer than before. 3\. The size of the manuscript has also been reduced particularly in the introduction and discussion sections: a\. the contextual definition sub-section has been taken out and key definition incorporated into the introduction. b\. the formula for the calculation of stillbirth and perinatal mortality ratios has also been removed from the manuscript. c\. the Tables has been reduced and others merged as necessary. We now have 5 Tables instead of 7 as contained in the original submission. 4\. You noted that some aspects of this manuscript appear to have been published somewhere else. This is not exactly clear to us. However; a\. the original protocol design of the study was published by BMC reproductive health journal and assigned DOI (<https://doi.org/10.1186/s12978-020-01011-9>) which was just the methodology of the full protocol at its design stage and did not include any results. The conceptual framework in the protocol appears similar to the one in the current manuscript, but not the same. There has been a significant modification as the study progressed which is what has been submitted in the current manuscript. b\. In preparing this manuscript, aspects of the qualitative results were presented by research gate as a pre-publication text (not peer-reviewed), usually meant to gather useful comments to enrich the manuscript (and does not constitute a publication of our current results) and thus, the wording may be similar. Accordingly, we can confirm that no part of this manuscript results has been published in any journal as a research finding and should be considered for publication in PLOS Journal as an original piece of work. 5\. (We also acknowledge the cooperation and support of Ghana Health Service and the Directors who granted entrée to hospitals and health centres for this study.) 6\. The above statement (item 5) was quoted to us, and intimating that we may have received funding support from Ghana Health Service or another agency. Perhaps, we did not convey the appropriate impression in the choice of words and terms such as “support of Ghana Health Service…”. However, for the avoidance of doubt, no funding support was received from Ghana Health Service or any agency of local or international in nature regarding this study. This manuscript is part of the PhD research work of the first author during his candidature in the University of Ghana School of Public Health, where the 2nd, 3rd and 4th authors are faculty and supervising committee members of the 1st author. The first author did receive tuition fee support (not research) from the Ghana Education Trust Fund (GetFUND) and this has been duly acknowledged in the manuscript. We are willing to modify our statement if the journal finds the GetFUND tuition fee as funding support for the research work. 7\. Figure 2 (Fig. 2) attached in the manuscript is an original map professionally constructed for this study and about the selected study site from two regions of Ghana. The map is an original work designed for it purpose following the commencement of the study and has neither been used by anyone nor published anywhere before. To this end, we do not find the need to follow your guideline on copyright concerning third party use of a figure (the map), as this does not apply to us in this case. We are willing to remove the particular figure (Fig. 2) if there is evidence to the contrary. 8\. The aspect of the ethical statement has been revised accordingly in the method section of the manuscript. The full name of the ERB, Ghana Health Service Ethical Review Board has been stated under the sub-section ‘Ethical consideration’. 9\. All Tables and Figures have been duly referred to i.e. Table 7 which was cited as not referred to has been corrected. The particular Table was referenced as multiple Tables in the manuscript. However, in the revised version, Table 7 (now Table 6) is attached as a supplementary file. 10\. The figures (Fig 1 and 2) have been uploaded to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic accordingly to meet PLOS requirements. 11\. Humbly submitted. 10.1371/journal.pone.0274573.r003 Decision Letter 1 Lupattelli Angela Academic Editor 2022 Angela Lupattelli 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 May 2022 PONE-D-21-35524R1Evaluating the impact of maternal health care policy on stillbirth and perinatal mortality in Ghana: a mix method approach using two rounds of Ghana demographic and health survey data sets and qualitative design techniquePLOS ONE Dear Dr. Azaare, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. Please submit your revised manuscript by Jun 30 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, Angela Lupattelli, PhD Academic Editor PLOS ONE Reviewers' comments: Reviewer's Responses to Questions 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: 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: Thank you for a good work. Please see manuscript text for comments. The purpose of the study has to be rewritten and stated clearly as per the title of study. Reviewer \#2: the manuscripts makes an important contribution to the field but needs major revisions. the writing style is not very much scientific\> the authors would benefit much by reading latest publication in the area and adjust the manuscript accordingly. the discussion and conclusion section in their current state need further revisions as they do not meet the minimum journal standards \*\*\*\*\*\*\*\*\*\* 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: **Emmanuel Ugwa 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.0274573.r004 Author response to Decision Letter 1 29 Jun 2022 Dear EDITOR, RESPONSE TO REVIEWERS COMMENTS Manuscript Title: Evaluating the impact of maternal health care policy on stillbirth and perinatal mortality in Ghana: a mix method approach using two rounds of Ghana demographic and health survey data sets and qualitative design technique Abstract All comments under the ABSTRACT sub-section have been addressed as follows; 1\. Background: This has been revised appropriately and the WHO/UNICEF, 2020 report referenced. 2\. Methods: The quantitative data was analyzed first, then the findings necessitated the qualitative study. The IDIs, especially the expert views were born out of the findings of the quantitative findings. The interview guide is attached as supplementary file. 3\. Results: Although the results of the qualitative findings provide a certain perspective to the quantitative finings, this study does not seek to establish causal inferences between the results of the quantitative analysis and the qualitative interviews. Its aim was to seek understanding within what contest the ‘free’ policy was expected to produce the desire impact. 4\. Conclusion: The qualitative results did not report exactly what was or what was not part of the ‘free’ maternal health care policy. It reported on what context factors affected the ‘free’ policy implementation and therefore, a consequence the stillbirth outcomes, despite the ‘free’ policy. This has been clarified in the introduction and mothed section. 5\. Key works; “policy evaluation” been included accordingly. Introduction 1\. Comment: Stillbirth and perinatal mortality are part of the maternal and child health indicators which are very sensitive health system indicators reflecting negatively on the overall performance of the health systems. This needs to come out clearly. Response: Sentence has accordingly been revised. 2\. Comment: Stillbirth and perinatal mortality have received relatively good policy attention. E.g. new born health has received renewed attention and it has the potential of addressing perinatal mortality. Stillbirth too has received relative attention. The ENAP set targets for national governments to adopt towards addressing the two. You should reflect on what Ghana has done in line to achieve the ENAP targets. Response: Although not explicitly stated, the ‘free’ maternal health care is implicitly implemented to achieve ascertain level of access to antenatal care and facility delivery utilization, which essentially feed into the purpose of ENAP. This has been explained in the opening introduction leading to stating the purpose of the study. 3\. Comment: The global burden of 2.6 stillbirth is an old statistic. The current burden is 2 million annual stillbirths according to UNICEF/WHO report titled “The neglected tragedy; the global burden of stillbirth 2020. The authors need to use Up to date references. Response: This has been revised. 4\. Comment: Technically, the WHO defines stillbirth as intrauterine death after conception”. This statement is not true. The definition only applies to fetal death after 28 weeks. Revise as appropriate. Response: Comments revised as suggested. 5\. Comment: “Access to maternal health care is proven to reduce stillbirth”. It is not access to any maternal healthcare but rather “quality and timely maternal healthcare” you need to revise this to reflect the recommendation. Response: “Quality and timely access” added to said sentence as recommended. 6\. Comment: Related to the above, the authors need to highlight the global burden, regional picture and country status to bring out the problem clearly in a funnel-like approach. Response: revised. 7\. Comment: Did that policy explicitly say that it targeted to address stillbirth and perinatal mortality? You need to reflect it in the introduction. Response: The policy included comprehensive antenatal care, facility level delivery and post-delivery care up to 10 days. During this period care, one of the gains expected by implication is reduced stillbirths and perinatal death. This is the bases for this study, to evaluate the ‘free’ policy contribution to late pregnancy outcomes and new-born care. 8\. Comment: Policy components of the free maternal healthcare policy need to be highlighted for the reader. Right now it is not known. Response: This has been revised in the introductory section. 9\. Comment: Other than the social health insurance, the main maternal and child health intervention since 2000 has been the MDGs with specific goals for maternal and child health whose intervention impacts directly on the indicators you are studying in this study. This relationship needs to come out explicitly. Response: Since MDGs were reviewed to SDGs, this is perhaps, a more appropriate reference to make which is what we made in our introduction section. 10\. Comment: As early stated, other than quoting the 2015 statistics, it’s better to use recent data from the 2020 UNICEF/WHO report. Response: The UNICEF/WHO reported has been used to update the figures and appropriated referenced. 11\. Comment: Also, instead of quoting Nigerian data it would be important to quote the stillbirth burden in Ghana straight away. Response: Ghana data used instead of Nigeria. 12\. Comment: The definition of stillbirth is not clear; key issues to note are the 28 weeks’ threshold as the official position of WHO, and birth with no sign of life either fresh or macerated. This needs to come out explicitly and when using the official definition of WHO, it’s better to use their own sources. Response: Statement revised. 13\. Comment: Other definitions which stretch to less than 28 weeks of gestation are not official but rather country and study specific definitions based on period of viability. Response: Definitions has been edited and revised. 14\. Comment: There are contextual factors which need to come out clearly in the background. For example, the MDGs were a cornerstone for improvement of maternal and child health services globally. Also, the UN secretary general strategy Every Woman Every Child which was operationalized through every New Born Action Plan (ENAP) need to be highlighted when investigating stillbirth. They are the back born for the renewed interest in global stillbirth campaigns. Response: Comments appropriately incorporated. 15\. Comment: An explanation of the “free maternal health policy within the national health insurance scheme needs to be done in the introduction. Was it only aimed at improving access through addressing the healthcare costs? Or there are maternal health interventions within such as increased access to EmOC, MPDSR among others. Response: The ‘free’ maternal health care policy explained and the national health insurance explained. 16\. Comment: “Studies have shown that maternal age, rural/urban area of residence”. This paragraph ignores the important contribution of the health systems factors in addressing stillbirth risks and yet the said policy was implemented within the health systems. It would be important to strike a balance by highlighting the health systems related factors too. Response: indeed, the policy is being implemented with certain health systems factors, which was the focus of the qualitative study. Accordingly, the sentence has been revised. 17\. Comment: Contextual definitions: Stillbirth; kindly check this definition. It lacks the basics of what would constitute a stillbirth such as after the 28 weeks cut off. Response: Definition revised. 18\. Comment: Free maternal healthcare policy: This is simplistic. I suggest an elaboration of the ingredients of this policy and the aspects of maternal health that were covered. Its coverage within the Ghanaian population and any other important information. Response: This has been revised and scope and package included in the introduction section. 19\. Comment: Page 13 “local contest” did you want to mean “context”? if so change as appropriate. Response: Yes, this refers to ‘context. Appropriately edited. 20\. Comments: “we analyze the outcomes of stillbirth and perinatal deaths among mothers in2008, when the policy had just started, compared to 2014 when the policy was fully rolled out”. This should speak directly to the title of the paper and objective of the study in the introduction of your abstract. Response: Comments revised accordingly. Methods 1\. Comment: Study design; the section is rather speaking more to the study setting other than the design. Be explicit and intentioned right away. Response: This comment has been addressed. 2\. Comment: Study setting; some information given is irrelevant to this paper, rather give information that adds value. For example, the geographical location may not be that needed, instead data on stillbirth and perinatal mortality may be more important here. Response: Data on stillbirth for the qualitative study sites have been included. 3\. Comment: The sampling procedure for the qualitative component is missing. Response: Perhaps, this was clear enough. It has been revised accordingly. 4\. Comment: Details about the qualitative tool are missing. It would be important to take the reader through the kind of information the tool elicited. Response: A little more detail has been added on the qualitative information. Tools also attached as supplementary file. 5\. Comment: Instead of attaching the ethics review letter, better quote the reference number for the protocol from the ethics committee. Response: protocol number quoted under the section ‘ethical consideration’ 6\. Comment: No mention of the inclusion and exclusion criteria for FGD participants. It needs to come out. Response: This had been revised to include inclusion and exclusion criteria of FDGs. 7\. Comments: On the qualitative data analysis, how was the codebook developed, who coded the data? How many people were involved in the analysis? Response: The first author analysed the qualitative data and supervised by the second and third authors. Coding essentially was guided by the deductive themes of stillbirth, and the context factors affecting maternal health care utilization under the ‘free’ policy and how they affect ‘stillbirth’ outcome. The emerged significant statements were grouped as inductive sub-themes and reported as the study results with verbatim quotes. Statements were reviewed thoroughly for significance through multiple reading and grouped as affecting quality of care and perhaps negatively influencing stillbirth. 8\. Comment: How was saturation attained? Response: When subsequent IDIs added no particular new information, we determined that saturation was reached, and suspended interviews. However, FGDs continued as planned and all 8 groups of FGDs were carried out and recorded views transcribed analysed in unison. 9\. Comment: Attribution of the reduction to the policy Response: This comment is not particularly clear. However, the study found that perinatal mortality rate declined between 2008 and 2014, while stillbirth rate increased within the same period. Yet, both perinatal mortality and stillbirth were more likely to occur in the ‘free’ maternal health care policy group compared to the no ‘free’ maternal health care policy group. 10\. Comments: The authors need to clarify explicitly how the two methods of data collection were integrated and informed each other. They should also state if this did not happen and why. Response: data collection and integration has been revised under the methods section 11\. Comment: I would suggest that you follow the COREQ Checklist to report on your qualitative findings better. Response: Although, the Consolidated Criteria for Reporting Qualitative research (COREQ) was not particularly used in the study, nonetheless, aspects in the manuscript have addressed issues the requirements of the COREQ checklist. The qualitative study conducted mainly by the first author, as a PhD candidate of the University of Ghana School of Public Health, under a thesis supervisory committee led by the second author, received ethical approval from the Ghana Health Service Ethical Review Committee (number quoted in main test of manuscript). Analysis was content in nature, where sub-themes were constructed to report the study results. Results 1\. Comment: “On risk of stillbirth, pregnancies in the ‘Free’ Maternal Health Policy (FMHCP) group were 1.64 times more likely to result in stillbirth compared to the no FMHCP group”. This needs to come out explicitly in the methods section how these groups were created. Response: Group creation has been revised accordingly in the design section. Also, Table 2 has been revised to report the values of the two groups. 2\. Comment: Qualitative study participants: the unit of analysis for the FGDs is the group. Therefore, when describing your study participants, its better to refer to how many FGDs were conducted other than the total number of participants within those FGDs. Response: descriptive section of FDGs revised appropriately. 3\. Comment: “The majority of pregnant women respondents had given birth previously, 41 (90.1%). Only 4 (8.9%) pregnant women participants were having children for the first time”. To me these appear to belong to the same group unless if you want to say that some were pregnant at the time of the interview and reference is being made to the index pregnancy. Revise statement to bring out clarity. Response: Yes, we agree with your comment. The section/sentence has be revised. 4\. Comment: Table 6: the column on average duration is less significant. This description should be in the narrative and refer to the average duration of each interview. In its current form it appears to be the total duration of all interviews in each category. Kindly revise as appropriate. Response: We agree with the reviewer. The particular column has be deleted appropriately. 5\. Comment: “…A few cases also dodge the hospital may they have two previous CS and knows that if they come to the hospital, there will be CS, so they avoid the hospital, when there are complications, then they quickly come...” (Doctor 1, IDI, UER). Start to use “signposts” in your write up. Whereas the introductory statement was speaking to increased number of macerated stillbirth, this statement seem to be off. Revise as appropriate. Response: Statement/section revised appropriately. 6\. Comment: “…and when they come we manage them at postnatal care…. they say they didn’t know that labour had started, some say they don’t want the examination. One woman was frank, she said when they come, they put fingers on her vagina and that one she doesn’t like it…” (Doctor 2, IDI, UER). Refer to comment above for this quotation. Response: Also revised. 7\. Comment: Service providers also observed a culture of little or no attention to stillbirth issues. Public or media lack of attention on stillbirth was cited as a principal reason compared to maternal mortality. On the other hand, pregnant women midwives’ snobbish attitude to them during labour, as a contributory factor to the rising stillbirths. The following quotes explains further. Was this part of your study objective or scope of the study? To me it appears not. Response: Yes, these emerged from the IDIs, as some of the contextual factors that bothers on stillbirth relative the ‘free’ policy in practice. 8\. Comment: “Use of local herbs for ‘rapid’ uterine contraction” this can’t be a subtheme of the study since its not part of the policy. It is rather inadequate implementation of the policy that is resulting into this. The write up should reflect that. Its either poor access to available services or poor enforcement of the policy that is resulting into this Response: This has been revised and reported under ‘stillbirth’. Essentially, the IDIs explained the consequences of taking local herbs which was a common practice in some of the study sites. Thus, making it a relevant observation, as a factor associated with the poor outcome of stillbirth, despite the ‘free’ policy. 9\. Comment: The qualitative findings only talk about the high numbers of stillbirth which have been explained to be caused by poor use of partograph for pregnancy monitoring and shortage of medicines. I highly doubt this is comprehensive and can solely explain the quantitative findings. More exploration of the qualitative data for more reasons to explain the high numbers. Response: Indeed, poor use of partograph as found in the current study is one of the context factor which may be explaining the stillbirth outcomes in the study site. Other context factors that emerged included; late reporting/arrival, poor, shortages of folic acid/multivitamin, and lack of interest in stillbirth ‘matter’. The section has accordingly been revised. 10\. Comment: “Service providers are of the view that the increase in utilization puts pressure”. Ensure to write the results section in past tense since activities and comments happened sometime back. Response: The results section has been read and revised. 11\. Comment: On the other hand, pregnant women midwives’ snobbish attitude to them during labour”. Revise sentence as appropriate. Response: sentence has been revised. 12\. Comment: Just so to confirm are these FGD quotes from mothers that experienced a stillbirth? If so it should be reported as such and if not you need to clarify in the methodology section. Response: Not only mothers who experienced stillbirth, but mothers previous experience while giving birth at the facility level and use of using antenatal care. 13\. Comment: Due to the facility level shortages, pregnant women regularly asked to purchased medicines outside of the health facility”. Revise statement as appropriate, it is not clear in its current form. Response: Statement revised accordingly. 14\. Comment: “This was not only a problem to the pregnant women, but also frustrates the quality of processes of the health care professionals” use past tense as appropriate. Response: Manuscript read and edited accordingly. Discussion 1\. Comments: Revise the discussion section accordingly after effecting changes in the results section Response: Discussion section revised. 2\. Comment: “improvement in perinatal mortality” is it improvement in perinatal mortality or reduction in perinatal mortality? Response: reduction in perinatal mortality. This has been revised accordingly. Conclusion 1\. Comment: Revise the discussion section accordingly after effecting changes in the results section Response: Relevant revisions effected, including the conclusion section. John Azaare (PhD Candidate, Thesis defended successfully) Lead/Corresponding Author 10.1371/journal.pone.0274573.r005 Decision Letter 2 Lupattelli Angela Academic Editor 2022 Angela Lupattelli 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 2022 PONE-D-21-35524R2Evaluating the impact of maternal health care policy on stillbirth and perinatal mortality in Ghana: a mix method approach using two rounds of Ghana demographic and health survey data sets and qualitative design techniquePLOS ONE Dear Dr. Azaare, 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 Sep 04 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, Angela Lupattelli, PhD 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. Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation. Reviewer \#1: All comments have been addressed Reviewer \#2: (No Response) \*\*\*\*\*\*\*\*\*\* 2\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 3\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 4\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 5\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 6\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: Thank you for the attempted revision. However, the reviewer would like to read changes made. You can either copy and paste on your comments or provide page and line numbers so that reviewer can easily identify such changes. Comments such as ''revision has been made as appropriate'' are not required. Reviewer \#2: Second round of review and the authors have done a good job to address earlier comments. more additional comments to enhance clarity are herewith attached. regards \*\*\*\*\*\*\*\*\*\* 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: **Emmanuel Ugwa 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.0274573.r006 Author response to Decision Letter 2 25 Aug 2022 THE ACADEMIC EDITOR PLOS ONE JOURNAL Dear Editor, RE: Evaluating the impact of maternal health care policy on stillbirth and perinatal mortality in Ghana: a mix method approach using two rounds of Ghana demographic and health survey data sets and qualitative design technique. Thank you for reviewing this manuscript. Below are the responses to the issues raised. General comments. 1\. The definition of stillbirth should come upfront within the first paragraph of the introductory section. Referring to it as a neglected tragedy is not a definition parse but rather what it is. Response: This has been addressed in Page 1, paragraph 1. 2\. Paragraphs two and three are a repetition with one showing absolute numbers and the other showing percentages of global stillbirths. Consider deleting one. Response: Paragraph three has been deleted, while paragraph two revised. 3\. The definition of stillbirth is inadequate and its contextual. Rather use the WHO definition which restricts stillbirth to loss after 28 weeks for global comparisons. It is clear that in developed countries the definition can go as low as 20 weeks due to viability but that is not the standard definition. Response: This has been revised in page 4, (under ‘Context definition section). 4\. It may not necessarily be history of abortion “Studies have shown that maternal age, rural/urban area of residence, twin pregnancy, history of abortion” but rather negative pregnancy history. It may be pre-term delivery, miscourage, abortion or even stillbirth. You may consider revising the sentence to bring out this fact. Response: “history of abortion” revised appropriately in paragraph four (under ‘conceptual framework’ section). 5\. Under contextual definitions, you definition of stillbirth is not appropriate. The standard definition is well known and if any adjustments to the definitions were done in this study, kindly indicate so. Key to the definition is the cut off of 28 weeks and being born with no sign of life. You may need to revise or operationalise what you call “dying within day zero after birth” for which I would highly discourage you from using. Response: This has been revised in page 6 (under context definition). 6\. Study design: Are you sure the mixed methods design was in “two prongs”? To me it appears it was sequential mixed methods because you started off with the quantitative secondary data and later the qualitative. If that is the case, then revise as appropriate. Response: This has been revised in paragraph one, page 8 (under the ‘design section’). 7\. Statement not clear “who were not necessarily stillborn babies”. Did you want to say the mothers who participated in the FGDs were not necessarily “mothers” to stillborn babies? If that is the case, then revise as appropriate. Response: This has been revised in paragraph two, page 8 (under the ‘design section’). 8\. Qualitative study setting: other than giving a geographical description, I would rather you write more about the health systems characteristics especially as they relate to maternal health service access and stillbirth burden, or characteristics like to lead mothers to having a stillbirth. Still more about the policy under review would work more compared to giving geographical descriptions. The map can be retained though for visual clarity about the study location. Response: The two regions are similar in structure and organization relative to health system characteristics, albeit with different outcomes of stillbirth and facility utilization. This has been reported in the manuscript (Table 1) and captured as Figure 3 (page 15). 9\. Inclusion and exclusion criteria: in the inclusion you say you included women 15-49 years and in exclusion you indicate women below 16 years. Where does that put the ones aged 15 years you indicated under inclusion criteria? Were they included or excluded? Response: Age “15-49’ as captured in the inclusion criteria is in reference to the age range of the secondary data sets as contained in the DHS survey data sets. However, age 16 years and below were excluded from the in-depth interview in line with the legal age limit for females to have consensual sexual decision, thus, pregnancy. To avoid any confusion for readers, the age range of ‘15-49’ in respect of the Ghana DHS survey data set has been removed under the inclusion criteria section. 10\. Besides exclusion is not the direct opposite of inclusion criteria but that those respondents who met the inclusion criteria in the first place and for some reason were excluded from the sample later such as those that met the inclusion requirements but had missing data for other variables. Revise as appropriate. Response: This has been revised under the section ‘Inclusion criteria’. 11\. Data analysis-qualitative: how was the code book developed? who coded the data? how was quality control ensured? how did you choose the quote to represent others in the results section? Response: The first author coded the data and was supervised by the second, and third author. Codes were deductive (prior to conducting in-depth interviews) based on the study objectives of stillbirth and context factors of service utilization. The second and third author checked for quality of codes, clarity and uniqueness, as supervisors of the first author. Afterwards, common themes were constructed inductively through multiple reading of the transcripts. This has been explained in paragraph on of the ‘qualitative analysis’ section. 12\. Ethics statements are repeated under “tools and pre-testing” and “ethical consideration”. Consider removing one to avoid repeating yourself. Response: The ‘ethical statement’ under the “tools and pre-testing” has been removed. 13\. Results: source of data presented in table 1, how is that data from the field for you to label it as “field data”. It is clearly from the DHIMS and should be labelled as such. Response: This has been corrected in Table (page 15). 14\. Quantitative findings. Is data from table 1 not part of the quantitative findings? If so then the sub heading of quantitative findings should be shifted up to include results from table one. Response: This comment has been addressed under the ‘Results’ section in page 14. 15\. Rates of stillbirth increasing between the two timepoints this is a unique finding that needs to be explored more. Why the risk appeared to be more in the free maternal health care policy group also needs to be explored. Response: We agree with the reviewer. The qualitative aspect of the study was aimed to address this concern, thus, the results as presented in the qualitative section. Of course, further exploration will be useful perhaps in future studies. 16\. Although there was an overall decline between the two timepoints, the risk of perinatal mortality was also high in the free maternal healthcare policy compared to the control group. This phenomenon needs to be explored and discussed in relation to the national context. It calls for interrogation of the quality of secondary data used. Otherwise the conclusion would be that the free maternal healthcare policy instead led to an increase in both stillbirth and perinatal mortality. Response: The secondary data used were two-rounds of Ghana DHS data sets (baseline and end line). DHS is a standardly collected data sets using complex design by the DHS Programme, funded by the USAID (independent of this study). The DHS data sets are nationally representative and regularly used in evaluating national programmes or project. Our analysis took clustering and stratification into consideration and applied sample weighting and also adjusted for confounding to minimise internal bias. The study findings of rising stillbirths could perhaps be attributed to increased utilization which may have affected quality of care, and this was corroborated by the qualitative study participants during the IDIs and FGDs. 17\. Centrally to what is known, the education level in this study seem to disfavour the outcome variables (that the higher one’s education level the more likely they are to have negative outcomes). This needs to be interrogated too in the context of national maternal and perinatal mortality service delivery. Response: We agree with the reviewers that education appears a disincentive as per the current findings. However, education as reported in this study is a confounding variable. Perhaps, analysing education as the main outcome variable could yield different results, but that is outside the scope of the current study. 18\. Table 7: group discussion the unit is the focus group. Kindly indicate it and in bracket show the total number of participants in those FGDs which is 44. Revise as appropriate. Response: This has been revised appropriately in Table 7. 19\. “Antepartum (macerated) stillbirths” these happen before the onset of labour and any stillbirth happening after the onset of labour whether outside the health facility is fresh stillbirth. Revise the sentence as appropriate to reflect that death before arriving at the facility is not the cut off to determine macerated stillbirth. Response: This has been revised accordingly in paragraph two, page 27. 20\. “The reasons for the delay in arriving in the health facility”. What I infer from this and subsequent statements is that the pathway to seeking delivery services has a number of bottlenecks which include actors (Traditional birth attendants) and negative cultural practices (use of local herbs) this needs to come out clearly other than amplifying the “kaligutiem” Response: Yes, cultural practices such as herbal intake affects the pathway to seeking care during labor. However, our use of the term ‘kalgutiem’ has been chosen to reflect the local context and observations by the service providers. However, this has been revised in pages 29 and 30. 21\. The effects of data capture may need to be brought into the picture here. The increased number/rates of stillbirth may have been due to improvements in data capture that almost all cases are captured. Needs to be discussed if you think it may have an effect on the observed trends between the two timepoints. Response: The effect of data capture was brought out by the service providers during the IDIs as reasons for the high figures. The current study did not independently analyse data capture impact to ascertain its influence on the totality of the stillbirth rate. Our analysis of the outcomes of stillbirth based on the secondary data obtained from the DHS programme. 22\. “Both stillbirth and perinatal deaths were prevalent among” did you try to explore the role of data capture this might play on the observed outcomes? It may be that proper data capture in the free policy group will show many cases which may be the opposite in the other group. Response: Although this study did not independently analyse the role of data capture, the service providers did mention ‘data capture’ as influencing the high figures during the IDIs, and this has been reported as part of the qualitative findings in page 27, paragraph two and also reflected in the discussion, paragraph nine (page 36). 23\. “The findings imply that stillbirth failed to show a decline in 2014”. You may need to highlight the limitations of this study wherein the paper objectives were never the main aim of the policy so there is a likelihood that it may have affected the observed trends. Response: This comment has been addressed in paragraph one under ‘strength and limitation’. SIGNED John AZAARE (Corresponding Author) 10.1371/journal.pone.0274573.r007 Decision Letter 3 Lupattelli Angela Academic Editor 2022 Angela Lupattelli This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 31 Aug 2022 Evaluating the impact of maternal health care policy on stillbirth and perinatal mortality in Ghana: a mix method approach using two rounds of Ghana demographic and health survey data sets and qualitative design technique PONE-D-21-35524R3 Dear Dr. Azaare, 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, Angela Lupattelli, PhD Academic Editor PLOS ONE 10.1371/journal.pone.0274573.r008 Acceptance letter Lupattelli Angela Academic Editor 2022 Angela Lupattelli 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 Sep 2022 PONE-D-21-35524R3 Evaluating the impact of maternal health care policy on stillbirth and perinatal mortality in Ghana; a mixed method approach using two rounds of Ghana demographic and health survey data sets and qualitative design technique. Dear Dr. Azaare: 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. Angela Lupattelli Academic Editor PLOS ONE [^1]: The authors have declared that no competing interest exist.

# Introduction Gastric cancer is the fifth most common cancer and the third most fatal cancer worldwide. In 2018, there were more than one million cases of GC, which caused an estimated 783,000 deaths. As such, GC is a public health-related burden for developed and developing countries. Almost half of the world’s GC cases and deaths occur in China, with an estimated 500,000 deaths in 2015. GC is a multi- factorial disease, and both environmental (68%) and genetic (22%) factors have been reported to associate with its etiology. Personal lifestyle choices, such as insufficient fruits and vegetables in the diet, excessive alcohol consumption and high intake of salt, have been demonstrated to associate with GC. In addition, a family history of GC and *Helicobacter pylori* infection both increase the risk of developing GC. As most patients have non-specific symptoms in the early stages of the disease, GC can progress to advanced stages, which is when most patients are diagnosed. Therefore, it is important to identify reliable diagnostic and prognostic markers of GC, with the hopes of detecting GC in the early stages of the disease. Long non-coding RNAs (lncRNAs) are a group of non-coding RNAs of more than 200 nucleotides in length without open reading frames. LncRNAs have a variety of functions, including post-transcriptional regulation, gene regulation and intercellular signaling, and dysregulation of lncRNAs is involved in a variety of diseases such as cancer. Accumulating evidence has revealed that lncRNAs are essential for cancer progression and metastasis. Recently, new lncRNAs have been identified, and bioinformatics approaches have been used to understand the roles of lncRNAs. However, the function of LINC00922 in the development of GC is unclear. Thus, we investigated the role of LINC00922 in GC using data mining and in vitro experiments. # Results ## Expression and prognostic value of LINC00922 in GC based on TCGA data A flow chart of the study design is shown in. In total, we extracted 407 samples from TCGA datasets, including 375 tumor and 32 non-tumor samples. As shown in, LINC00922 expression was increased in GC tissues (2.908 ± 0.077) and decreased in adjacent non-tumor tissues (0.790 ± 0.189) (P \< 0.001). However, LINC00922 expression was not correlated with age, gender, race, tumor depth, lymphnode invasion, remote metastasis, and pathological stage. Next, we used GEPIA to determine the prognostic value of LINC00922 in GC. As shown in, increased LINC00922 expression was significantly correlated with poor overall survival (P = 0.0056) and disease-free survival (P = 0.0091), indicating that LINC00922 functions as an oncogene in GC. ## Prediction of genes related to LINC00922 To identify the genes related to LINC00922, the MEM database and TANRIC were used. For the MEM database, 4,360 related genes were identified when the threshold was set to 10⁻⁶. For TANRIC, 1,025 related genes were identified. Among these, 336 genes were overlapping. ## GO and KEGG analyses We used Database for Annotation, Visualization and Integrated Discovery (DAVID) to conduct GO and KEGG analyses. For the GO database, there are three categories of functional annotations, namely, molecular function (MF), cellular component (CC) and biological process (BP). As shown in, there were 10, 10 and 3 significant terms in BP, CC and MF, respectively (false discovery rate (FDR) \< 0.05). For MF, cell adhesion, biological adhesion and cell motion were mostly enriched. For CC, plasma membrane, actin cytoskeleton and contractile fiber part were mostly enriched. For BP, actin binding, cytoskeletal protein binding and growth factor binding were mostly enriched. In the KEGG pathways, there were 11 significant terms (P \< 0.05). Among these, the TGF-β signaling pathway, dilated cardiomyopathy and focal adhesion were the top three terms. ## Construction and analysis of protein-protein interaction network As shown in, the protein-protein interaction network contained 335 nodes and 167 edges, and the degree of connectivity was analyzed by Cytoscape using these features. The genes with more than three degrees were defined as hub genes. As shown in, LAMC1, LTBP1, IGFBP4, IGF1, PENK, SPARCL1, PRSS23, FSTL3, PAK3 and ADRB2 were the top ten hub genes. Next, we examined the expression levels and prognostic values of the most significant hub genes in GEPIA. The results indicated FSTL3 was highly expressed in GC tissues, and correlated with overall survival and disease-free survival, while LAMC1 expression was increased in GC tissues, but not correlated with overall survival and disease-free survival of GC patients. Thus, FSTL3 and LAMC1 are the likely targets of LINC00922. ## LINC00922 is essential for GC cell proliferation To examine the function of LINC00922, we assessed the expression of LINC00922 in GC cell lines HGC27, SGC7901 and AGS as well as the normal gastric cell line GES-1. As shown in, all three GC cell lines expressed a high level of LINC00922, and the HGC27 cell line expressed the highest level of expression. Next, we silenced LINC00922 in HGC27 cells. To assess cell proliferation, the MTT assay was used, and the results showed that LINC00922 knockdown could significantly affect cell viability in the ShRNA group (P \< 0.05). Furthermore, decreased LINC00922 expression could increase HCG27 cell apoptosis (P \< 0.05). These findings indicate that LINC00922 is important in cell proliferation. ## LINC00922 associates with gastric cancer cell invasion and migration The Transwell assay was performed to assess cell invasion. As shown in, decreased LINC00922 expression could inhibit cell invasion. Next, the wound- healing assay was used to evaluate the effects of LINC00922 on cell migration. As shown in, LINC00922 knockdown could significantly decrease cell migration. These results reveal that LINC00922 is an important regulator of cell invasion and migration. # Discussion According to The National Central Cancer Registry of China, the mortality rate of patients with GC is the second highest among all cancers in China. Although advances have been made in the treatment of GC, such as surgery, chemotherapy and radiotherapy, the prognosis of GC patients still remains poor. Among the prognostic and therapeutic biomarkers, HER2 is the only biomarker screened in clinical settings. However, HER2 is overexpressed in only 9%–38% of GC patients. Thus, a better understanding of GC will lead to the identification of new biomarkers. Our results indicated that LINC00922 was overexpressed in GC tissues. Recently, several studies have demonstrated the association of lncRNAs with a variety of cancers. For example, the increased expression of lncRNA HOXA11-AS has been reported to associate with cell cycle progression, invasion and migration in GC. Another lncRNA MALAT1, regulates VE-cadherin/β-catenin, ERK/MMP and FAK/paxillin pathways to facilitate angiogenesis and promote metastasis. Here, we demonstrated that LINC00922 was overexpressed in GC tissues and correlated with poor prognosis in GC patients. These findings are consistent with those of other studies, which reported several lncRNAs, such as PEG10, MT1JP and BLACAT1, to associate with cancer patient survival. We identified 366 genes related to LINC00922 by the MEM database and TANRIC. By GO and KEGG analyses, adhesion-related processes, such as cell and biological adhesion, were significantly enriched. Cell migration was also related to LINC00922, which was confirmed by in vitro experiments that showed LINC00922 to be required for cell invasion and migration. To date, several studies have revealed that lncRNAs play critical roles in metastasis. For example, Sun et al. demonstrated that YAP1 is highly up-regulated in GC, accelerates tumor growth and metastasis through ERK1/2 phosphorylation and regulates lncRNAs, including HOTAIR, MALAT1, LATS2 and LATS2-AS1-001. HOTAIR and LINC00261 can regulate the epithelial-mesenchymal transition by modulating the expression of E-cadherin, N-cadherin and vimentin. It is well established that an imbalance between cell proliferation and survival is critical for cancer progression. For example, H19 is up-regulated in GC and suppresses the expression of p53 and p53 target B-cell lymphoma-associated X protein, which induce GC cell proliferation. H19 also regulates GC cell proliferation by acting as a competing endogenous RNA of miR-141 and competing for binding to its target genes, including insulin-like growth factor (IGF) 1, IGF receptor 2 and zinc finger E-box-binding homeobox 1. Gastric carcinoma high expressed transcript (GHET1) interacts with IGF2 mRNA binding protein 1 (IGF2BP1), which enhances the interaction between IGF2BP1 and c-Myc and increases the expression of c-Myc, thereby promoting cell proliferation. Our findings demonstrate that LINC00922 inhibits cell apoptosis and promotes cell proliferation, indicating that the expression of LINC00922 might disrupt several cancer-related processes and signaling pathways, and contribute to GC. We investigated the prognostic significance of LINC00922 in GC using GEPIA, and the results revealed that increased LINC00922 expression correlated with shorter overall survival and disease-free survival. In addition, we obtained the predicted hub genes, FSTL3 and LAMC1, of LINC00922 using STRING and Cytoscape. FSTL3 is a member of the follistatin family, which binds to and inactivates activin to regulate cell growth and differentiation. FSTL3 has been reported to participate in tumorigenesis and associate with nuclear grade and tumor size in breast cancer. LAMC1 encodes laminin γ1, which is involved in a variety of processes such as normal tissue development, cancer cell invasion and metastasis. In hepatocellular carcinoma (HCC), LAMC1 competes for miR-124 binding and acts as a trans-regulator to induce the expression of CD151. Zhang et al. reported that overexpression of LAMC1 in HCC enhances invasion and migration and predicts poor prognosis. Taken collectively, these findings suggest that FSTL3 and LAMC1 are involved in tumor biology and targeting LINC00922 may be an effective therapy in GC patients. There were several limitations in this study. First, lncRNAs play a variety of roles in GC, many of which were not examined here. As such, the functions of LINC00922, such as the ceRNA mechanism, should be further explored and combined with bioinformatics approaches and in vitro and in vivo experiments. Second, a larger cohort and a computational model are needed to assess the potential of LINC00922 as a biomarker for GC. Despite these limitations, our results indicate that LINC00922 is critical in the progression, invasion and migration of GC cells and can be used as a potential target for the treatment of GC. # Materials and methods ## Data collection GC tissue and non-tumor tissue gene expression profiles were obtained from TCGA dataset (<https://portal.gdc.cancer.gov/>). To normalize the expression levels of lncRNAs in different samples, the data were log2-scaled. Differential expression analysis of lncRNAs was performed with the Limma package (<http://www.bioconductor.org/>). P-value \< 0.05 and fold change (FC) \> 1 were set as the threshold. Among the differentially expressed lncRNAs, LINC00922 was selected for further study. ## GEPIA dataset analysis GEPIA is an online web server that includes 9,736 tumor tissue and 8,587 normal samples tissue specimens in TCGA and genotype-tissue expression datasets. Differential gene expression analysis and Kaplan-Meier analysis of overall survival and disease free survival were conducted by GEPIA. ## Genes potentially related to LINC00922 in gastric cancer Online bioinformatics databases MEM and TANRIC were used to predict the genes potentially related to LINC00922. In this study, overlapping genes from the aforementioned programs were recorded as target genes of LINC00922 in GC. ## Analysis of associated gene-enriched pathways and their functions DAVID (<https://david.ncifcrf.gov/>) was used to conduct GO and KEGG analyses. The results were considered to be significant when the FDR was \< 0.05 in GO analysis and the P-value was \< 0.05 in KEGG analysis. ## Protein-protein interaction network construction and confirmation of hub genes To identify the interactions between overlapping genes, STRING 11.0 ([http://www.string-db.org](http://www.string-db.org/)) was used to construct the protein-protein interaction network. Genes with combined scores \> 0.9 were obtained. The degree of connectivity in the protein-protein interaction network was analyzed with Cytoscape software 3.0, and the hub gene network of LINC00922 was constructed. In the protein-protein interaction network, each node indicated a protein, and each color corresponded to a cluster. The edges represented the functional associations of prediction, and the thickness of lines represented the strength of evidence. ## Cell culture and transfection HGC27, SGC7901 and AGC cell lines were obtained from the Cancer Hospital of the Chinese Academy of Medical Sciences (China). The GES-1 cell line was obtained from the American Type Culture Collection (USA). The cells were cultured in Roswell Park Memorial Institute (RPMI) medium supplemented with 10% fetal bovine serum (FBS) (Gibco, USA) and 100 μg/ml streptomycin (Invitrogen, USA), and the cells were cultured in conditions of 5% CO₂ at 37˚C. The LINC00922 sequence was obtained from GenBank, and shRNAs against LINC00922 were synthesized with the following sequences: shRNA-1 (5’-`CCTGCACCTACAGATCTACACCTCGAGGTGTAGATCTGTAGGTGCAGG`-3’), shRNA-2 (5’-`TGCAGGAAGTGTTCATCTAAGCTCGAGCTTAGATGAACACTTCCTGCA`-3’). ShRNAs against LINC00922 or the negative control were transfected into the pSICOR-GFP plasmid (Shanghai Geneline Bioscience, China). Flow cytometry was used to confirm knockdown. The cells with the highest knockdown efficiency were used in subsequent experiments. ## Real-time polymerase chain reaction (RT-PCR) Total RNA was isolated from cells using TRIzol reagent (Invitrogen, USA) according to the manufacturer’s instructions, and cDNA was synthesized. Next, the mRNA transcripts were amplified in the Eppendorf PCR system (China). The relative expressions of each target gene was normalized to that of GADPH. The primer sequences were as follows: LINC00922 (5’-`CAAGACCGCCAAGCATTAAGAT`-3’ \[forward\] and 5’-`AGTGGCTCCTCCACCACCTCCT`-3’ \[reverse\]) and GADPH (5’-`GTCAACGGATTTGGTCTGTATT`-3’ \[forward\] and, 5’-`AGTCTTCTGGGTGGCAGTGAT`-3’ \[reverse\]). The 2-ΔΔcq method was used to calculate the relative expression level of LINC00922. ## MTT assay and apoptosis analysis For the MTT assay, cells (1 × 10³ cells/well) were seeded in 96-well plates and transfected with shRNAs or controls, and cell proliferation was measured every 1 to 4 days. MTT (20 μl; stock concentration, 5 mg/ml) (Solarbio, China) was added to each well, and the plate was incubated at 37˚C for 4 h. The absorbance at 540nm was measured. For the apoptosis assay, cells were harvested and stained with Annexin V-FITC/PI and analyzed with an Accuri C6 flow cytometer (Becton-Dickinson, USA) and Flow Jo 7.6.1 software (Tree Star Inc., USA). ## Transwell assay and wound-healing assay The invasion assay was performed using Transwell chambers (Corning, USA). In brief, 1 × 10⁵ cells were added into the upper chamber, and the lower chamber was filled with medium supplemented with FBS (10%, 600 μl), which served as the chemo-attractant, and the plates were incubated for 24h at 37˚C. Next, the cells on the lower membrane were fixed with 100% methanol and stained with 0.1% crystal violet. Five fields were randomly selected and viewed under 200× magnification. The number of cells was counted. For the wound-healing assay, cells (1 × 10³) were added into 6-well plates and cultured until confluence. Next, a 200-μL pipette was used to create a wound. The cells were imaged after 24 hours. ## Statistical analysis GraphPad Prism 6 (USA) was used for statistical analysis. The data were presented as mean ± standard deviation (SD). Student’s t-test was used for comparisons between two groups, and one-way ANOVA was used for comparisons among three or more groups. All experiments were repeated three independent times. P-vales \< 0.05 were statistically significant. [^1]: The authors have declared that no competing interests exist.

# Introduction In speech communication, pronunciation variation is ubiquitous. A well- documented phenomenon of contextual variation is assimilation, by which a particular phoneme adopts/adapts to the feature(s) of an adjacent phoneme, such as the production of the underlying segment /n/ in green boat as a surface \[m\]-like sound. In tonal languages, variation can also occur at the supra- segmental level. That is, a lexical tone assigned to a word or morpheme can be realized with different pitch contours contingent upon adjacent tones and discourse context. The question that has received much attention in the literature is how the human brain deals with such variability in the speech signal and what are the neural mechanisms that underlie the representation and processing of speech sounds and their variants. ## Models of Pronunciation Variation in Lexical Representation Different models have been proposed to deal with pronunciation variation, which differ in the level as well as the nature of variation representation and processing. Here, we limit our attention to how regular contextual variation is incorporated into lexical representation. Three main types of accounts have been proposed: the underspecification account, the multi-variant account, and the inference account. The most representative model for the underspecification account is the featurally underspecified lexicon model (FUL), which claims that each word in the mental lexicon is associated with a highly abstract phonological representation with non-distinctive phonological features unspecified, which consequently enables ready accommodation of pronunciation variants in speech. This view has been supported by a series of behavioral and neurophysiological studies, in particular for the underspecification of the \[coronal\] place of articulation in both perception and production. At the other end of the spectrum, the multi-variant account (hereafter MV) assumes that contextual variants are stored in the mental lexicon. The nature of these variants have been proposed to be episodic and detailed or abstract with varying degrees of strength in memory, which reflects the listener’s frequency of exposure to each particular variant form. For example, it was found that American English listeners exhibited a sensitivity to word-specific frequency in their identification of the onset segment (*b* or *p*) of word-nonword continua (e.g., *pretty*-*bretty*), which were embedded in either a flap or a *t* variant carrier word. This suggests the inclusion of the flap variant of spoken words as a stored lexical representation in the mental lexicon. While the FUL and MV accounts differ in how much speech variability is abstracted away in the mental lexicon, one thing they have in common is that they both locate pronunciation variation at the level of feature or higher-level lexical representation. Both therefore differ from the context-based inference account (hereafter CI), which assumes a single abstract lexical representation. Within the CI, pronunciation variation is resolved at the pre-lexical level via an inference process based on their context (e.g.; see also for evidence of feature cue parsing of subphonemic modifications which are important for lexical processing). What is important for this and similar accounts is that speech context is essential for the recognition of pronunciation variants so that inference can be made on-line during speech perception and lexical access. ## Lexical Tone and Tonal Variation It is worth noting that the aforementioned studies all examined speech variants at the segmental level. In addition to consonants and vowels, many languages of the world (60–70%) use pitch variation, together with other (mostly secondary) acoustic features such as phonation, to convey word meanings, known as tonal languages. Unlike consonants and vowels, lexical tones are considered supra- segmental, and their acoustic attribution extends over the whole syllable or multiple syllables. For example, Standard Chinese has four distinctive lexical tones. Produced in isolation, T1 is a high level tone, T2 a high rising tone, T3 a low dipping tone (often accompanied by creakiness), and T4 a high falling tone. Following the tradition of representing lexical tones numerically along five levels of a speaker’s pitch range (with 1 indicating the lowest level and 5 the highest one), we may transcribe the tones as 55, 25, 214, and 53 respectively. The tone-bearing unit in Standard Chinese is the syllable, which typically is also a morpheme. In connected speech, T3 is realized with a rising pitch contour (hereafter Sandhi Rising or T3V) when immediately followed by another T3. Acoustically speaking, T3V is very similar though slightly different from the lexical rising tone (T2). The question that interests us is the consequence of such allotonic variation on the representation and auditory processing of lexical tones in general. Zhou & Marslen-Willson, to our knowledge, is the first perception study that investigates the representation of context-conditioned allophonic tonal variants. They started with the assumption that T3 changes to T2 before another T3 (following). They investigated the representational relationship between T2 and T3 in an auditory-auditory priming lexical decision task. The reaction time results showed that the processing of bisyllabic T3T3 compounds were significantly facilitated by a T3TX prime (where X refers to any of the four tones except for T3), which belongs to the same lexical tonal category but surfaces with a different pitch contour. For the same listeners, T3T3 was also inhibited by a T2TX prime, which belongs to a different lexical tonal category but surfaces with a similar pitch contour, when compared to the condition with control primes which do not share lexical tone or segments. On the other hand, for T2TX targets, T3 (i.e., different tone and different contour) showed a significantly more inhibitory effect than T3T3 (different tone but similar contour) and T2 (the same tone and pitch contour), which in turn showed a significantly more inhibitory effect than primes with no segmental or tonal sharing. Taken together, their results point to the possible representational relationship of T3V = T3 and T3V≠T2, but T2≠T3 and T2 = T3V. However, given their assumption that T3T3 undergoes categorical tone sandhi change into T2T3, they were left with the conundrum of the contradictory results of T2 = T3 but T2≠T2/T3. This challenges the assumption that T3 is changed to T2 when followed by another T3, calling for further studies to clarify the nature of T3 contextual sandhi variant. The more specific goal of our study was therefore to seek neurophysiological evidence for the mental representation and auditory processing of lexical tones and tonal sandhi variants. In particular, we were interested in whether and how in Standard Chinese, the existence of allophonic tonal variant (i.e. T3V) might affect the way a lexical tone (i.e. T3) is represented and processed in general, as compared to the other lexical tones (i.e. T1 and T2). To this end, we will examine whether the processing of T3 would activate automatically its allophonic variant T3V, even when there is no tone sandhi context. Evidence as such should indicate the long-term representation of T3V (together with the canonical T3) in the mental lexicon. ## Neurophysiological and Neuroimaging Studies of Tone Perception A few studies have investigated the neural basis underlying lexical tone perception. Those using neuroimaging techniques such as functional magnetic resonance imaging (fMRI) or positron emission tomography (PET) demonstrate that the left hemisphere is involved in between-category functional processing of lexical tone. However, for lower level acoustic processing of lexical tone, right hemisphere is more often recruited. For example, in an fMRI study, Zhang et al. found that, relative to between-category tonal variation, within-category tonal variation elicited weaker activation in the left middle temporal gyrus (MTG) and, more importantly, stronger activation in the right middle superior temporal gyrus (STG), reflecting low-level acoustic processing. Consistent with this fMRI study, the electroencephalography (EEG) experiment conducted by reported evidence of early right-lateralized mismatch negativities (MMNs) for lexical tone processing, and put forward a two-stage model in which the acoustic signal of a lexical tone is initially processed in the right hemisphere and then the mapping of its functional information (i.e. tonal category) takes place in the left hemisphere. The EEG study conducted by Xi et al. further found that the MMNs elicited by within-category tonal variation are reduced over the left hemisphere as compared to between-category tonal variation. The EEG study in adds further evidence that MMNs evoked by within-category tonal contrast (e.g., pitch level differences within the lexical Tone1) can be lateralized completely to the right hemisphere. Thus, with some differences in the degree of lateralization, the consensus that has emerged from the existing literature is that the left hemisphere is more sensitive to between-category functional processing of lexical tone while the right hemisphere is more sensitive to low- level acoustic processing of pitch variation within a lexical tone. The above studies made important contributions to the neural basis associated with the perceptual processing of lexical tones in general. However, it is important to note that thus far, the existing literature, which examines either within-or between-lexical tonal contrasts, sheds little light on the possible effects of contextual tonal variation on the neural processing of lexical tones in general. Another line of research worth noting is EEG studies which examine how acoustic pitch distances influence lexical tone processing. For example, behavioral results showed that in Standard Chinese, T2 (high rising; 25) and T3 (low dipping; 214) are perceptually more similar and more difficult to discriminate than the pair T3 and T1 (high level; 55). Chandrasekaran, et al. used synthesized pitch contours over the same segmental syllable to examine the MMN responses to large deviant T1/T3 (standard/deviant) vs. small deviant T2/T3 tonal pairs. Their results showed that, for native Chinese speakers, the MMN elicited by the T1/T3 contrast peaked earlier and had larger amplitude than the MMN elicited by the T2/T3 contrast. What remains unclear is whether the above results can be extended to neural processing of naturally produced lexical tones, where multiple cues, in addition to pitch contours, signal lexical tonal contrasts. ## The Present MMN Study The main goal of the present study was to investigate the role of tonal variants in the representation and neural processing of lexical tones in general. Specifically, we were interested in whether T3, given its tone sandhi variant T3V, is 1) underspecified, 2) represented as its canonical tonal form, or 3) represented as both the canonical form and its variant T3V? To this end, a passive oddball MMN paradigm was used to tap into the neural processing of lexical tones. Within the oddball MMN paradigm, infrequent acoustic events (deviant) occasionally occur among frequently repeated sounds (standard). The standard stimulus is generally believed to create a sound representation which corresponds, to a large extent, to the high-level long-term memory traces while the deviant stimulus evokes a low-level acoustic representation of the speech perception. MMN can thus be used to study not only the low-level sensory or acoustic contrast but also the difference between the surface form, extracted from the deviant, and the underlying representation, activated by the standard. Over the last decade, there has been ample evidence that MMNs reflect the phonetic and phonological properties of the sound system of the listeners’ native language. For example, Dehaene-Mambretz showed that given the same acoustic distance, a large mismatch negativity was observed only when the deviant crossed a phonemic boundary (and not with a within-category deviants). Näätänen, et al. reported enhanced MMN responses to phonemic prototypes in listeners’ native languages, relative to when the deviant is a non-native sound. Such native-language, phoneme-related MMN response enhancement suggests the existence of language-dependent memory traces of speech sounds, which can serve as the basis for auditory mismatch detection. More recently, a body of studies has also established that MMNs can be employed as a valuable tool to study the neural correlates of lexical access (see reviews in). These studies lend support to the view that word memory traces are strongly connected assemblies of neurons; the speed and magnitude of their activation is linked to the strength of internal connections in a memory circuit, such that words generate stronger MMN responses than acoustically similar pseudo-words and high-frequency words elicit stronger and earlier MMNs relative to low-frequency ones. We extended the paradigm to examine the effect of tone sandhi variants on the representation and neural processing of lexical tones in general, inspired by a series of studies that examined MMN responses to tap into the representation of segmental features and their related allophonic alternations (e.g.,). We know that MMNs reflect not only high-level long-term memory (e.g. phonological representation in the mental lexicon) but also low-level sensory contrast (e.g. acoustic pitch contour differences of a lexical tone). Given a set of lexical tones (e.g. T1, T2, and T3), low-level sensory contrasts are expected to exist in all of the possible oddball conditions, since T1, T2 and T3 are distinct lexical tones with different pitch contours and native speakers are able to identify lexical tones (produced in isolation) with near ceiling-level accuracy. To tap into the high-level long-term memory representation with regard to lexical tone and tonal variants, a reversed design was adopted, where standard and deviant stimuli in one condition were reversed as deviant and standard in another condition. In other words, we employed four oddball conditions with two tonal pairs (i.e. T1/T3 vs. T3/T1 and T2/T3 vs. T3/T2; 85%/15%). In this way, the standard-deviant acoustic-phonetic contrast, an important factor determining the MMN responses, was identical in both conditions. The specific question that we addressed is whether upon hearing the T3 produced in isolation, its T3V (which only surfaces in the specific T3T3 sandhi context) is also activated as part of the high-level long-term memory of T3. Given that the allophonic variant of the lexical Low tone (T3V) has a comparable acoustic distribution to that of the lexical rising tone (T2), T3V and T2 can be perceptually ambiguous in the allophonic T3T3 context (against T2T3). The near- neutralization of T3V and T2 has also been reported to enhance the perceptual similarity of T2 and T3 to native Chinese ears, compared to their greater perceptual distance to American English ears. T1 and T3, however, are rather distinct in all contexts. Importantly, our design only involved monosyllabic words, which does not introduce the T3 sandhi context. The interesting question that arises here is whether the auditory input of T3 produced in isolation would nevertheless invoke the representation of the sandhi T3V variant (if it is stored as long-term presentation in the mental lexicon), which would then render the T3 and T2 to be less distinguishable. A critical comparison of our design is between the MMNs elicited in the T2/T3 condition and that in the reversed T3/T2 condition, as compared to the MMNs due to the contrast between T1 and T3. In the T1/T3 condition, a conflict always occurs, since the mental representation activated by the standard T1 (lexical T1 with high level pitch contour) would be different from the acoustic low pitch contour information extracted from the deviant T3. A similar conflict exists in the T2/T3 condition, since the mental representation (lexical T2 with rising pitch contour) activated by the standard T2 would also differ from the acoustic low pitch contour extracted from the deviant T3. Different models of lexical representation, however, would make different predictions for the reversed T3/T2 and T3/T1 conditions. Within the FUL model, T3, given its variability in context, is expected to be underspecified in the mental lexicon. Therefore, in both T3/T1 and T3/T2 conditions, no conflict (related to long-term memory representation) is expected. Previous studies demonstrated that MMN has a higher magnitude if the mapping involves a conflicting rather than a non-conflicting situation. So, we expect to observe reduced MMNs when T3 serves as a standard, and correspondingly, asymmetrical MMNs in both T1/T3 vs. T3/T1 condition and T2/T3 vs. T3/T2 condition. With regard to the hemisphere dominance underlying tonal processing, if indeed T3 is underspecified (FUL), we expect more left-lateralized patterns of hemisphere processing in the T1/T3 and T2/T3 conditions, given their between- category tonal processing, but not necessarily so in the T3/T1 and T3/T2 conditions, given the underspecified representation of the standard T3. According to the MV account, multiple tonal variants can co-exist in the mental lexicon for T3; both its canonical form and its variant (with a rising pitch contour) are therefore expected to be activated by the standard T3. The latter is dissimilar to the pitch contour extracted from the deviant T1 and similar to that extracted from the deviant T2. Therefore, conflict is expected in the T3/T1 condition but not in the T3/T2 condition. Symmetrical MMNs are expected in T1/T3 vs. T3/T1 conditions while the T3/T2 condition is expected to show reduced MMNs compared to the T2/T3 condition. Under the assumption that multiple tonal variants are stored, we expect that the T1/T3, T3/T1, and T2/T3 conditions should involve clear between-category tonal contrasts and therefore left-hemisphere dominant processing. The T3/T2 condition, however, may recruit different, right lateralized neural basis of processing, given that the standard T3 may activate T3V, which is similar to the pitch contour extracted from the deviant T2; and consequently, T3V and T2 may be processed in a way similar to within-category difference. As for the CI account, its core essence is that the processing of speech variants is context-specific. Within this possibility, T3V is not stored in the lexical representation but computed on-line from the underlying T3 during the speech encoding process. Given that our study uses monosyllabic morphemes produced in isolation, it is then clear that there should be no licensing context for tonal variants. The long-term memory representation activated by the standard T3 is therefore expected to be different from both the deviant T1 and the deviant T2. Hence symmetrical MMNs are expected in the T1/T3 and T3/T1 conditions as well as in the T2/T3 and T3/T2 conditions. This third possibility (CI) predicts more left-lateralized patterns of hemisphere processing for all conditions given that T3 is represented as its canonical tonal form, and the standard and deviant stimuli should then be consistently perceived as different lexical tones across conditions and induce between-category functional processing. One additional issue, ancillary to our central research question but nevertheless important, can be addressed with our design. This concerns the effect of acoustic distances between different lexical tonal pairs on the neural processing of lexical tones. By comparing the MMNs elicited in T1/T3 vs. T2/T3 conditions or T3/T1 vs. T3/T2 conditions, we expect to replicate, with naturally-produced stimuli, the findings of which used synthesized stimuli, on the timing and magnitude of MMN effects introduced by different acoustic distances of lexical tonal pairs. # Method ## Ethics statement All participants provided written informed consent in accordance with the Declaration of Helsinki. The ethics committee of the Institute of Psychology at the Chinese Academy of Sciences approved this study, its participant-recruitment procedure and its methodology. ## Participants Twenty right-handed subjects, as measured by the Edinburgh handedness questionnaire (18–24 years old; 18 females), participated in the experiment. All of them were university students and native speakers of Mandarin dialects (from Beijing and its surrounding areas). None of them had any neurological impairment, neurological trauma, or used neuroleptics. ## Stimuli The auditory stimuli were a set of three Mandarin Chinese words that are distinguished by lexical tones only: ma^T1 ‘mother’ \[T1\], ma^T2 ‘linen’ \[T2\], and ma^T3 ‘horse’ \[T3\]. The words were produced by a female speaker of Standard Chinese who was born and grew up in Beijing. Tonal durations for the three naturally-elicited words ‘ma¹’, ‘ma²’, and ‘ma³’ are 556 ms, 552 ms, and 554 ms respectively. Their voice amplitudes are 72 dB, 72 dB, and 70 dB respectively. As shown in, T1 has a high Level *f0* contour (onset: 295 Hz; offset: 312 Hz). T2 has a high rising *f0* contour (onset: 234 Hz; offset: 281; turning point: 224 Hz around the midpoint of the tone-bearing syllable). T3 has a dipping *f0* contour (onset: 233 Hz; offset: 204 Hz; turning point: 172 Hz, slightly after the midpoint of the tone-bearing syllable). The experiment consisted of four oddball conditions (standard/deviant, 85% /15%). In the first two oddball conditions, T3 was always used as the deviant stimuli, with the standard stimuli being T1 in one condition (T1/T3, standard/deviant) and T2 in the other condition (T2/T3). For the other two oddball conditions, the standards and deviants were reversed. One included T3 as standard and T1 as deviant (T3/T1) and the other T3 as standard with T2 as deviant (T3/T2). In addition, there were three control conditions (i.e., with just one type of stimuli): T1 (100%), T2 (100%), and T3 (100%). ## Procedure After the electrodes were positioned, subjects were seated in an acoustically and electrically shielded room, facing a computer screen. They were instructed to watch a silent movie while listening to the auditory stimuli passively. To ensure that subjects focused their attention on the silent movie, they were informed that questions about this movie would be asked afterwards. The whole experiment, with a total of 7 blocks, lasted about 70 minutes. For the four oddball conditions, each block consisted of a standard (p = 0.85, 567 trials) and a deviant (p = 0.15, 100 trials). For the three control conditions, the deviants in the four oddball conditions were presented alone: T1, T2, and T3 (100%, 667 trials). The inter-trial interval was 400 ms. All oddball blocks preceded the control blocks. The order of the respective blocks was counterbalanced among participants. Within each oddball block, the order of stimuli presentation was pseudo-random, i.e., with at least two standard stimuli preceding each deviant. Furthermore, every block began with an additional 10 trials, which were all standard stimuli. ## EEG acquisition EEG was recorded (0.05–100 Hz, sampling rate 500 Hz) from 64 Ag/AgCl electrodes mounted in an elastic cap (Neuroscan Inc.), with an on-line reference to the nose. EEG and EOG data were amplified with AC amplifiers (Neuroscan). Vertical eye movements were monitored via a supra- to sub- orbital bipolar montage. A right to left canthal bipolar montage was used to monitor horizontal eye movements. All electrode impedance levels (EEG and EOG) were kept below 5 kΩ. ## ERP analysis Data of four participants were excluded from further ERP analyses due to excessive artifacts. The remaining data from a total of 16 participants were subsequently analyzed. The first ten trials of all blocks were not included in the analysis. The raw EEG data were first corrected for eye-blink artifacts using the ocular artifact reduction algorithm in the Neuroscan v. 4.3 software package. Then, the EEG data were filtered with a band-pass filter of 0.8–30 Hz. Subsequently, the filtered data were divided into epochs ranging from 100 ms before the onset of the words to 800 ms after the onset of the words. A time window of 100 ms preceding the onset of the words was used for baseline correction. Trials contaminated by eye movements, muscle artifacts, electrode drifting, amplifier saturation, or other artifacts were identified with semiautomatic artifact rejection (automatic criterion: signal amplitude exceeding ±75μV, followed by a manual check). Trials containing the above-mentioned artifacts were rejected (12.5% overall). Rejected trials were evenly distributed among conditions. To guarantee that the results of our study can be directly compared to previous studies, two methods were used to calculate MMN. One is to subtract the stimuli presented in the 100% control condition from the same stimuli that were presented as deviant in the oddball condition (MMN-a) (following e.g.,). The other is to subtract the standard stimuli in one oddball condition from the same stimuli that were presented as deviant in the reversed oddball condition (e.g., the standard T3 in T3/T1 was subtracted from the deviant T3 in T1/T3) (MMN-b; also known as identity MMN) (following e.g.,). To match the number of deviant trials (100 trials) in the oddball condition, for MMN-a, 100 trials (which were at the same position as the deviants in the oddball sequences) in the 100% control condition were selected. For MMN-b, the standards immediately preceding and following the deviants in the oddball condition were deleted, and then, 100 trials were randomly selected from the remaining standard trials. Both MMN-a and MMN-b can therefore effectively control for acoustical differences between stimuli, as the deviant was subtracted by the same stimulus. It may be necessary to point out that short-term habituation might influence MMN amplitude, since the deviant stimuli (e.g., T3 in T1/T3) was only presented 100 times in one block, but the same stimuli coming from the reversed oddball block (e.g., T3 in T3/T1) or the 100% control block was presented 567 or 667 times. It is important to note, however, that in the present study, we are not interested in the absolute MMN amplitude in a specific oddball condition, but interested in the relative MMN amplitude. That is, the present study hinged upon the similarity or differences between the MMNs elicited in the different oddball conditions, as described in the last part of the introduction section. The short-term habituation effect, if it exists, would be the same in the different oddball conditions, ruling out its potentially confounding effect. As shown in, the deviant stimuli in the four oddball conditions all elicited larger negative deflections than the same stimuli presented in the 100% control condition or in their corresponding reversed oddball conditions. The MMNs (for both MMN-a and MMN-b) in the four conditions (T1/T3, T3/T1, T2/T3, and T3/T2) peaked after the stimulus onset around 192 ms, 212 ms, 278 ms, and 300 ms respectively (see Figs). For statistical analysis, we employed the cluster-based permutation test in the Fieldtrip (<http://fieldtrip.fcdonders.nl>) soft-ware package to examine whether MMNs were elicited by the oddball conditions and whether MMN amplitudes were significantly different across different conditions. This non-parametric statistical procedure optimally handles the multiple-comparisons problem. Frontal-central electrodes were included in this test, since auditory MMN usually has a frontal-central distribution. Two kinds of permutation test were performed within 0–500 ms post-word onset in steps of 2 ms: over all of the 43 frontal-central electrodes (permutation test-1); over the right (19 electrodes) or left (19 electrodes) electrodes respectively (permutation test-2). For every data point (‘electrode by time’) of two conditions, a simple dependent-samples *t* test was performed. All adjacent data points exceeding a preset significance level (0.05%) were grouped into clusters. Cluster-level statistics were calculated by taking the sum of the *t*-values within every cluster. The significance probability of the clusters was calculated by means of the so- called Monte Carlo method with 1000 random draws. Second, based on the results of the cluster-based permutation test (which suggest the window latencies that demonstrated the strongest MMN effect) and visual inspection of the ERP waveforms, traditional ANOVAs (analyses of variance) were conducted within the specific time windows on a selection of two midline electrodes (Fz, Cz) and four lateral electrodes (F3/F4; C3/C4). First, to estimate whether MMNs were elicited by the oddball conditions, ANOVA-1 was conducted with Condition (T1/T3, T3/T1, T2/T3, and T3/T2), Stimulus type (deviant, control), Laterality (left, midline, right), and Anteriority (frontal F3/Fz/F4, central C3/Cz/C4) as independent factors. Second, to compare the amplitude or peak latency of MMNs elicited by the four oddball conditions, ANOVA-2 was performed based on four different waveforms (T1/T3, T3/T1, T2/T3, and T3/T2), with Condition, Laterality, and Anteriority as independent factors. When the degree of freedom in the numerator was larger than one, the Greenhouse- Geisser correction was applied. Whenever applicable, Bonforroni correction was used to counteract the problem of multiple comparisons. # Results ## The existence of MMN effects The cluster-based permutation test over all frontal-central electrodes (permutation test-1) revealed that, for MMN-a, the MMN effect reached significance within around 100–280 ms, 140–280 ms, 240–340 ms, and 240–360 ms for T1/T3 (“deviant T3 from T1/T3” vs. “T3 from 100%”, *p* \<.0001), T3/T1 (“deviant T1 from T3/T1” vs. “T1 from 100%”, *p* \<.01), T2/T3 (“deviant T3 from T2/T3” vs. “T3 from 100%”, *p* \<.005), and T3/T2 (“deviant T2 from T3/T2” vs. “T2 from 100%”, *p* \<.05) conditions respectively. The MMNs in the T1/T3, T3/T1, and T2/T3 conditions have a broad distribution, while the MMN in the T3/T2 condition has a right-hemisphere distribution. Permutation test-2, which examines the MMNs over the left and right electrodes separately, provided further evidence that, for the T1/T3, T3/T1, and T2/T3 conditions, the MMN effects reached significance over both hemispheres; in contrast, for the T3/T2 condition, the MMN effect reached significance only over the right electrodes. For the MMN-b, permutation test-1 revealed that, the MMN effect reached significance around 140–260 ms, 120–260 ms, and 220–320 ms for T1/T3 (“deviant T3 from T1/T3” vs. “standard T3 from T3/T1”, *p* \<.0001), T3/T1 (“deviant T1 from T3/T1” vs. “standard T1 from T1/T3”, *p* \<.0001), and T2/T3 (“deviant T3 from T2/T3” vs. “standard T3 from T3/T2”, *p* \<.005) conditions respectively. However, the T3/T2 condition (“T2 from T3/T2” vs. “T2 from T2/T3”) did not elicit significant MMN effects in permutation test-1 or in permutation test-2. Furthermore, permutation test-2 demonstrated that, for the T1/T3, T3/T1, and T2/T3 conditions, the MMN effects reached significance over both left and right hemispheres. Based on the results of the permutation test (for window latencies that demonstrated the strongest MMN effect) and visual inspection of the ERP waveforms, ANOVA was conducted within the window latencies of 172–212 ms, 192–232 ms, 258–298 ms, and 280–320 ms for the MMNs elicited in the T1/T3, T3/T1, T2/T3, and T3/T2 conditions respectively. For MMN-a, the ANOVA-1 analysis revealed a significant main effect of Stimulus type (*F*_{(1, 15)} = 44.35, *p* \<.0001, *d* =.75), indicating that the deviant stimuli in the oddball condition evoked a larger negative deflection (MMN) than the same stimuli in the control condition (magnitude difference: -1.85 μV). The effect of Stimulus type was qualified by a two-way Condition × Stimulus type interaction (*F*_{(3, 45)} = 6.96, *p* \<.005, *d* =.32) as well as a three-way Condition × Stimulus type × Hemisphere interaction (*F*_{(6, 90)} = 2.88, *p* \<.05, *d* =.16). Subsequent simple analyses showed that, for T1/T3, T3/T1, and T2/T3 conditions, the MMN effects reached significance over left, middle, and right hemispheres (T1/T3: *F*_{left(1, 15)} = 30.99, *p* \<.0001, *d* =.68, *F*_{middle(1, 15)} = 35.66, *p* \<.0001, *d* =.70, and *F*_{right(1, 15)} = 34.22, *p* \<.0001, *d* =.69; T3/T1: *F*_{left(1, 15)} = 24.37, *p* \<.0001, *d* =.62, *F*_{middle(1, 15)} = 19.05, *p* \<.001, *d* =.63 and *F*_{right(1, 15)} = 11.40, *p* \<.005, *d* =.43; T2/T3: *F*_{left(1, 15)} = 29.28, *p* \<.0001, *d* =.66, *F*_{middle(1, 15)} = 34.97, *p* \<.0001, *d* =.70, and *F*_{right(1, 15)} = 29.18, *p* \<.0001, *d* =.66). However, for the T3/T2 condition, the MMN effect reached significance only over the right electrodes (*F* _{(1, 15)} = 14.08, *p* \<.005, *d* =.48). For MMN-b, the ANOVA-1 also revealed a significant main effect of Stimulus type (*F*_{(1, 15)} = 42.19, *p* \<.0001, *d* =.74), suggesting that the deviant stimuli evoked a larger negative deflection (MMN) than the same stimuli presented as standard in the reversed oddball condition (magnitude difference: -1.62 μV). The two-way Condition × Stimulus type interaction (*F*_{(3, 45)} = 3.2, *p* \<.05, *d* =.18), three-way Condition × Stimulus type × Anteriority interaction (*F*_{(3, 45)} = 5.15, *p* \<.01, *d* =.26), and the four-way Condition × Stimulus type × Anteriority × Hemisphere interaction (*F*_{(6, 90)} = 2.60, *p* =.055, *d* =.15) all reached significance or marginal significance. Further simple analyses showed that, for the T1/T3, T3/T1, and T2/T3 conditions, the MMN effects reached significance over all of the six brain areas (T1/T3: *F*_{frontal-left(1, 15)} = 23.13, *p* \<.0001, *d* =.61, *F*_{frontal-middle(1, 15)} = 24.82, *p* \<.0001, *d* =.62, *F*_{frontal-right(1, 15)} = 28.63, *p* \<.0001, *d* =.66, *F*_{central-left(1, 15)} = 14.33, *p* \<.005, *d* =.49, *F*_{central-middle(1, 15)} = 14.81, *p* \<.005, *d* =.52, *F*_{central-right(1, 15)} = 16.76, *p* \<.001, *d* =.53) (T3/T1: *F*_{frontal-left(1, 15)} = 31.60, *p* \<.0001, *d* =.68, *F*_{frontal-middle(1, 15)} = 40.89, *p* \<.0001, *d* =.73, *F*_{frontal-right(1, 15)} = 28.72, *p* \<.0001, *d* =.66, *F*_{central-left(1, 15)} = 18.46, *p* \<.001, *d* =.55, *F*_{central-middle(1, 15)} = 25.98, *p* \<.0001, *d* =.63, *F*_{central-right(1, 15)} = 15.86, *p* \<.001, *d* =.51) (T2/T3: *F*_{frontal-left(1, 15)} = 46.54, *p* \<.0001, *d* =.76, *F*_{frontal-middle(1, 15)} = 31.86, *p* \<.0001, *d* =.68, *F*_{frontal-right(1, 15)} = 27.53, *p* \<.0001, *d* =.65, *F*_{central-left(1, 15)} = 10.50, *p* \<.005, *d* =.41, *F*_{central-middle(1, 15)} = 13.84, *p* \<.005, *d* =.48, *F*_{central-right(1, 15)} = 8.93, *p* \<.01, *d* =.37); however, for the T3/T2 condition, the MMN effect again reached significance only over the midline frontal (*F*_{(1, 15)} = 5.36, *p* \<.05, *d* =.25) and right frontal electrodes (*F* _{(1, 15)} = 5.60, *p* \<.05, *d* =.25). In sum, both MMN-a and MMN-b showed that the deviants in the four oddball conditions all elicited a MMN effect, with the MMNs in the T1/T3, T3/T1, and T2/T3 conditions present over both left and right hemispheres while the MMN in the T3/T2 condition present only over the midline and right electrodes. Although for MMN-b, T3/T2 condition elicited no significant MMN in the cluster-based permutation test, this condition did elicit significant MMNs over the midline and right electrodes in the ANOVA analysis. ## MMN mean amplitude Similar permutation tests, but with MMN amplitude as the dependent factor, were performed to examine whether MMNs were different between every two oddball conditions. The Permutation test-1 revealed that MMN-a in the reversed T2/T3 condition was significantly larger than that in the T3/T2 condition around 240–300 ms (*p* \<.05), which mainly occurred over the left hemisphere. However, the MMN-a in the T1/T3 condition was not different from that in the T3/T1 condition (*p* =.20). Furthermore, permutation test-2 showed that, for both MMN-a and MMN-b, the T2/T3 condition elicited larger MMNs than the reversed T3/T2 condition around 240–300 ms over the left hemisphere (both *ps* \<.05), but not over the right hemisphere (*p* =.06 for MMN-a, *p* =.09 for MMN-b); the difference between MMNs in the T1/T3 and T3/T1 conditions did not reached significance over the left hemisphere or the right hemisphere (*ps* \>.20). For ANOVA-2, the MMN mean amplitude was computed based on the four difference waveforms (T1/T3, T3/T1, T2/T3, and T3/T2) in their corresponding latency windows (172–212 ms, 192–232 ms, 258–298 ms, and 280–320 ms respectively). For MMN-a, ANOVA-2 resulted in a significant main effect of Condition (*F*_{(3, 45)} = 6.96, *p* \<.005, *d* =.32) and a significant two-way Condition × Hemisphere interaction (*F*_{(6, 90)} = 2.88, *p* \<.05, *d* =.16). The simple main effect of Condition reached significance over the left hemisphere (*F*_{(3, 45)} = 8.08, *p* \<.001, *d* =.35), midline electrodes (*F*_{(3, 45)} = 7.47, *p* \<.005, *d* =.33), and right hemisphere (*F*_{(3, 48)} = 4.39, *p* \<.05, *d* =.23). Follow-up pairwise comparisons revealed that the MMN amplitude in the T1/T3 condition was not different from that in the reversed T3/T1 condition (*p* = 0.30; *p* = 0.08; *p* = 0.20 for left, midline, and right areas respectively); in contrast, the MMN amplitude in T2/T3 condition was significantly larger than that in the reversed T3/T2 condition over the left and midline electrodes (left: *p* \<.05, magnitude difference: -1.75 μV; Middle: *p* \<.05, magnitude difference: -1.76 μV). In addition, the other pairwise comparisons did not reach significance. For MMN-b, ANOVA-2 resulted in a significant main effect of Condition (*F*_{(3, 45)} = 3.20, *p* \<.05, *d* =.18) and a significant two-way Condition × Anteriority interaction (*F*_{(3, 45)} = 4.82, *p* \<.01, *d* =.23) and a significant three-way Condition × Anteriority × Hemisphere interaction (*F*_{(6, 90)} = 2.60, *p* =.055, *d* =.15). The follow-up pairwise comparisons over the six brain areas revealed that the MMN amplitude in the T2/T3 condition was larger than that in the reversed T3/T2 condition only over the left frontal electrode (*p* \<.05; magnitude difference: -1.71 μV). The other comparisons didn’t reach significance. Thus, for both MMN-a and MMN-b, the MMN amplitude in the T1/T3 condition was not different from that in the reversed T3/T1 condition. However, the MMN amplitude in T2/T3 condition was larger than that in the reversed T3/T2 condition over the left hemisphere. In addition, there was no significant difference between the MMNs in the T1–T3 pair (larger acoustic distance) and T2–T3 pair (smaller acoustic distance) conditions. ## MMN peak latency Finally, we examined the timing characteristics of MMNs evoked in the four oddball conditions. Peak latency of the negative deflection in the four MMNs (T1/T3, T3/T1, T2/T3, and T3/T2,) was measured by determining individual peak latencies in the following latency windows: 120–260 ms, 120–260 ms, 220–350 ms, and 220–350 ms after the onset of the deviant stimulus respectively. This time- window was set after visual inspection of the window latencies for the corresponding MMNs in the present result. For MMN-a, ANOVA-2 with peak latency as dependent factor resulted in a significant main effect of Condition (*F*_{(3, 45)} = 55.04, *p* \<.0001, *d* =.79). The subsequent pairwise comparisons revealed that neither the difference between the T1/T3 and T3/T1 conditions (p = 1.000) nor the difference between the T2/T3 and T3/T2 conditions (p =.542) reached significance. However, MMNs in the T1/T3 condition (195 ms) peaked earlier than that in both the T2/T3 condition (273 ms) (p \<.0001) and the T3/T2 condition (290 ms) (p \<.0001); MMNs in the T3/T1 contrast condition (203 ms) also peaked earlier than that in both the T3/T2 (p \<.0001) and the T2/T3 conditions (p \<.0001). For MMN-b, ANOVA-2 with peak latency as dependent factor also resulted in a significant main effect of Condition (*F*_{(3, 45)} = 76.997, *p* \<.0001, *d* =.84). The subsequent pairwise comparisons also showed that neither the difference between the T1/T3 and the reversed T3/T1 (p = 1.000) conditions nor the difference between the T2/T3 and the reversed T3/T2 (p =.864) conditions reached significance. However, the MMN effects in the T1/T3 condition (198 ms) peaked earlier than that in the T2/T3 (278 ms) condition (p \<.0001) and that in the T3/T2 (291 ms) condition (p \<.0001); the MMN effects in the T3/T1 condition (204 ms) also peaked earlier than that in the T3/T2 (p \<.0001) and the T2/T3 conditions (p \<.0001). To summarize, for MMN peak latency, neither the difference between T1/T3 and the reversed T3/T1 nor the difference between T2/T3 and the reversed T3/T2 reached significance. However, MMNs elicited in conditions with big acoustic differences (i.e. T1/T3 and T3/T1) peaked earlier than that with relatively smaller acoustic differences (i.e. T2/T3 and T3/T2). # Discussion With a passive oddball MMN paradigm, the present study examined the representation and neural processing of lexical tones in Standard Chinese. A reversed design was adopted to control the standard-deviant acoustic contrast so as to tap specifically into the possible effects of allotonic variation on tonal representation and neural processing in general. Significant MMN effects were found for the T1/T3, T3/T1, and T2/T3 conditions over both the left and right hemispheres. For the T3/T2 condition, however, the MMN responses were much more reduced and had a right hemisphere lateralization. More importantly, the comparison of these MMN effects revealed symmetrical MMN effects for the T1–T3 pair but asymmetrical MMNs for the T2–T3 pair. Before we consider these results in more detail, it is important to note that different acoustic distances between lexical tonal pairs (i.e. T1–T3 vs. T2–T3) did introduce different MMN effects. As introduced earlier, in Standard Chinese, the pitch contour of T3 is acoustically more similar to T2 than to T1. Our results showed that the MMNs evoked by the tonal pair T1–T3 peaked earlier than the MMNs evoked by the tonal pair T2–T3 where there is smaller acoustic difference and later identification point. The present study thus replicated the results reported in which showed earlier-peaked MMNs in the T1/T3 condition compared to the T2/T3 condition. This also echoes the general observation that MMN responses originate from the magnitude of acoustic contrasts between the standard and deviant stimuli. Note that Chandrasekaran et al. also found a significantly larger amplitude in the T1/T3 condition than that in the T2/T3 condition. This, however, was not replicated in our study. Note that their study used synthesized pitch contours while our stimuli were produced naturally by a native speaker of Standard Chinese. The lack of amplitude difference in our study is likely due to the richer acoustic cues in the naturally produced lexical tones which can probably cue more effectively the lexical tonal contrasts, and consequently led to greater MMN effects in our T2/T3 condition. This enhanced MMN effects in the T2/T3 condition in our study, however, stands in stark contrast to the weakened and non-significant MMN effects in the T3/T2 condition. What cries for explanation is thus the two intriguing patterns of asymmetry that have emerged in our study. First, the T2–T3 pair showed different MMN effects in the T3/T2 vs. T2/T3 condition, in contrast to the comparable MMN effects for the tonal pair T1–T3 (which have rather distinct pitch contours in all tonal contexts). Second, the MMN effects in the T3/T2 condition were mainly observed in the right hemisphere, in contrast to the broad (both right and left hemispheres) MMN effects observed in the T2/T3 condition, as well as in the T1/T3 and T3/T1 conditions. The present results thus demonstrated that with the same amount of acoustic distance between the standard and deviant auditory stimuli, T1/T3 elicited symmetrical MMNs as the reversed T3/T1 condition. The T2–T3 pair, however, despite that the holistic acoustic distance in the T2/T3 condition remains the same as those in the T3/T2 condition, showed significantly different MMN effects not only in terms of MMN amplitude, but also in terms of their scalp distribution. These differences clearly cannot be explained by mere acoustic features, especially given our standard-deviant reverse design. In the following, we will argue that such asymmetries shed light on the effect of allophonic tonal variants on the representation (Section 4.1) and neural processing (Section 4.2) of lexical tones in general. Implications of these results on proper modeling of pronunciation variation in speech processing will be taken up and discussed in Section 4.3. ## The effect of allophonic variation on lexical tone representation The central goal of the present study was to investigate the effect of allophonic lexical tone variants on the representation of lexical tones in general. To recapitulate, in Standard Chinese, T3 has an allophonic tone sandhi variant T3V, which surfaces only in specific tonal context (i.e. in a T3T3 sequence). T3V has a rising pitch contour, which is acoustically similar to (and often perceptually undistinguishable from) the rising pitch contour of the lexical T2. T3 and T3V, however, are always quite distinct from the lexical T1, which has a level pitch contour. Note that in our experiment, listeners were only exposed to T3 produced in isolation, which are categorically different from both T1 and T2 with near ceiling-level accuracy in tonal identification. We were interested in whether, without the specific tonal sandhi context (i.e. T3 produced in isolation), T3V can be activated, which should constitute as convincing evidence for the effect of tone sandhi variant (T3V) on the memory representation of lexical T3 in general. Results of the oddball paradigm showed significant and widely distributed MMN effects in the T1/T3, T3/T1, and T2/T3 conditions across the left, middle, and right electrodes, as one would have expected for lexical tone processing. In the T3/T2 condition, however, we observed much more reduced and restricted MMN effects, which was only significant in the right hemisphere. Further comparisons showed that for the tonal pair T1–T3, MMN in the T1/T3 condition did not show any significant difference from that in the reversed T3/T1 condition both in terms of MMN amplitude and peak alignment; however, for the tonal pair T2–T3, MMN was significantly larger when T2 was the standard and T3 the deviant (T2/T3), as compared to the reversed T3/T2 condition. One may be tempted to attribute the asymmetrical MMNs to the different directions of pitch change between standard and deviant stimuli in the four oddball blocks. Gomes et al., however, showed that low-to-high and high-to-low pitch changes of pure tones do not introduce significant differences in the magnitude and scalp distribution of MMNs, suggesting that pitch direction change should not have brought the observed asymmetry. Furthermore, we know that T1 has a high level pitch contour, T2 a high rising contour, and T3 a low rising contour. As is clear from , in the T2–T3 pair, their difference only lies in the magnitude of pitch falling from high to low. In the T1–T3 pair, however, the pitch falls from high to low in T1/T3 and rises from mid to high in T3/T1. If indeed pitch direction change could have contributed to different MMN effects, we should have observed greater asymmetry of MMNs in the T1–T3 pair, and not in the T2–T3 pair. Taken together, the pitch-change direction account seems untenable for our observed asymmetry. A related alternative is that the asymmetrical MMNs are due to other acoustic factors such as spectral and temporal differences, or frequency differences of the stimuli. We think this is also highly unlikely. First, in the present study, T1 and T2 have the same intensity (72 dB). Asymmetric MMNs were observed in the T3–T2 pair but not in the T3–T1 pair. Furthermore, the durational differences in the two tonal pairs were both about 2 ms (T1: 556 ms; T2: 552 ms; and T3: 554 ms), which makes it doubtful to have induced the observed MMN asymmetry. Lexical/tonal frequency distribution patterns also raise further doubts to the third possibility. We know that ma^T3 (0.02663) has a higher frequency than both ma^T1 (0.00825) and ma^T2 (0.00438). Furthermore, T3 is known as the least frequent tone among the four lexical tones. According to *Junda Chinese Text Computing*, 16% of the 9933 unique characters in modern Chinese (listed on the site) have T3 but the percentages of T1 and T2 characters are higher (25% and 26% respectively). We know that words with higher frequency occurrence in a language show a more pronounced and earlier lexical MMN response than its low-frequency counterpart. Regardless of whether and which of these two types of frequency may have differential MMN effects, the fact that asymmetrical MMNs were observed only in the T3–T2 pair renders versions of frequency account implausible. A third possibility is to relate this asymmetry to the more general mechanisms of asymmetrical perception reported in an auditory domain, such as infant vowel perception, in a visual domain, such as color perception, as well as in other domains, such as abstract concept categorization, which all show that perceptual asymmetries exist given different orders of presentation for the same stimulus pair. Such perceptual asymmetries have been attributed to the unequal perceptual saliency of the stimuli resulting from, for example, their different frequencies, distributional patterns, or familiarity, with the relatively more frequent, familiar, or more peripheral ones (e.g. in the case of vowel) typically acting as the perceptual magnet or default and therefore inducing a more discriminable ‘pop out’ effect. Following the line, one may argue then that T3 is the default lexical tone in the language, which gives rise to the perceptual saliency when T2 acts as the standard in the oddball paradigm, introducing stronger MMN effects in the T2/T3 condition than that in the T3/T2 condition. The question left unanswered, however, is the symmetrical MMN effects in the T3/T1 vs. T1/T3 conditions. We propose that such an asymmetry can be better explained by taking into consideration the possible activation of T3V even though listeners in our experiment were exposed only to the canonical T3 production, without the presence of the T3 allophonic context. Specifically, when T3 serves as the standard, its long-term memory representation may be activated, which includes not only the canonical low tone (T3) representation but also its rising sandhi variant (T3V) representation, which has a similar rising pitch contour as T2. Thus, when listeners encountered the deviant T2, there was little conflict between the activated phonological representations (i.e. T3 and T3V) and the acoustic parameters extracted from the deviant T2. This then accounts for the weakened MMN effects in the T3/T2 condition. When T3 is the deviant and T2 standard (T2/T3), however, the pitch contour extracted from the deviant T3 (with a low pitch contour) mismatches the perceptual long-term representation activated by the standard T2 (i.e. a lexical rising tone). This contrast then gave rise to the observed asymmetrical MMNs between T2 and T3. This account is also compatible with results from behavioral production studies on the phonological encoding of lexical tone and tonal variants in Standard Chinese with the implicit priming paradigm and the picture-word interference paradigm, both of which suggest the co-storage of T3 and T3V in the mental lexicon. Further evidence on representation of multiple tonal variants comes from which investigated the lexical access of a type of tonal variability in Jinan Mandarin − lexically non-contrastive in specific words but potentially contrastive in other words. Results of their auditory lexical decision experiment in a medium-term auditory priming paradigm showed that lexically non- contrastive word-specific tonal variability induced facilitation, in contrast to the inhibition effect found for lexically-contrastive primes, which suggests storage of word-specific multiple tonal variants. As we will show below, such a view of the long-term memory representation of T3 together with its sandhi variant T3V also helps to understand the neural basis of lexical tonal and tonal variant processing we have observed in the study. ## The effect of allophonic variation on the neural processing of lexical tones Moving onto the effect of allophonic variation on the neural processing of lexical tones in general, it might be helpful to recapitulate that the existing literature suggests that the left hemisphere is more sensitive to between- category functional processing of lexical tones while the right hemisphere is more sensitive to low-level acoustic processing of within-category tonal differences. In the present study, the MMNs elicited in the four oddball conditions suggest asymmetrical patterns of hemisphere lateralization. Specifically, the deviant in the T1/T3, T2/T3, and T3/T1 conditions all elicited bilateral MMNs. Moreover, relative to the reversed T3/T2 condition, the MMN in the T2/T3 condition was enhanced only over the left hemisphere. In other words, the MMNs elicited in the T1/T3, T2/T3, and T3/T1 conditions showed the same pattern of left-hemisphere dominant lateralization with an overall more global MMN distribution (extending over to the midline and right hemisphere). This suggests that in these three conditions, the standard and deviant stimuli were perceived as categorically different lexical tones and correspondingly induced between-category functional processing of the lexical tones. In the T3/T2 condition, however, the MMNs were more lateralized to the right hemisphere with no significant neural signatures of processing at all in the left-hemisphere. The results thus point to the possibility that both the lexical low tone T3 and its rising variant T3V were activated by the standard T3, with the latter having a similar acoustic distribution as the deviant T2. T3 as the standard, with T2 as the deviant, is then likely to be processed by the native listeners as having only acoustic difference, similar to within-category tonal processing. Consequently, the tonal processing in the T3/T2 condition showed a rightward scalp distribution, different from the other three oddball conditions. It is worth noting that in the present study, the majority of the participants were females. One may question whether the observed patterns of lateralization can be due to just gender difference in neural processing. Previous studies have showed differential degree of hemisphere specialization in the MMNs. Ikezawa and colleagues found that although for low-level pure-tone MMNs, there is no significant gender differences in hemispheric lateralization, males (relative to females) exhibited greater left-hemisphere lateralization with relatively high- level phonetic MMNs. In line with Ikezawa and colleagues’ results, a recent study on white matter microstructure also found that males have increased language left-lateralization, as indicated by higher FA (fractional anisotropy) in the left superior longitudinal fasciculus, and females have more efficient inter-hemispheric transfer, as indicated by higher FA in corpus callosum. These findings predict that if more male listeners participated in the present study, the MMNs elicited in the three conditions T1/T3, T3/T1, T2/T3 would show increased left-hemisphere lateralization as compared to the present results, since the three conditions involved relatively high-level functional or linguistic difference; the MMN elicited in the T3/T2 condition would show a similar pattern of results as the present ones, since this condition mainly involve low-level acoustic difference. Therefore, it is less likely that the hemispheric lateralization observed in the present study can be reduced to gender-specific differences. However, since the present study was not designed specially to examine the gender difference and did not directly manipulate the high-level lexical-semantic variation of the lexical tone, the gender or acoustic/functional effects on the hemisphere lateralization of tonal variation processing needs to be further studied in the future. To summarize, although ERP responses recorded on the scalp are not always induced by a definitely stronger left/right lateralization in the cortex, the more rightward scalp distribution of MMNs in the T3/T2 condition, as compared to the bilateral scalp distribution in the other three oddball conditions, is consistent with the neural basis of cross-category vs. within-category lexical tone processing. This, in turn, provides further evidence for the hemispheric preference and their functional roles as related to high-level cognitive and lower-level sensory properties of lexical tone processing, along the line of the model put forward by Gandour and his colleagues. ## The three pronunciation variation models Results of this study provide a new window for evaluating models of the representation and processing of pronunciation variation at the supra-segmental level. Following the FUL model, we would assume that lexical items are associated with one highly abstract phonological representation in the mental lexicon, where no detailed or predictable variation is stored in the mental representation. Following the MV account, we expected that multiple tonal variants can be jointly stored in the mental lexicon. The CI account, however, assumes a single abstract lexical representation and the processing of tonal variants is context-specific so that tonal variants should be only inferred given its proper context. They make different predictions for the MMN effects in the four oddball conditions as laid out in. Our results showed an asymmetrical MMN effect in the processing of the T2–T3 pair but no difference in the T1–T3 pair. This counters the prediction of both the FUL account (which predicts asymmetrical MMNs for both pairs when T3 serves as the standard) and the CI account (which expects no asymmetry in either pair). On the other hand, our results fit well with the MV account, granted that the standard (i.e. frequently presented) T3 indeed activates the memory representations of both its canonical low tone and its rising sandhi variant (which is acoustically similar to T2), thereby inducing asymmetrical MMNs in the T2–T3 pair but not in the T1–T3 pair. Earlier studies such as have lent convincing support to the FUL account with evidence from underspecified featural representation. The difference between the current study on tonal variants and the earlier studies might be related to the different types of stimuli used. The earlier studies focused on featural specification in relation to segmental variation, such as the coronal place of articulation. The present study, however, investigated allophonic variation at the super-segmental level, namely, lexical tone variants. In Standard Chinese, there are only four lexical tones. Tones therefore play a very important role in distinguishing lexical meaning. Our results, in which the repeated presentation of a lexical tone activated its multiple allophonic variants, are hard to be reconciled by simply assuming T3 as the default tonal category in the language which leads to asymmetry in processing. Instead, the present study squares much better with the possibility that multiple tonal variants of a lexical item are jointly stored in the mental lexicon, leading to processing asymmetry that arises from acoustic similarities of the co-activated representations. Our results also suggest that the activation of multiple tonal representations can be free of context. In Standard Chinese, allophonic tonal variation mainly depends on lexical tonal context. For example, only when immediately followed by another T3 in the connected speech, is the Low dipping tone (T3) realized with a rising pitch contour. In the present study, all stimuli consisted of only monosyllabic morphemes produced in isolation and were presented with an inter- trial interval of 400 ms, which, to the native ears, indisputably gives rise to the perception of two separate *ma*^*T3* morphemes produced in isolation. It is also important to note that the integration of two occurrences of *ma*^*T3* would result in a nonsense sequence (i.e. ma^T3ma^T3 which literally means ‘horse horse’). While further experimentation may be beneficial to exclude the possible phantom T3 sandhi context in a sequence of standard T3s, we feel it is on the safe side to conclude that no appropriate phonological context is licensed for tonal alternation to argue for the context-dependent inference (CI) account, where phonological variation is inferred via the presence of variation-licensing context. This view, however, does not deny the possible role that licensing context and active inference processes can play in the perceptual treatment of pronunciation variation in speech. # Conclusions In conclusion, the present study used the reversed MMN paradigm to investigate the effect of lexical tonal variants on the representation and neural processing of tones in general. The current results suggest the activation of the memory representations of both canonical lexical tones and their allophonic tonal variants, even when free of allophonic tonal context. Our results provide supporting evidence for the storage of tonal variants in the mental lexicon, compatible with the multiple variant representation account. The activation of the allophonic tonal variants can lead to dominant right-hemisphere processing of lexical tones, which are otherwise categorically processed via recruiting both left and right hemispheres. Although tonal variability is ubiquitous in speech, very little is known about their effects on the representation and processing lexical tones in general. Few neurophysiological studies examining tonal variants exist (but seeon the neural encoding of tonal variants in speech production). Furthermore, none uses the oddball paradigm with MMN effects, which is known as an effective tool to tap into the early neural processing of linguistic representations. Future studies are necessary to further understand the internal structure of lexical representations in particular with regard to allophonic tonal variants in a diverse range of languages with different patterns of tonal variation. We acknowledge the financial support from the National Natural Science Foundation of China (31271091) to XQL, the European Research Council (Starting Grant-206198) to YC, and the Netherlands Royal Academy of Sciences (KNAW China Exchange Program 13CDP012) to YC & XQL. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Thanks also to Jodie Mann for proofreading the manuscript. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: YC XL. Performed the experiments: XL. Analyzed the data: XL. Contributed reagents/materials/analysis tools: YC XL. Wrote the paper: YC XL.

# Introduction Flower and inflorescence characteristics are thought to influence pollinator behavior. Floral orientation, the angle between a flower’s main axis and the horizontal, is thought to affect pollinator attraction –, foraging behavior, and pollen transfer. However, most previous studies are on plant species that produce single flowers, relatively little is understood about the effect of floral orientation on pollination in more complex inflorescences. In vertical bee-pollinated inflorescences packaged in racemes or spikes, flowers typically have a horizontal orientation This trait is explained by hypotheses regarding pollinator behaviors and pollination precision. Firstly, pollinator recognition: The bumble bees were reported to have a preference for horizontal flowers. Sprengel (1793) proposed that bees are mainly attracted to horizontal- facing flowers from the front, and forcing this course of approach may facilitate the recognition of flower patterns. The inflorescences with horizontal flowers are thought to receive more visits due to enhanced recognition. Secondly, consistency of pollinator movement on inflorescences: bees have a tendency to move upwards in vertical inflorescences,, and a horizontal orientation is thought to ensure consistency of pollinator movements on inflorescences. Finally, pollination precision: A horizontal orientation may function to control the access of landing sites and ensure the contact between floral reproductive organ and pollinator body. Deviation from horizontal may result in decreased pollen transfer. These hypotheses predicted the significance of floral orientation on plant reproductive success from different aspects. However, the effects of horizontal orientation on pollinator recognition and pollination precision have been investigated for relatively few plant taxa pollinated by certain pollinator groups, e.g. syrphid fly, hummingbird. In addition, relatively less is known about the effect of floral orientation on pollinator attraction of inflorescences. Moreover, experimental investigations are needed to directly test the effect of floral orientation on the consistency of pollinator movement on inflorescences. In this study, we test the hypotheses by using a species of *Corydalis*, a genus characterized by zygomorphic, bee-pollinated, horizontal-facing flowers arranged in racemes. By artificially manipulating the orientation of individual flowers to prepare inflorescences with different floral orientations, we attempt to detect whether change of floral orientation affects (1) the capacity of an inflorescence to attract pollinators, (2) the number of successive probes and consistency of pollinator movement on inflorescences, (3) efficiency of pollen removal, and (4) seed production. Since pollinator attraction is associated with pollen transfer and mate diversity, the movement of pollinators on vertical inflorescences would affect geitonogamy and pollen export. We predict that a deviation from the horizontal (natural) orientation will result in decreases in both male and female fitness, due to declines in pollinator attraction and consistency of pollinator movement. # Materials and Methods ## Ethics Statement This study was conducted in accordance with all People’s Republic of China laws. No specific permits were required for the described field studies. No specific permissions were required for access to the locations described in this study. The location is not privately owned and neither is it protected in any way. This study species is also not protected by any law. ## Study Species and Site *Corydalis sheareri* S. Moore f. *sheareri* is one of the most common spring ephemerals in China. Each plant generally produces 2 to 10 inflorescences, with flowers arranged in racemes. The flowers are protandrous, with the lower (older) flowers in the female stage and upper (younger) flowers in the male stage. There is a considerable overlap in anthesis among the flowers of a single inflorescence, and the display size mostly ranges from 1 to 10. The flowering period of an inflorescence ranges from 9 to 23 days. Each flower is attached at the side of a branch by a pedicel, and the flower display forms a one-sided raceme. The pollen production per flower is 20273.72±3712.72 and ovule number is 22.55±2.14 (N \>30, mean ± SD). The pollen-ovule ratio (ranges 750–1067) of this species matches that of facultative xenogamy. The flowers of this genus show secondary pollen presentation, and the pollen is released on stigmatic area before flowers open. However, bagged or hand self-pollinated flowers yield no seeds, and the self-pollen tubes often do not reach the ovules (unpublished data), indicating that pollinator-mediated cross-pollination is required for successful seed and fruit set. The flowers have a near horizontal orientation that is on average −6.80° ±10.63 (mean ± SD) to the horizontal. The corolla has four petals, and the upper petal extends to form a spur in which nectar collects. The stigma and anthers are concealed in the inner petals, when legitimate pollinators visit from the front of the flowers, they depress the inner petals and expose the reproductive surfaces. We conducted the study in a natural population located in Sandouping, Hubei Province, China (30° 49′ 46.3″ N, 111° 04′ 03.3″ E, 150 m in altitude). The population consisted of more than 1000 flowering individuals, forming mono- specific patches surrounding forest fragments. Flowering period began in March and lasted to late April. Seeds matured in May. In this population, bee species, primarily leafcutter bees (*Megachile* spp.) and bumblebees (mostly *Bombus richardsi*) are the major pollinators. *Veronica persica* and *Petasites tricholobus* also flower at the same time, but they are usually visited by syrphid flies and honeybees, respectively (personal observation). Our field study was conducted from 10 March to 5 May 2013, encompassing the flowering peak at the study site. ## Flower Manipulation We presented three types of inflorescences with different floral orientations: (i) unmanipulated inflorescences with horizontal-facing flowers (Unmanipulated), (ii) inflorescences with flowers turned upward (Up), and (iii) inflorescences with flowers turned downward (Down). We conducted pollination observation by using individual flower patch. Each flower patch contained at least nine flowering individual plants and was 3 m in minimum separation between each other. Inside the patch, nine inflorescences with newly open flowers were selected from different plants and randomly assigned to the treatments (three of each type). The angle of individual flowers was altered by leaning the pedicel and taping the terminal of the spur to the inflorescence stalk. These manipulations were carefully conducted to ensure not to damage flower structure. We avoided shaking of flowers in order to not cause pollen loss. To minimize the influence of display size on pollinator behavior, , each studied inflorescence was trimmed to five flowers. ## Pollinator Observation For each flower patch, we observed pollination on focal inflorescences for a period of 15 minutes. We conducted a total of 165 observation periods from 0800 h to 1600 h on sunny days when pollinators were active. Moreover, different flower patch was used for each observation period. We scored two types of pollinator behaviors on inflorescences: approaches and visits. An approach occurred when a pollinator found an inflorescence and approached it from the front side of the flowers. A visit occurred only when the pollinator probed at least one flower in an inflorescence. Each visit on an inflorescence was tracked and we recorded the number of successive probes within an inflorescence, as well as the direction (upward/downward) of movements between flowers of the same inflorescence. To detect changes of pollinator attraction with different floral orientation, we counted the number of inflorescences that were approached and visited per observation period. To detect changes of pollinator behavior within inflorescences, we used the number of successive probes within an inflorescence during a visit and the direction of vertical movements on an inflorescence. ## Pollen Removal and Seed Set For each treatment (Unmanipulated, Up, Down), we measured pollen removal during a visit by using at least 30 enclosed inflorescences which were exposed to pollinators. The flowers were collected after they received a single visit from bees. We used individual flowers as replicates (N = 93 for leafcutter bees and N = 29 for bumble bees, respectively). Additionally, we randomly collected 20 flower buds to count pollen production per flower. All collected flowers were separately stored in vials containing 1 ml 70% ethanol. Pollen removal was estimated as the number of pollen grains removed: *P*₀–*P*_v. *P*₀ refers to pollen production per flower, and *P*_v is the number of pollen grains remaining on visited flowers. Since in *Corydalis*, pollen is secondarily presented on stigmas, the number of pollen grains left in anthers and on stigmas of the flower constitutes the number of pollen remaining in each visited flower. The anthers and stigmas were squashed to dislodge pollen grains, and the number of pollen grains in five 20-µL subsamples from each flower was counted under light microscopy (×4). For each treatment (Unmanipulated, Up, Down), we measured seed production per inflorescence by using at least 20 enclosed inflorescences each from different plants. The inflorescences were trimmed to five flowers and exposed to pollinators. When the seeds matured, we counted seed production per inflorescence as the total number of seeds produced by each inflorescence. We harvested a total of 49 inflorescences to collect seed data, because some of the focal inflorescences were damaged in field. ## Data Analyses Analyses of pollinator behavior were conducted for leafcutter bees and bumble bees separately, because pollinator taxa may differ in their foraging preference and direction of movement on inflorescences. We used the data from observation periods during which at least an approach was observed. To test whether a horizontal orientation facilitated pollinator recognition and attraction, we compared the number of inflorescences that were approached or visited per observation period among treatments (Unmanipulated, Up, Down), using observation periods as replicates. We applied generalized linear models (GLMs; with Poisson error and logarithmic link). The number of approached or visited inflorescences was treated as the response variable and the treatment as a fixed effect. To test the effect of floral orientation on occurrence of visits following approaches, we applied generalized linear mixed models (GLMMs) with binomial errors and logit-link function. The treatment was considered as a fixed effect and the response variable was the presence/absence (1/0) of a visit after an approach. Because the observation period was the source of replication and the approaches occurred in the same period were not independent of each other, we included observation period as a random term. We only used the data from inflorescences that received at least an approach. We used GLMMs (with Poisson error and logarithmic link) to examine the effect of floral orientation on the number of successive probes within an inflorescence per visit. The number of successive probes was considered as a response variable. The treatment was considered as a fixed effect and observation period as a random term. Only data from inflorescences received at least a visit was used. To analyze the consistency of pollinator movement on inflorescences, we compared the direction percentage of vertical movements (upward and downward) to the total vertical movements among treatments using Pearson chi-squared test. We also compared the pollen removal per visit per flower and seed set per inflorescence among treatments using GLMs (with Poisson error and logarithmic link), using individual flowers and inflorescences as unit of replication, respectively. All statistical analyses were conducted in SPSS 19.0 at significant level of 0.05. # Results ## Pollinator Behavior Compared to the unmanipulated inflorescences, the number of inflorescences that were approached and visited by leafcutter bees decreased in both Up and Down treatments. The approaches by bumblebees decreased only in the Up treatment, however, the number of inflorescence visited by bumble bees decreased for both Up and Down treatments. Compared with Unmanipulated inflorescences, the occurrence of visits following approaches decreased in manipulated inflorescences for both pollinator types (GLMM, leafcutter bee, Up: *b* = −1.433±0.309, *z* = −4.638, *P*\<0.001; Down: *b* = −0.698±0.275, *z* = −2.538, *P* = 0.011; bumble bee, Up: *b* = −2.455±0.669, *z* = −3.669, *P*\<0.001; Down: *b* = −1.769±0.354, *z* = −4.997, *P*\<0.001; Unmanipulated is used as the baseline and a negative value of *b* implies a negative effect of the treatment). Leafcutter bees made 1.69±0.83 (mean ± SD), 1.59±0.50, 1.75±0.87 successive probes during a visit on Unmanipulated, Up and Down inflorescences. For bumble bees, the values were 1.74±0.98, 1.50±0.54 and1.29±0.46, respectively. The number of successive probes didn’t differ among treatments for leafcutter bees (GLMM, Up, *b* = −0.058±0.071, *z* = −0.817, *P* = 0.413; Down, *b* = 0.037±0.079, *z* = 0.468, *P* = 0.635). Compared with Unmanipulated inflorescences, the bumble bees made significantly less probes during a visit on Down inflorescences (Up, *b* = −0.151±0.141, *z* = −1.071, *P* = 0.286; Down, *b* = −0.305±0.118, *z* = −2.585, *P* = 0.010). Changes of floral orientation significantly affected the consistency of movement on inflorescences for leafcutter bees, as the proportion of upward movement was significantly greater in the horizontal than in the Up and Down inflorescences. The direction of movement by bumble bees was not analyzed since their movements on Up and Down inflorescences were relatively infrequent. ## Pollen Removal and Seed Set The leafcutter bees removed 69.34±20.78% of pollen per flower per visit from Unmanipulated inflorescences, 55.91±26.90% from Up inflorescences and 58.73±26.28% from Down inflorescences. For bumble bees, the values for Unmanipulated, Up and Down inflorescences were 59.47±21.72%, 37.41±19.10% and 42.72±38.40%, respectively. Both pollinator types removed significantly less pollen grains per visit from Up and Down inflorescences, relative to Unmanipulated inflorescences (GLM, leafcutter bee, Up: *b* = −0.215±0.002, *z* = −107.500, *P*\<0.001; Down: *b* = −0.166±0.002, *z* = −83.000, *P*\<0.001; bumble bee, Up: *b* = −0.464±0.005, *z* = −92.800, *P*\<0.001; Down: *b* = −0.331±0.005, *z* = −66.200, *P*\<0.001;). Additionally, deviation from the horizontal orientation also significantly decreased seed production per inflorescence (GLM, Up: *b* = −0.379±0.048, *z* = −7.896, *P*\<0.001; Down: *b* = −0.320±0.044, *z* = −7.273, *P*\<0.001;). # Discussion We tested three hypotheses concerning the role of horizontal orientation in affecting pollinator behavior and pollen transfer. These hypotheses predict higher pollinator attractiveness, higher pollination efficiency and greater consistency of pollinator movement for inflorescences with horizontal-facing flowers. Our results using bee-pollinated *C. sheareri* inflorescences support the three hypotheses. For the manipulated inflorescences, the decreases in approaches indicate that bee recognition of the floral pattern reduced. The decreases in occurrence of visits following approaches indicate that changing floral orientation may impair the function of floral landing platforms. In addition, the declines in pollen removal and consistency of bee movement may be attributed to changes in postures when bees foraged on flowers of different orientation. Legitimate pollinators typically approached the one-sided racemes and foraged on the horizontal-facing zygomorphic flowers from the front. Altering floral orientation could affect visibility of floral patterns and impair the function of floral landing platforms, resulting in decreased pollinator recognition and diminished inflorescence attractiveness. The argument was supported by our results that changes in floral orientation led to significant decreases in the number of approaches and visits to inflorescences. The inflorescences with downward-facing flowers received as many approaches by bumble bees as the unmanipulated inflorescences, suggesting bumble bee recognition of foraging opportunities may not be affected by turning flowers downwards. This is consistent with the fact that bumble bees do specialize on some plant species with downward flowers, and may not dislike downward-facing flowers. However, the decreases in visits after approaches for inflorescences with downward flowers indicate that downward flowers may have a reduced attractiveness. Hummingbirds which specialize on plant species with pendant flowers were also reported to have a tendency to prefer less-pendent flowers to pendent ones. This was explained that horizontal flowers may be more visible and accessible because their entrances are displayed in the surface of the inflorescence. Additionally, downward orientation is also a character of wilting flowers in some plant species, including *C. sheareri*. Since wilted flowers are presumably less rewarding, inflorescences composed of downward flowers are not as attractive as horizontal flowers. Leafcutter bees typically move upwards on an inflorescence, as has been shown by Iwata et al. (2012). Racemes enhance greater consistency of pollinator movement compared to panicles and umbels. Nevertheless, changes in floral orientation significantly increased the proportion of downward movements by leafcutter bees, supporting the hypothesis that a horizontal orientation ensures consistency of pollinator movement on inflorescences. Corbet et al.(1981) proposed that foraging posture may affect the direction of pollinator movement. Pollinators usually fly and forage in an upright position, and moving upwards rather than turning around is considered to incur less cost in time and energy. This is corresponding with the fact that pollinators who forage in a head-down position tend to move downwards on inflorescences. Additionally, the upward movement is also explained by that pollinators may have a better vision above than below when they position themselves upright. The effect of floral orientation on direction of pollinator movement may be attributed to the influences on both foraging posture and the view of pollinators. When pollinators forage on horizontal or downward-facing flowers, they generally maintain an upright position,. Therefore, the pollinators would have a better view of entrance to flowers above, and tend to move upwards rather than turning around. By contrast, pollinators forage on upward-facing flowers in a head-down position, they would have a better view of flowers below and probably make more downward movements. Inconsistent with these ideas, we found that the proportion of downward movements increased in both artificially Up and Down inflorescences. Interpretation is complicated by the fact that altering the floral orientation also altered the pollinator behavior when they switched between flowers within the inflorescences. The pollinators can only fly to next flower when flowers face horizontal, but changing floral orientation decreased the space between neighboring flowers and enabled pollinators to crawl between neighboring flowers. A higher directionality of upward movement is predicted in flights because insects may have a better control of their flight when they move against gravity. By contrast, inflorescence characters that facilitate crawling between flowers would reduce the consistency of pollinator movement, and crawling between neighboring flowers would reduce flight from one flower to another. Therefore, we propose that the effect of floral orientation on consistency of pollinator movement may be associated with flower arrangement. Floral orientation is thought to affect the efficiency of pollen transfer due to changes in pollinator landing behavior or mechanical fit between plant reproductive organs and pollinators. Foraging on downward flowers requires bees to remain in hanging position and foraging on upward flowers requires a head- down position. Though both leafcutter bees and bumblebees are relatively skilled to forage on *C. sheareri* flowers with different orientations, the changes in foraging posture may cause variations in foraging behavior and the fit between bees and floral reproductive organs. Since the pollinator attraction is associated with pollen transfer and mate diversity, reductions on attraction may reduce both male and female functions. The movement of pollinators on inflorescences may mediate pollen transfer, affecting geitonogamy and pollen export. In our study, bumble bees made less successive probes on inflorescences with downward-facing flowers, relative to unmanipulated inflorescences. However, the directions along which pollinators move usually combine with the direction of dichogamy and order of development in inflorescences, affecting geitonogamous self-pollination and pollen-stigma interference. In protandrous racemes of *C. sheareri*, lower (older) flowers always display female function while upper (younger) flowers are in the male stage. The upwards direction of bee movements functions to facilitate outcrossing, because the pollinators start to visit lower stigma-presenting flowers and deposit cross-pollen from previously visited plants. Meanwhile, pollen export can be enhanced as the pollinators left the inflorescences from upper pollen-presenting flowers. By contrast, a higher percentage of downward movements would result in more pollen transferred from higher (male) flowers to lower (female) flowers. For self-incompatible *C. sheareri*, though self- fertilization is impossible, the self-pollen may occupy the stigmatic area and restrict the deposition of cross-pollen, as well as reduce the growth of cross- pollen tubes. Moreover, the amount of pollen transferred to other plants would be also reduced. In our study, the decreased seed set in manipulated inflorescences may be the result of reduced pollinator attraction and pollination efficiency, as well as increased interference by self-pollen. Functional aspects of floral orientation in pollinator behavior and pollination precision have been investigated in individual flowers. In our study, we confirmed that a horizontal orientation has adaptive significance in ensuring consistency of pollinator movement on inflorescences. Since manipulation of pollinator behavior to promote outcrossing and pollen export is critical in the evolution of floral and inflorescence characters, the orientation of flowers within inflorescences may be under selection by pollinators. Though inflorescence architecture and floral orientation are traits showing quite a bit of phylogenetic conservatism, our results verify that floral orientation may act together with certain inflorescence architecture in controlling pollinator movement. Further research is necessary to examine the effect of floral orientation on pollen flow within inflorescences, in order to understand the influences of a horizontal orientation on geitonogamy and outcrossed siring success. In addition, the generality of the effects of floral orientation on pollinator behavior calls for investigations in plants pollinated by different pollinator types, with different floral design and inflorescence architecture. We thank Prof. Paul Wilson for greatly improving the quality on an earlier manuscript, Prof. Wulfila Gronenberg and two anonymous reviewers for their detailed and constructive comments, Robert W Gituru for language improvement, Chang-Long Xiao and Xue-Gang Zhu for assistance in fieldwork, Xiao-Xia Li for helpful suggestions. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: HW CFY YHG. Performed the experiments: HW ST. Analyzed the data: HW CFY. Wrote the paper: HW CFY DY.

# Introduction Transfer RNAs are fundamental biological molecules that read cognate codons in mRNA to complete the flow of genetic information from DNA to protein. A ubiquitous feature of tRNAs is chemical modification, the most prominent being methylation (of the base or ribose sugar) as catalyzed by tRNA methyltransferase enzymes (Trms). Notably, tRNA methylation occurs post-transcriptionally, and can thus be viewed as an epitranscriptomic feature at the molecular level. In general, methyl-based modifications of tRNA function to ensure translational fidelity, although how cells accomplish this depends upon the chemical nature of the modification and its position within the tRNA molecule. For example, modifications within the anticodon stem loop (ASL) at position 34 (decoding the wobble base) and 37 influence both the accuracy and efficiency of translation through degenerate codon selection. Modifications within other loop regions of tRNA can influence base interactions that direct folding, increase stability, or alter the structure of the tRNA molecule in ways that regulate aminoacylation or interactions with the ribosome. This study is motivated by two observations that support a role for tRNA modification systems (i.e., Trm enzymes, their modifications, and target transcripts) in the oxidative stress response. First are studies in several eukaryotic cells demonstrating that deleting Trms results in sensitivity to one or more toxicants that induce oxidative stress, or damage DNA. Second are observations that tRNA modifications are dynamic, changing in response to various toxicants that induce oxidative stress; moreover, these changes include methylation of the ribose sugar moiety at the 2’-OH position (i.e., 2’-O-ribose methylation). Collectively, these observations are a good indication that Trms and their corresponding modifications play a role in the cellular response to stress. However, questions remain about how cells engage these tRNA modification systems to survive and recover from oxidative stress after toxicant exposure. Here we focused on *Saccharomyces cerevisiae* (budding yeast) 2’-O-ribose methyltransferases because 2’-O-ribose modified bases are known to increase under conditions of oxidative stress. For example, 2’-O-methylcytosine (Cm), 2’-O-methyladenosine (Am), and 5-methoxycarbonylmethyl-2'-O-methyluridine (mcm⁵Um), and unpublished observations). Furthermore, methylation of tRNA at the 2’-OH position of the ribose sugar is generally thought to increase the stability of tRNA via mechanisms that protect against spontaneous hydrolysis or nuclease digestion (e.g., in non-helical regions) and reinforce intra-loop interactions that stabilize the tertiary structure of the molecule\]. Such stabilization strategies are essential for thermophilic organisms where tRNAs must carry out protein translation at denaturing temperatures, an idea supported by observations in diverse archaea that show an abundance of 2’-O-methylated nucleosides among global populations of modified tRNAs. In mesophilic prokaryotes and eukaryotes, 2’-O-ribose modification also plays a structural role in the non-duplex loop regions of the D- and T-arm. The importance of this type of methylation event in mesophilic eukaryotes is not fully understood, though it seems plausible that it serves a stabilizing function in environments that are destabilizing to tRNA molecules (e.g., increased intracellular reactive oxygen species, or ROS). *S*. *cerevisiae* tRNAs undergo methylation of the 2’-OH position of the ribose sugar at three sites in the body of the tRNA molecule (i.e., 4, 18, and 44), and at two sites within the ASL (i.e., 32 and 34, the latter decoding the wobble base). At least four Trms are known to carry out these methylation events: Trm3 → Gm18, Trm7 → Cm32 and Nm34, Trm13 → Cm4, and Trm44 → Um44. To investigate a role for 2’-O-ribose methylation of tRNA in the oxidative stress response, we assessed corresponding *trm* deletion mutants for their sensitivity to oxidative stress after exposure to hydrogen peroxide, rotenone, and acetic acid. In all assays, hydrogen peroxide was the most potent inducer of oxidative stress sensitivity. Accordingly, we characterized methylation chemistries in 2’-O-ribose mutants after acute exposure to hydrogen peroxide and found a complex profile of methylation events in exposed cells, which included increases in Um, Cm, and Gm in wild type cells. Notably, 2’-O-ribose associated modification chemistries were attenuated in *trm3*, *13*, and *44* mutants. Potential roles for these modifications in the oxidative stress response are discussed further in the context of their position on the tRNA molecule. To test the idea that tRNA modification systems targeting the ASL regulate the oxidative stress response, we used Trm7 → tRNAPhe^(GmAA) as a model for preferred UUC codon selection. Using a bioinformatics approach, we identified UUC-enriched transcripts in *S*. *cerevisiae* as potential Trm7-influenced translational targets and found several with biological functions linked to the oxidative stress response. *GNP1* stood out as a UUC- enriched transcript with significant oxidative stress-sensitivity in corresponding *gnp1* mutants. In light of recent evidence that *trm7* mutants activate the general amino acid control (GAAC) response pathway, we propose that attenuated codon-biased translation of *GNP1* is a component of this response, contributing to the oxidative stress sensitivity phenotype of the *trm7* mutants, which is a focus for future study. Overall, this work solidifies a role for 2’-O-ribose methylation in the oxidative stress response and supports a model in which *S*. *cerevisiae* uses this type of modification event in an epitranscriptomic strategy that combines increased tRNA stability and enhanced translational effects post-exposure. # Materials and methods ## *S*. *cerevisiae* cell culture and transformation Deletion mutants were retrieved from an *S*. *cerevisiae* library (Open Biosystems, Thermo Scientific, Hudson, NH) established by the *Saccharomyces* Genome Deletion Project and are on a BY4741 background (i.e., haploid genotype: *MAT****a*** *his3Δ1 leu2Δ0 met15Δ0 ura3Δ0*). In this library, ORFs have been replaced by a KanMX cassette, selected for by growth in yeast extract- peptone-dextrose (YPD) broth supplemented with G418 (200 μg/mL) prior to toxicant exposure. Open reading frames (ORFs) for *TRM3*, *7*, and *13* were obtained in the galactose-inducible BG1805 plasmid vector from the Yeast ORF Collection (Dharmacon^TM, Horizon Discovery Group, Cambridge, UK). ORF plasmids were introduced into their respective *trm* mutant strains by transformation using a one-step buffer containing 40% polyethylene glycol (PEG), 200 mM lithium acetate and 100 mM dimethyl sulphoxide. *TRM*-ORF transformants were selected by plating on synthetic complete medium: yeast nitrogen base/complete amino acid supplement mixture (YNB/CSM) minus uracil, plus 2% dextrose (SC-Ura, Sunrise Science Products, TN) and growth at 30°C for three days. For ORF induction, single overnight colonies were initially grown in SC-Ura, and then diluted 1:100 in YNB/CSM minus uracil plus 2% raffinose (ACROS) and allowed to reach stationary phase. These cultures were then diluted three-fold with concentrated media to either induce or repress ORF expression. That is, 3 x YNB/CSM-Ura plus 6% galactose (SIGMA) or 6% dextrose (SIGMA), respectively. After growth to an OD_600nm of 2.0, cultures were split into separate tubes for one-hour control (water) or hydrogen peroxide (20 mM) treatments at 30°C, followed by plating on SC-Ura (i.e., 100 μL of a 10⁻⁵ dilution) and colony counting after three days of growth at 30°C. ## Growth, viability, and colony-forming assays For the plate-based growth assays, individual mutants were inoculated in liquid, incubated overnight at 30°C with shaking (i.e., 225 rpm), and then diluted in a ten-fold series (i.e., 10⁻¹ to 10⁻⁵). Five μL of each dilution was then plated on either YPD, or YPD supplemented with hydrogen peroxide (2 mM), rotenone (400 μM), and acetic acid (20 mM). Plates were incubated at 30°C for 3 days and imaged for growth using a digital camera system (Bio-Rad, CA). Also, quantitative measurements of growth over a twenty hour period were performed on overnight cultures grown from single colonies using an automated microplate reader set at 30°C with orbital shaking (TECAN). To assess viability post-exposure, cultures of *trm* mutants were grown to log-phase from single colonies in YPD + G418, and then diluted to 3 x 10⁶ cells/mL in YPD or YPD supplemented hydrogen peroxide (40 mM), rotenone (400 mM), and acetic acid (80 mM); BY4741 were similarly treated and served as the control strain. The number of viable cells was determined by staining with trypan blue, four and eight hours post-exposure; cell death was determined by dividing the number of trypan blue positive cells by the total number of cells and is calculated as a percentage. For the colony-forming assay, similar liquid cultures of mutant single colonies and BY4741 were grown to stationary phase in rich media (i.e., YPD) and diluted to a uniform OD_600nm of 2 (\~ 6 x 10⁷ cells/mL). Hydrogen peroxide (10 mM), rotenone (100 μM) or acetic acid (20 mM) were added directly to 1 mL of these cultures, with an equivalent volume of water serving as a control. These cultures were incubated for 1 hour at 30°C with shaking, serially diluted to 10⁻⁵, and plated on YPD (100 μL). After 3 days of growth at 30°C, the number of colonies was counted. Similar colony forming assays were performed on *trm* mutants transformed with their respective ORFs, albeit with different growth conditions for ORF selection and expression as described above. ## Quantitative real-time reverse-transcription PCR (qRT-PCR) Total RNA was first isolated from BY4741, and *trm3*, *7*, *13*, and *44* mutant strains using TRIzol® reagent (ThermoFisher Scientific) and then 1 μg was converted into complementary DNA (cDNA) using an iScript^™ cDNA synthesis kit (Bio-Rad) according to the manufacturer's instructions. cDNAs were diluted 5 fold and 4 μL was used to detect *TRM3*, *7*, *13*, and *44* gene expression with TaqMan® gene-specific expression assays via qRT-PCR on a StepOnePlus real-time PCR system (ThermoFisher Scientific). *TRM* expression in the *trm* mutants was determined relative to BY4741 using the delta-delta cycle threshold method (2^-ΔΔCt), with actin serving as the internal control. ## Transfer RNA modification analysis Transfer RNA was isolated from yeast spheroplasts generated by digestion with lyticase (from *Arthrobacter luteus*, Sigma-Aldrich, MO) digestion using a micro RNeasy RNA isolation kit, modified to prevent the loss of covalent modification chemistries. Specifically, at the first step of cell lysis, QIAzol reagent was supplemented with the deaminase inhibitor tertrahydrouridine (0.2 mM, Millipore Sigma, MO), and anti-oxidants deferoxamine and butylhydroxytoluene (both at 0.1 mM, Millipore Sigma, MO). At least 50 μg of total micro RNA was isolated by this method (containing \> 85% tRNAs) for subsequent mass spectrometric (MS) analysis of modification chemistries. The MS analysis was performed on tRNAs isolated from *trm* mutant strains, before and after acute exposure to hydrogen peroxide (20 mM for 1 hour), as previously described in detail, and for distinct cellular stress responses. Briefly, isolated tRNAs were digested with select nucleases to obtain analogous mononucleotide mixtures for analysis by direct infusion electrospray ionization (ESI) to detect variable levels of expression in individual modification chemistries. Expression levels were calculated using the Abundance versus Proxy method (AvP), which determines the relative abundance of each post-translational modification (PTM) normalized against the sum of the intensities of the four canonical bases. Statistically significant differences in AvP values (exposed versus unexposed) were determined using an unpaired Student *t*-test (n = 3), and those with a p-value \< 0.05 are shown in and appear in color in and Tables. The same tRNA isolates were also digested with a combination of select nucleases and phosphatases to obtain individual mononucleotides for Tandem MS analysis, carried out in both the positive and negative modes. This approach facilitated the differentiation of isobars detected during PTM analysis, including Am, Um, Cm, and Gm. ## Intracellular ROS detection *S*. *cerevisiae trm* mutants and BY4741 strains were inoculated into YPD from single colonies. When the culture was at mid-log phase (OD_600nm \~ 0.5), cells were pelleted by centrifugation (2500 rpm for 5 min), washed once with phosphate-buffered saline (PBS), and then re-suspended in PBS containing 10 μg/mL 2’,7’-dichlorofluorescin (DCFDA, Molecular Probes-Invitrogen, CA). A stock of DCFDA was prepared fresh on the day of the experiment by dissolving in dimethyl sulfoxide to 10 mg/mL. The cell-DCFDA suspensions were then incubated for 2 hours at 30°C with shaking, washed once with PBS and re-suspended in 1 mL of PBS for direct analysis by flow cytometry using an Image Stream ISX100 (Amnis, WA). The level of reactive oxygen species was measured based on the emission of oxidized DCFDA dye in the green channel (517–527 nm). ## Bioinformatics screen of *S*. *cerevisiae* ORFs Whole-genome ORFs of *S*. *cerevisiae* were surveyed for UUC codons using the CUT Codon Utilization bioinformatics tool. The CUT tool uses existing genome and transcript data for *S*. *cerevisiae* to quantitate the number of times a transcript uses an individual codon or a run of a particular codon (i.e., UUC or UUCUUC). CUT then determines whether or not the codon, or codon run, occurs more or less than the average genome frequency. Statistically significant codon usage was determined by Z-score, with a cutoff of +1.96 representing the 95% confidence level, or an associated P-value of 0.05. # Results ## 2'-O-ribose *trm* mutants are sensitive to oxidative stress Considering the ability of oxidative stress to induce changes in 2’-O-ribose methylation events in *S*. *cerevisiae*, together with observations of stress sensitivity phenotypes in other *S*. *cerevisiae trm* mutants (i.e., Trm9 and Trm4, respectively), we investigated whether 2’-O-ribose *trm* mutants are also sensitive to oxidative stress. Four 2’-O-ribose methyltransferases were identified in the *Saccharomyces* Genome Database (SGD), *TRM3*, *7*, *13*, and *44* (). Quantitative real-time reverse transcription PCR (qRT-PCR) using *TRM*-specific TaqMan® gene expression assays confirmed that the deletion mutants do not express their respective *TRMs*. Interestingly, in the *trm3*, *7* and *44* mutants, alternate 2'-O-ribose *TRMs* are up-regulated to varying degrees relative to the BY4741 wild type strain, supporting that compensatory mechanisms may exist at the transcriptional level. *S*. *cerevisiae* strains lacking the above methyltransferases were retrieved from a haploid single-gene deletion library and grown on plates containing hydrogen peroxide, rotenone, and acetic acid to induce oxidative stress. We found that all 2'-O-ribose methyltransferase mutants were sensitive to these toxicants relative to the wild type (control) BY4741 strain. Notably, this sensitivity phenotype was most apparent on plates supplemented with hydrogen peroxide and was greatest for the *trm3* mutant. These plate-based assays represent a chronic exposure setting in which it can be challenging to differentiate the effect of the toxicant from the impact of the gene deletion on growth if any. Indeed, a slow-growth phenotype has been reported for *trm7* mutants, and this phenotype was also detected here for the *trm7* mutant strain, while the growth of the *trm3*, *13*, *and 44* mutant strains was intermediate between *trm7* and wt BY4741 cells. To further assess the oxidative stress sensitivity of the 2'-O-ribose *trm* mutants, we performed two acute exposure assays in which cells are normalized with respect to cell number prior to exposure. The resulting data reflect survival and recovery post- exposure rather than continued growth in the presence of the toxicant. First, the effects of exposure to hydrogen peroxide, acetic acid and rotenone on cell death was determined by trypan blue staining to distinguish non-viable cells (i.e., positive for trypan blue) from viable cells. An eight-hour exposure to hydrogen peroxide produced high percentages of non-viable cells for the *trm3* and *trm7* mutant strains (top panel). Exposure to rotenone for eight- hours significantly increased cell death across all strains, with the wild type BY4741 strain being largely refractory to this toxicant, which disrupts mitochondrial function to increase intracellular ROS (middle panel). A significant increase in cell death was only observed for the *trm7* strain after an eight-hour exposure to acetic acid (bottom panel). Overall, using viability assays as a measure of oxidative stress sensitivity, *trm7* mutants were the most sensitive to all toxicants, followed by *trm3* mutants (sensitive to hydrogen peroxide and rotenone), whereas *trm13* and *trm44* mutants were only sensitive to rotenone. Second, we used colony-forming assays to assess the ability of the 2'-O-ribose *trm* mutants to recover from an acute toxicant dose, sub-lethal to the wild type BY4741 strain. In this assay, all colonies formed post-exposure were counted, and the data indicate an ability to recover from toxicant exposure. We found that after acute exposure to hydrogen peroxide colony formation was significantly decreased for all 2’-O-ribose *trm* mutants compared to wild type cells. Colony formation was similarly decreased after acute exposure to acetic acid and rotenone in *trm3* and *trm13* mutants. ## Reintroduction of 2’-O-ribose methyltransferases attenuates sensitivity to oxidative stress Since the 2’-O-ribose *trm* mutants were most sensitive to hydrogen peroxide, we used this toxicant to assess colony formation after introducing the open reading frames (ORFs) of the *TRM* genes under the control of a galactose (GAL) inducible promoter (i.e., *trm*-Δ + *TRM-ORF*). ORF constructs were available from a yeast open reading frame collection for *TRM3*, *7*, and *13*, and their expression was selected for by initial growth in the absence of uracil. Under non-inducible growth conditions for ORF gene expression (i.e., + Dex: *TRM* “off”) the colony-forming ability of the 2’-O-ribose mutants was reduced after exposure to hydrogen peroxide as expected. However, under growth conditions that induce ORF gene expression (i.e., + Gal: *TRM* “on”) the number of colonies that survived hydrogen peroxide exposure significantly increased. Although glucose is the preferred carbon source for *S*. *cerevisiae*, under +Gal-untreated conditions, the *trm3* and *trm7* mutants produced more colonies relative to the +Dex-untreated (results not shown), suggesting that some reprogramming occurs during the carbon source switch; thus, the data is presented relative to the same growth conditions concerning the carbon source. These results demonstrate some degree of rescue of oxidative stress sensitivity phenotypes by introducing 2’-O-ribose methyltransferase genes into their respective deletion mutants. ## Oxidative stress-induced nucleoside modification profiles in 2’-O-ribose *trm* mutants We used direct infusion electrospray ionization mass spectrometry (ESI-MS) to quantitate the levels of ribonucleoside modifications in wild type BY4741 cells and 2’-O-ribose *trm* mutants before and after exposure to hydrogen peroxide, focusing on Am, Um, Cm and Gm. Exposure-induced changes are shown relative to unexposed wild type cells , which were calculated from discrete values determined using the abundance versus proxy method. This method measures the absolute signal intensity of the modified ribonucleoside over the sum of the absolute intensities of the canonical bases, taking into account any changes unmodified nucleosides. We detected increases in Um, Cm, and Gm in wild type cells after exposure to hydrogen peroxide. These toxicant-induced increases in Um, Cm, and Gm were attenuated across all 2’-O-ribose *trm* mutants. Notably, in the cases of *trm3*, and *13*, this lack of induction occurred on a background of higher levels of Um, Cm and Gm, relative to wild type cells. Am was also higher in *trm13* mutants. These interesting (and unexpected) results lead us to question whether there is an oxidative stress phenotype in 2’-O-ribose *trm* mutants prior to exposure. We investigated this possibility by staining cells with the 2’,7’-dichlorofluorescein (DCFDA), a dye that oxidizes in the presence of intracellular reactive oxygen species (ROS), emitting fluorescence that is detected in the “green” channel using flow cytometry. The mean DCFDA intensity was significantly increased in all 2’-O-ribose *trm* mutants, with the *trm3* and *trm13* strains showing the most substantial increases compared to wild type cells. These preliminary results support that some redox mechanism(s) is deregulated in the *trm* mutants and are not indicative of the type of reactive oxygen species elevated, or hydrogen peroxide levels, which will be determined in future. In the case of *trm7* mutants, 2’-O-ribonucleosides were not induced by exposure to hydrogen peroxide. In fact, in cells treated with hydrogen peroxide, Um, Cm and Gm were decreased, and Am was undetectable. Overall, these results show a complex profile of 2’-O-ribose methylation events in *S*. *cerevisiae* exposed to hydrogen peroxide and suggest that some of these modifications act as part of both the oxidative stress response and in ROS maintenance under normal growth conditions. This complexity is further reflected in our global analysis of ribonucleotide modifications that are changed in the 2’-O-ribose *trm* mutants under both normal growth conditions and in response to hydrogen peroxide treatment ( and Tables). ## Identification of candidate transcripts influenced by 2’-O-ribose methylation Considering the oxidative stress sensitivity of the 2'-O-ribose *trm* mutants, we were interested in identifying transcripts that are translationally regulated by this type of methylation event. Based on previous work, we expect that stress-associated methylation of tRNAs enhances the translation of stress- responsive gene transcripts. If this is true, then genes involved in the stress response will use codons whose translation is influenced by methylated tRNAs, including those methylated at the 2'-O-ribose position. We tested this idea using tRNA^Phe(GAA) as a model since it is a biologically relevant target of Trm7, and is methylated at the wobble position (underlined), a site that is known to influence codon selection (reviewed in). In addition, phenylalanine decoding represents a straightforward "two-choice" decision compared to other degenerate codons since it is decoded by either UUU or UUC, and *S*. *cerevisiae* Trm7-modified tRNA^Phe(GmAA) is thought to promote cognate UUC base-pairing interactions that favor the translation of UUC over UUU. To identify transcripts that may be influenced by Trm7-mediated tRNA^Phe(GmAA) modifications, we took a bioinformatics approach using the gene-specific Codon UTilization (CUT) tool. This tool identified a set of transcripts that use UUC to decode phenylalanine at a higher frequency when compared to all other transcripts of the *S*. *cerevisiae* genome. Moreover, these UUC-enriched transcripts correspond to genes that are involved biological processes associated with the oxidative stress response, as well as others that are involved in processes that regulate cell growth and metabolism. Interestingly, these transcripts not only use UUC at a higher frequency compared to other transcripts, they also have UUC in duplicate or triplicate codon "runs", which is a rare event considering that there are only 24 reported open reading frames in *S*. *cerevisiae* that contain a UUCUUCUUC run, representing only 0.36% of the coding genome for this organism. ## 2'-O-ribose translational targets are oxidative stress-sensitive Our bioinformatics approach identified transcripts that are potential targets of Trm7-mediated 2’-O-ribose methylation, providing support for the idea that UUC- enriched transcripts rely on 2'-O-ribose methylation to enhance their translation in response to stress. As a first step towards testing this theory, we retrieved *S*. *cerevisiae* strains deficient for a subset of UUC-enriched genes (i.e., *CRT10*, *FRE5*, *HXT2*, *HIR3*, and *GNP1*) from our library of deletion strain mutants, and assessed colony survival after exposure hydrogen peroxide. We found significant sensitivity, to varying degrees, across all of the UUC-enriched gene mutants investigated thus far. These findings support that UUC-enriched gene transcripts encode proteins that play a functional role in recovering from oxidative stress and point to a potential role for Trm7-mediated regulatory effects at the translational level, which fits into our broader hypothetical model of how 2'-O-ribose methylation contributes to the oxidative stress response. # Discussion Here we have shown that the deletion of 2’-O-ribose methyltransferase genes in *S*. *cerevisiae* leads to sensitivity to oxidative stress-inducing toxicants. The degree of individual 2’-O-ribose *trm* mutant sensitivity was dependent upon assay endpoint (i.e., cell viability or cell survival to colony formation), and hydrogen peroxide produced the highest degree of oxidative stress sensitivity by colony assay across all *trm* mutants. Notably, although rotenone significantly increased cell death across all *trm* mutants, mitochondrial dysfunction may also be a contributing factor to a decrease in viability. Our analysis of oxidative stress-induced nucleoside modifications in the *trm* mutants revealed a complex profile of 2’-O-ribose methylation events, raising new questions. For example, why do cells deficient in *TRMs 3* and *13* increase overall levels of 2’-O-ribose modified nucleosides while exhibiting no observable growth impairment? Alone, these mutants were not deficient in any single 2’-O-ribose methylation event, arguing for compensatory mechanisms of 2’-O-ribose methylation that are sufficient for growth, but insufficient to overcome oxidative stress. Compensatory mechanisms further suggest an underlying stress phenotype, which included elevated levels of intracellular ROS in the *trm* mutants. We postulate that to offset the destabilizing effects of this ROS on tRNA molecules, 2’-O-ribose *trm* mutants undergo widespread increases in 2’-O-ribose methylation. Further research is required to define the molecular mechanisms behind this ROS-associated phenotype. The *trm3* mutant strain showed a high degree of sensitivity across all toxicants relative to other 2’-O-ribose mutants. Trm3 is responsible for the formation of 2’-O-methylguanosine (Gm) at position 18, which occurs in several tRNA species including isotypes of tRNA^Leu, tRNA^His, tRNA^Tyr, tRNA^Ser, and tRNA^Trp; in these isotypes, Gm18 is adjacent to guanosine at position 19. Our modification analysis shows an increase in Gm after exposure to hydrogen peroxide that was absent in the *trm3* mutants, raising the possibility that Trm3-mediated Gm18 formation may be a component of the oxidative stress response. A caveat to this interpretation is that our modification data is not positional, and so we cannot attribute exposure-induced Gm solely to the activity of Trm3 at position 18. A likely role for Trm3-mediated Gm modification in response to oxidative stress is to enhance the stability of tRNA molecules after exposure to toxicants that elevate ROS. Gm at position 18 of the D loop interacts with Ψ at position 55 and, together with other D- and TΨ-loop interactions (i.e., G19-C56 and G53-m1A), acts to stabilize the three-dimensional core of the tRNA molecule. A parallel for this exists in thermophilic organisms where 2’-O-ribose methylation and 2-thiolation of t54, acts as a thermal adaptation to increase tRNA stability. In either environment—oxidative stress or extreme temperature—tRNA molecules would be susceptible to denaturation, particularly in non-duplex regions, such as the D-loop where G18 resides. It follows that Trm3-mediated Gm modification could represent an oxidative stress adaption of protein synthesis. Trm7 interacts with Trm732 and Trm734 for respective 2’-O-ribose methylations of C32 and N34 to form Cm32, and Gm34 in tRNA^Phe, Cm32 and Cm34 for tRNA^Trp, and ncm5Um for tRNA^Leu(UAA). As these Trm7-dependent modifications occur in the anticodon stem loop (ASL), they can influence anticodon: codon interactions. By extension, the oxidative stress sensitivity seen here for the *trm7* mutants may be due to impaired codon selection, which negatively impacts the translation of stress-responsive transcripts (modeled). Reports of tRNA^Phe genomic sequences with an AAA anticodon are rare, and whether or not these tRNA^PheAAA genes are transcribed is not clear. So, eukaryotes seem to rely upon a single anticodon in tRNA^Phe, GAA, to decode both UUC (cognate at the wobble) and UUU (non-cognate at the wobble). Thus, Gm methylation as a mechanism for promoting cognate codon recognition could represent a critical mechanism for degenerate codon selection in certain cellular contexts, including oxidative stress. Several UUC-biased genes were identified here, and their corresponding *trm* mutants were sensitive to oxidative stress. *gnp1* mutants were the most sensitive relative to other UUC-biased mutants—a notable result since *GNP1* encodes an amino acid transmembrane transport protein, and *trm7* mutants have recently been shown to activate the general amino acid control (GAAC) response pathway. These observations suggest that impaired translation of *GNP1* in *trm7* mutants is a component of the GAAC response and contributes to the pronounced oxidative stress sensitivity phenotype of these mutants. Trm13 is responsible 2’-O-ribose methylation of position 4 in at least three species of *S*. *cerevisiae* tRNAs: tRNA^Gly(GCC)-Cm₄, tRNA^Pro-Cm₄, and tRNA^His-Am₄. Despite the overall elevated levels of Am in the *trm13* mutants, we detected a small but significant attenuation of the Am signal as induced by hydrogen peroxide. Together with the *trm13* mutant stress sensitivity phenotype, this suggests that there may be a Trm13 → Am₄ component to the oxidative stress response. Position 4 resides within an acceptor stem region that is engaged base-pairing interactions with residue 69 to form a stable duplex region; thus, a role in structural stability is not predicted for 2’-O-methylation at this site. Another possibility is that Trm3-mediated position 4 methylation influences other processes of translation, such as translocation or aminoacylation. Indeed, in *E*. *coli*, 2’-O-ribose methylation within the acceptor stem can negatively influence translation, and some aminoacyl tRNA synthetases rely on tRNA features within the acceptor stem or variable loop region to determine tRNA identity for charging. *trm44* mutants demonstrated the least amount of oxidative stress-sensitivity relative to the other 2’-O-ribose *trm* mutants studied herein. Trm44 is responsible for Um 2’-O ribose methylation at position 44, and although U occurs at position 44 in isotypes of tRNA^Cys, tRNA^Tyr, and tRNA^Leu, to date, the only tRNA species targeted by Trm44 identified in *S*. *cerevisiae* is cytoplasmic tRNA^Ser. Attenuated increases in Um modifications in response to hydrogen peroxide treatment were detected in the *trm44* mutants, suggesting a Trm44 → Um component to the oxidative stress response. Um44 modification is thought to increase tRNA stability by maximizing base stacking and base interactions that promote the C3’-endo form (preferred for stability in thermophiles) of RNA. Another role for Trm44-mediated Um44 modification could be linked to charging of tRNA^Ser by seryl-tRNA synthetases, which rely on features within the acceptor stem and variable loop regions for recognition. Thus, both stability and accurate tRNA^Ser charging could be affected in *trm44* mutants, producing sensitivity to oxidative stress. While this study substantiates a role for 2'-O-ribose tRNA methylation in the oxidative stress response, the question of how oxidative stress induces *trms* to methylate tRNAs remains. At least one eukaryote Trm, human TRM9L, is targeted for serine phosphorylation by the ERK-RSK pathway, as induced by hydrogen peroxide; thus, 2'-O-ribose Trms may be similarly targeted for activating phosphorylation events, either by this or other stress-responsive kinase pathways. This effect could be further augmented by the oxidative inactivation of dual-specificity phosphatases, which dephosphorylate protein phospho-serine -threonine, and -tyrosine, effects that are known to be induced by ROS. It is also possible that some increases in 2'-O-ribose methylation events do not come from an increase in *trm* activity per se, but are the result of changes in the pools of tRNAs available for *trm*-specific methylation (e.g., through transcriptional up-regulation or increased turn-over of unmethylated tRNA species). Overall, the toxicants used in this study generate diverse intracellular ROS, capable of damaging tRNAs as well as other biomolecules. The non-helical regions of tRNAs may be particularly susceptible to ROS since they are more likely to undergo spontaneous hydrolysis relative to double-stranded helical regions due to the “in-line” geometry of the phosphate and adjacent 2’-OH moiety, which can act as an attacking nucleophile. This type of geometry does not occur in double-stranded helical regions, which are well-conserved among tRNAs and rarely methylated. Further, methylation of the 2’-OH sugar favors a 3’-endo sugar pucker in the helical region of tRNAs, which prevents the type of in-line geometry that is susceptible to spontaneous hydrolysis. Thus, we propose that oxidative stress-associated 2’-OH ribose methylation represents a dual strategy that both maintains the structural stability of tRNAs and exerts translational effects post-exposure. In some instances, this strategy may involve promoting the translation of stress-responsive transcripts with a codon- bias (e.g., Trm7 → tRNA^Phe), while in other instances, it promotes accuracy in charging (e.g., Trm44 → tRNA^Ser). Both situations represent combined epitranscriptomic strategies for oxidative stress recovery. # Supporting information First we would like to thank all members of the Department of Biology and Chemistry at SUNY Polytechnic Institute for their support in enabling this research. In addition, we are grateful to Dr. E. Phizicky (University of Rochester) for providing an independent *trm7* mutant strain with an inducible Trm7 ORF to validate our observations. Also, we thank Dr. T.J. Begley (University at Albany) for providing access to his Open Biosystems deletion strain library. Dr. M. Fasullo provided advice on general techniques and specific protocols for *S*. *cerevisiae* growth and transformation, and generously offered his time to critically review this manuscript. 10.1371/journal.pone.0229103.r001 Decision Letter 0 Preiss Thomas Academic Editor 2020 Thomas Preiss 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 Nov 2019 PONE-D-19-29059 2'-O-ribose methylation of transfer RNA promotes recovery from oxidative stress in Saccharomyces cerevisiae PLOS ONE Dear Dr. Endres, 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 reviewers raise a series of valid points, most of which you should be able to address by a thorough revision of the manuscript as such. It would be advantageous if you could also perfomr the experimental controls suggested by one reviewer; if not possible please argue carefully what the limits of your data are and how this could potentially affect any of your conclusions. We would appreciate receiving your revised manuscript by Dec 28 2019 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, Thomas Preiss, PhD Academic Editor PLOS ONE Journal Requirements: 1\. When submitting your revision, we need you to address these additional requirements. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at <http://www.journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_m ain_body.pdf> and <http://www.journals.plos.org/plosone/s/file?id=ba62/PLOSOne_f ormatting_sample_title_authors_affiliations.pdf> \[Note: HTML markup is below. Please do not edit.\] 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: Partly \*\*\*\*\*\*\*\*\*\* 2\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes Reviewer \#2: No \*\*\*\*\*\*\*\*\*\* 3\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes Reviewer \#2: 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 \*\*\*\*\*\*\*\*\*\* 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: Endres et al. present a study on the role of selected yeast methyltransferases (Trms) in response to oxidative stress. They focus on trm 3, 7, 13 and 44 deletions in the context of hydrogen peroxide, rotenone and acetic acid treatment. They show impaired viability of these deletion mutants to oxidative stress as well as impaired relative increase of 2-0-Methylation (albeit the background methylation in analysed mutants is increased). Finally, they use bioinformatic approach to suggest that trm7 could play a role in tRNA- Phe(GGA) 2-0-Methylation, which can potentially influence translation of UUC codon-enriched mRNAs. Overall, it is a well-executed study and well-written manuscript. However, I do have two major points that need to be addressed: 1\. It is important to confirm deletions and overexpressions by western blot or other quantitative proteomic methods. This is especially relevant due to unexpected increase of 2-0-Methylation in trm deletion strains. One potential explanation is that deletion of one trm results in the increase of others. 2\. Figure 7 is very speculative and does not summarise the study well. The statement that 'methylation event enhances translation of transcripts that preferentially use UUC..' is unfounded. The experiments showing trm7 involvement in regulation of translation of selected mRNAs are missing. As it stands, Fig. 7 is way too hypothetical as the major message of the paper is that deletion of trm genes influence response to oxidative stress. Reviewer \#2: This manuscript by Endres et al, investigates the role of tRNA methyl transferases (Trms) in S. cerevisiae oxidative stress response and reveales that several Trms are necessary for optimal survival in response to H2O2. Most of the survival data and changes in tRNA modifications in response to H2O2 in the Trm mutant strains is sound to demonstrate that tRNA modifications are affected and important in the oxidative stress response in S. cerevisiae. However, there are some issues related primarily to the representation of data and statistical analysis that require revision, which are listed below. 1\. Figure 2: the data from supplemental figure S1 should be incorporated and quantified in Figure 2 of the manuscript, as this is an important point in regards to previous discrepancies in the literature. Moreover, the representative figure of 2A top panel suggests that the trm3 mutant is less viable than wt. 2\. The data of Figure 3B is not adequately represented. If the authors are going to statistically compare TRM-on vs TRM-off in response to H2O2 all data should be expressed relative to TRM-off-untreated. This will take into consideration any changes in survival that are elicited by TRM-on-untreated conditions. Further, ANOVA statistical analysis should be performed to compare all 4 experimental groups. It is assumed that gray shaded bars represent H2O2 treatment, but this was not indicated in the figure legend. 3\. It is assumed that Figure 4 represents averages of 3 independent experiments (according to Tables S2 and S3 in supplemental data). However, it is unclear how statistical analyses were performed to obtain p\<0.05, as stated in the legend. Error bars should show the experimental variability. Moreover, tables in supplemental data should list +/- SD to indicate variability between experimental replicates. It is unclear what the difference in statistical significance is between cyan and dark blue squares. Authors should review the legends of supplemental figures for more detail, grammatical errors, and reference to supplemental table 1 rather than S2. 4\. The authors should acknowledge that measurements of DCFDA have several caveats. As illustrated here: <https://www.ncbi.nlm.nih.gov/pubmed/22027063>. If possible, it would be advised that another method is employed to demonstrate that different strains have an altered basal oxidative stress level, and/or use negative and positive controls for DCFDA measurements. 5\. In Figure 6, the data of survival of untreated mutants should also be represented relative to wt untreated cells to show if there is significant loss of survival even in unchallenged conditions (as one would expect for genes that are of vital importance). Which of these genes are only important to stress response (ie H2O2 challenge) rather than overall survival? This could also inform selection of genes to follow up in point 6. below. 6\. In Figure 6 the authors demonstrate that several identified genes with UUC enriched codon use are also necessary for survival in response to H2O2, however the direct link to Trms is not made experimentally. The authors should demonstrate that protein expression of at least some of these target proteins are increased in response to H2O2 and that this is abrogated in response to Trm7 loss. 7\. The discussion nicely describes the role tRNA modifications may have in stress response. However, some redundant discussions of the data from the results section could be shortened. Instead the authors should elaborate on potential mechanism by which Trms sense oxidative damage. It is not discussed how H2O2 induces Trms to methylate tRNAs. Is it known if Trms are transcriptionally increased in resonse to H2O2? Or are they directly modified at the protein level to increase Trm activity, such as direct oxidation of Trm cysteines, or in response to upstream response to redox modified signaling pathways (e.g. oxidized and inactivated phosphatase?). Moreover, would it be expected that Trms identified here, such as Trm7, more susceptible to these types of redox dependent regulation? Alternatively, as discussed in the last paragraph, is it possible that tRNAs are degraded that are not methylated by Trm7, and hence the ratio of methylated to unmethylated tRNAs increases? In this case, is it possible that there is not net change in Trm7 activity in response to H2O2? This should be discussed. \*\*\*\*\*\*\*\*\*\* 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.0229103.r002 Author response to Decision Letter 0 20 Jan 2020 Reviewer \#1: Endres et al. present a study on the role of selected yeast methyltransferases (Trms) in response to oxidative stress. They focus on trm 3, 7, 13 and 44 deletions in the context of hydrogen peroxide, rotenone and acetic acid treatment. They show impaired viability of these deletion mutants to oxidative stress as well as impaired relative increase of 2-0-Methylation (albeit the background methylation in analysed mutants is increased). Finally, they use bioinformatic approach to suggest that trm7 could play a role in tRNA- Phe(GGA) 2-0-Methylation, which can potentially influence translation of UUC codon-enriched mRNAs. Overall, it is a well-executed study and well-written manuscript. However, I do have two major points that need to be addressed: 1\. It is important to confirm deletions and overexpressions by western blot or other quantitative proteomic methods. This is especially relevant due to unexpected increase of 2-0-Methylation in trm deletion strains. One potential explanation is that deletion of one trm results in the increase of others. Response: We agree that TRM expression needed to be assessed in the mutant strains, which could provide relevant information concerning potential compensatory mechanisms. To address this shortcoming, we performed qRT-PCR on reverse-transcribed cDNAs from each trm deletion mutant with TRM-specific TaqMan gene expression assays. After comparing TRM expression in the mutants to that in the wildtype we have: i) confirmed that the deletion mutants do not express their respective Trms, and ii) identified several TRMs that are up-regulated the mutant strains, suggesting that compensatory mechanisms may exist, at least at the transcriptional level. This new data appears in S1 Fig., and is discussed in the results section. These results will inform proteomic studies, which are currently limited by the availability of commercial antibodies to our TRMs of interest (i.e., only available for TRM13 from Biorbyt). We would like to note that yeast strains with TAP-tagged open reading frames (ORFs) are widely used to study the yeast proteome using anti-TAP antibodies for western blotting, which could be a factor in the limited availability of antibodies for specific Trms in yeast. Further, the TAP-tagged strains could not be used here to address compensatory mechanisms as they do not harbor the relevant TRM deletion, having instead a TAP-tagged ORF. Our lab is not equipped for antibody production; however, as antibodies for Trms 3,7 and 44 become available, we aim to perform the suggested western blots. 2\. Figure 7 is very speculative and does not summarise the study well. The statement that 'methylation event enhances translation of transcripts that preferentially use UUC..' is unfounded. The experiments showing trm7 involvement in regulation of translation of selected mRNAs are missing. As it stands, Fig. 7 is way too hypothetical as the major message of the paper is that deletion of trm genes influence response to oxidative stress. Response: This figure has been completely revised to summarize the major message of the paper that 2'-O-ribose methylation acts as a response to oxidative stress. The biological processes involved in this response (as suggested by our bioinformatic screen) are now represented as purely hypothetical. In retrospect, the model for enhanced translation of UUC-enriched transcripts was designed to guide prospective studies, reflecting where we are going and not where we are in this paper. We thank the reviewer for this sound criticism. Reviewer \#2: This manuscript by Endres et al, investigates the role of tRNA methyl transferases (Trms) in S. cerevisiae oxidative stress response and reveales that several Trms are necessary for optimal survival in response to H2O2. Most of the survival data and changes in tRNA modifications in response to H2O2 in the Trm mutant strains is sound to demonstrate that tRNA modifications are affected and important in the oxidative stress response in S. cerevisiae. However, there are some issues related primarily to the representation of data and statistical analysis that require revision, which are listed below. 1\. Figure 2: the data from supplemental figure S1 should be incorporated and quantified in Figure 2 of the manuscript, as this is an important point in regards to previous discrepancies in the literature. Moreover, the representative figure of 2A top panel suggests that the trm3 mutant is less viable than wt. Response: The previous Figure S1 is now incorporated into Figure 2. To quantitate the effects of TRM deletion on the growth of the individual mutant strains, we developed a new assay to monitor growth over time with an automated microplate reader. This quantitative growth-curve data appears in a new Figure S2, and confirms the previous reports of a slow-growth phenotype for trm7. Also, we show that the growth of trm3, 13 and 44 is also mildly impaired (i.e., being intermediate between wildtype and trm7). Together, we feel that this addresses discrepancies in the literature concerning the growth effects of trm3, and also strengthens work by adding new information about the growth phenotypes of trm13 and 44. 2\. The data of Figure 3B is not adequately represented. If the authors are going to statistically compare TRM-on vs TRM-off in response to H2O2 all data should be expressed relative to TRM-off-untreated. This will take into consideration any changes in survival that are elicited by TRM-on-untreated conditions. Further, ANOVA statistical analysis should be performed to compare all 4 experimental groups. It is assumed that gray shaded bars represent H2O2 treatment, but this was not indicated in the figure legend. Response: The absence of a figure legend indicating that the gray shaded bars are the H2O2 treated group was an unintentional omission. This figure is now corrected. For robust induction of the galactose-inducible TRM open reading frames, trm transformants were cultured first to stationary phase in raffinose (to deplete intercellular pools of glucose) followed by a switch to galactose to induce expression ("on") or dextrose to repress expression ("off"). Glucose is the preferred carbon source for S. cerevisiae; however, we noted that under gal- untreated conditions, the trm3 and trm7 mutants seemed to switch their preference to galactose, which was not the case for trm13. These results suggest that some reprogramming occurs during the carbon source switch that is not uniform among the trm mutants. Although this is an intriguing observation, it would require much further experimentation to explain taking us outside of the context of the experiment, which was to determine whether the reintroduction of TRM expression rescues H2O2-associated decreases in survival under similar growth conditions. We feel that the data as represented is best since it reflects the same growth conditions concerning the carbon source. Additional text has been added to the results section to make this point clear. 3\. It is assumed that Figure 4 represents averages of 3 independent experiments (according to Tables S2 and S3 in supplemental data). However, it is unclear how statistical analyses were performed to obtain p\<0.05, as stated in the legend. Error bars should show the experimental variability. Moreover, tables in supplemental data should list +/- SD to indicate variability between experimental replicates. It is unclear what the difference in statistical significance is between cyan and dark blue squares. Authors should review the legends of supplemental figures for more detail, grammatical errors, and reference to supplemental table 1 rather than S2. Response: We determined standard error for the fold change data, which is now added to Figure 4 as error bars to show experimental variability. During this analysis, we identified an inadvertent duplication of the fold change data for trm44 untreated and have corrected the figure and revised the results and discussion sections accordingly. The data in Figure 4 were calculated using the AvP values from supplemental tables S3 and S4, and an explanation of the statistical analysis (i.e., 95% confidence) has been added to the table descriptions, referencing the methods section. Concerning the mean AvP values in tables S3 and S4, standard deviations (+) have been added. Also added is an explanation of the difference between the values appearing in the cyan and dark blue shaded cells, with an additional color code (red) to indicate post-translational modifications detectable in some strains but not in wt untreated cells. Please note that these tables have been shifted to accommodate new experimental data. Legends have been reviewed and corrected for grammatical errors. 4\. The authors should acknowledge that measurements of DCFDA have several caveats. As illustrated here: <https://www.ncbi.nlm.nih.gov/pubmed/22027063>. If possible, it would be advised that another method is employed to demonstrate that different strains have an altered basal oxidative stress level, and/or use negative and positive controls for DCFDA measurements. Response: Agree. We have added text describing the limitations of our observations with the DCFDA stained cells. Namely, that our results only support that some redox mechanism(s) is deregulated in the trm mutants. They do not indicate the type of reactive oxygen species elevated, or provide a direct measurement of intracellular hydrogen peroxide. We undertook the DCFDA staining to provide some explanation that accounts for the increase in oxidative stress- associated tRNA modifications under unchallenged conditions. As stated, this was an interesting and unexpected result, and we feel that the results were presented this way (i.e., as a preliminary and requiring future investigation to explain). We appreciate this point because discussing the limitation of the DCFDA results further emphasizes their preliminary status. In future, both the nature and source of the ROS need to be explored using specific probes and assays (e.g., dihydrorhodamine to measure peroxynitrite, 8-Isoprostane to detect lipid peroxidation, Mito-SOX to source mitochondrial superoxide, catalase activity to measure hydrogen peroxide, etc.), which constitute in an independent investigation. 5\. In Figure 6, the data of survival of untreated mutants should also be represented relative to wt untreated cells to show if there is significant loss of survival even in unchallenged conditions (as one would expect for genes that are of vital importance). Which of these genes are only important to stress response (ie H2O2 challenge) rather than overall survival? This could also inform selection of genes to follow up in point 6. below. Response: The data in Figure 6 no longer shows survival relative to untreated individual mutants. It is now presented as percent survival relative to untreated wt in which the mutants with a loss of survival under unchallenged conditions are apparent (i.e., hxt2, hir3, and fre5). To convey the effects of hydrogen peroxide on mutant colony survival (outside of the loss of survival due to the genetic mutation), we determined the magnitude of the decrease in survival. That is the percent decrease in survival; untreated minus treated conditions. This percent decrease in survival of the mutants post-exposure is compared to wt and presented as an inset to the original figure. 6\. In Figure 6 the authors demonstrate that several identified genes with UUC enriched codon use are also necessary for survival in response to H2O2, however the direct link to Trms is not made experimentally. The authors should demonstrate that protein expression of at least some of these target proteins are increased in response to H2O2 and that this is abrogated in response to Trm7 loss. Response: We acknowledge that we have not made any direct links between the genes that are UUC-enriched and survival downstream of Trms. This point was also raised by reviewer 1. To address this problem, we have revised the model in Figure 7 to reflect the significant findings of the paper rather than our hypothetical model of Trm7-associated UUC-biased codon translation, which at this stage of our investigation is not supported by the data. Assessing protein expression of the transcripts implicated as Trm7 targets is an essential step towards linking them to oxidative stress survival in our model. We are somewhat restricted by the availability of antibodies, finding only anti- Gnp1 (ThermoFisher) and -Hir3 (Abcam), which would need to be optimized and validated for detection in yeast. Instead, we intend to use yeast strains with TAP-tagged open reading frames (ORFs) to study protein expression and are currently optimizing a western blot protocol for use with a yeast anti-TAP antibody. The caveat to this approach is that each trm deletion mutation must be introduced into the TAP-tagged strains. At this stage, this research effort is ongoing, and considering that we presented the UUC-enriched transcripts as only candidate Trm7 targets, we think that their further characterization represents the next chapter of this story. 7\. The discussion nicely describes the role tRNA modifications may have in stress response. However, some redundant discussions of the data from the results section could be shortened. Instead the authors should elaborate on potential mechanism by which Trms sense oxidative damage. It is not discussed how H2O2 induces Trms to methylate tRNAs. Is it known if Trms are transcriptionally increased in resonse to H2O2? Or are they directly modified at the protein level to increase Trm activity, such as direct oxidation of Trm cysteines, or in response to upstream response to redox modified signaling pathways (e.g. oxidized and inactivated phosphatase?). Moreover, would it be expected that Trms identified here, such as Trm7, more susceptible to these types of redox dependent regulation? Response: These are excellent questions, and got us thinking about potential upstream events that activate Trm activity after hydrogen peroxide exposure. New ideas about potential mechanisms of Trm activation have now been added to the discussion and take the place of some sections that were redundant with the results (see below). Namely, we discuss the idea that at least one eukaryote Trm, human TRM9L, is targeted for serine phosphorylation by the ERK-RSK pathway, as induced by hydrogen peroxide (Gu et al., 2018); thus 2'-O-ribose Trms may be similarly targeted for activating phosphorylation events, either by this or other stress-responsive kinase pathways. We further speculate that this effect could be augmented by the oxidative inactivation of dual-specificity phosphatases, which dephosphorylate protein phospho-serine -threonine, and -tyrosine; effects that are known to be induced by ROS (Ostman, et al., 2011). Overall, our focus in this paper was to implicate 2'-O-ribose tRNA methylation in the oxidative stress response and discuss the biological importance of this type modification from two perspectives: i) its effects on the tRNA molecule itself, and ii) its potential ability to influence the translation of genes with the potential to act in oxidative stress repair/recovery. Thus, we have not concentrated on upstream activation mechanisms, but from this input, we plan to do so now. Accordingly, we have recently requested access to the TAP-tagged strains from our collaborator to fully characterize changes in both TRM transcript and Trm protein levels after hydrogen peroxide exposure. Alternatively, as discussed in the last paragraph, is it possible that tRNAs are degraded that are not methylated by Trm7, and hence the ratio of methylated to unmethylated tRNAs increases? In this case, is it possible that there is not net change in Trm7 activity in response to H2O2? This should be discussed. Response: Yes, unmethylated tRNAs could be degraded, affecting the methylated: unmethylated ratios. In fact, our model predicts that 2'-O-ribose methylation increases tRNA stability acting in concert with anti-codon modifications (e.g., as catalyzed by Trm7) to enhance translation. So, it is certainly possible that while Trm7 activity remains the same, the pools of tRNAs targeted by Trm7 are protected from degradation, increasing the methylated: unmethylated ratio. Also, there could be de novo transcriptional increases in Trm7 target tRNAs, not requiring an increase in enzyme activity to be methylated. This scenario would also increase the methylated: unmethylated ratio. As a first step to addressing these possibilities, a global assessment of tRNA transcript expression before and after hydrogen peroxide exposure is needed. We have revised the discussion to address these excellent points in more detail. Redundant sections removed or reworded to accommodate new discussion points: \- lines 446 to 449, "Um, Cm and Gm increased in wild type cells exposed to hydrogen peroxide, a response that was attenuated in trm3, 13, and 44 mutants; however, this attenuation occurred on a background of elevated Um, Cm and Gm before exposure, a curious result that raises new questions." \- lines 456 to 458, "Initial studies of trm3 mutants did not reveal a growth impairment (42), and we confirmed that the growth of trm3 mutants is comparable to that of the wild type strain, BY4741. However, when challenged with oxidative stress (chronic and acute)," \- lines 485 to 487, "Here we suggest that under conditions of oxidative stress, Trm7-dependent tRNAPhe(GmAA)) promotes UUC translation of mRNA transcripts with a role in the oxidative stress response (Fig 7)" \- lines 493 to 496, "There are conflicting reports on whether or not trm13 mutants have impaired growth (46, 47), and the growth of trm13 mutants here was comparable to wild type cells. However, upon toxicant exposure, the growth of trm13 mutants was impaired relative to wild type cells in both plate-based and colony-forming assays." 10.1371/journal.pone.0229103.r003 Decision Letter 1 Preiss Thomas Academic Editor 2020 Thomas Preiss 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 Jan 2020 2'-O-ribose methylation of transfer RNA promotes recovery from oxidative stress in Saccharomyces cerevisiae PONE-D-19-29059R1 Dear Dr. Endres, 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, Thomas Preiss, 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 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: 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 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: All my comments have been addressed. I think it is an interesting study that deserves to be published. Reviewer \#2: (No Response) \*\*\*\*\*\*\*\*\*\* 7\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No Reviewer \#2: No 10.1371/journal.pone.0229103.r004 Acceptance letter Preiss Thomas Academic Editor 2020 Thomas Preiss 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. 4 Feb 2020 PONE-D-19-29059R1 2'-O-ribose methylation of transfer RNA promotes recovery from oxidative stress in *Saccharomyces cerevisiae* Dear Dr. Endres: 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 Prof Thomas Preiss Academic Editor PLOS ONE [^1]: The authors have declared that no competing interests exist.

# Introduction *Pneumocystis jirovecii* pneumonia remains one of the most important causes of pulmonary opportunistic infections in HIV-infected patients in the era of combination antiretroviral therapy (cART), accounting for more than half of the pulmonary complications in patients whose CD4 lymphocyte count is less than 200 cells/µL. The mortality rate of *P. jirovecii* pneumonia is approximately 10% to 12% in the HIV-infected patients. While trimethoprim/sulfamethoxazole (TMP/SMX) has been the recommended treatment of choice for *P. jirovecii* pneumonia in HIV-infected patients, TMP/SMX administered at therapeutic doses may cause several adverse effects such as allergy, hepatotoxicity, bone marrow suppression, hyperkalemia, and nephrotoxicity. Use of TMP/SMX in treatment of *P. jirovecii* pneumonia has been recently shown to cause acute psychosis in HIV-infected patients and transplant recipients and the incidence increases with the dose administered. Hepatotoxicity has been reported to occur in less than 10% of the patients who received TMP/SMX in treatment of pneumocystosis in clinical trials, and most cases of hepatotoxicity occurred on the 2^nd to 12^th day after initiation of TMP/SMX. Two mechanisms have been proposed for TMP/SMX- related hepatotoxicity: allergic response and metabolite-related toxicity. For the latter mechanism, the hepatotoxic metabolite of TMP/SMX, hydroxylamine, is produced after TMP/SMX enters the metabolic pathway through cytochrome protein 450 (CYP450) subtype 2C9 in the liver. There are two major patterns of hepatotoxicity: hepatocellular and cholestatic, with the latter being more commonly reported in previous studies,. The occurrence of hepatotoxicity may vary with the pharmacogenetics of different ethnicities enrolled in the studies, and concurrent use of multiple medications with drug-drug interactions may pose challenges in determining the culprit of hepatotoxicity. For example, concomitant use of fluconazole for oro-esophageal candidiasis is common in HIV-infected patients who present with *P. jirovecii* pneumonia, which may raise concerns because fluconazole may potentially increase the risk of hepatotoxicity when combined with TMP/SMX. In this multicenter study, we aimed to investigate the incidence of hepatotoxicity with the use of Roussel UCLAF Causality Assessment Method (RUCAM) scale ; to identify its associated factors; and to investigate the role of concomitant use of fluconazole in hepatotoxicity in HIV-infected adult patients who received TMP/SMX in the treatment of *P. jirovecii* pneumonia at referral hospitals for HIV care around Taiwan. # Materials and Methods ## Study setting and population This multicenter, retrospective cohort study was conducted at 6 hospitals designated for HIV care around Taiwan, where inpatient or outpatient HIV care, including cART and monitoring of plasma HIV RNA load and CD4 count, are provided free-of-charge. We reviewed the medical records of the HIV-infected patients, aged 20 years or greater, who presented with *P. jirovecii* pneumonia and received TMP/SMX from July 2009 to February 2013. The cases of *P. jirovecii* pneumonia were identified from the computerized databases of the participating hospitals. The diagnosis of *P. jirovecii* pneumonia was made based on identification of *P. jirovecii* by Giemsa stain of the sputum or bronchoalveolar lavage specimens, or histopathology of transbronchial or surgical lung biopsy specimens; detection of *Pneumocystis* 16S rRNA in the sputum or bronchoalveolar lavage specimens plus a typical clinical history and chest radiography that was consistent with interstitial pneumonitis; or a typical clinical history and computed tomography of the chest consistent with interstitial pneumonitis plus clinical response to TMP/SMX or clindamycin plus primaquine. The patients who did not have at least 1 follow-up liver function value within 1 month after initiation of TMP/SMX or had no baseline liver function value were excluded. The study was approved by the Research Ethics Committees of the participating hospitals (registration no. 201205032RIB) and written or oral informed consent was waived. The health records of the patients were anonymized prior to use. ## Data collection A standardized case record form was used to collect information on demographic and clinical characteristics such as age, sex, risk factors for HIV infection, smoking status, alcohol use, comorbidity (diabetes mellitus, hypertension, chronic lung disease, chronic kidney disease, malignancy, or tuberculosis), serostatus of hepatitis B and C viruses, concurrent medications including steroids, antimicrobials, and cART before and during the treatment course of TMP/SMX, weight and body-mass index (BMI) (expressed in kilogram divided by height in square meters), dose and treatment duration of TMP/SMX, complications of *P. jirovecii* pneumonia (respiratory failure or admission to the intensive care unit), and outcome. In this retrospective cohort study, we collected all the laboratory test results that were available within 1 month before or nearest to the start of TMP/SMX (baseline) and during the treatment course from the computerized databases of the participating hospitals. The data collected included total and direct bilirubin, aspartate aminotransferase (AST), alanine aminotransferase (ALT), and alkaline phosphatase (ALP), CD4 count and plasma HIV RNA load, serum creatinine and electrolyte levels, and lactate dehydrogenase (LDH). The highest values of serial data of aminotransferases and bilirubin following the initiation of TMP/SMX were used to categorize the severity (grading) of hepatotoxicity. To analyze the relationship between the acetylator genotypes and TMP/SMX-related hepatotoxicity, the available data of *NAT1* (*N*-acetyltransferase-1) and *NAT2* (*N*-acetyltransferase-2) genes were collected from 141 patients who were subsequently enrolled in a pharmacogenetic study for antiretroviral therapy at the National Taiwan University Hospital (Hung CC, unpublished data). The data were collected by the HIV-treating physicians at the 6 participating hospitals, which were rechecked by the principal investigators (Yang JJ and Hung CC) to resolve and correct the inconsistencies and errors identified. ### Laboratory investigations Plasma HIV RNA load was quantified using the Cobas Amplicor HIV-1 Monitor test (Cobas Amplicor version 1.5, Roche Diagnostics Corporation, IN) with a lower detection limit of 40 copies/mL, and CD4 count was determined using FACFlow (BD FACS Calibur, Becton Dickinson, CA). Anti-hepatitis A virus antibodies were determined using chemiluminescence immunoassay (CIA) (Abbott Diagnostics: CMIA- ARCHITECT i-2000); hepatitis B surface antigen (HBsAg), anti-HBs antibody, and hepatitis B core antibody (anti-HBc antibody) using enzyme immunoassay (Abbott Laboratories, Abbott Park, IL); and antibodies to hepatitis C virus using a third-generation enzyme immunoassay (Ax SYM HCV III, Abbott Laboratories, North Chicago, IL). *NAT1* and *NAT2* acetylator types were determined with the use of polymerase-chain-reaction (PCR) restriction fragment length polymorphism to differentiate common single-nucleotide polymorphisms (SNPs) predictive of the acetylator phenotypes. ### Definitions of hepatotoxicity Hepatotoxicity was defined as 2-fold or greater increase of aminotransferase or total bilirubin level from baseline based on the reports of international consensus meetings. RUCAM scale was used to evaluate whether the hepatotoxicity was related to the use of TMP/SMX. The RUCAM scale includes the evaluation parameters of time course, other risk factors for hepatotoxicity, concomitant medications, survey for non-drug causes, previous recorded hepatotoxicity for the drug, and response to re-administration. A score of RUCAM scale of 6 to 8 is considered as probably related, and a score of greater than 8 as highly probably related. Grading of hepatotoxicity was determined based on the Division of AIDS table for grading the severity of adult and pediatric adverse events (2004 version 1, 2009 clarification) and the grading was determined using the ratio of the follow-up aminotransferase or total bilirubin data divided by the baseline values if the baseline liver function profiles were abnormal. The table defines grade 1 through 4 hepatotoxicity as follows: grade 1, elevation of aminotransferases by 1.25–2.5× upper limit of normal (ULN); grade 2, 2.6–5.0× ULN, grade 3, 5.1–10.0× ULN and grade 4, \>10.0× ULN\]; or grade 1, elevation of total bilirubin by 1.1–1.5× ULN; grade 2, 1.6–2.5× ULN; grade 3, 2.6–5.0× ULN; and grade 4, \>5.0× ULN. The type of hepatotoxicity was categorized by the equation (R = \[ALT/ULN (upper limit of normal)\]/\[ALP/ULN\]) as hepatocellular (ALT≧3× ULN and R≧5), cholestatic (ALP≧2× ULN and R≦2), or mixed types (ALT≧3× ULN, ALP≧2× ULN and 2\<R\<5). ## Statistical analysis All statistical analyses were performed using the statistical package SAS 9.3 (SAS Institute Inc., Cary, North Carolina, USA). The Fisher exact or the Chi square tests were used to compare categorical variables and the Wilcoxon test to compare continuous variables between the patients with TMP/SMX-related hepatotoxicity and those without hepatotoxicity. All comparisons were two-tailed and a *P* value \<0.05 was considered significant. We included the variables with a *P*-value \<0.1 in the univariate analysis such as BMI, diabetes mellitus, CD4 count at the start of TMP/SMX, and presence of hyperkalemia or psychosis, and variables that were of biological significance such as possible hepatotoxic agents (fluconazole or antiretroviral agents) in the multivariate logistic regression models. # Results During the study period, 302 HIV-infected patients with *P. jirovecii* pneumonia who received TMP/SMX were enrolled. Overall, 18 patients (6.0%) were excluded: 3 with sulfonamide allergy; 10 without liver-function test results at baseline; 3 without follow-up liver-function tests after initiation of TMP/SMX; 1 without a complete medical record; and 1 receiving a suboptimal daily dose of TMP/SMX. Of the 286 treatment courses that administered to 284 patients, 152 cases of hepatotoxicity (53.1%) were detected and 47 (16.4%) were classified as TMP/SMX- related hepatotoxicity by RUCAM scale. Of those 47 cases of TMP/SMX-related hepatotoxicity, 42 (89.4%) were classified as hepatocellular type, 1 (2.1%) cholestatic type, and the other 4 (8.5%) mixed type. Twenty of the cases (42.6%) were of grade 3 to 4 hepatotoxicity, and 3 (6.4%) developed jaundice. The interval between initiation of TMP/SMX and development of hepatotoxicity ranged from 2 to 24 days (median, 11 days). Among the 47 patients, 41 (87.2%) had normalization of liver-function profiles after dose reduction or discontinuation of TMP/SMX. Twenty patients (42.6%) had to discontinue TMP/SMX and change to other regimens (clindamycin plus primaquine or dapsone), and 16 (34.0%) had to reduce the dose of TMP/SMX; the other 11 patients had been treated with TMP/SMX for at least 10 days, and decisions to discontinue TMP/SMX were made by the treating physicians. There were 15 (31.9%) and 28 patients (20.9%) in the hepatotoxicity group and control group (patients without hepatotoxicity), respectively, that required respiratory support in the intensive care unit (*P* = 0.13); 4 (8.5%) and 12 patients (9.0%) of the respective group died within 30 days of hospitalization (*P* = 0.999). The median length of hospital stay for the hepatotoxicity and control group was 18 days (interquartile range \[IQR\], 12–27) and 15 days (IQR, 10–27), respectively (*P* = 0.17). None of them developed acute liver failure with coagulopathy, and none of the deaths resulted from hepatotoxicity. shows the clinical characteristics of the patients in the control group (n = 134 episodes) and those (47) with TMP/SMX-related hepatotoxicity. Compared with the patients in the control group, patients with hepatotoxicity had a higher BMI (20.3 vs 19.5, *P* = 0.070), were less likely to have previous exposure to antiretroviral therapy (8.5% vs 28.4%, *P* = 0.006), and were more likely to have diabetes mellitus (10.6% vs 2.2%, *P* = 0.029). However, both groups had similar daily dosage of TMP/SMX (median dose, 14.2 vs 14.4 mg/kg/day, *P* = 0.479). In total, 96 and 128 of the 141 patients enrolled at the National Taiwan University Hospital had data on *NAT1* and *NAT2* genes, respectively. The proportion of patients with *NAT1* and *NAT2* genes predictive of slow acetylation phenotype was not statistically significantly different between the patients who developed TMP/SMX-related hepatotoxicity and those who did not. Concomitant medications and treatment responses of *P. jirovecii* pneumonia to TMP/SMX are shown in. Compared with patients without hepatotoxicity, patients with hepatotoxicity had a higher CD4 count at the start of TMP/SMX (median, 48 vs 29 cells/µL, *P* = 0.083), were less likely to have concomitant use of cART (55.3% vs 73.1%, *P* = 0.031) and fluconazole (51.1% vs 61.9%, *P* = 0.191), and were more likely to develop respiratory failure (29.8% vs 18.7%, *P* = 0.110) in univariate analysis. Other TMP/SMX-related adverse events are shown in. Compared with patients without hepatotoxicity, those with hepatotoxicity were more likely to experience acute psychosis (19.6% vs 11.1%, *P* = 0.046), hyperkalemia (36.2% vs 15.8%, *P* = 0.003), and hyponatremia (84.1% vs 66.2%, *P* = 0.024) during the treatment with TMP/SMX in the univariate analysis. There were no statistically significant differences between the two groups of patients in terms of age, daily dose of TMP/SMX, serostatus of hepatitis B or C virus, plasma HIV RNA load, alcohol ingestion, smoking, use of steroids, baseline total bilirubin, ALT, AST, and LDH level, mortality, development of skin rash, nausea or vomiting, and leukopenia. In logistic regression analysis, concomitant use of fluconazole for oro- esophageal candidiasis was significantly associated with a lower risk of hepatotoxicity (adjust odds ratio, 0.372; 95% confidence interval, 0.145 to 0.957, *P* = 0.040). If we included only those 20 cases of patients with hepatotoxicity of grades 3 and 4, the usage of fluconazole remained a significant protective factor in multivariate logistic regression compared with the control group, with an adjusted odds ratio of 0.262 (95% confidence interval, 0.072–0.951). The results of comparisons between the patients with all-cause hepatotoxicity and those without hepatotoxicity in multivariate logistic regression analysis are shown in supplementary, in which a heavier weight, being antiretroviral- naive, no concomitant use of ritonavir, and respiratory failure during the treatment course were associated with a higher risk of hepatotoxicity. The incidence of TMP/SMX-related hepatotoxicity decreased with an increasing daily dose of fluconazole until the daily dose reached 4 mg/kg. After excluding those patients who received fluconazole at a daily dose of greater than 4 mg/kg, the protective role of fluconazole in TMP/SMX-related hepatotoxicity was accentuated(adjusted odds ratio, 0.263; 95% confidence interval, 0.096 to 0.718; *P* = 0.009) (data not shown). Similar trends were also observed when all of the 152 cases of hepatotoxicity were included. # Discussion In this multicenter, retrospective study, we found that the estimated incidence of TMP/SMX-related hepatotoxicity in HIV-infected Taiwanese patients who received TMP/SMX for *P. jirovecii* pneumonia was 16.4%. Hepatotoxicity was mainly of the hepatocellular type and concomitant use of fluconazole for candidiasis attenuated the risk of hepatotoxicity. The incidence of TMP/SMX-related hepatotoxicity observed in this study in the HIV-infected patients with *P. jirovecii* pneumonia is higher than that observed in several randomized clinical trials that compared the efficacy between TMP/SMX and other regimens in western countries. In those studies, the incidence of TMP/SMX-related hepatotoxicity was consistently lower than 10%. The discrepancy in the incidence of hepatotoxicity may be explained by the differences in populations studied, concurrent medications, and criteria used in defining hepatotoxicity. The higher rate of hepatotoxicity in the current study might be because we used 2-fold or greater increase of aminotransferase or total bilirubin level from baselines to define hepatotoxicity and inclusion of some patients who continued TMP/SMX at a lower dose. If we use 5-fold or greater increase of aminotransferase levels (grade 3 or higher) to define hepatotoxicity, the incidence would be 7.0% in our study. TMP/SMX-induced hepatotoxicity may occur as a result of two postulated mechanisms. Hypersensitivity to TMP/SMX possibly mediated by glutathione metabolism may cause hepatotoxicity. Wang et al found SNP rs761142 in the glutamate cysteine ligase catalytic (GCLC) subunit to be associated with reduced GCLC mRNA expression. By catalyzing a critical step in glutathione biosynthesis, it was associated with SMX-induced hypersensitivity in HIV-infected patients. The metabolites of TMP/SMX may also cause hepatocellular damage or cholestasis. Hepatic *N*-acetyltransferase plays a role in transforming TMP/SMX to non-toxic metabolites, and, therefore, the rate of acetylation may contribute to the risks of TMP/SMX-related hepatotoxicity. A previous study by Smith et al has shown that HIV-infected patients with slow acetylation phenotypes had a higher incidence of adverse reactions to TMP/SMX. In an ethnic Chinese population, the frequency of the slow acetylation genotypes is about 25% , whereas it is above 50% in Caucasians. Hence the acetylation phenotype cannot explain the higher rates of hepatotoxicity observed among HIV-infected ethnic Chinese in our study compared to that previously reported by studies conducted in western countries. The alternative pathway of TMP/SMX metabolism is via the cytochrome protein 450 (CYP) 2C9 pathway, wherein sulfamethoxazole is metabolized to hydroxylamine by the CYP 2C9 pathway. Hydroxylamine plays an important role in hepatotoxicity. The frequency of poor metabolizers for CYP 2C9 in the Taiwanese population is about 8.2%, which is significantly lower than that for Caucasians (about 20%). Therefore, the likelihood of TMP/SMX being metabolized rapidly to hydroxylamine will be higher among Taiwanese, which may contribute to the higher incidence of hepatotoxicity observed in our study. Fluconazole is a potential inhibitor of CYP2C9 activity. In a previous pharmacokinetic study in healthy volunteers who concurrently received fluconazole and TMP/SMX, Mirta et al demonstrated that use of fluconazole decreased the area-under-the-curve (AUC) of hydroxylamine by 37%. Therefore, we postulate that our findings of reduced risk of TMP/SMX-related hepatotoxicity in association with concomitant use of fluconazole may be explained by reduced formation of hydroxylamine by fluconazole, though more pharmacologic studies are warranted to confirm our hypothesis. Several previous reports have shown that fluconazole is in itself hepatotoxic in a dose-dependent manner. The incidence of hepatotoxicity during the treatment course of fluconazole was approximately 10%, and only 0.7% of the patients needed discontinuation of therapy due to severe hepatotoxicity. The liver function profiles generally recovered after discontinuing or tapering the dose of fluconazole. Given the intrinsic hepatotoxicity potential of fluconazole and inhibition of CYP2C9 by fluconazole that may provide protective effects in patients receiving TMP/SMX for *P. jirovecii* pneumonia, it is not clear with respect to the appropriate dose of fluconazole that may balance the risk of potentiation of hepatotoxicity with protective effects. As shown in and, the incidence of TMP/SMX-related hepatotoxicity decreased from 31% in patients not receiving concomitant fluconazole to 17% in those receiving fluconazole at a daily dose of 2–4 mg/kg, but the trends reversed when the daily fluconazole dose increased to greater than 4 mg/kg in the patients we studied. While we do not have good explanation why the trends reversed with higher doses of fluconazole than 4 mg/kg, the potential risk of hepatotoxicity with further increase of fluconazole doses may exceed the potential benefit conferred by inhibition of formation of hepatotoxic hydroxylamine by CYP2C9. Antiretroviral therapy, especially nevirapine, could cause hepatotoxicity. However, fewer patients in the hepatotoxicity group had a history of previous antiretroviral exposure or were concurrently taking antiretroviral therapy when TMP/SMX was continued, which suggests that antiretroviral therapy might play a minor role, if any, in hepatotoxicity detected in those patients with TMP/SMX- related hepatotoxicity. There are several limitations of our study and interpretation of our results should be cautious. First, this is a retrospective study and the timing and frequency of follow-up of liver-function profiles were not standardized, which may preclude us from knowing exactly when TMP/SMX-related hepatotoxicity developed. Second, while we recorded alcohol ingestion, the amount of alcohol consumed on a daily basis and the use of concurrent medications that may cause hepatotoxicity such as statins was not recorded. In our study population, the mean age was 34 years, and, only 8.5% in the hepatotoxicity group had previous exposure to antiretroviral therapy, rendering the use of lipid-modifying agents for cART-related dyslipidemia uncommon. Third, 105 patients with hepatotoxicity were excluded from analysis and all the patients were HIV-infected ethnic Chinese, hence the findings may not be generalizable to patients of other ethnicities or non-HIV-infected patients who may have different pharmacokinetic and pharmacogenetic profiles. Fourth, the tests for acetylator types were only performed in a subgroup of patients. Therefore, the exact role of *NAT* genes could not be appropriately examined in our study. Fifth, while we used the RUCAM scale to increase the reproducibility of our findings, misclassifications may still occur because the cause of elevated aminotransferase levels in the HIV- infected patients are often multifactorial. However, a recent study has shown that RUCAM scale had similar accuracy, but better reproducibility (agreement between the observers) than the NARANJO scale, though the bias of RUCAM scale utilization should still be considered. Sixth, none of our patients underwent liver biopsy. Non-alcoholic steatohepatitis (NASH) is commonly associated with abdominal obesity and metabolic disorders for HIV-infected patients, and it could evolve to severe liver injuries including nonalcoholic cirrhosis and hepatocellular carcinoma. Last, the serum concentrations of TMP/SMX and its metabolites such as hydroxylamine were not measured and the potential drug-drug interactions between TMP/SMX and antiretrovirals were not examined in our study population. However, of all antiretrovirals, only efavirenz may inhibit CYP 2C9, which may decrease the concentration of hydroxylamine. In both groups, the proportions of patients who subsequently received efavirenz while receiving TMP/SMX were similar (21.3% vs 26.9%, *P* = 0.449) (data not shown). In conclusion, we found that the incidence of TMP/SMX-related hepatotoxicity was estimated 16.4% among HIV-infected patients who are ethnic Chinese and received TMP/SMX for treatment of *P. jirovecii* pneumonia. Concomitant use of fluconazole at a dose not exceeding 4 mg/kg was associated with reduced risk for TMP/SMX-related hepatotoxicity. # Supporting Information We would like to thank the Centers for Disease Control, Taiwan for the research grant support, Professor Victor L. Yu, Department of Internal Medicine, University of Pittsburgh, for the review of the manuscript and Aristine Cheng, National Taiwan University Hospital Hsin-Chu Branch for the English editing of the revised manuscript. **Footnote:** Preliminary analyses of these data were presented as a poster abstract no. 1055 in the *52nd Annual Interscience Conference on Antimicrobial Agents and Chemotherapy* held in San Francisco, 9–12 September, 2012. [^1]: The authors declare that no competing interests exist. [^2]: Conceived and designed the experiments: J-JY C-EL H-JT Y-HC P-LL C-CH. Performed the experiments: J-JY C-HH C-EL H-JT C-JY Y-CL K-YL M-ST. Analyzed the data: J-JY M-ST S-WL. Contributed reagents/materials/analysis tools: S-WL Y-HC P-LL C-CH. Contributed to the writing of the manuscript: J-JY C-HH P-LL C-CH.

# Introduction More than 850 million people from over 100 countries rely on the exceptional biodiversity of coral reefs, providing crucial economic, cultural, social, and aesthetic goods and services. These reefs are mostly supported by small colonial, calcifying organisms, the hermatypic scleractinian corals, which create complex three-dimensional habitats offering a variety of shelter and food for thousands of organisms. Coral assemblages often exhibit a marked spatial heterogeneity from local to geographic scales (i.e., from within and among reef habitats to among regions and ocean basins), reflecting the contrasting life history strategies and functional traits of coral species as well as the biological and physical processes that influence their biology and vary in frequency, intensity, and spatial scale. Since corals are particularly sensitive to changes in environmental conditions, reef ecosystems are highly vulnerable to both chronic and acute stressors and may change rapidly in their structure and functioning. There is concern that the frequency and severity of large-scale disturbances, such as coral bleaching events associated to thermal stress, cyclones, or outbreaks of the coral predator *Acanthaster* spp. in the Indo-Pacific, have increased over the last four decades. Humans have further contributed to this declining trend by contributing multiple, direct anthropogenic disturbances that kill reef organisms, such as overharvesting of reef organisms, destructive fishing methods, uncontrolled tourism and recreational impacts, and increased sedimentation and pollution associated with dredging, coastal development, deforestation, and intensive agriculture. In response to these threats, coral reefs have been severely impacted by widespread mortalities of foundation species and degraded habitat, in a number of cases undergoing a striking phase shift involving the replacement of corals by macroalgae or other non-reef- building benthic organisms, an undesirable state providing fewer goods and services to human populations. In the context of this “coral reef crisis”, evaluating the vulnerability, adaptability, and resilience of reef communities and their dependent human societies is urgently needed. Encouragingly, management and conservation actions have already been taken to preserve the diversity, structure, and functions of coral reefs and to maintain the goods and services they provide to humans. The Marine Protected Area (MPA) has been one of the most common tools to support the health of coral reefs and to enhance their resistance and resilience to disturbances. Although the capacity of MPAs to maintain fisheries productivity and sustainable livelihoods has been demonstrated in many cases conflicting results and uncertainties regarding their effectiveness in protecting the transformed reefs that are expected in future decades have also been highlighted. For corals, management of fishing activities within MPAs is intended to maintain a sufficient herbivory rate to remove algae that are competitive with corals, in turn enhancing coral recruitment. However, this mechanism is not always observed, and a lack of positive effects of MPAs on coral replenishment capacities has been recorded. Among the major drivers of MPA effectiveness are the duration of protection, size, and connectivity, together with the type and compliance of regulation measures. Madagascar is one of the largest islands in the world, with a coastline extending over \~14° of latitude (11° 47’ to 25° 35’ S) and, with \~2400 km² of coral reefs, is a major biodiversity hotspot in the Western Indian Ocean. The high diversity of reef organisms, including \~380 scleractinian coral species, is partly linked to the size, morphological diversity, and contrasting environmental conditions of these reefs. Studies on the diversity and structure of coral assemblages have taken place since the 1960s with the establishment of the marine research center at Toliara on the southwest coast. However, most of these studies have been restricted to either local or regional scales, or have focused on documenting the general health status of these reefs. In contrast, no multiscale analysis of spatial patterns in coral assemblages has been conducted, and the drivers of such variability remain poorly understood. As most of the world’s coral reefs, those of Madagascar have confronted large- scale disturbances and human-induced local stressors, most notably bleaching events, sedimentation, and overfishing. Consequently, coral assemblages have declined since the 1980s at several reefs around the island, as documented along the Great Reef of Toliara, though a return to a healthier coral community has recently been documented for this reef. In response to the overall degradation of coral reefs, the Malagasy authorities have implemented mitigation actions, resolutely investing in the establishment of MPAs since 1989, with the creation of the first one, the Mananara Nord Biosphere Reserve on the east coast. Twenty- two MPAs currently cover an area of 14,451 km², representing 1.26% of the Exclusive Economic Zone. Involvement of primary users, most notably fishermen and sea farmers, have been encouraged through the implantation of Locally Managed Marine Areas (LMMA) that have shown some success in effectively managing fisheries. However, the effects of these MPAs on coral assemblages have not been rigorously examined, leaving the benefit and effects of such a management tool on the overall reef biodiversity, functions, and services difficult to assess in Madagascar. In this context, the aim of this study was to analyze the spatial variation of coral assemblages at multiple scales around Madagascar. Composition, generic richness, abundance, life history strategies, and cover of coral assemblages were compared among 18 stations located at three regions around the island. We analyzed the potential influence of several environmental factors, including algal cover, substrate rugosity, herbivorous fish biomass, and geographic location on the spatial patterns of coral variables. By comparing stations in unfished areas (MPAs) and stations where fishing is allowed, we also investigated the effect of fishing protection level on coral community structure. Although this study is a snapshot in a highly dynamic system, the data set and questions addressed in this study provide a valuable contemporary baseline to understand the drivers of coral community structure and to monitor future changes of these reefs, and may also help identify effective conservation and management actions. # Materials and methods ## Study area and sampling strategy Fieldwork was conducted from March to October 2020 at three regions around Madagascar. Masoala, on the northeast coast (15°59’08” S, 50°09’27” E), is composed of fringing reefs of 0.5 to 3 km wide with several passes to the open ocean. This region includes the Masoala Marine Park, created in March 1997. Nosy-Be, in the northwest (13°24’21” S, 48°16’31” E), is composed by several small islands with narrow (\< 1 km) fringing reefs. This region comprises the Lokobe Marine Park and the Nosy Tanihely Marine Park, both created in September 2011. Salary Nord, along the southwest coast (22°33’17” S, 43°17’10” E), is a region with fringing reefs of 1.7 to 4.9 km wide, with a channel (\~5 m depth) that allows the circulation of boats. These reefs are integrated within the Soariaka MPA, created in April 2015. In recent decades, these reefs have been impacted by the bleaching event of 1998, and 2016. The major cyclones that had a potential impact on coral assemblages are Haruna at Salary Nord in 2013, Enawo 2017 at Masoala and Nosy-Be in 2017, and Belan at Nosy-Be in 2019. However, the impact of these disturbances on coral assemblages in our three regions has not been quantified. At each region, six sampling stations were surveyed, including three stations in unfished areas and three stations in areas where fishing is allowed. The location of our stations took into account logistical considerations (accessibility of sites), meteorological constraints during sampling, and the directives of local authorities (Madagascar National Park, Wildlife Conservation Society, and local managers). All 18 stations were located on the outer reef slope at \~10 m depth, where direct anthropogenic disturbances are lower, and where diversity and abundance of coral assemblages is generally higher compared to other reef habitats. Stations codes are abbreviated as follow: the first two letters indicate the region (NE for Masoala in the northeast, NW for Nosy-Be in the northwest, and SW for Salary Nord in the southwest), the number (1 to 6) differentiates the six stations from each region, and NTZ (“No Take Zone”) is for stations in unfished areas. Field work was carried out with a research permit granted by the Malagasy Ministry of Environment and Sustainable Development (75/20/MEDD/SG/DGEF/DGNRE). At each station, generic richness (GR) and abundance of coral assemblages (scleractinian corals and the calcareous hydrocoral *Millepora*) were estimated using three randomly replicated belt-transects of 10 m² (10 × 1 m), laid parallel to depth contours and separated by \~2 m. To complement the characterization of the diversity of coral assemblage composition, the Shannon diversity index (H’) was calculated at each station using log₂ and colony abundance data for each coral genus. In order to estimate the functional trait diversity of coral assemblages, we assigned one of the four life history strategies, as defined by Darling et al., to each coral colony encountered in the belt-transects: competitive, generalist, stress-tolerant, and weedy. For those coral genera that encompass species spanning several of these life history strategies, we have assigned proportional values based on life history strategy of the species typically found in the region (; for example, *Pocillopora* was estimated to include 75% generalist and 25% weedy species). The percent cover of corals, macroalgae, turf algae, and crustose coralline algae (CCA) was estimated at each station using three line intercept transects (LIT) of 10 m, placed in the middle of the belt-transects. Biomass of herbivorous fishes was estimated at each station using underwater visual censuses on three randomly replicated belt-transects of 250 m² (50 × 5 m) laid parallel to depth contours and separated by \~2 m. All herbivorous fishes were identified at the species level, counted, and measured according to their total length. Abundance values were converted to kilograms of biomass per unit area of reef (kg.ha^-1) using species-specific length-weight equations: W = a⋅L^b, where *W* is the weight (in g), *L* is the total length (in cm), and parameters *a* and *b* are species-specific constants that have been extracted from FishBase. At each station, we estimated the substrate rugosity with a visual assessment of reef topography, graded from 0 to 5 (0 = no vertical relief; 1 = low and sparse relief; 2 = low but widespread relief; 3 = moderately complex; 4 = very complex with numerous fissures and caves; 5 = exceptionally complex with numerous caves and overhangs), following Polunin & Roberts. Cover of algae (macroalgae, turf, and CCA), rugosity, herbivorous fish biomass, fishing protection level (fished *vs*. unfished areas; i.e., we did not measure fishing pressure at each station), and geographic location (belonging to one of the three regions) have been recorded to estimate their potential influence on the spatial variation of coral assemblages, and will be referred to as “explanatory variables” hereafter. ## Data analysis The spatial variation in the composition of coral assemblages was analyzed using nonmetric multidimensional scaling (nMDS), based on the Bray-Curtis dissimilarity index of the abundance of coral genera recorded at each station. The nMDS was performed using the “metaMDS” function in the “vegan” R package. The potential influence of explanatory variables was analyzed with PERMANOVA on distance matrices using the “Adonis” function in the “vegan” R package. Additionally, pairwise post-hoc multilevel comparisons were conducted to test for significant differences in the composition of coral assemblages between the three regions, using the “pairwise.Adonis2” function in the “pairwiseAdonis” R package. We used linear mixed effect models to determine which explanatory variables were influencing the spatial variation of the coral assemblage descriptors (GR, H’, total abundance, abundance of the four life history strategies, and overall coral cover). Prior to analysis, Spearman rank correlations were calculated between all pairs of explanatory variables (cover of macroalgae, turf algae, CCA, herbivorous fish biomass, and rugosity) to identify potential collinearity between these variables. As no strong correlations (-0.7 \< overall ρ \< 0.7) were recorded, all our explanatory variables were considered as independent and included in the models. All explanatory variables were centered to a mean of zero and scaled to a standard deviation of 1 in order to allow for direct comparisons of their effects. Linear mixed effect models were selected as they take into account the hierarchical structure of the data set: here, the non- independence of data points (transects) belonging to particular stations, in turn belonging to the same region. We used a multi-model information-theoretic approach to compare the explanatory variable contributions and their most important interactions in describing the variation in coral descriptors. To ensure that all model assumptions were met and to take into account overdispersion, we modeled GR, H’, abundance and cover data using a Gaussian distribution, while abundance of the four life history strategies were modeled using a Poisson distribution. Models were fitted by maximum likelihood estimation using the “lmer” and “glmer” functions in the “lme4” R package. We estimated how much of the variation in coral descriptors was explained by the variables included in the models using the marginal R² (“r.squaredGLMM” function in the “MuMIn” R package;). For each coral variable, we selected the most parsimonious combinations of fixed effects by comparing models with all possible combinations of the predictor variables using the corrected Akaike’s information criterion (AICc;). The subset of best models was selected as the ones with ΔAICc value \< 2 (the difference between each model’s AICc and the lowest AICc) using the “dredge” function in the “MuMIn” R package. All models with a ΔAICc \< 2 were considered as having similar levels of support from the data, thus belonging to the group of best models (i.e., equally parsimonious;). We used Akaike weights (wAICc), derived from the AICc, to evaluate the relative likelihood of each model given the data set and the set of models considered as well as to estimate the relative importance of each explanatory variable by summing these wAICc across the models in which they were included. Akaike weights were directly interpreted as each model’s probability of being the best at explaining the data. Subsequently, the effect sizes (lm coefficients) of the predictions of the individual models selected at the previous step were averaged using the Akaike weights as weighting coefficients. Finally, we investigated the differences among factor modalities in the mixed models using Tukey post-hoc tests (“emmeans” function in the “emmeans” R package;). All statistical analyses were performed using R 4.1.0. # Results ## Composition of coral assemblages A total of 7,638 coral colonies, representing 16 families and 40 genera (30 at Masoala, 38 at Nosy-Be, and 29 at Salary Nord), were recorded at the 18 stations. The nMDS ordination showed a substantial geographical clustering in the composition of coral assemblages. Among the seven potential explanatory variables examined, geographic location (belonging to one of the three regions) was the strongest contributor to the overall dissimilarity in the composition of coral assemblages, explaining \~27% of the variance (PERMANOVA, F = 3.825, p = 0.001; R² = 0.27) and followed by rugosity, though to a lesser degree (PERMANOVA, F = 2.444, p = 0.021; R² = 0.08;). In fact, the differences in the composition of coral assemblages between regions were all significant (pairwise post-hoc tests, Masoala × Nosy-Be, F = 2.384, p = 0.013; Masoala × Salary Nord, F = 2.817, p = 0.029; Nosy-Be × Salary Nord, F = 3.380, p = 0.006). The first two axes of the nMDS differentiated the stations of Masoala, characterized by higher abundances of *Acropora*, *Pavona*, *Hydnophora*, and *Leptastrea*, from the stations of the two other regions. An exception is station NE2-NTZ, where lower abundance of *Acropora* was recorded and which was more similar to Salary Nord stations. Most Nosy-Be stations were located in the upper part of the nMDS plot, and were characterized by the presence of *Diploastrea*, *Euphyllia*, *Isopora*, and *Tubastrea* that were recorded only in this region. Station NW4 at Nosy-Be was however more similar in its composition to several Masoala and Salary Nord stations. Stations from Salary Nord, in the right portion of the nMDS plot, were characterized by higher abundance of *Leptoseris*, *Pachyseris*, and *Stylophora* at several stations. ## Generic richness Generic richness was highly variable among stations within the regions, but the range of variation was similar among the three regions, with values from 10.00 ± 1.00 genera.10 m^-2 (mean ± se) to 21.66 ± 0.33 at Masoala, from 10.66 ± 1.20 to 19.00 ± 2.08 at Nosy-Be, and from 11.66 ± 1.76 to 20.66 ± 0.66 at Salary Nord. The selection procedure identified four models to explain the variation in GR (ΔAIC \< 2; Tables and and and and Figs), with fishing protection level and herbivorous fish biomass having the strongest contribution. The biomass of herbivorous fish was included in all four models and had the maximum value of relative importance (1.000). Models indicated that the number of coral genera per 10 m² increases by 1.33-fold with an increase of 1 kg.ha^-1 in herbivorous fish biomass. Fishing protection level also had a strong influence on the variation of GR (relative importance: 0.736), with 2.55 times more coral genera per 10 m² in unfished stations compared to fished ones. Cover of CCA and turf algae weakly predicted the variation in GR (relative importance: 0.191 and 0.154, respectively), with a positive coefficient for CCA (0.39) and a negative one for turf algae (-0.28). All other explanatory variables were not selected by the models to account for the variation in GR. ## Shannon diversity index The Shannon diversity index was also variable among stations within regions, particularly at Masoala, with H’ values from 1.54 ± 0.03 (mean ± se) to 2.85 ± 0.00, compared to Nosy-Be (1.89 ± 0.06 to 2.49 ± 0.06) and Salary Nord (1.98 ± 0.08 to 2.76 ± 0.02;). Five models were selected as the most parsimonious to explain variation of H’ (Tables and and and and Figs). CCA cover was the most important predictor (relative importance: 1.000), followed by herbivorous fish biomass (0.581) and fishing protection level (0.424), with all three variables having a positive contribution (coefficient of 0.08, 0.07, and 0.15, respectively). Rugosity was also selected in some models, but its contribution to the spatial variation in H’ was lower (relative importance: 0.118; coefficient: 0.03). ## Coral abundance Abundance of coral colonies was highly variable among stations within region, notably for Masoala, with 81.70 ± 8.09 to 345.66 ± 25.01 colonies.10 m^-2 (mean ± se), compared to Nosy-Be (96.70 ± 2.03 to 205.33 ± 31.40 colonies.10 m^-2) and Salary Nord (75.00 ± 7.77 to 144.33 ± 7.26 colonies.10 m^-2;). Three models were selected to explain the variation in coral abundance (Tables and and and and Figs). Herbivorous fish biomass had the highest relative importance (1.000), followed to a lesser degree by fishing protection level (0.250) and CCA cover (0.205). The effect of herbivorous fish biomass on coral abundance was strong and positive, with a coral abundance increase by 26.81-fold with an increase of 1 kg.ha^-1 in herbivorous fish biomass. The models also suggested that 30.68 times more coral colonies were recorded in unfished stations, whereas the effect of CCA cover on variation in coral abundance was negative (coefficient: -3.35). The abundance of the 10 major coral genera at each region, as well as other, less represented genera (“others”), was highly variable among stations and regions. *Acropora* was the most abundant coral at eight of the 18 stations, with particularly high values at NE1-NTZ and NE4 located at Masoala and, to a lesser degree, at the unfished stations of Nosy-Be (NW1-NTZ, NW2-NTZ, NW3-NTZ). Several genera were frequently recorded at certain regions while greatly less abundant in others. This is the case for *Galaxea*, which was abundant at Masoala and Nosy-Be stations but less represented at Salary Nord, and *Pocillopora*, which was abundant at Masoala and Salary Nord but not at Nosy-Be. *Porites* and *Seriatopora* were recorded at almost all stations, being among the dominant corals at several stations (NW6 and SW2-NTZ for *Porites*, NW5, SW1-NTZ, and SW5 for *Seriatopora*). Several genera were part of the 10 dominant corals at only one of the three regions, as *Favia* at Masoala, *Leptoria*, *Fungia* and *Merulina* at Nosy-Be, and *Leptoseris* and *Stylophora* at Salary Nord. Coral assemblages were dominated by stress-tolerant taxa at 11 of the 18 stations, whereas competitive taxa were the most represented life history strategy at NE1-NTZ and NE4 in Masoala and, to a lesser degree, at NW1-NTZ in Nosy-Be ( and and Tables). Generalist and weedy taxa were generally less abundant except at stations NE5 in Masoala and SW1-NTZ and SW2-NTZ in Salary Nord. Five models were selected as the most parsimonious to explain the variation in the abundance of the four life history strategies (Tables and and). Region, fishing protection level, life history strategies, herbivorous fish biomass, and the interactions of region × fishing protection level, region × life history strategies, fishing protection level × life history strategies, and region × fishing protection level × life history strategies had the maximum value of relative importance (1.000). Rugosity (0.322), macroalgae cover (0.526), turf (0.169), and CCA (0.169) were also selected by the models, but their relative importance was lower. The models underlined the higher abundance of competitive corals at some Masoala stations, the higher abundance of competitive corals in unfished stations at Nosy-Be, and the higher abundance of weedy corals in unfished stations at Salary Nord. ## Coral cover The overall coral cover was highly variable within and among regions. Higher cover values were recorded at Nosy-Be (from 42.6% ± 2.9 to 69.8% ± 4.2, mean ± se), compared to Salary Nord (from 25.8% ± 1.3 to 49.7% ± 3.0) and Masoala (from 26.5% ± 0.5 to 46.2% ± 5.6;). The spatial variation in coral cover was supported by seven models (Tables and and and and Figs). Fishing protection level and rugosity had the maximum relative importance (1.000), followed by region (0.603). CCA, turf, and the interaction region × fishing protection level had a lower relative importance (0.474, 0.200 and 0.094, respectively). The models suggested that coral cover increases by 7.04% with an increase of rugosity by 1 unit, and by 8.04% in unfished areas compared to fished ones, this effect being more pronounced at Nosy-Be (12.16% increase;). # Discussion In the context of increasing degradation of coral communities around the globe, as well as the localized threats that Madagascar is facing, our results demonstrate the occurrence of reefs with high diversity, abundance, and cover of corals, including the competitive *Acropora*, which has decreased among other coral reefs. Our study also underlines a positive effect of MPAs on most coral variables, but with varying intensity among regions. These outcomes support Madagascar as a biodiversity hotspot and offer a strong argument for the need to maintain and strengthen conservation and management actions. ## Spatial heterogeneity of coral assemblages Our survey highlights the marked spatial variability of coral assemblages in Madagascar, with variation at either or both regional and local scales for all coral descriptors. This multiscale spatial heterogeneity is consistent with several previous studies on coral reefs worldwide, and thus represents one of the major characteristics of this ecosystem. As in previous studies in the Western Indian Ocean, we found a marked regional-scale variation in the composition of coral assemblages, whereas changes at the local scale (among stations within regions) were less pronounced. Despite this regional dissimilarity in species composition, the local variation of generic richness showed a similar amplitude among regions, with up to two-fold differences among the six stations at each region. This local heterogeneity in generic richness was followed by a similar pattern for the Shannon diversity index, with a particularly high variability at Masoala. Our results also underline a marked heterogeneity at both regional and local scales in the abundance and cover of coral assemblages, with variable contribution of the four life history strategies. Masoala was characterized by the high abundance of coral colonies (81–345 colonies.10 m^-2), most notably of the competitive *Acropora* and *Pocillopora as well as* the stress-tolerant taxa *Porites* and *Galaxea* at some stations. The highest abundance values ever recorded at Masoala are greater than the one recorded during this study at Salary Nord, Nosy-Be, or at Toliara on the southwest coast (150–220 colonies.10 m^-2;). The range of coral cover (\~26–46%) that we recorded at Masoala was higher than the one reported by previous studies in this region (\~20–35% in 1998,; \~13% in 2005,). Our results support the outcomes of McClanahan, who suggested that good ecological conditions of these reefs, with a diverse and abundant coral assemblage and a high dominance of *Acropora* reported in 1996, is probably related to the moderate human impacts in this region. Moreover, the high hydrodynamic and water circulation in the northeast coast likely facilitates the larval dispersal of broadcast spawning corals such as *Acropora* and their subsequent regional recruitment. However, this hypothesis should be investigated by a connectivity survey among reefs in this region, which is currently lacking. The good ecological state of the reefs at Masoala seem nevertheless conditioned to the fact that sedimentation and nutrients from rivers do not increase in the coming years. In contrast, Salary Nord was distinguished by lower abundances of coral assemblages, notably with a depauperate population of competitive taxa, mainly *Acropora*. In contrast to the two other regions, a higher proportion of stress- tolerant taxa, such as massive *Porites*, and weedy corals, such as *Seriatopora* and *Stylophora*, were recorded at most stations. The abundance of these taxa, known to colonize and often dominate disturbed environments, supports the general consensus that coral reefs of the southwest coast of Madagascar are the most threatened of the island. This lower diversity and abundance of southwestern coral assemblages has been attributed to high-resource extraction and sedimentation. However, the range of coral cover that we recorded at Salary Nord in 2020 (\~25–49%) was higher compared to those found by Randriamanantsoa et al. in 2008 (\~21–24%). Thus, though comparisons between studies with different sampling methodologies should be taken with caution, our results suggest that Salary Nord reefs have not shown a clear declining trend in recent years. Coral assemblages at Nosy-Be were characterized by a higher diversity (38 coral genera), higher abundances of several taxa, including *Diploastrea* and *Merulina*, that were much less abundant or not even recorded at the other two regions, and also by the high coral cover (\~42–70%) recorded at unfished stations. These high coral cover values, among the highest ever recorded at Madagascar, are similar to what was found by Bigot et al. in 1999 (\~68% on the outer slope of Nosy Tanihely), but higher than those recorded by Webster & McMahon in the same year (\~28–51%). Even if they are confronted by several threats, coral reefs of Nosy-Be are generally considered to be healthier compared to most other reefs around Madagascar, with a high capacity to recover from disturbances. In fact, Nosy-Be is located within the high diversity center in the Northern Mozambique Channel. The high diversity and abundance of coral assemblages in this region is generally associated with the high connectivity among reefs generated by meso-scale eddies, and to less frequent storms. This region is therefore considered a priority area for conservation. The present study has focused on the regional and local scales and has provided a valuable contemporary baseline to determine temporal trajectories of coral assemblages, which is crucial for implementing effective management and conservation actions. Given the size of Madagascar and its diversity in environmental conditions, our survey should be reinforced with, for example, the addition of other reef habitats such as shallow back reefs and mesophotic depths, as well as other reefs around the island, notably those from less studied regions such as those in the Kirindy-Mite and Barren Isles in the west, Ambodivahibe in the northeast, and Anosy in the southeast. ## Influence of management status and environmental conditions Our results clearly underline the positive effects of fishing protection levels on most descriptors of coral assemblages. Stations localized in unfished areas (MPAs) of the three regions showed higher values of generic richness, Shannon diversity index, colony abundance, and cover. The relative contribution of the four life history strategies was also influenced by fishing protection level with, as an example, higher abundance of competitive coral taxa in unfished areas. In fact, species composition was the only component of the coral assemblages that was not related to fishing protection level and rather linked to geographic location. This “MPA effect” was particularly evident at Nosy-Be, with the highest contrast between fished and unfished stations being in terms of abundance and cover of coral colonies, with notably higher abundance of encrusting and massive coral growth forms and reduced macroalgae cover at unfished stations. Size, age, depth and connectivity are major factors in the effectiveness and success of MPAs. Here, we suggest that the relatively small size and accessibility of the MPAs are key aspects of the MPA effect at Nosy-Be, facilitating management in terms of monitoring, control, and awareness-raising. Moreover, MPAs at Nosy-Be are the oldest in this study, and this greater time of protection may also explain the highest MPA effect recorded here, consistent with other coral reefs worldwide. Moreover, with several small, scattered islands, and favorable current conditions, the connectivity in the Nosy-Be region is likely a factor that contributes to this positive MPA effect on the higher diversity and abundance of coral assemblages, as recently documented. The biomass of herbivorous fishes was also linked to the spatial variation in generic richness, Shannon diversity index, life history strategies, and abundance of coral assemblages, with higher values of these coral descriptors at stations with high biomass of herbivorous fishes. Herbivory on coral reefs is considered a key function because it often mediates coral-algal interactions in favor of reef-building corals. This expected negative correlation between herbivorous fish biomass and cover of turf and macroalgae was recorded at Masoala and Nosy-Be, but not at Salary Nord. This herbivorous-algal interaction probably explains the negative link that we also recorded between algal cover (turf and macroalgae) and the generic richness, cover, and life history strategies of coral assemblages. These outcomes reinforce the demonstrated mechanism of a high biomass of herbivorous taxa limiting the cover of algae, in turn benefitting coral assemblages, as fleshy algae compete for space and use allelopathy to reduce recruitment, growth, and survivorship of corals. This mechanism has been effectively demonstrated in several MPAs though could not be confirmed in others. Crustose coralline algae (CCA) cover was positively related to spatial variation of the Shannon diversity index and, to a lesser degree, of generic richness and cover. These relationships could be explained by the fact that some species of CCA induce the metamorphosis and settlement of larvae of several coral species, and also tend to prevent the establishment and growth of macroalgae by producing antifouling chemical signals. The fact that CCA is a composite metric with not all species inducing coral settlement likely explains the relatively weak, although statistically significant, relationship that we found here, notably for generic richness and cover. Furthermore, the negative link between CCA cover and abundance of coral colonies that we recorded is difficult to interpret. Due to the correlative nature of our analysis, this relationship does not necessarily imply a strong underlying ecological mechanism. Biochemical and habitat-specific cues not considered in this study likely influence the density of coral populations at local scales. Substrate rugosity was strongly and positively linked with coral cover and composition as well as, at a reduced level, the Shannon diversity index and the relative abundance of the four life history strategies. High rugosity typically provides a larger surface available for sessile organisms as along with more protection from environmental stressors and ecological niche space for marine organisms. For several reef invertebrates, including corals, high substrate micro-topography promotes larval settlement and recruitment by reducing mortality by predation up to a point where this factor represents a bottleneck in the replenishment of local populations. Consequently, the reduction of substrate rugosity, notably after storms, cyclones, or gleaning activities, is associated with reef degradation. The outcomes of our survey thus reinforce the positive link between rugosity and the diversity and abundance of reef invertebrates and fishes previously recorded. However, because the substrate rugosity was visually assessed in the present study, it should be interesting to examine the relationship between rugosity and the structure and dynamics of coral assemblages with the recently developed underwater 3D photogrammetry approach, which allows a much more precise measurement of the structural complexity of reef habitats. Since the interactive effects of multiple biotic and abiotic drivers govern the structure and dynamics of coral assemblages, it is evident that other environmental factors not selected in the present study could also have a major influence on the spatial patterns revealed here. Sedimentation, light penetration, nutrients dynamics, and temperature regime, all of which have been identified as structuring factors for other coral reefs in the SWIO and around Madagascar, should be considered in a future survey. ## Conclusions and perspectives One of the major results of this study is the identification of reefs with high diversity, abundance, and coverage of corals, including the competitive *Acropora*. These original outcomes are particularly unexpected in the current context of increasing degradation of many coral reefs around the world. However, these optimistic results should be tempered as our survey is only a snapshot in a highly dynamic system and especially as large-scale disturbances, such as thermally induced bleaching events and local stressors associated with human growth, are expected to increase in the near future, with possible shifts in coral assemblage distribution, diversity, and structure. Nevertheless, our survey represents a valuable contemporary baseline to document changes of these reefs and may help to determine if they will provide refuge from future environmental changes. The positive effects of MPAs on coral assemblages have been clearly demonstrated here. The outcomes of our study strongly support the implementation of MPAs where fishing should be reduced or banned, notably for herbivorous fishes, for which we demonstrated a positive role for the abundance and diversity of coral assemblages, in line with previous studies. Conservation actions should also focus on reducing all fishing activities, such as gleaning and using destructive gears, that strongly alter the reef habitat rugosity and structural complexity that is critical for the health of reef communities. Alternative sources of income, such sea cucumber and seaweed farming, that may reduce the negative effects of overfishing on coral communities, should be also encouraged. The strong spatial variability this study reveals on coral assemblages at the local scale suggests that MPAs should integrate a sufficiently larger scale to capture such heterogeneity but that, in contrast, small reserves are easier to manage in Madagascar given the reduced logistic and human resources generally allotted to them. In fact, priority should be given to the implementation of locally managed marine areas with strong primary user involvement, as has been successfully done for fisheries in the Toliara region and elsewhere, such as in Fiji. The outcomes of our study also confirm the need to maintain and strengthen existing MPAs, as older MPAs are known to be more effective in preventing coral decline. This effort should notably focus on the MPAs of Nosy-Be, as this high diversity area may act as a larval source for other “sink” reefs and may favor their replenishment and resilience. It is also timely to establish long-term interannual monitoring of the reef communities, environmental conditions, and threats at several reefs of various protection levels to precisely examine and improve the effectiveness of coral reef conservation at Madagascar. # Supporting information We thank the Centre National de Recherches Océanographiques, the Madagascar National Park, and the Wildlife Conservation Society Madagascar for their technical and logistical supports in the field. We would also thank the Malagasy Ministry of Environment and Sustainable Development, and the local managers for providing us with the necessary authorizations for field work. We gratefully acknowledge critical comments made by Jane Ballard, and two anonymous reviewers. This research is product of the JEAI ACOM and LMI Mikaroka. [^1]: The authors have declared that no competing interests exist.

# Introduction The increasing prevalence of myopia and its associated complications due to ocular stretching necessitates appropriate intervention for myopia control. Based on several studies, time spent outdoors is considered protective for myopia. Recently published systematic review and meta-analyses indicated time outdoors to be protective for myopia or to delay the onset of myopia. While Sherwin et al. reported an additional hour per week of outdoor activities can reduce odds of becoming myopic by 2%, Ho et al. suggested 120 minutes of daily outdoor exposure during school hours as the most effective intervention in controlling myopia. Studies that quantified the light exposure pattern by estimating illuminance using a light tracker reported that myopic children spent most of their time in an indoor environment with illuminance level \<1000 lux. A recent study reported reduction in odds of developing myopia when exposed to an illuminance level of \>3000 lux every day. Children at school mostly spend their time in an indoor environment such as classroom and relatively less time in an outdoor environment such as playground. There can be several variations in an indoor and outdoor environment. For example, a playground can have an open-top or will be covered by a big tree shade or a roof where children play. Similarly, a typical indoor environment such as a classroom can be with no window, a single window, or multiple windows, along with a single or multiple light sources. School corridors can be open on one side or completely closed from both sides and so are the other places where children spend their time. There are several other conditions such as weather (sunny or cloudy), geographical locations, elevations, seasons (summer or winter) and time of the day (morning, noon, afternoon or evening) that might alter the illuminance level in different locations. The direction of the sunlight towards sight of children while playing can also affect the illuminance levels reaching the eye level. For example, illuminance levels can be higher when facing the sun than that recorded opposite to the sun. The trend of using hat/cap while going outdoors to protect skin burn and other ultraviolet (UV) related damage to eye can also impact the amount of illumination reaching the eye. Dharani et al. performed a pilot study using a pendant type light meter and reported high illuminance levels in an outdoor location on a bright sunny day (278919–30311 lux) compared to that of a dark cloudy day (3896–7559 lux) and least in indoors (\<1000 lux). Lanca et al. reported that use of sunglasses/hat still provided illuminance (at the eye level) 11–43 times higher than that of indoors. The outdoor locations investigated by Lanca et al. were limited to an open field environment, under the shade of tree and street, and indoor locations were limited to a room with the fluorescent lamp and open window, and a room with cool light-emitting diode (LED) without window. Considering that multiple factors can affect light levels reaching the eye, understanding illuminance levels in different locations, where children spend most of their time would be beneficial in making appropriate recommendations while suggesting time outdoors as an anti-myopia strategy. To the best of our knowledge, quantification of illuminance levels in different outdoor and indoor locations where school-going children spend most of their time has not been studied extensively. This study investigated how ambient illuminance levels reaching the eye vary with weather (sunny/cloudy), time of a day, sensor source position (eye position in relation to source), sun protection with hat/cap and seasons (summer/winter) in 9 outdoor and 4 indoor locations. # Methods This is an experimental study conducted in the first week of June and November in the year 2019 at Hyderabad, which is the capital city of Telangana state that is in southern part of India. Illuminance levels in ‘lux’ were obtained using a factory-calibrated digital lux meter (Sinometer LX-1330B with 0 to 200000 lux output range) by a single examiner in nine outdoor and four indoor locations. To maintain accuracy of recording illuminance levels at each location, the display screen was auto zeroed by closing the cap of the sensor as recommended in the manual provided by the lux meter manufacturer. The measurements of illuminance levels were obtained by keeping the sensor of a lux meter at the level of examiner’s eye (5.6 feet above the ground) to obtain a closer value of illuminance level that enters the eye. Whereas, illuminance levels obtained in a glass bowl location were recorded with the sensor of the lux meter facing upwards. Three consecutive readings were noted for each measurement condition in all the locations and average of these readings was used for further analysis. ## Locations used in the study The description of all locations is given in along with the pictorial representation. Different measurement conditions influencing these locations are shown in. Nine outdoor locations were “open playground”, “under a translucent artificial-shade”, “under a porch facing east”, “under a porch facing south”, “under a big tree”, “between three buildings”, “within 4 buildings”, and “canopy”. As a ninth outdoor location, “Under a glass bowl” in the outdoor location was used as a simulation for “glass classroom model” and measurement was taken at the floor level only to determine in overall the illuminance conditions with glass covered on all sides. Likewise, 4 indoor locations were a “room with multiple large windows”, “room with combination light source”, “room with multiple artificial lights”, and “room with single artificial light”. ## Time of the day, seasons, and weather The illuminance level was recorded across four different time points in a day (7:00–8:00, 10:00–11:00, 13:00–14:00 and 16:00–17:00 hours) each on two sunny and two cloudy days. Measurements were obtained in both winter (November 2019) and summer (June 2019) seasons to compare the variance of illuminance level between the two most observed seasons in major parts of the country. The forecast of the day (as sunny or cloudy) was recorded from the default weather application on an android smartphone (AccuWeather application, <https://www.accuweather.com/en/in/hyderabad/202190/weather-forecast/202190>). Out of 9 outdoor locations, illuminance levels was recorded only in 3 outdoor locations during the summer season (“open playground, “under a glass bowl”, and “under a translucent artificial shade”). Indoor locations were mostly used in summer season as a comparison for classroom levels so, the outdoor locations were limited in illuminance measurements. ## Different positions of sensor relative to the source (sensor-source position) The measurements in outdoor locations were obtained by positioning the sensor of the lux meter in three different directions relative to the source of light, i.e. i) facing towards the source—TS, ii) facing opposite to the source—OS, and iii) facing intermediate to the source—IS which is facing 90 degrees midway from both TS and OS. Likewise, for indoor locations, measurements were obtained for two different positions of lux meter sensor relative to the location of artificial light source i.e. i) under the source- US (directly under the source of light with the sensor facing towards the ceiling), and ii) away from the source- AS (depending on the size of indoor location and placement of sources of light on the ceiling, measurements were obtained at a distance of 1–2 meters away). ## Measurements using sun protection To investigate the influence of using sun protection with hat or cap white in colour on the illuminance level reaching to an eye-level, all the above measurements were obtained under three conditions in the outdoor locations: i) direct exposure to sunlight, ii) after wearing a round brimmed white colour hat, and iii) after wearing a cap. ## Statistical analysis The data were recorded and analysed using MS-Excel 2016 (Microsoft Corporation, USA) and was presented as median; Inter Quartile Range (IQR) for different outdoor and indoor locations. The term “overall” was defined as the illuminance levels recorded in all locations depending on different conditions (time of the day, weather, seasons, relative sensor-source position and using sun protection with hat or cap). # Results ## Illuminance levels in outdoor and indoor locations The overall median illuminance level recorded across all outdoor locations was 8 times higher than that of the indoor locations (1175; 197–5400 lux vs. 179; 50–333 lux). Overall, the highest median illuminance levels under all the conditions in outdoor locations was recorded in an “open playground” (median: 9300; 4100–16825; maximum: 93500 lux), followed by “under a translucent artificial shade (median: 8180; 4200–13300; maximum: 79900 lux) and the lowest in “within 4 buildings” (median: 11; 6–20; maximum: 102 lux). Of all these included outdoor locations, only two outdoor locations, i.e. “open playground”, and “under a translucent artificial shade” recorded median high illuminance levels ≥ 1000 lux irrespective of different measurement conditions previously stated in. Three outdoor locations namely, “under a porch facing south”, “under a porch facing east” and “under a big tree although recorded overall median high illuminance levels \>1000 lux, even in conditions such as time of the day in sunny or cloudy weather. These locations also showed low illuminance levels (\<1000 lux) with different sensor source positions and thus lacked in maintaining high illuminance levels consistently (Figs). The median illuminance for “under a porch facing east or south” and under a big tree varied by 600 lux with overall illuminance levels ranging from 1500 to 2100 lux in these 3 locations. The “under a glass bowl” location which was used as a simulation for “glass classroom model” in outdoor location recorded illuminance levels (median: 13300; 4075–20550; maximum: 80200) similar to the range obtained in an “open playground”. Outdoor locations such as “between three buildings”, “canopy” and “within four buildings” showed overall low median illuminance levels of \<1000 lux in all conditions (Figs). Indoor locations showed an overall low median illuminance levels (i.e. \<1000 lux) irrespective of different measurement conditions. However, the overall median illuminance levels in a “room with multiple large windows” crossed 1000 lux at specific time points such as 10:00 and 13:00 hrs. on both sunny and cloudy weather conditions and showed the highest illuminance among all indoor locations (median: 179; 50–333; maximum: 28500 lux). ## Illuminance levels at different time points of a day The median illuminance levels of all the 9 outdoor locations together were \>1000 lux during 07:00–08:00 hours (1300; 200–5250 lux), 10:00–11:00 hours (1400; 10–6000 lux), and 13:00–14:00 hours (1400; 200–5650 lux), but dropped to 1000 lux between 16:00–17:00 hours (1000; 167–4900 lux). In the indoor locations, the illuminance level in “a room with multiple large windows” and “a room with combination light source” changed with the time in a day. The illuminance in a “room with multiple large windows” gradually decreased from morning 07:00–08:00 hours to 16:00–17:00 hours, but always remained \>1000 lux. Cloudy weather shifted the high illuminance levels from time point of 07:00 hrs to 10:00 hrs. Rooms with single artificial light did not show any change in illuminance levels with time in a day. ## Illuminance levels in sunny and cloudy days The overall median (IQR) illuminance level recorded between the two sunny days or between two cloudy days across outdoor and indoor locations varied by 12–13% and 1–2%, respectively. The median illuminance level of all the outdoor locations together on a sunny day was 2.4 times higher than that of a cloudy day (1920; 300–7125 lux vs. 800; 150–3825 lux). The top three outdoor locations in always showed illuminance levels \>1000 lux in both sunny and cloudy days irrespective of different conditions. However, illuminance level under porch facing east, under porch facing south, and under a big tree although showed illuminance level \>1000 lux in most of the measurement conditions during sunny days, dropped to \<1000 lux on cloudy days. In the indoor locations, there was no effect of weather except for the illuminance level at multiple large windows which was \< 1000 lux in the morning on a cloudy day (a sunny day at 07:00 hours—875 lux vs. 375 lux in a cloudy day). The median illuminance level in “open playground”, “under a glass bowl” and “under a translucent artificial shade” is 17 times higher than indoor locations in the sunny day (3,295 vs. 190 lux respectively) and 12 times higher in the cloudy days (1,150 vs. 98 lux respectively). ## Illuminance levels with sun protection, different sensor source positions and season The median illuminance level of all the 9 outdoor locations was 1.4–1.7 times higher when the measurements were obtained directly without sun protection (1600; 308–6200 lux) than that recorded with hat (1185; 219–5200 lux) or cap (900; 160–4400 lux) (Figs). The sensor source position did not affect the overall illuminance level in outdoor locations (TS: 1400; 200–5795 lux; IS: 1240; 200–5400 lux; and OS: 1145; 192–5375 lux). In the indoor locations, the median illuminance level was similar in both US (180; 54–337 lux) and AS (169; 50–333 lux) sensor source positions. Median (IQR) illuminance levels in each of the 3 outdoor locations where the illuminance was determined in summer and winter did not indicate the influence of seasons (illuminance level changing from high to low level or vice-versa) as shown in (summer vs. winter values in “Outdoor playground”: 13600 (2538–33425) lux vs. 16100 (7250–25475) lux; “under a glass bowl”: 11800 (4350–15900) lux vs. 15150 (10600–29700) lux; and “canopy”: 800 (267–900) lux vs. 210 (124–358). # Discussion This study described the median (IQR) illuminance levels in 9 different outdoor and 4 indoor locations in various conditions. High illuminance levels were recorded in an “open playground” and “under a translucent artificial shade” irrespective of the various conditions such as time of day, weather, sun protection and type of the day. The median illuminance levels in indoor locations were below 500 lux and were about 8 times lesser than outdoor locations. Outdoor locations such as “under a canopy”, “between three buildings” and “within four buildings” recorded median illuminance levels comparable to that of indoor locations. Low illuminance levels were recorded in all indoor locations and different conditions with the median illuminance levels \<500 lux in the current study, which are in agreement with the values reported previously. Lanca et al. reported the illuminance level measured at an eye level using a mannequin in an open field environment to range from 11,080–18,176 lux and was approximately 100 times brighter than indoor locations (112–156 lux) during cloudy days in Singapore. Besides, the illuminance levels were reported to vary from 5556 lux to 7876 lux under the tree. Wu et al. through a portable light tracker sensor reported illuminance of \>100000 lux in an open field, 7480 lux under a tree shade and 7600 lux in a hallway. The maximum and range of illuminance values observed in the current study (“open playground” without sun protection:1120–93500 lux, “under a translucent artificial shade”: 3200–33900 lux) or “under a big tree”: 96–15000 lux) is much higher than that reported in Singapore which could be due to differences in the type of a day (we measured illuminance level on sunny days also while Lanca et al. measured illuminance level only on cloudy days), and sensor-source position (we recorded illuminance in 3 different positions which have sensors facing towards the source, opposite to the source and in an intermediate position in 9 diverse outdoor locations). Another possible reason for higher values in the current study could be the placement of the sensor with respect to location of the eye. We manually placed the sensor in front of the eye whereas Lanca et al. placed the sensor inside the eye of the mannequin limiting exposure of sensor to light from periphery. This explanation can be supported by the findings of Dharani et al. who reported a greater range of illuminance values (278919–30311 lux) using a pendant type light meter instead of a mannequin. Few outdoor locations such as “canopy”, “between three buildings” and “within four buildings” had low illuminance levels (median: \<500 lux), possibly due to the blockage of light reaching the level of eye/sensor by solid structures/thick leaves or shade-like roof. These locations showed illuminance levels \>1000 lux only at specific time points and in a certain sensor position. Contrastingly, among the indoor locations, “room with multiple large windows” showed the highest illuminance level, and the overall median (IQR) illuminance level ranged from 350–28500 lux with median illuminance value of 2650 lux. Although a “room with multiple large windows” is considered as an indoor location, the illuminance level exceeded 1000 lux throughout the day except in the evening time (17:00 hours). This could be due to the position of the sun (which was the main source of illuminance) in relation to the room. The windows of this location were facing towards the east direction and during evening time when the sun was present in the west, illuminance level dropped below 1000 lux. The findings suggest that not all locations may be having adequate illuminance for preventing myopia onset and progression, raising the currently employed oversimplified recommendation of time outdoors as anti-myopia strategy. There are a few outdoor locations that show poor illuminance levels and at the same time a few indoor locations show high illuminance that might help in myopia control. In the realistic world and technically, for myopia control based on the illuminance, a few of the outdoor locations may be acting like an indoor and thus the definition of time outdoors for myopia control needs more specifications with regards to time of the day, duration, and the type of locations may be beneficial. While only 3 locations in this study showed median high illuminance of \>10,000 lux, three other outdoor locations namely, “under a porch facing south”, “under a porch facing east” and “under a big tree recorded overall median high illuminance levels \>1000 lux, irrespective of the time and weather condition (sunny or cloudy). Previous studies have indicated that illuminance level \>1000 lux itself might have beneficial effects in either preventing myopia or delaying the onset of myopia. In addition, Wu et al. based on finding from school-based cluster randomized trial reported that exposure to high illuminance (strong sunlight) may not be necessary for myopia prevention and instead indicated that spending greater time (close to 200 minutes per day) in locations with illuminance levels \>1000 lux may be sufficient to slow the progression or prevent myopia. Therefore the potential of any indoor, artificial or semi- natural locations that has illuminance levels \>1000 lux cannot be ruled out for myopia control in humans. In the current study, sunlight was a major source of light in all the 9 outdoor locations, and therefore the cardinal position of the sun can play a major role in various time points of a day. The positional impact of the sun at a time can be higher in areas such as artificial structures that can block sun rays. In this study, five out of nine outdoor locations were of such kind, i.e. “under a translucent artificial shade”, “under a porch facing east”, “under a porch facing south”, “between three buildings” and “within four buildings”. The other factor that can play a role in illuminance levels is the geographic location of the country in relation to the equator of the earth (i.e. location of country above or below the equatorial will influence the ambient light levels reaching the surface). This study was conducted in Hyderabad (capital of state Telangana in India with GPS coordinates: 17° N and 78° E) which is positioned north to the equatorial line and thus the extrapolation/generalizability of results from this study to other regions should be made with caution. The median illuminance levels “under a porch facing east” were lesser by 1.5 times than “under a porch facing south” although both were outdoor locations with similar size and nature of porch. This could be due to the higher exposure of sun rays which is positioned relatively towards the south direction of “under a porch facing south”. The weather conditions, seasons (cloudy day), time of the day (07:00 hours and 16:00 hours), sensor source position (away from the source), and sun protection with cap or hat although showed a reduction in the illuminance levels, did not shift any location from being high illuminance level category (\>1000 lux) to low (\<1000 lux), which are in agreement with the previous studies. While the difference in illuminance levels between the two sunny or two cloudy days in outdoors varied by a small margin of 12–13%, a greater variation in illuminance levels between sunny and cloudy day was noted. The illuminance levels in outdoor locations on a sunny day were 50% higher than that on a cloudy day. The outdoor location indicated median low illuminance levels on a cloudy day (800 lux). However, “Open playground” and “under a translucent artificial shade” always showed illuminance level \>1000 lux in both sunny and cloudy days irrespective of different measurement conditions. Given that parents may raise a question that on which time of a day is better to be outside and if the light levels are high in specific time to ensure if myopia can be prevented with time outdoors, we have chosen to record measurements at four-time points which includes closer to timings in before or/and after school hours. The findings of this study indicate that time of the day after 7.00 hours and before 17:00 hours will still have light levels \>1000 lux in the southern state of India and children might be benefitted by spending time outdoors before and after school hours. Considering the eye and skin related problems due to direct sun exposure and the finding from the current study indicating lesser variation in illuminance level due to use of any sun protection with hat or cap (illuminance with and without sun protection differed by 1.4 times), it should be noted that spending time in outdoors with sun protection during the day through hat or cap may not interfere with myopia prevention activity based on outdoor light exposure. While the current study did not investigate the influence of sunglasses on illuminance, Lanca et al. indicated that combination of hat and sunglasses in the evening could reduce the light levels at the eye and showed values similar to that of indoors and thus one should choose the outdoor location wisely so that the impact of sun protective gear (hat/cap with sunglasses) on illuminance will be minimal. Considering the indication for bright classroom model with high illuminance as an anti-myopia strategy, we have also measured the illuminance “under glass- bowl” simulating the outdoor glass room to check how illuminance level will be different to that of outdoor conditions. Zhou et al. developed a bright classroom which measured the median (IQR) illuminance level of 2,540 lux (1,330–4,060 lux). The use of de-polished (light-diffusing) shatterproof clear glass with blinds to the walls could have led to relatively reduced median (IQR) values then that reported here. The intention was to show that illuminance “under a glass bowl” (median (IQR): 13300; 4075–20550; maximum: 80200 lux) will be similar to that of the measurements obtained in an “open playground”. Given that children spend most of their time in a school classroom during the day, modifications to classrooms might be worth to provide better ambient light to prevent myopia. The modified classroom can expose elevated light levels and spectra closer to the outdoor locations. In consideration the cost of building glass classroom in reality, the option of an indoor classroom with multiple large windows could be considered to ensure children get exposed to ambient light levels of \>1000 lux. The greatest strength of this study is that we investigated illuminance levels in 9 outdoor and 4 indoor locations which are common/regular places where children are likely to be spending their time. Different conditions (four-time points in a day, sun protection, summer and winter seasons, and sunny and cloudy weather) were included while recording illuminance levels to improve understanding of variation in outdoor and indoor environments with regards to the ambient lighting. Illuminance levels recorded at these locations were at an eye level with three relative sensor positions (lux meter) considering a child might face different directions with respect to light sources while doing daily activities. This study was limited by the following conditions: -i) data of only three outdoor locations were reported during the summer season, while both summer and winter season are equally experienced in India. ii) illuminance level was recorded only in Hyderabad city, which is in the southern part of India, but the geographical spread of India is wide where the weather and season can vary with different states. For example- the Himalayan range in the northern region experience lesser clouds compared to other regions. Likewise, states near the northern border of India may experience more foggy weather and blocks sunlight in the winter season as compared to the southern states. Further investigations are warranted for reporting illuminance levels in different parts of India. The current study cannot make recommendations about locations that can be beneficial for myopia given that this study aimed to document how illuminance varies across different locations and conditions but did not investigate its effect on myopia. In conclusion, it should be noted that not all outdoor locations may provide adequate light exposure for myopia prevention. Based on the illuminance levels recorded with subject to all conditions, the investigated outdoor locations can be labelled as high illuminance regions that recorded illuminance \>1000 lux consistently, moderate illuminance regions simulating semi-outdoor locations where the illuminance level varied between being indoors to outdoors and low illuminance locations that recorded illuminance levels \<1000 lux in a majority of environmental conditions. It is worth highlighting that illuminance levels reported in the study did not vary with sun protection, time of the day, weather or seasons and thus children should be encouraged to spend time outdoors with sun protection even in the morning or the evenings. Keeping in view of the variation in illuminance in different locations and other environmental factors, it should be noted that the children and parents need to be wisely provided with more details while recommending time outdoors as anti-myopia strategy which are otherwise oversimplified currently. # Supporting information The authors thank the administration of L V Prasad Eye Institute (Hyderabad) for permitting the use of various sites in the campus to execute this study. [^1]: The authors have declared that no competing interests exist.

# Introduction The circadian clock possesses two core properties. The first one is the endogeneity of circadian rhythms, and consequently the existence of a well- defined natural period, which expresses itself under constant environmental conditions. The second property is the capability to synchronize to a periodic external Zeitgeber by establishing a precise phase relation to it. This paper applies mathematical modeling to connect these two properties and to explain in a systematic way how the timing between the clock and its Zeitgeber is determined by clock and Zeitgeber parameters. Without external time cues, which are called Zeitgebers, a clock runs with its characteristic period, which may deviate from 24 h. On the other hand, many Zeitgebers on the Earth are quite precise and possess a “dian” period of exactly one day. If the strength of the Zeitgeber is capable to overcome the period mismatch, the Zeitgeber enforces the periodicity of 24 h in the clock. This situation is called synchronization or entrainment. The range of the period mismatches for which synchronization occurs is called entrainment range. In fact, it is the *difference* of both periods and (but not the single periods *per se*) that determines whether the clock would be synchronized or not for a given Zeitgeber strength. In a synchronized situation, the intrinsic clock adopts a stable phase relation to the Zeitgeber. The difference between the phase of the Zeitgeber and the phase of the clock attains a well-defined value called “phase of entrainment”. A large body of research has been accumulated on the phase of entrainment, since it is critical for the coordination of daily rhythms in physiology, metabolism, and behavior. Variations of the phase of entrainment has been extensively studied in the past decades. From an evolutionary perspective, the phase of entrainment is a parameter under selection and, hence, a tight regulation is expected. The goal of this paper is to study systematically how the phase of entrainment depends on the clock and Zeitgeber properties. We approach this problem by theoretical studies of analytically tractable models, supported by numerical simulations. Our main result relates the phase of entrainment to the period mismatch, Zeitgeber strength, and the range of entrainment. A cornerstone of this theory is the “180° rule” asserting that within the range of entrainment, the entrainment phase can vary over a range of about 180°. A similar behaviour of entrainment phase has been already noticed in earlier studies. Here we explain how the 180° rule can be rigorously derived from three quite different modeling approaches. We additionally discuss the applicability of the rule and point out its limitations. The 180° rule implies that clocks with a narrow range of entrainment exhibit high sensitivity of entrainment phase to the mismatch between endogenous and Zeitgeber periods. We argue that this high sensitivity of entrainment phase contributes to the wide range of chronotypes found in humans. Furthermore, our conceptual framework provides insight how midnight-locked, dusk-locked, and dawn-locked entrainment phases can be obtained. Our theory predicts also the dependence of entrainment phase on Zeitgeber strength (see for experimental data). On the theoretical side, our results bridge discrete and continuous approaches to circadian entrainment. Both time-discrete, PRC/PTC-based models and time- continuous models for the phase of clock lead essentially to the same dependencies of entrainment phase on oscillator and Zeitgeber parameters. ## Mathematical Model A hierarchy of mathematical models has been developed to describe circadian oscillators ranging from van der Pol oscillators to comprehensive biochemical models. Basic features of entrainment, however, can be studied with the help of simple yet generic amplitude-phase models. We assume that the oscillator of interest is characterized by its amplitude, its intrinsic period, and its stability with respect to amplitude perturbations. These general oscillator properties can be parameterized by the Poincaré oscillator, which is given in polar coordinates with radius and phase by Here, the parameters and denote amplitude and intrinsic period of the oscillator. The ratio is the angular frequency of the oscillator. The parameter quantifies the relaxation of perturbations to the stable limit cycle with. Small values of correspond to slow and large values of to fast amplitude relaxation to the stationary value. Two characteristic values of were used: for “strong” oscillators and for “weak” oscillators, see below. We have shown recently that oscillators with large amplitude and large relaxation rate exhibit narrower entrainment ranges. Such oscillators will be termed “strong” in this paper (compare). We will show below that strong oscillators are characterized by quite flexible entrainment phase. # Results ## Entrainment Phase for Strong and Weak Oscillators The framework that we are going to present can be applied to the circadian clock on the organismic level as well as in different tissues. In both cases, a large range of entrainment phases has been reported. Recently, temperature cycles were described as a universal entrainment clue for different tissues. It has been shown that the mammalian pacemaker, the suprachiasmatic nucleus (the SCN), is quite resistant to phase resetting. Contrarily, pituitary and lung cultures were found to be easily entrainable. The low and high susceptibility of circadian oscillators to Zeitgebers leads to the concept of strong and weak oscillators. Weak oscillators have a wide entrainment range, tolerating large mismatches to Zeitgeber periods. Lung tissues, as an example of a weak oscillator, can be entrained by alterations between 37°C and 35°C using Zeitgeber periods of 20 and 28 h. In contrast, the suprachiasmatical nucleus is quite entrainment-resistant, and has, consequently, a narrow range of entrainment. The distinction of strong and weak oscillators can be also applied to clocks on the organismic level. For example, Aschoff and Pohl in suggest that vertebrates exhibit narrow entrainment ranges compared to unicellular organisms and plants. These differences might be due to differential responsiveness to the entrained signal and/or due to different feedback couplings. Mathematically, strong oscillators are characterized by large oscillation amplitudes, fast relaxation of perturbations, and small-amplitude PRCs. As discussed above, the value of the period mismatch determines whether the clock would be entrained or not. Thus we can on equal footing study mismatches due to varying endogenous period (different phenotypes) as well as experimental variations of Zeitgeber periods. Positive mismatches result, e.g., from driving the circadian oscillator with a short Zeitgeber period h. Negative mismatches correspond to. For example, a patient with the Familial Advanced Sleep Phase Syndrome (FASPS) had a short endogenous period of h leading to a mismatch of h. shows the phase of entrainment as a function of the mismatch for strong and weak oscillators. The points and mark the borders of the entrainment range. Our simulations reproduce the observations that the entrainment phase increases with mismatch. This implies, for example, that organisms with a short endogenous period have earlier activity phases. For a given endogenous period, we find early peaks for Zeitgeber period longer than 24 h. Both strong oscillators (left) and weak oscillators (right) reproduce the expected increase of with mismatch. An obvious difference is the steepness of the increasing function. The strong oscillator (left) exhibits a high sensitivity of the entrainment phase to changes of the mismatch. In the following, we relate this finding to properties of the oscillator and its Zeitgeber. The slope of the curves in is determined by the width of the entrainment range and the vertical phase span. The vertical variation is in both cases about 12 h. The entrainment range, however, is quite different. Consequently, the slope of the curve is nearly reciprocal to the width of the entrainment range. In other words, strong oscillators with a narrow entrainment range exhibit a quite flexible entrainment phase. This strong dependence of the phase sensitivity on the entrainment range is based on the observation that both oscillator types exhibit a 12 h range of entrainment phases. A 12 hours variation corresponds to 180° for periods of about 24 hours. Our simulations in suggest that a 180° variation of phase within the entrainment range is a common feature of periodically driven oscillators. Empirical data from many organisms support the theoretical prediction that the entrainment phase spans a range of about 180°. We derive below this “180° rule” using discrete and continuous models of entrainment. ## Is there a 180° Rule? The most striking result of the previous section is that for both strong and weak oscillators, the phase can vary over the same range of about 180° across the entrainment range. Here, we provide an explanation for this fact based on three different approaches: phase response curves, the Kuramoto equation for the phase of entrainment and a forced linear damped oscillator. All three approaches result in similar phase flexibility under variation of the period mismatch. Note that these considerations are applied to oscillators with relatively narrow entrainment ranges, characteristic for many vertebrates including humans. ### Phase response curves Phase response curves (PRCs) are an established theoretical tool in chronobiology. For a circadian oscillator, a PRC describes the amount of phase shift, caused by a short Zeitgeber pulse in dependence on when the Zeitgeber pulse is applied. In we plot numerically obtained PRCs of the strong and weak oscillators from. In both cases, there are positive and negative phase shifts: During the first half of the circadian cycle, a short Zeitgeber pulse instantaneously advances the clock, whereas in the second half a Zeitgeber pulse delays the clock. The strong oscillator (on the left) has a PRC with a smaller amplitude compared to the weak one. Being phase variables, phase shift and the relative position of the pulse can be specified either in hours or in degrees or radians. In order to entrain stably an oscillator by a -periodic (i.e. with a period of hours) sequence of Zeitgeber pulses, each pulse in the sequence must result in a delay or an advance that would compensate the period mismatch. This requirement conditions the entrainment phase. Pulses must occur at the phase where the corresponding value of the PRC equals the mismatch. Thus, the largest mismatches that can be compensated correspond to the maximum and the minimum values of the PRC. For the PRC of the strong oscillator in (left), we can compensate mismatches of up to h. Negative mismatches of (clock faster than Zeitgeber) can be compensated by aligning the phases of the clock and the Zeitgeber in such a way, that the Zeitgeber pulse comes approximately at 270°. This allows for the desired phase delay of h, compensating for the faster clock. Analogously, for a clock slower than the Zeitgeber (mismatch positive), the phases of the clock and the Zeitgeber must be aligned such that the light pulse arrives at approximately 90°, thus advancing the phase by about h. The same logic applies to the weak oscillator in (right). Here the PRC suggests that light pulses can compensate for larger mismatches up to approximately h. We have now illustrated that the maximum and the minimum of the PRC correspond to the maximum and minimum mismatches, which can be compensated in entrainment, i.e. they correspond to the borders of entrainment range. The corresponding phases of pulses thus give the entrainment phases at the borders of the entrainment range. Both PRCs in are nearly sine-like, which implies that the maximum and minimum are nearly 180° apart and, consequently, the phase of entrainment can vary over a range of 180°. This is exactly the result from we were seeking to explain. The 180° rule will be modified if the PRC is not well approximated by a sine- curve. Still, the maximum and minimum of the PRC determines the range of the entrainment phases. In “” section, we provide mathematical details regarding the phase response curve/phase transition curve approach. We emphasize that rodent and human PRCs resemble sine-curves for long durations of light pulses. On the other hand, experimental PRCs based on short light pulses often clearly deviate from a sinusoidal shape. Thus, a more general theory of phase response is required to deal with natural light-dark cycles, as described below. ### Kuramoto phase equation A serious limitation of the aforementioned PRC approach is the assumption that the Zeitgeber acts as a sequence of short pulses, each of them instantaneously introducing a phase delay or advance to the oscillator. For a continuously varying Zeitgeber, such as the daily variation of light intensity, this might be not the best representation. Fortunately, there is another theoretical tool which allows to overcome this limitation: The so-called Kuramoto equation describes the phase difference between the oscillator and its Zeitgeber for arbitrary Zeitgeber waveform. The Kuramoto equation accounts for the Zeitgeber waveform by integrating its product with the PRC, compare Eq. (6). The integral produces sine-like waveforms even for PRCs deviating strongly from a sine-curve. As outlined above and more detailed in “” section, the phase of entrainment is essentially determined by the compensation of the mismatch by the effective Zeitgeber strength. This implies that the Zeitgeber strength determines the allowed range of mismatches, for which entrainment occurs. This agrees with the intuition that larger Zeitgeber strength implies broader entrainment range (see next section). The exact expression for entrainment phase from the Kuramoto theory iswhere is the frequency mismatch between the Zeitgeber frequency and endogenous frequency. The parameter represents the effective Zeitgeber strength. For the phase of entrainment, Eq. (2) has the following consequences. First, it limits the choice of frequency mismatches where entrainment is possible to the interval of (recall that the -function is defined for arguments with the absolute value less than one). This results in the Arnold tongue, which is discussed in detail in the next section. Secondly, the unique values of the function span a range of 180°, which supports our 180° rule. The extrema of the entrainment phase are assumed at the border values of, or, equivalently, in. There, the function produces values of ±90°, correspondingly. Curiously, for a sinusoidal Zeitgeber, the expression for phase of entrainment obtained from the Kuramoto model coincides with the one for an oscillator with sinusoidal PRC under -periodic sequence of light pulses. The explanation for this fact can be obtained by considering a continuous sinusoidal Zeitgeber as a superposition of many short pulses with a sinusoidal amplitude. A generalization of PRC theory leads to an integral of the product of PRC and the Zeitgeber, see section “”, compare also. When any of those two are sine-functions, the integral produces another sine-function, whose maximum and minimum are necessarily 180° apart. This again results in a 180° range of entrainment phase within the entrainment range. The Kuramoto theory can be thus also used to describe and predict the response of a circadian oscillator with a non-sine PRC to a periodic Zeitgeber. ### Periodically driven damped oscillator It has been shown that linear damped oscillators can be adequate models for the circadian rhythms in single cells. This simple model allows an exact calculation of the phase difference between the external driving force and the oscillator, see “” section. When the period of the driving Zeitgeber is close to the intrinsic period of the oscillator, a resonance occurs – an increase of the amplitude of the driven oscillations. At the same time, the phase difference between Zeitgeber and the oscillator shows an interesting behavior. For a slow Zeitgeber (with), the phase difference is nearly zero – the oscillator and the Zeitgeber are in phase. In other words, the forced oscillator can follow the slow driver without any delay. On the other hand, if the Zeitgeber is fast compared to the oscillator (with), the phase difference approaches 180° and the oscillator lags in antiphase behind the Zeitgeber. Again, between a fast and a slow Zeitgebers the phase difference can vary over a range of 180°, which nicely agrees with the aforementioned 180° rule. For periodically driven damped oscillator with the eigenfrequency and damping rate, the phase difference between driving force with frequency and oscillator is given by The function produces values in the range of \[0°,180°\]. The lower limit 0° is achieved when the argument of the function becomes close to zero, i.e. when Zeitgeber is much slower than the endogenous clock. If, on contrary, is much larger than, the argument of the function becomes negative and assumes the value of 180°. This is exactly our 180° rule – the phase difference varies over a range of 180°. Even though a driven linear oscillator has not an entrainment range in a strict sense, there is still a 180° phase difference between the oscillator response to slow and fast Zeitgeber periods. In summary, three different approaches indicate that for strong oscillators, a phase range of about 180° can be expected within the entrainment range. This is not a strict rule since deviations from sinusoidal waveforms can cause variations of the 180° range. ## Entrainment for Variable Zeitgeber Strength So far, we have considered Zeitgebers with a constant strength. A Zeitgeber can vary in its strength, like the amount of perceived daylight over different seasons or on different geographical latitudes. The Zeitgeber strength, among other factors, determines the phase of entrainment. The main result of this section is that the 180° rule approximately holds for any Zeitgeber strength. As Zeitgeber strength increases, entrainment range spreads over larger mismatches and so does the observed 180° flexibility of entrainment phase. Plotting the entrainment range versus increasing Zeitgeber strength produces a wedge-shaped entrainment region, see, termed “Arnold tongue” (referring to the Russian mathematician V.I. Arnold). As intuitively expected, small Zeitgeber strength can overcome only small mismatches. With increasing Zeitgeber strength, the Arnold tongue opens up and entrainment ranges over several hours can be observed. In, we plot the phase of entrainment inside the Arnold tongue by different grades of blue. Close to the borders of the tongue, the entrainment phase assumes values of about h. For small Zeitgeber strength, the phase of entrainment can vary a lot for small period mismatch changes due to a narrow entrainment range. This implies a high sensitivity of entrainment phase to mismatch. Contrarily, for large Zeitgeber strength, the entrainment range is large and the sensitivity of entrainment phase to mismatch is lower. The effective Zeitgeber strength scales reciprocally with the oscillator amplitude. Thus, the interaction of a given oscillator with a relatively small Zeitgeber strength is similar to the behavior of a strong oscillator. On the other hand, a strong Zeitgeber acting on a reference oscillator resembles a weak oscillator. Near the tip of the Arnold tongue where the entrainment range is narrow, the situation corresponds to entrainment of a strong oscillator. For large Zeitgeber strength, the entrainment range is wider, bearing a similarity to a weak oscillator. Based on the Kuramoto equation, we show in “” section that the phase of entrainment depends only on the ratio between normalized Zeitgeber strength and mismatch. The condition for a constant entrainment phase can be expressed as In coordinates “Zeitgeber strength – mismatch”, this is an equation for a straight line, going through the origin. This finding nicely corresponds to the numerically obtained, showing the lines of constant phases (isophases) as nearly straight lines, originating at zero Zeitgeber and mismatch. The above result implies that the phase of entrainment remains unchanged, if both Zeitgeber strength and mismatch are varied proportionally to each other. ## Entrainment Phase Flexibility and Chronotypes We can use our previous theoretical results to explain how small variations in endogenous periods can result in large variations in chronotype. The endogenous period in human populations has been found to be fairly precise with a standard deviation of about 12 minutes. On the other hand, the mid-sleep time associated with entrainment phase has been found to have a much larger variability of h. It has been shown in that the phase of entrainment is strongly determined by the intrinsic period. As an extreme example of high entrainment phase sensitivity, a patient with Familial Advance Sleep Phase Syndrome (FASPS) with a moderate mismatch h has a 4 hour phase advance. A question arises: How is it possible that a few minutes faster clock advances activity rhythms by more than an hour? An answer is given by inspection of (a): for oscillators with a narrow entrainment range the slope of the function is quite large. Consequently, small differences in lead to large changes of the entrainment phase. For an entrainment range of h we obtain, for instance, a slope of 3 due to the 12 h range of the entrainment phase. As an approximation, we can use the relationconnecting period changes to variations of the entrainment phase. The numerator is fixed to be about 12 h according to the 180° rule. Consequently, the denominator governs the slope. A narrow entrainment range leads to large slopes of the function. from the Aschoff survey summarizes the experimentally found relations between the phase of entrainment and the range of entrainment in 15 species. From that survey slopes between 0.5 (for unicellular organisms) and 4.5 (for birds) can be extracted. We illustrate our explanation of the relation between and chronotype in. We simulated two ensembles with strong and weak oscillators, correspondingly, with a Gaussian distribution of with the mean of 24 h and a standard deviation of 0.2 h. For strong oscillators (left) we obtain a wide range of entrainment phases whereas a weak oscillator (, right) leads to a narrow distribution of. The standard deviation of entrainment phase of strong oscillators was found to be nearly three times larger than for weak ones. Thus, for strong oscillators including the human clock, small variations in lead to highly variable entrainment phases. # Discussion ## Flexibility of Entrainment Phase We provide a conceptual framework how oscillator properties control the entrainment phase. Strong oscillators with a narrow entrainment range exhibit more flexible entrainment phases. This implies that small variations of the endogenous period lead to different phases, associated with different chronotypes. The core of our theory is the “ 180° rule” formulated already by Wever : under general assumptions, the entrainment phase varies within the entrainment range by 180° (or 12 h given a period of 24 h). This rule is rigorously derived using 3 different approaches: (i) phase response curves, (ii) phase model (Kuramoto equation) and (iii) resonance theory. Numerical simulations of limit cycle models as in and confirm the 180° rule. As discussed in “”, Eq. (6), the integral of the product of Zeitgeber and PRC determines the phase dynamics. Even for non-sinusoidal PRCs, the 180° rule can hold, as long as the integral in Eq. (6) produces a sine-like function. Indeed, already Aschoff and Pohl suggested that the 180° rule is approximately valid in many organisms, even though most PRCs deviate from a sinusoidal shape. ## Generalizations of the Theory Our central) that connects phase variations to period changes and the entrainment range is a linear approximation of the functions. Inspection of reveals that there are clear deviations from linearity. Sigmoidal functions have been reported earlier. Our calculations predict curved relationships near the borderlines of the entrainment range. The theory of driven damped oscillators leads directly to sigmoidal functions. Sinusoidal PRCs and Kuramoto phase equation result in functions resembling the simulation results in the left graph of. Consequently, our theory is also capable to explain deviations from a linear function , which have been experimentally found. In, we have implicitly assumed that the intrinsic period is not changing with Zeitgeber strength. It is known, however, that light intensity may change the endogenous period. This would lead to skewed Arnold tongues. Furthermore, seasonal variations implicitly change Zeitgeber strength and induce period and entrainment phase variations,. These issues will be addressed systematically in a forthcoming study. ## Theoretical Predictions As visualized in, our theory predicts the dependence of the entrainment phase on mismatch and Zeitgeber strength. For, an increase of the Zeitgeber strength leads to earlier entrainment phases. This effect has been shown recently in temperature entrainment studies. The phase of entrainment of lung tissue was decreased by 3 h by increasing Zeitgeber strength (compare Figure S5). Comparable predictions relate light intensity and chronotypes. For night owls with we expect that more light leads to earlier phases. Considering the left half of, large Zeitgeber strength leads to later phases for. Altogether, the theory predicts that stronger Zeitgebers lead to narrower distributions of chronotypes. This prediction can be tested comparing chronotype distributions from different countries or in summer and winter. ## Evolutionary Aspects of Entrainment Phase Flexibility Temporal coordination of physiology and behavior with extrinsic Zeitgebers leads to evolutionary benefits of a properly functioning circadian clock. In particular, a suitable phase of entrainment most likely provides selective advantages. It is by no means obvious whether or not a stable entrainment phase as observed for weak oscillators or a flexible phase for strong oscillators is advantageous. For example, bird and butterfly navigation can be supported by a phase marker that is precisely aligned to Zeitgeber phase. If an organism can track noon independent of season and latitude, the highest point reached by the sun always defines south direction in the northern hemisphere. In many other situations a flexible phase of entrainment is required. The prime example is the adaptation to seasonal changes. Frequently the circadian clock allows to track dusk and dawn. This can be achieved partly by masking, but the observation of activity peaks tracking sunrise and sunset indicates successful adaptations of the entrainment phase to seasonal changes. Across many organisms narrow and wide ranges of entrainment are observed. This implies that the slopes of the function varies from 0.5 (unicellular organisms) up to 4 (vertebrates) since the “180° rule” associates entrainment ranges with the phase variability described by the slope. A wide entrainment range corresponds to a high responsiveness to Zeitgeber clues. For example, the clock of unicellular organisms is heavily influenced by inputs. In mammals, a hierarchy of circadian oscillators exists. The suprachiasmatic nucleus (SCN) receives direct light input and orchestrates peripheral clocks. The SCN is a quite strong oscillator with a relatively narrow range of entrainment. This has been shown recently for temperature as a universal Zeitgeber, but applies also to light inputs since phase response curves have typically relatively small advance and delay portions. In contrast, peripheral clocks have large PRCs and wide entrainment ranges. These observations might reflect a design principle of mammalian clocks: the pacemaker subject to noisy inputs is quite strong whereas peripheral clocks constitute weak oscillators that can be easily entrained by neuronal and humoral signals or body temperature. The theory of entrainment phase control, as discussed in this paper, emphasizes that the robustness of the SCN implies flexible entrainment phases due to large slopes of the function. The wide spread of chronotypes reflects this large phase flexibility. Moreover, the sensitivity of the entrainment phase allows adaptation to seasons and latitudes. In a forthcoming study we will show how oscillator properties rule seasonal adaptations as observed in the classical work of Daan and Aschoff. # Methods – are based on simulations of the amplitude-phase oscillator Eq. (1) with. Periodic forcing is simulated by adding the term to the -coordinate of the oscillator with Eq. (1) re-formulated in the Cartesian coordinates. The parameter is termed “Zeitgeber strength” in. The PRC in is obtained with pulses of strength lasting 30 min. The entrainment phases visualized in are based on simulations with varying period mismatch and Zeitgeber strength. For each parameter set, 24 initial conditions were chosen equally distributed along the limit cycle. The median entrainment phase after 50 days was plotted using contour lines. Numerical simulations were performed in MATLAB (2007a, The MathWorks, MA, Natick). Below we consider generic models of periodically driven oscillators. We present calculations showing that under quite general assumptions, the entrainment phase varies over about 180° (or 12 h for circadian periods). The derivation of this “180° rule” points also to limitations and generalizations of this finding. ## Sinusoidal Phase Response Curve In many cases the effects of external driving forces can be studied successfully using Phase Response Curves (PRCs). If the relaxation of a perturbation is fast compared to the period of the oscillator, PRCs can be used to calculate the phase of entrainment for periodical stimulations. This discrete approach to the entrainment of circadian clocks is mathematically related to stable fixpoints of the associated Phase Transition Curve (PTC). For a sinusoidal PRC we find the following PTC :Here the phase variable is normalized to the range, where is the endogenous period. Further, denotes the period of the Zeitgeber and the parameter represents the effective Zeitgeber strength. The upper graphs in show a sinusoidal PRC and its associated PTC for the frequency detuning and Zeitgeber strength. The PTC can be considered as an iterated map, i.e. subsequent perturbations lead to a series of iterated phases. Fixpoints of this map are given by the intersection with the diagonal. A stable fixpoint (e.g. the full circle at h) corresponds to the entrainment phase. Increasing the frequency mismatch leads to a loss of entrainment, since there are no more intersections of the PTC with the diagonal. This happens at the marked point via a saddle-node bifurcation (see for details). The disappearance of a stable fixpoint marks the borderline of the entrainment region (compare). Similarly, negative values of the frequency mismatch lead to a loss of entrainment at point. The critical entrainment phases at and correspond to the extrema of the PRCs in, since the PTC is essentially a rotated version of the PRC. For sinusoidal PRCs, these extrema are half a period (or, equivalently, 180°) apart. This implies that the phase entrainment varies by 12 h as illustrated in. Note that – are obtained from simulations of amplitude-phase oscillators with sinusoidal periodic forcing. Consequently, the PTC-theory discussed above is only approximately valid. Thus, the entrainment range is larger and more asymmetric than the advances and delays in in the main text. A more general approach of periodically driven limit cycle has been developed by and will be discussed below. ## Entrainment Phase of Periodically Driven Limit Cycles Kuramoto studied extensively the phase dynamics of driven oscillators. His theory is intimately related to the continuous approach of Aschoff. The following differential equation for the phase difference between the periodic force and oscillator can be derived:Here, has the same meaning as the frequency mismatch between the Zeitgeber and the oscillator as above. The integral represents the cumulative action of the Zeitgeber as “sensed” by the PRC. If PRC or Zeitgeber are sinusoidal, the integral can be computed analytically and the above equation simplifies to what we know nowadays as the Kuramoto equation: The equation looks quite similar to the PTC discussed in the previous section. Again the extrema of the sine-function (180° apart) correspond to the borderline of the entrainment range. A stationary phase determined by the condition equalswithin the entrainment range. This implies that the entrainment phase varies again by 180° within the entrainment range. The -function resembles the simulation result in, left graph. In the middle, its slope is approximately given by, i.e. for small entrainment ranges large slopes occur. Due to the periodicity of Zeitgeber and PRC, the integral in Eq. (6) is a -periodic function in phase. A special case of certain shapes of Zeitgeber and PRC is worth to be discussed separately: If one of the functions or PRC contains no higher harmonics (i.e. is either a pure sine or a cosine function or a linear superposition of such), the integral would contain no higher harmonics either. As a consequence, the 180° rule would hold exactly in this case. Under a milder assumption that both Zeitgeber and PRC do contain higher harmonics, the integration would suppress them. In comparison to the PRC approach, the Kuramoto theory provides a more general framework for entrainment. The entrainment phase can be determined for different Zeitgebers and PRCs. According to the above calculation, any combination of Zeitgeber and PRC that produces a phase equation similar to Eq. (7) implies the 180° rule. In summary, the phase equation according to Kuramoto leads to a similar conclusion as the PRC approach. Narrow entrainment ranges imply strong dependences of the entrainment phase on the mismatch. ## Periodically Driven Damped Oscillator Even though the circadian clock has been proven to be a self-sustained oscillator, weakly damped oscillations exhibit somewhat similar features when periodically driven. Amplitudes increase via resonance and phases vary with the mismatch between intrinsic angular frequency and driver frequency. For a linear oscillatorthe amplitude and the phase of the driven oscillation can be calculated shows the corresponding functions. For slow external frequencies, the forced oscillator is in phase with the driving force, whereas it lags behind it by 180° for fast drivers (with high). This model shows large phase shifts near the resonance resembling entrainment phase variations of driven limit cycle oscillators. It has been shown recently that weakly damped oscillators are fairly good approximations of circadian single cell rhythms. If the amplitude shows resonance behavior as observed in lung tissues and skin cells, the theory of weakly damped oscillators can serve as a reasonable approximation. We do not claim that the “180° rule” derived here can be considered as a universal law. Obviously, deviations from a sinusoidal PRC shape, phase resetting by large pulses, and periodically driven relaxation oscillators require a more general theory. Nevertheless, the “180° rule” is supported by quite different mathematical approaches (PTC, Kuramoto’s phase equation, resonance theory) and can be successfully applied to strong oscillators with type-1 PRCs and narrow entrainment ranges. These assumptions are reasonable for many vertebrates including humans. We thank Ute Abraham, Pål O. Westermark, Ueli Schibler, and Anna Erzberger for stimulating discussions. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: HH AK. Performed the experiments: AG GB. Analyzed the data: AG GB. Contributed reagents/materials/analysis tools: AG GB. Wrote the paper: AG GB HH AK. [^3]: Current address: Department of Systems Biology, Harvard Medical School, Boston, Massachusetts, United States of America

# Introduction Ubiquitination is a posttranslational modification that is involved in most aspects of cell biology. It controls the activity of proteins by promoting proteasomal or lysosomal degradation or by modulating the activity of targeted proteins. Many neuron-specific processes are affected by ubiquitination including axon outgrowth, synapse formation and elimination, and synaptic transmission. Ubiquitin ligases catalyze the final step of the ubiquitination reaction. Several hundred predicted ubiquitin ligases are encoded in the genomes of multicellular organisms, and many ubiquitin ligases appear to have a small number of specific target proteins. Determining the function of specific ubiquitin ligases is crucial to understand how ubiquitination controls neuronal function. However, a large fraction of predicted ubiquitin ligases has not yet been studied. SCF ubiquitin ligases are a subfamily of ubiquitin ligases (SCF stands for three subunits, Skp1, cullins, and F-box proteins). The F-box subunit mediates specificity of ubiquitination by direct interaction with target proteins. The genomes of *C. elegans*, flies and mammals encode about 520, 20 and 100 F-box proteins, respectively. Only eight of these F-box proteins appear to be evolutionarily conserved from *C. elegans* to mammals. Two of the conserved F-box proteins, FSN-1 (FBXO45 in mammals) and LIN-23 (β-TrCP), have roles in axon outgrowth, synapse formation and regulation of glutamate receptors in *C. elegans* –, and homologs of FSN-1 in flies (DFsn) and mammals are involved in axon outgrowth, synapse formation and synaptic transmission. SEL-10 (FBXW7) functions in the developmental elimination of synapses in *C. elegans*, and the mammalian F-box protein Scrapper (FBXL20) ubiquitinates the active zone RIM1 to promote degradation of RIM1 and control neurotransmitter release. Thus, conserved F-box proteins have diverse functions in the nervous system. In *C. elegans*, mutations in the conserved F-box protein *mec-15* (FBXW9) result in defects in mechanosensation and synapse formation in touch receptor neurons. We further investigated the neuronal functions of MEC-15. *mec-15* is widely expressed in the nervous system, including both cholinergic and GABAergic motor neurons. Using behavioral, electrophysiological and imaging approaches, we found that MEC-15 promotes neurotransmitter release from GABAergic, but not cholinergic, motor neurons, possibly by controlling the abundance of SNB-1 at synapses. # Results ## Absence of MEC-15 in GABAergic Motor Neurons causes Behavioral Defects To study the role of MEC-15 in the *C. elegans* nervous system, we obtained an allele of *mec-15(tm2691)* with a 352 base pair deletion at the N-terminus that is predicted to result in a frame shift and early stop codon and behaves like a complete loss of function of *mec-15*. We tested *mec-15(tm2691)* mutants in a behavioral assay that measures the rate of paralysis of animals in the presence of the acetylcholine esterase inhibitor aldicarb. When exposed to aldicarb, wild-type animals paralyze over a time course of 1–2 hours due to accumulation of acetylcholine in the extracellular fluid and subsequent permanent contraction of all body wall muscles. Changes in the rate of paralysis in this assay can result from changes in synaptic transmission in cholinergic or GABAergic motor neurons innervating body muscles. For example, mutant animals with reduced release of acetylcholine paralyze more slowly since extracellular accumulation of acetylcholine takes longer. Conversely, mutant animals with increased release of acetylcholine paralyze faster. Mutant animals with defects in GABA release also paralyze faster in the aldicarb assay since inhibition of muscle contraction by GABA signaling is reduced. We found that *mec-15* mutants paralyze faster than wild-type control animals. To determine in which cells MEC-15 acts, we expressed *mec-15* specifically in GABAergic motor neurons of *mec-15* mutants using the promoter of the *unc-25* gene. Expression of this transgene in *mec-15* mutants completely rescued the fast paralysis, suggesting that MEC-15 acts in GABAergic motor neurons to affect GABAergic synaptic transmission. Next, we tested if *mec-15* is normally expressed in GABAergic motor neurons. We expressed GFP driven by the *mec-15* promoter (*Pmec-15*::GFP) in animals co- expressing the red fluorescent protein mCherry in GABAergic motor neurons with a cell-specific promoter (*Punc-25*::mCherry). Most, if not all, GABAergic motor neurons expressed *Pmec-15*::GFP. In addition, P*mec-15*::GFP was expressed in many non-GABAergic motor neurons in the ventral nerve cord, indicating that *mec-15* is also expressed in cholinergic motor neurons. P*mec-15*::GFP was also expressed in neurons in the head and tail as well as in other tissues as reported previously. ## GABA Release is Reduced in *mec-15* Mutants To confirm a specific function of MEC-15 in GABAergic neurons, we measured synaptic activity at the NMJ in dissected animals by patch-clamp recordings from body wall muscles. In these dissected animals, basal nervous system activity drives endogenous excitatory and inhibitory postsynaptic currents. *mec-15* mutants had a significantly lower endogenous IPSC rate than wild-type animals. Importantly, this defect is due to a role of MEC-15 in GABAergic motor neurons, since it could be rescued by expressing *mec-15* specifically in these neurons using the *unc-25* promoter. The amplitude of endogenous IPSCs was not affected. Together, these results suggest that release of GABA is reduced in *mec-15* mutants while the GABA content of synaptic vesicles and muscle responsiveness to GABA are normal. In contrast, both the rate and amplitude of endogenous excitatory postsynaptic currents (EPSCs) were similar in wild-type and *mec-15* mutants. These results are consistent with the results from the aldicarb assay and suggest that MEC-15 controls GABA release in GABAergic motor neurons, but does not affect cholinergic synaptic transmission. ## Changes in the Abundance of the Synaptic Vesicle Protein SNB-1 (Synaptobrevin) at GABAergic Synapses and in Cell Bodies in the Absence of MEC-15 Neurotransmitter release is controlled by a variety of mechanisms. To begin to elucidate how MEC-15 affects synaptic transmission at GABAergic synapses, we determined the abundance of the GFP-tagged synaptic vesicle protein synaptobrevin (SNB-1-GFP) stably expressed in GABAergic motor neurons. In wild- type animals, SNB-1-GFP localizes to punctate structures in the dorsal nerve cord that correspond to presynaptic sites in GABAergic motor neurons. In *mec-15* mutants, the density of presynaptic puncta was not affected, but the fluorescence intensity of SNB-1-GFP at presynaptic sites was significantly reduced. SNB-1-GFP is also diffusely localized between the punctate structures. This pool of SNB-1 is in the plasma membrane of neurites as part of the synaptic vesicle cycle. The diffuse neurite fluorescence of SNB-1-GFP was also significantly reduced in *mec-15* mutants compared to wild-type animals (1+/−0.024, n = 26 and 0.74+/−0.027, n = 34, for wild-type and *mec-15* mutants, respectively, p\<0.001). The reduced fluorescence intensity of SNB-1-GFP at punctate structures could be rescued to wild-type levels by expressing *mec-15* specifically in GABAergic motor neurons, indicating that MEC-15 functions in these neurons to control SNB-1-GFP abundance at synapses (rescue 1). We also analyzed a second rescue line (rescue 2) that showed a significant increase of SNB-1-GFP fluorescence compared to wild-type, possibly due to overexpression of MEC-15 in this line. We also analyzed if the synaptic abundance of SNB-1-GFP in cholinergic motor neurons was affected in *mec-15* mutants using transgenic animals expressing SNB-1-GFP specifically in cholinergic motor neurons under a promoter fragment of the *unc-129* gene. In contrast to the reduced synaptic abundance of SNB-1-GFP in GABAergic neurons, we found that neither SNB-1-GFP fluorescence nor presynaptic density was affected in cholinergic motor neurons, consistent with the results from the aldicarb experiments and electrophysiological recordings. Changes in the abundance of SNB-1-GFP at synapses could result, among other things, from changes in protein expression, trafficking or degradation of SNB-1. Decreased trafficking could result in accumulation of SNB-1-GFP in GABAergic cell bodies. Therefore, we measured fluorescence intensity of SNB-1-GFP in the cell bodies of two GABAergic neurons, DD5 and VD10. In *mec-15* mutants, SNB-1-GFP abundance was significantly increased in these cells bodies, consistent with the idea that MEC-15 controls the trafficking of SNB-1 between cell bodies and synapses. To determine if the distribution of other synaptic proteins was affected in *mec-15* mutants, we analyzed the synaptic vesicle protein RAB-3 and the active zone protein SYD-2 (liprin-α) in GABAergic neurons. Fluorescently-tagged RAB-3, SYD-2 and SNB-1 co-localize at punctate structures that correspond to presynaptic sites. The fluorescence intensity of punctate structures of RAB-3-GFP and SYD-2-GFP was not changed in *mec-15* mutants, and there was no change in the densities of these punctate structures. Together, these data suggest that overall formation of GABAergic synapses occurs normally in *mec-15* mutants, and that MEC-15 specifically affects the distribution of SNB-1. # Discussion We analyzed the evolutionarily conserved and predicted SCF ubiquitin ligase subunit MEC-15 (FBXW9) in *C. elegans* to gain a more complete understanding of the role of specific ubiquitin ligases in the nervous system. Our results indicate that MEC-15 controls GABAergic synaptic transmission, possibly by regulating the abundance of the synaptic vesicle SNARE protein SNB-1 at synapses. Four lines of evidence support this. First, MEC-15 is expressed in GABAergic motor neurons in the ventral nerve cord. Second, *mec-15* mutants paralyze faster than wild-type animals in the aldicarb assay. This effect is rescued by expressing *mec-15* specifically in GABAergic motor neurons, suggesting that the fast paralysis is due to reduced GABA release in *mec-15* mutants. Third, electrophysiological recordings show a reduced endogenous IPSC rate in *mec-15* mutants, and this is rescued by expressing *mec-15* in GABAergic motor neurons. In addition, IPSC amplitude was not affected, indicating that signaling in postsynaptic muscle cells is not affected in *mec-15* mutants. Fourth, the abundance of the synaptic vesicle protein SNB-1 at GABAergic synapses, but not the density of synapses, is significantly reduced in *mec-15* mutants. Since the density of SNB-1-GFP punctate structures is not altered in the absence of *mec-15*, the decreased IPSC rate of *mec-15* mutants is unlikely caused by reduced synapse numbers, but instead could be due to reduced synaptic abundance of SNB-1. Reduced SNB-1 abundance could, in principle, result from defects in synapse assembly. This is not very likely since the density and abundance of two other presynaptic proteins, RAB-3 and SYD-2, is not changed in *mec-15* mutants. Instead, we found that SNB-1-GFP accumulates in cell bodies in *mec-15* mutants. Together, this suggests that MEC-15 affects the abundance of SNB-1 at synapses by controlling trafficking of SNB-1 between cell bodies and synapses; for example, export of SNB-1 from cell bodies could be reduced in the absence of MEC-15. The normal synaptic abundance of another synaptic vesicle protein, RAB-3, could hint that, unexpectedly, trafficking of SNB-1 and RAB-3 might be achieved by distinct mechanisms. In contrast to the function of MEC-15 in GABAergic neurons, we did not detect a role for MEC-15 in cholinergic motor neurons. The fast paralysis in the aldicarb essay was efficiently rescued by expressing *mec-15* in GABAergic neurons indicating that potential changes in cholinergic motor neurons do not significantly contribute to this phenotype. Furthermore, endogenous EPSC rate and amplitude as well as abundance of SNB-1-GFP at cholinergic synapses are unchanged in *mec-15* mutants. Since MEC-15 is expressed in cholinergic neurons, it may have a function in these neurons that is not revealed in our experiments. *mec-15* mutants were previously found to have diverse phenotypes in touch receptor neurons, including defects in synapse formation, enlarged cell bodies and reduced touch sensitivity. Touch receptor neurons in *mec-15* mutants had reduced numbers of synapses in the ventral nerve cord, reduced accumulation of GFP-tagged RAB-3 in the nerve ring, and enlarged cell bodies accumulated more GFP-tagged RAB-3. In contrast, synapse densities in GABAergic and cholinergic motor neurons are not changed in *mec-15* mutants and fluorescence intensity of RAB-3-GFP is similar to wild-type animals. Furthermore, defects of touch receptor neurons in the absence of MEC-15 are modulated by mutations in two tubulin genes that function specifically in these neurons. Together, this could indicate that the function of MEC-15 in touch receptor neurons is distinct from the function in GABAergic and cholinergic motor neurons. For example, MEC-15 could have different ubiquitination targets in these three types of neurons. Alternatively, it is possible that the distinct phenotypes of *mec-15* mutants in these neurons are the result of the same underlying function of MEC-15 but differences in redundant or compensatory mechanisms. Since chemical synapses in touch receptor neurons are not required for the touch response, the reduced touch sensitivity in *mec-15* mutants did not allow conclusions about functional changes of chemical synapses. Here, we provide behavioral and electrophysiological evidence that MEC-15 promotes synaptic transmission in GABAergic motorneurons. MEC-15 interacts with SKR-1 (Skp1), a common subunit of SCF ubiquitin ligases, in a yeast two-hybrid assay, supporting a function as part of an SCF ubiquitin ligase. Definitive evidence will require biochemical analysis of MEC-15 and identification of ubiquitination targets. This will also be crucial to further define how MEC-15 functions in the nervous system. Since MEC-15 (FBXW9) is one of a few evolutionarily conserved F-box proteins, it should be informative to explore the function of FBXW9 in neurons of other species. # Materials and Methods ## Strains, Mutants, Transgenes and *mec-15* Constructs Animals were maintained at 20°C and fed OP50 *E. coli* as described. The wild- type reference strain was N2 Bristol. Other strains and transgenes used in this study (strain information can be found at <http://www.wormbase.org>): *mec-15(tm2691)* was kindly provided by Shohei Mitani and outcrossed four times, *yuEx30* (P*mec-15*::GFP), *yuEx24* (P*unc-25p::mec-15*), *juIs1* (P*unc-25*::*SNB-1*-GFP), *hpIs3* (P*unc-25*::GFP-*syd-2*), *wyIs202* (P*flp-13*::GFP-*rab-3* and P*flp-13::*mCherry), *nuIs152* (P*unc-129::snb-1-*GFP). The expression pattern of *mec-15* was determined using a 750 bp fragment from the start ATG of the *mec-15* gene to the 3′ end of the upstream gene as a promoter. This promoter was fused to GFP by PCR and directly injected into transgenic animals stably expressing mCherry in GABAergic neurons driven by the *unc-25* promoter (*yuIs10*) as described. To rescue the *mec-15* mutant phenotypes, *mec-15* was amplified by PCR from cDNA and ligated into the *unc-25* promoter construct pSC325 (pIY73). Sequences of these constructs and PCR primers can be obtained upon request. Transgenic strains were generated by injecting N2 or *mec-15(tm2691)* animals with expression constructs (10–25 ng/µl) and the co-injection markers P*ttx-3*::GFP (40 ng/µl) or pKP1368 (P*myo-2*::NLS-DsRed, 5 ng/µl). Microinjections were performed using standard techniques as previously described. ## Aldicarb Assay For analysis of sensitivity to the inhibitor of acetylcholinesterase aldicarb, paralysis of young adult worms was scored every 10 min, starting at 30 min, using 1 mM aldicarb (Chem Services), as previously described. For each experiment, 20 worms per genotype were placed on NGM plates supplemented with aldicarb. Genotypes were blind to the scorer, and the analysis was repeated at least three times. ## Fluorescence Microscopy Fluorescence microscopy experiments were performed as previously described. Briefly, animals were immobilized with levamisole (10 mg/ml; Sigma) before imaging. Image stacks of young adult animals were acquired and maximum intensity projections generated using Metamorph software (Molecular Devices). Images of synapses were taken from the dorsal cord near the posterior gonad bend with the dorsal cord oriented towards the objective. Linescans of dorsal cord fluorescence were generated with Metamorph and analyzed in Igor Pro (WaveMetrics) using custom-written software to determine the fluorescence intensity, density and width of presynaptic puncta. Images of GABAergic cell bodies DD5 and VD10 were acquired in a similar way, but with the ventral side of animals oriented towards the objective. Cell body fluorescence was determined using Metamorph by measuring fluorescence from cell bodies and a directly adjacent area as background. In all experiments, images of animals with different genotypes were acquired in parallel. Images of 20–30 animals were used for quantification. Values reported in the figures are means ± SEM. Statistical significance was determined using the Student’s t test. ## Electrophysiology Electrophysiology was done on dissected adults as described, and recording conditions were as described previously. Briefly, worms were superfused in an extracellular solution containing 127 mM NaCl, 5 mM KCl, 26 mM NaHCO₃, 1.25 mM NaH₂PO₄, 20 mM glucose, 3 mM CaCl₂, and 3 mM MgCl₂ (330 mOsm at pH 7.2), bubbled with 5% CO₂ and 95% O₂ at 20°C. For endogenous acetylcholine EPSCs, whole-cell patch-clamp recordings from body wall muscles were carried out at −60 mV using an internal solution containing 105 mM CH₃O₃SCs, 10 mM CsCl, 15 mM CsF, 4 mM MgCl₂, 5 mM EGTA, 0.25 mM CaCl₂, 10 mM HEPES, and 4 mM Na₂ATP (315 mOsm, adjusted to pH 7.2 using CsOH). For endogenous GABA IPSCs, whole-cell recordings were carried out at 0 mV. Statistical significance was determined using Student’s t test. We would like to thank Josh Kaplan for his generous support and advice, Alison Frand for discussions and comments on the manuscript, and Mei Zhen, Kang Shen and Shohei Mitani for kindly providing strains. The Caenorhabditis Genetics Center provided some strains used in this study. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: YS LD. Performed the experiments: YS ZH. Analyzed the data: YS ZH YG LD. Wrote the paper: LD.

# Introduction Substance use disorders (SUD) have been declared one of the top public health priorities in the United States (U.S.), with 185 SUD-related deaths, on average, each day in 2018. SUD disorders are considered a subgroup of the addiction diseases that are deemed as mental health conditions in which a person repeatedly uses substances or engages in behaviours with the knowledge of their harmful consequences. In the U.S., it is estimated that one in five people aged 12 years or older used an illicit drug, and 8.1 million had an illegal drug use disorder in 2018, with 67,367 reported deaths by drug overdose in the same year. Overall, the U.S. mortality rate related to SUD reached 20.7 deaths per 100,000 inhabitants in 2018, with West Virginia (51.5), Delaware (43.8), Maryland (37.2), Pennsylvania (36.1), Ohio (35.9), and New Hampshire (35.8), having the highest mortality rates at the state level. Several studies have examined multiple characteristics of the addiction epidemic in the U.S. These studies have reported a significant increase in mortality rates from 2010, with its highest peak in 2017, and a considerable demographic and spatial heterogeneity of the epidemic being attributed, in part, to the uneven distribution of several demographic and socioeconomic factors and health comorbidities across the country. However, previous studies have not fully explained the reasons behind the unequal spread of the SUD epidemic and there remains a need to reduce the high level of SUD-related mortality rates in the country. As a result, several sociological studies have suggested the need for implementing a socio-ecological framework to conceptualize the drivers of addictive behaviours according to their level of influence in order to design effective strategies. These studies highlight the importance of the interconnection between individual and broader social and environmental domains as essential to understanding the SUD epidemic. Within this framework, individual, family, neighborhood, and community-level attributes have been identified as potential drivers of the current SUD epidemic. Furthermore, our preliminary study conducted in Ohio identified different spatial and demographic distributions associated with the opioid overdose deaths in the state, such that the epidemic is concentrated in specific demographic groups and locations, with multiple spatial and temporal sub-epidemics emerging at distinct time periods. Successful approaches like the “Know your epidemic, know your response” framework implemented for counteracting the malaria and HIV epidemics worldwide have resulted in mitigation policies that shifted from intervention strategies (i.e. vaccines, medical treatment) to targeted prevention plans (i.e. modifying behavioral response of individuals). The core of the “Know your epidemic, know your response” approach is the identification of the environmental, socioeconomic, and demographic drivers of an epidemic. These drivers become the cornerstones of the design and implementation of prevention measures that target vulnerable populations under their unique social, environmental, and epidemiological circumstances. Moreover, “Know your epidemics, know your response” approach highlights the role of the individual awareness of the risk in the ability to respond with appropriate mitigation strategies, allowing to focus on education efforts and mitigation of risk factors, more than in allocating resources for intervention policies. Similar to malaria and HIV, addiction disorders are characterized by complex spatial hierarchical structures caused by multiple concurrent sub-epidemics of different intensities among different populations. However, in the case of SUD-related mortality rates, the link between community-level factors and risk of death is not well understood. In addition, the vulnerable populations suffering the highest burden of the SUD epidemic driven by specific socioeconomic characteristics and comorbidities are still not well characterized. Epidemiologic research to resolve these complexities should address the spatial and hierarchical nature of the epidemic to estimate associations between individual- and community-level attributes and SUD-related mortality. Against this background, we used data from the U.S. Centers for Disease Control and Prevention (CDC) on individual mortality from 2005–2017 to analyze the demographical, spatial, and temporal structure of the SUD epidemic and its associated risk factors in the U.S. In accordance with the “Know your epidemic, know your response” approach, the aim of this study is two-fold: (i) to identify and characterize the demographic groups at highest risk of death by SUD, and (ii) to describe the spatial and temporal dynamics of the SUD epidemic in the U.S. We aimed to identify the key demographic factors associated with the epidemic, and the vulnerable populations and places where the burden of the epidemic is concentrated. A nationwide description of the epidemic would assist in the design and implementation of targeted policy interventions for addiction mitigation campaigns through an understanding of the spatial variability and epidemiological profiles in the U.S. # Research methods ## Data sources description, sampling, and demographic analysis Data were provided by the CDC from restricted-use vital statistics micro-data files for the period of January 2005 to December 2017, which is the latest available mortality data at the time of the analysis. Available data included the date and county of death, demographic characteristics of individuals (sex, race, age, marital status, and educational level) and the International Classification of Diseases, 10^th Revision (ICD-10) code for the cause of death. We extracted information about drug overdose deaths for individuals aged 5 to 84 years from ICD-10 codes for unintentional substance poisoning. Monthly death rates by county were computed as the ratio of the number of SUD deaths to the number of total deaths and were scaled by 1,000. Community-level factors related to health behaviours and physical and mental health at the county level were retrieved from the County Health Rankings & Roadmaps program from 2010 to 2017. These covariates corresponded to social and health risk factors that have been associated with SUD in previous studies at the community level. We included the self-reported number of days per month under physical and mental distress, excessive adult drinking, and tobacco consumption from the Behavioral Risk Factor Surveillance System (BRFSS). We also included the percent of children living in poverty and the population without health insurance in each county as potential socioeconomic factors associated with the SUD epidemic. In addition, from the complete data set provided by the CDC, we performed stratified random sampling with strata given by year and state of death occurrence to avoid requiring excessive computational resources for regression analysis. Finally, SUD death rates by demographic groups were visualized using time series graphs and heat maps to describe the temporal dynamics of the SUD epidemic from 2005 to 2017. We computed death rates by race, gender, and age group to determine the groups most affected by the epidemic. Demographic analysis was conducted using the complete data and also data from the stratified random sampling. Institutional Review Board Approval was not necessary for this study because all data were deidentified and publicly available. ## Risk factors associated with mortality caused by substance use disorders We conducted logistic regression analyses of data collected from stratified random sampling to identify individual- and community-level factors associated with the odds of SUD-related mortality. The binary outcome variable for each study subject was death by SUD (y = 1) or death by other causes (y = 0). Individual-level covariates were age group (by quinquennial), race (White, Black, other), sex (female, male), educational level (primary, secondary, college or higher), and marital status (never married, currently married, and previously married). The logistic regression model was implemented using a mixed effects generalized additive model (GAM) that allowed for nonlinear trends for all of the community-level covariates (individual-level covariates are all categorical). Our primary analysis used a logistic regression GAM mixed model for evaluating associations between individual- and community-level covariates and SUD-related mortality without including interaction terms. A supplementary analysis added interaction terms between individual- and community-level covariates (mental and physical health) to the model. All logistic regression models included a random effect for county. All sampling operations were conducted using Python 3.8, and Spark 4.1 with the pyspark package, and statistical analyses were conducted using R version 3.5.2 (R Project for Statistical Computing) with the mgcv 1.8–31 package. ## Cluster analysis and spatiotemporal risk estimation Spatial clusters of SUD-related deaths were identified using scan statistics implemented in the SaTScan software. Locations in the U.S. where the number of deaths due to SUD was higher than expected under the null hypothesis of a homogeneous distribution of SUD related deaths were classified as hotspots. The number of SUD-related deaths from the complete dataset at the county level from 2005 to 2017 were analyzed using a Poisson model with the total number of deaths from any cause by county included as an offset. Resulting hotspots were selected based on having p-values less than 0.05 and filtered to contain at least three counties and non-overlapping clusters. Community-level covariates were computed for each hotspot, all hotspots combined, and non-hotspot areas. In addition, we assessed the spatial and spatiotemporal dynamics of the relative risk (RR) of SUD-related mortality using a Bayesian zero-inflated Poisson regression model to accommodate excess zero counts in sparse area data in the context of a Besag-York-Mollie (BYM) model. The spatial analysis was computed by counties within the contiguous U.S. with available community-level information and was applied to the total number of deaths from 2005 to 2017, while the spatiotemporal study used the deaths by county, aggregated by semester from 2005 to 2017. The model was fitted using an integrated nested Laplace approximation implemented in the R-INLA software package. Results of these analyses were mapped using the R statistical software along with the ggplot2 library for spatial visualization. Extended details of the methods can be found in the. # Results ## General demographic profile of the SUD epidemic in the U.S. presents the distribution of deaths caused by SUD in the selected demographic groups, with 463,717 SUD-related deaths (2.04%) among the total number of deaths (22,705,614) registered in the U.S. from 2005 to 2017. Males had a higher proportion of SUD-related deaths (2.38%) compared to females (1.61%) in all racial groups. Additionally, the proportion (2.14%) of SUD-related deaths for the White population was higher than that for the Black population (1.60%), and other races (1.37%). illustrates the temporal trajectories of SUD-related death rates per 1,000 total deaths by race and sex, with the White male population consistently having the highest SUD-related mortality rates from 2005 to 2017. However, SUD-related death rates for Black males have increased sharply since 2014, going from 18.91 (2014–01) to 38.65 (2017–12) with a percentage change (PC) of 104,38%, compared to the White males increase from 26.46 (2014–01) to 39.77 (2017–12), PC: 50.30%. In addition, the heat maps in show the temporal patterns of SUD death rates by race, sex, and age groups, and indicate a concentration of SUD-related deaths among individuals aged 15 to 39 in both sexes and all race groups, with an additional clustering of deaths in Black males aged 40–49. illustrates the SUD-related mortality rates peaking for white population during the first semester of 2017 with the highest rates on White young males (350 SUD-deaths per 1,000 total deaths), in contrast to the Black young males with 140 SUD-related deaths per 1,000 total deaths. The substance discrimination analysis, which identified different substances leading the epidemic in different populations, is included in the. ## Socioeconomic factors and comorbidities associated with the SUD epidemic Results from the multilevel mixed effect logistic regression GAM model over the stratified sample are presented in for the individual covariates and in for the county-level variables. The statistical characteristics of the stratified sample are described in. Five percent of the total number of registered deaths in the U.S. from 2005 to 2017 were included in the sample (1,111,199 deaths, with 22,483 or 2.02% prevalence of SUD-related deaths). We found that age, race, educational level, and marital status were significantly associated with the odds of death by SUD. Individuals aged 25–29 years had the highest odds of SUD- related mortality (odds ratio \[OR\]: 3.71, 95% confidence interval \[CI\]: 3.31–4.16) compared to individuals aged 15–19 years, followed by the 30–34 year old age group (OR: 3.65, 95% CI: 3.26–4.09) and 20–24 year old (OR: 2.58, 95% CI 2.30–2.91). Whites had more than double the odds of death by SUD compared to Blacks (OR: 0.45, 95% CI: 0.43–0.47) and other races (OR: 0.45, 95% CI 0.43–0.47). Those with a secondary level education had higher odds of death by SUD (OR: 1.24, 95% CI: 1.10–1.39) compared to those with a primary education. Married individuals had lower odds of SUD-related death than singles (OR: 0.59, 95% CI: 0.57–0.62) and divorced/widowed individuals (OR: 1.19, 95% CI: 0.57–0.62). There was no statistical evidence for a difference in the population odds of SUD-related death for males and females. The same logistic regression GAM analysis indicated that average number of mentally and physically unhealthy days, percentage of children living in poverty, and percentage of the uninsured population were community-level factors associated with the odds of SUD-related death. The average number of mentally and physically unhealthy days were directly (i.e., positively) associated with an increasing the odds of SUD deaths in individuals living in counties with an average of more than 4.0 of mentally and 4.5 of physically unhealthy days (, respectively). Children living in poverty and uninsured population percentages showed an inverse relationship, with decreased odds of SUD-related deaths in counties with a percentage population of more than 25% (children living in poverty) and 15% (uninsured population). Lastly, the effects of the average number of mentally and physically unhealthy days on each age group, sex, and race included in our supplement showed dissimilar effects of mentally and physically unhealthy days across demographic groups, especially in the age-group interaction model. ## Clustering analysis and spatio-temporal risk estimation We identified 25 clusters (hotspots) with a significant concentration of SUD- related deaths at the national level from 2005–2017. The hotspots contained 165,682 (35.73%) of the total reported SUD-related deaths in the U.S., but only included 527 (17.00%) of the 3,111 counties in our clustering and risk estimation analysis. The RR of the hotspots was 1.35 (observed vs. expected SUD deaths) and 2.76% (165,682/5,999,443) of SUD-related deaths relative to all deaths, in comparison to an RR of 0.87 and 1.78% (298,035/16,706,171) in the areas outside the hotspots. shows the location of the SUD-related mortality hotspots, and the spatial distribution of the RR of death by SUD. The estimated RR ranged from 0 to 5.6 and was classified as lowest risk areas (RR \< 0.60), low risk (RR: 0.60–1.0), intermediate-risk (RR: 1.00–1.50), high risk (RR: 1.50–2.50), and highest risk (RR\> 2.50). The highest density of SUD-related death hotspots was located on the border areas of the East North Central, Middle Atlantic, East South Central, and South Atlantic regions, including in the states of Ohio, Pennsylvania, Kentucky, West Virginia, Indiana, and Tennessee (Clusters 1, 3, 5, 8, 23, 25). From 2005 to 2017, 66,227 SUD-related deaths (14.28% of all SUD-related deaths) occurred in these hotspots (RR = 1.44). Additionally, a second area of high relative risk was found in the Pacific and Southwest, including in the states of California, Utah, Colorado, Arizona, Nevada, and New Mexico (Clusters 4, 7, 9, 10, 12, 13, 17, 20, 21, 22) with 38,348 SUD-related deaths (8.26%) and an average RR of 1.49. Finally, areas with the lowest RR were identified in the West North Central regions, with no clusters, and only 21,875 (4.71%) SUD-related deaths in these areas. The average RR in these areas was 0.50 for North Dakota, South Dakota, Nebraska, Iowa, Kansas, Missouri, and Minnesota. describes the temporal trends of SUD-related mortality risk by estimating the percent change between the RR for the first semester of 2005 and the RR for the last semester of 2017. Additional estimates of the RR of SUD-related mortality from the Bayesian Poisson spatial regression analysis are summarized in and a detailed description included in the results supplement. # Discussion We found substantial spatial and demographical variation of the SUD epidemic in the U.S. from 2005 to 2017, which was characterized by the emergence of several micro-epidemics of different intensities across demographic groups and locations within the country. We found that the White male population was the group experiencing the highest rates of SUD-related deaths during this timeframe, and according to our results, 33.82% of the total deaths in White males aged 30 to 34 were caused by unintentional drug-related poisoning during the first semester of 2017. The most vulnerable age-groups among White males were 25–29 (31.34% deaths by SUD), and 30–34 (30.71%) in the second semester 2017, which is the most updated data available in our analysis. However, although the White male population was suffering the highest burden of the epidemic during the study period, a striking surge of the epidemic emerged in the Black male population, particularly in ages 30–34 (12.01%), 35–39 (11.88%), 40–49 (11.59%), and 25–29 (11.37%) by the second semester, 2017. The demographic disparities identified in this study could be the result of a complex system of sub-epidemics fueled by different substances targeting specific demographic groups, and leading different phases of the epidemic. According to our results, the latest stage of the epidemic has been led by prescription opioids, and, since 2013, by synthetic opioids. Early in the epidemic, Black males were one of the most affected populations, impacted by crack-cocaine substances that were fueling this first wave of the SUD epidemic (during early 1990s), but the rapid increasing in the prescribing of opioids in the following phases of the epidemic boosted the SUD-related death mortality in the White population. However, the increased availability of illegal synthetic opioids and heroin has shifted again the epidemic towards the Black population, with an increase in SUD-related Black males’ deaths, particularly in Black males age 45 to 55, who have become one of the most vulnerable populations in the past few years. Additionally, mental and physical distress were found to be key community-level drivers of the SUD epidemic in the country. We found that an additional average day of mental distress might increase the RR of SUD-related mortality by 39% at the county level. Mental health and SUD comorbidity are known as co-occurring disorder or dual diagnosis is a long-known associated illness. Managing mental illness in SUD patients can be a key factor in the addiction mitigation, due to a higher probability of addiction relapsing in individuals with mental disorders. Moreover, our results suggest mental distress impacted young adults more commonly in locations where the average mentally unhealthy days exceeds 4.02. Furthermore, we found that an additional average day of physical distress might increase the RR of SUD-related mortality by 28%, and this factor was affecting more older adults with a more pronounced effect in the White population. Characteristics of the spatial distribution of physical distress suggest higher levels in the South and Midwest regions of the U.S., potentially associated with a high prevalence of chronic health conditions, smoking, obesity and physical inactivity, especially higher in women and populations with low SES characteristics. These findings have been previously discussed by other researchers. In particular, Case and Deaton’s “*Mortality and Morbidity of 21*^*st* *Century*” work included a wider examination of mortality rates of midlife population of the U.S. from 1999–2015. Among their findings, they reported an increase of death rates due to alcohol, suicide, and overdose related causes and their link with an increase of the physical and mental morbidity on the White population. Our study differs from that of Case and Deaton because we focused only on unintentional drug overdoses in a wider age- groups, which potentially limits the scope of age, income and education role on the SUD-related death risk. However, their study also highlights the role of marital status, and revealed a non-clear association of gender and wealth to the increase of the death rates, matching our results. Moreover, we included an updated data until 2017, that revealed the increasing trend of Black population SUD-related death rates during the last stage of the epidemic from 2015 to 2017. These findings suggest that decreasing physical distress by including preventive measures such as strategies to decrease morbidity of chronic conditions such as cardiovascular diseases, cancer, diabetes, and stroke may help lower SUD when used in conjunction with traditional approaches to prevent or treat SUD. The geographical patterns of the SUD-related mortality observed in our study revealed a series of spatially clustered sub-epidemics with different characteristics within the country. We found that areas in the Midwest surrounding the tri-state border of Ohio, Kentucky, and West Virginia had the highest RR of SUD-related mortality at national level. Counties within this hotspot had a risk of SUD-related death between 2.5 to 5.6 times higher compared to the rest of the country. Other areas with a significant spatial concentration of SUD-related deaths were found among the southern Pacific and mountain divisions in California, Nevada, Utah, Colorado, and New Mexico. The characteristics of the concentration of SUD-related deaths in these areas differ from the above-mentioned synthetic opioid sub-epidemic occurring in the Midwest. These differences included the substances driving the sub-epidemics as well as the temporal trending on SUD-related deaths (Southern Pacific trending decreasing while the Midwest is increasing). The spatiotemporal pattern of the RR of SUD-related deaths suggests a spread of the epidemic from Southwest to Northeast during the period of the study. This progression of the overdose mortality rates is attributed mainly to the interplay between illegal drugs coming from the southern boarders and prescription and synthetic opioids throughout the Midwest and Northeast States. While the epidemic in the Southern Pacific division was fueled by methamphetamines with a substantial amount of heroin overdoses in New Mexico from 2013 onwards, the Northeast region showed a significant increase in the RR of SUD-related deaths and like in the Midwest, this sub-epidemic is led by prescription and synthetic opioids. Both Southwestern and Northeastern areas reported high levels of physical and mental distress, which resulted positive associated to high risk of death by substance overdose in our analyses. Our study had several limitations worth noting. The main limitation comes from the nature of the data, which relies in the autopsies’ ability to detect and classify substances and circumstances causing the death. Firstly, we used deaths classified as unintentional substance poisoning (ICD-10 codes: X40, X41, X42, X43, X44) to estimate SUD mortality rates, assuming that this classification is a proxy for the mortality rates of the SUD epidemic. This assumption excludes death counts from the overdoses with no information of the self-awareness of harm (IDC-10 codes from Y10 to Y14), and deaths by intentional sef-harm/suicide by substance overdose (IDC-10 codes from X60 to X64), which can be difficult to classify in practice. In addition, drugs causing the overdoses are difficult to categorize, and approximately 20% of the overdose death certificates do not include the involved substance. Even when a drug is listed, a significant number of opioid-related poisonings were classified into the broader categories of other opioids (T40.2) or other and unspecified narcotics (T40.6). Multiple opioids deaths (which were the leading cause of deaths during the last periods) and opioids combined with other drugs were often involved in overdose incidences which did not identify the substance responsible for the overdose. Additionally, autopsies and death certificates can change among states, and our analysis did not take into consideration this variation in the classification for SUD-related mortality. Further efforts are needed to improve the quality of the characterizations of SUD-related deaths, and to standardize substance classification across states, as for example the inclusion of fentanyl into the ICD-10 codes. Another important limitation is the self-reported nature of the physical and mental distress data, which could produce correlation among covariates, and some bias in our estimations. The selection of our metrics was based on previous studies about the drivers of the addiction diseases, and the availability of the information at national level. Moreover, the BRFSS is designed to provide confident data about the mental and physical distress, and it is widely used by several studies because it includes two important independent health characteristics of the population. Finally, the last limitation is related to our analysis limited to 2017 due to the official source of data for mortality rates is provided always two years behind the current date, which corresponds to the data request process to the CDC which was conducted in 2019. Despite these limitations, our study is one of the first to conduct a multilevel spatial characterization of the key individual and community-level drivers of the SUD-related mortality in the U.S. Collectively, our results suggest that individual and community-level risk factors are unevenly distributed across different demographic groups, generating a series of sub-epidemics emerging at different times and locations within the country. Moreover, the epidemic has been fueled by the introduction of different substances at different times, impacting the SUD-related mortality rate at different phases of the epidemic. Federal, state, and local governments in the U.S. have implemented multiple intervention measures to decrease SUD-related mortality rates such as restrictions on the prescribing of opioids, efforts to restrict the flow of illicit opioids, and enhancing access to naloxone. Although these efforts, among others, have been relatively successful in decreasing overdose mortality rates in general, the identification of the vulnerable populations and areas that contain the multiple sub-epidemics would enhance the ability to design prevention campaigns, which have proven more effective in managing other diseases than intervention approaches alone. Aligned with the “Know your epidemic, know your response” approach, the detailed spatial and epidemiological description of the vulnerable populations at high risk of SUD-related mortality in the U.S generated in this study can be used to create targeted prevention strategies and to localize intervention campaigns. Microtargeting strategies based on the understanding of the spatial structure and the multifactorial nature of the addiction epidemic would facilitate the design of targeted integrated preventive therapies for early identification of diagnosis in the young adult population. # Supporting information The authors thank the U.S. Center for Disease Control and Prevention (CDC) for collecting and releasing the data used in this study. 10.1371/journal.pone.0251502.r001 Decision Letter 0 Alamian Arsham Academic Editor 2021 Arsham Alamian This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 1 Feb 2021 PONE-D-20-31970 “Know your epidemic, know your response”: Epidemiological assessment of the substance use disorder crisis in the United States PLOS ONE Dear Dr. Cuadros, 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 18 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, Arsham Alamian, PhD, MSc, FACE 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\) You indicated that ethical approval was not necessary for your study. We understand that the framework for ethical oversight requirements for studies of this type may differ depending on the setting and we would appreciate some further clarification regarding your research. Could you please provide further details on why your study is exempt from the need for approval and confirmation from your institutional review board or research ethics committee (e.g., in the form of a letter or email correspondence) that ethics review was not necessary for this study? Please include a copy of the correspondence as an "Other" file. 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: 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: “Know your epidemic, know your response”: Epidemiological assessment of the substance use disorder crisis in the United States This manuscript describes the substance use disorder epidemic that is currently occurring in the United States. Although SUD is occurring at very high rates, not enough information is known about the demographics or socioeconomic factors related to individuals using substances. This study used data obtained from the CDC from 2005-2017 to examine characteristics of SUD. Findings were that White males consistently had highest SUD mortality rate from 2005-2017. There was a surge of increase in mortality in Black males from 2014-2017. SUD clustering showed there were more clusters in the west and Midwest. The researchers also found that having mental health distress or conditions significantly increased the relative risk of acquiring SUD. Overall, this is a very interesting manuscript that contributes to the scientific literature. However, the writing could be more clear at times. Suggestions for improvement are provided below: Abstract: 1\. For this sentence in the conclusion “we found that this sad epidemic in the U.S. is characterized by…” I got confused with “sad” standing for substance abuse disorder – perhaps change to SUD or rephrase 2\. For the first sentence in the results section, consider spelling out “U.S.” to United States so it is more clear that the next sentence begins with “white”, otherwise this appears to be a run on sentence. Introduction: 1\. Regarding the definition of SUD, the manuscript states that the person that engages in the behavior has knowledge of the harmful consequences of engaging in excessive substance use. However, how is this measured and how do we know the individual is aware of the harm? In the research methods section the data collected is about substance use and behavior, not knowledge of the behavior. 2\. End of page 3, states that “know your epidemic, know your response” framework shifted from coercive strategies to targeted prevention strategies. What does this mean? How were they coercive before, how did the shift occur to targeted prevention, and clarify what this means? Methods: 1\. Page 5, second to last paragraph, last sentence should be “publicly” (not publicity) 2\. The statistical methods seem to be adequately explained and appropriately applied to the data set. Results: 1\. Page 7, paragraph 1 – for the sentence that says mortality in Black males has risen sharply since 2014, 40.00 SUD-related deaths per 1,000 total deaths is listed, however I would like to see a comparison number for white males. 2\. Page 7 – table 1, figure 1 and figure 2 use the terminology substance abuse disorder (SAD) while the rest of the manuscript has been using substance use disorder (SUD) – clarification is needed here 3\. Page 8, paragraph 1 – states that there is no statistical difference in population odds of SUD-related death for males and females, this needs clarification as males suffer from disproportionately high SUD and have shown highest mortality 2005-2017 stated multiple places elsewhere. Discussion: 1\. Middle paragraph on page 11 – the description of the epidemic in the southern Pacific/mountain region is a little unclear to me. 2-3 more sentences of elaboration would be helpful to contrast with the epidemic occurring in the Midwest and other regions. 2\. Appropriate limitations are mentioned. Figures: 1\. Figure 2 – description of the figure and short analysis contrasting A and B would be helpful/useful 2\. Figure 4 – B – clarify ‘intentional’ substance use disorders Recommendation: Accept subject to revision Reviewer \#2: Re: PONE-D-20-31970 Diego Cuadros “Know your epidemic, know your response”: Epidemiological assessment of the substance use disorder crisis in the United States 1\. Topic of considerable interest given Case and Deaton’s PNAS paper and book on “Deaths of Despair” and the noted increased mortality due to suicide, overdose, accidents in select groups in the last 5-8 years. 2\. The authors identify areas in the U.S. with increase deaths by SUD using CDC data 2005-2017. 3\. Methodology excellent 4\. Results Increase of death due to SUD Males White with a sharp increase in Blacks since 2014 Increase 25-29, higher education (Did they look at college/no college, single, divorced, poverty, physical illness?) Certain States, Mid-West followed by Southern Pacific, Mountain States, and North East States showed increase. The opportunity to check out the Case and Deaton findings on education is partially lost by the absence of 2.18% of the data on education. The figures are presented for primary/secondary/college level. Can the authors look further at college graduate vs. others? \*\*\*\*\*\*\*\*\*\* 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.0251502.r002 Author response to Decision Letter 0 23 Feb 2021 We appreciate the opportunity to resubmit our work to the journal. We would like to thank the editor and the reviewers for assessing our work and for their valuable feedback and suggestions that have improved our manuscript. Please find below a point-by-point reply that addresses each of the journal and the reviewers’ comments. We have also incorporated these suggestions in the revised manuscript as noted below. We would be pleased to address any further points that the editor or reviewers may find unsatisfactory. JOURNAL COMMENTS: 1\. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. Style requirements were followed in the revised version of the manuscript according to the guidelines provided by the journal, including the file naming patterns. 2\. You indicated that ethical approval was not necessary for your study. We understand that the framework for ethical oversight requirements for studies of this type may differ depending on the setting and we would appreciate some further clarification regarding your research. Could you please provide further details on why your study is exempt from the need for approval and confirmation from your institutional review board or research ethics committee (e.g., in the form of a letter or email correspondence) that ethics review was not necessary for this study? Please include a copy of the correspondence as an "Other" file. As we stated in our manuscript, we used for our analyzes the publicly available vital statistics micro-data from the Centers for Disease Control (CDC) compiled by the National Center for Health Statistics (NCHS) (<https://www.cdc.gov/nchs/data_access/vitalstatsonline.htm>). These data are neither identifiable nor private and thus do not meet the federal definition of “human subject” as defined in 45 CFR 46.102. Therefore, these research projects do not need to be reviewed and approved by an Institutional Review Board, (IRB). For your reference, a similar study using the same data sources as our study and published last year in the journal also stated that because their study involved analysis of existing, deidentified data, it was exempt from human subjects review (“Glei, Dana A., and Samuel H. Preston. "Estimating the impact of drug use on US mortality, 1999-2016." PloS one 15.1 (2020): e0226732.) 3\. Please include captions for your Supporting Information files at the end of your manuscript, and update any in-text citations to match accordingly. The supporting information captions were included at the end of the manuscript according to the style of the reference provided by the editor. (Page 20, Line 1, Supporting Information Section) REVIEWER COMMENTS: REVIEWER 1 Abstract: 1\. For this sentence in the conclusion “we found that this sad epidemic in the U.S. is characterized by…” I got confused with “sad” standing for substance abuse disorder – perhaps change to SUD or rephrase We appreciate the feedback of the reviewer. The term Substance Use Disorder (SUD) replaced all other related abbreviations along the manuscript for consistency, accordingly to the reviewer suggestion. (Page 2, line 3, Abstract section), (Page 2, line 16, Abstract section), (Page 8, line 16, Table 1 caption), (Page 8, line 18, Fig 1 caption), (Page 8, line 20, Fig 2 caption), (Page 10, line 4, Table 2 caption), (Page 10, line 7, Fig 3 caption), (Page 12, line 5, Fig 4 caption), (Page 13, line 2, Table 3 caption), (Page 12, line 5, Discussion section). 2\. For the first sentence in the results section, consider spelling out “U.S.” to United States so it is clearer that the next sentence begins with “white”, otherwise this appears to be a run on sentence. The suggestion was included in the reviewed version of the manuscript. (Page 2, line 11, Abstract section). Introduction: 1\. Regarding the definition of SUD, the manuscript states that the person that engages in the behavior has knowledge of the harmful consequences of engaging in excessive substance use. However, how is this measured and how do we know the individual is aware of the harm? In the research methods section the data collected is about substance use and behavior, not knowledge of the behavior. We appreciate the feedback from the reviewer, and we agree with the reviewer about the need of further clarification of the definition of Substance Use Disorders (SUD) mortality. Our research methods focus on the analysis of unintentional (accidental) deaths by substance overdose as defined by the ICD10 standard for causes of death codes X40, X41, X42, X43, X44. The other causes of deaths including codes X60, X61, X62, X63, X64 (Deaths by Intentional Self- harm/Suicide caused by substance overdose), and deaths with no information about the self-intention (codes Y10, Y11, Y12, Y13, Y14) were used as control events (outcome ‘0’) in our binary outcome. This choice made the assumption that death by accidental substance overdose is a proxy for the mortality of substance abuse disorders that does not include the intention of self-harm. This choice was based on the literature review and implies some limitations worth noting. As a result of the reviewer suggestion, we included in the revised version of the manuscript a mention to this limitation as the data source does not specify previous medical history of deceased individuals. (Page 14, line 14, Discussion section, limitations). 2\. End of page 3, states that “know your epidemic, know your response” framework shifted from coercive strategies to targeted prevention strategies. What does this mean? How were they coercive before, how did the shift occur to targeted prevention, and clarify what this means? “Know your epidemic, know your response” is a framework that prioritize preventive approaches (e.g., education campaigns, targeting cause of epidemics) more than reactive measures (e.g., vaccination, medical treatment) for mitigating epidemics. In principle, this framework has been proposed to target HIV, however several studies reported that components of this approach are also applicable to other epidemics. Additionally, this framework states that the individual’s knowledge about risk is critical for the ability to respond epidemics with risk reduction strategies, and the individual social behaviors are determinants on the prevalence of the epidemics. As a suggestion of the reviewer’s feedback, we include an updated bibliography highlighting the aforementioned elements, and rephrasing the concept of coercive strategies for clarifying in the reviewed version of the manuscript. (Page 4, line 4, Introduction section). Methods: 1\. Page 5, second to last paragraph, last sentence should be “publicly” (not publicity) We appreciate the careful examination of the reviewer. We included the suggestion in the reviewed version of the manuscript. (Page 5, line 20, Methods, Data source Description section). Results: 1\. Page 7, paragraph 1 – for the sentence that says mortality in Black males has risen sharply since 2014, 40.00 SUD-related deaths per 1,000 total deaths is listed, however I would like to see a comparison number for white males. As a result of the reviewer’s suggestion. We included in the revised version of the manuscript the comparison of the temporal trend of SUD-related deaths of black and white males. (Page 7, line 11, Results section). 2\. Page 7 – table 1, figure 1 and figure 2 use the terminology substance abuse disorder (SAD) while the rest of the manuscript has been using substance use disorder (SUD) – clarification is needed here We appreciate the feedback of the reviewer. The term Substance Use Disorder (SUD) was included along all the manuscript accordingly to the reviewer suggestion. (Page 2, line 3, Abstract section), (Page 2, line 16, Abstract section), (Page 8, line 16, Table 1 caption), (Page 8, line 18, Fig 1 caption), (Page 8, line 20, Fig 2 caption), (Page 10, line 4, Table 2 caption), (Page 10, line 7, Fig 3 caption), (Page 12, line 5, Fig 4 caption), (Page 13, line 2, Table 3 caption), (Page 12, line 5, Discussion section). 3\. Page 8, paragraph 1 – states that there is no statistical difference in population odds of SUD-related death for males and females, this needs clarification as males suffer from disproportionately high SUD and have shown highest mortality 2005-2017 stated multiple places elsewhere. Taking into account the suggestion of the reviewer, we added in the discussion section further mention about this result, and we included new bibliography that analyzed this topic with similar results as ours. Even though we agree with the reviewer that men suffer a disproportionately high SUD-related deaths, our association analysis resulted in a non-significant effect of gender on the risk of death by SUD likely caused by the relationship between gender, morbidity and other socioeconomic factors included in our adjusted model. (Page 13, line 14, Discussion section). Discussion: 1\. Middle paragraph on page 11 – the description of the epidemic in the southern Pacific/mountain region is a little unclear to me. 2-3 more sentences of elaboration would be helpful to contrast with the epidemic occurring in the Midwest and other regions. As a result of the suggestion of the reviewer, we add more details and updated bibliography regarding the differences of the epidemic occurring in the Midwest and North East compared to the southern Pacific/mountain regions. (Page 14, line 2, Discussion section). Figures: 1\. Figure 2 – description of the figure and short analysis contrasting A and B would be helpful/useful We added further comparison among the rates between black and white male population in Figure 2. We also rounded the morality rates to 2 decimals to compute the percentage of increasing rates for black and white males from 2014 to 2015. (Page 7, line 6, Abstract section). 2\. Figure 4 – B – clarify ‘intentional’ substance use disorders We appreciate the feedback of the reviewer and we agree with the need of clarification about the use of ‘intentional’ and ‘unintentional’ in the context of death by substance overdose. Our research methods include unintentional accidental deaths by substance overdose as defined by the ICD10 codes for causes of death X40, X41, X42, X43, X44. The other causes of deaths including codes X60, X61, X62, X63, X64 (Deaths by Intentional Self- harm/Suicide caused by substance overdose) and Y10, Y11, Y12, Y13, Y14 (Undetermined intent substance overdose death) were used as control cases in our binary outcome. This clarification is included in the reviewed version of the manuscript. (Page 14, line 14, Discussion section). REVIEWER 2 1\. Topic of considerable interest given Case and Deaton’s PNAS paper and book on “Deaths of Despair” and the noted increased mortality due to suicide, overdose, accidents in select groups in the last 5-8 years. We appreciate the interesting insights of the reviewer. We also agree that the reference the author is citing is an important piece of the literature related to our study, given the fact they used the same data source for the death counts (CDC Wonder micro-files), and validated some of our findings. As a result, we added this reference to our bibliographic review in the revised version of our manuscript and included more details to the discussion based on the findings of Case and Deaton’s work. (Page 19, line 14, Reference section). 4\. Results Increase 25-29, higher education (Did they look at college/no college, single, divorced, poverty, physical illness? The opportunity to check out the Case and Deaton findings on education is partially lost by the absence of 2.18% of the data on education. The figures are presented for primary/secondary/college level. Can the authors look further at college graduate vs. others? Our study reported the results from a statistical association analysis between the individual risk of death caused by unintentional substance poisoning and several demographic and socio-economic covariates, including (age group, gender, education, socio economic status and marital status), as well as self-reported days of physical and mental distress from the Behavioral Risk Factor Surveillance System averaged for the time period of 2010 – 2017 and aggregated at the county level. From the results of our association analysis, we agree with the reviewer that Case and Deaton’s research is an important piece of the literature to be discussed within the context of our study. However, Case and Deaton study shows several differences and similarities with our study that are worth noting because they limit our ability to compare both findings of both studies, especially in the relationship of educational level and SUD-related death risk. As a result of the reviewer’s suggestion, we summarized the similarities and differences of our study with the Case and Deaton’s work and included them in the revised version of our manuscript. (Page 13, Lines 10, Discussion section) Among similarities and differences, Case and Deaton’s research offers a wider range of death causes than the focus of our study. They also concentrated a significant portion of their study on midlife adults, and highlighted suicides, alcohol-related and overdose as one of the critical drivers of the mortality rates increase for whites non-hispanic (WNH) in the United States. In addition, they found that educational level was critical for several aspects of the death rate increment from 1999 - 2015. In our study, we excluded intentional self- harm, and other substances apart from (Heroin, Methadone, Other opioids, synthetic narcotics, cocaine, unspecified narcotics) and included a wider range of age groups (from 5 to 84 years). As age is related to educational level, this could explain that college education did not show statistical significance, but high school resulted in an associated factor for the increased risk of SUD- related death. Another difference is the time range of both studies. While Case and Deaton’s focused from 1999 to 2015, our study was conducted from 2005 to 2017. This is especially important because Case and Deaton study reported an increase of WNH death rates from 1999 to 2015, and a parallel decrease of death rates for Blacks and Hispanic population on the same period. Our study shows that Black SUD- related death rates rised sharply from 2013 and reach the WNH SUD-related death rates in 2017, period which was out of the scope of the Case and Deaton’s study. Finally, Case and Deaton were similar to our methods in the use of the same data sources for the computation of mortality rates, and a resulting increase of morbidity of physical and mental health linked to the increase of death rates. This offers an interesting insight to our study because they exposed the lack of job opportunities and education as a potential cause of the increasing death rates. In addition, they validated our spatiotemporal pattern which suggest a diffusion of the epidemic from the Southwest to the North Eastern and Mideastern states and an unclear association between gender and wealth with an increasing of alcohol, suicide and overdose related death rates that is a super set of the SUD-related mortality rates of our study. 10.1371/journal.pone.0251502.r003 Decision Letter 1 Alamian Arsham Academic Editor 2021 Arsham Alamian 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 Apr 2021 “Know your epidemic, know your response”: Epidemiological assessment of the substance use disorder crisis in the United States PONE-D-20-31970R1 Dear Dr. Cuadros, 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, Arsham Alamian, PhD, MSc, FACE 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: 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 \*\*\*\*\*\*\*\*\*\* 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: This version of the manuscript is much better. I believe both my comments and the other reviewer's comments appear to be addressed. Good detail that elaborates on the differing epidemics in South vs Midwest, helpful clarification was provided regarding statistical analysis as well. Overall good job. I feel that this manuscript needs no further revision at this point. \*\*\*\*\*\*\*\*\*\* 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: **Kathryn Gerber 10.1371/journal.pone.0251502.r004 Acceptance letter Alamian Arsham Academic Editor 2021 Arsham Alamian 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 May 2021 PONE-D-20-31970R1 “Know your epidemic, know your response”: Epidemiological assessment of the substance use disorder crisis in the United States Dear Dr. Cuadros: 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. Arsham Alamian Academic Editor PLOS ONE [^1]: The authors have declared that no competing interests exist.

# Introduction Microorganisms and their products cause the development of pulp and periradicular pathology. *Enterococcus faecalis* is the most isolated bacteria in secondary root canal infections. The incidence of *E*. *faecalis* in failure cases was nine times higher than primer endodontic infections. Moreover, *E*. *faecalis* is capable of biofilm formation on dentinal walls; therefore, it can resist decontamination procedures. Sodium hypochlorite solution (NaOCl) is the most preferred irrigant in root canal treatments for its tissue-dissolving and antibacterial abilities. The antibacterial properties of NaOCl are directly related to its concentration. However, NaOCl in high concentrations may irritate periapical tissues. Although NaOCl is the most preferred agent for irrigation, no single agent can succeed all the tasks required by irrigation. Therefore, irrigation devices and techniques would help to safe and effective irrigation such as EndoActivator and ultrasound. EndoActivator (EA) (Dentsply Tulsa Dental Specialties, Tulsa, OK) is based on sonic vibration of a noncutting polymer tip to vigorously agitate irrigation solutions during root canal treatment. Working frequency of the device can be increased up to 167 s^-1. In previous studies showed that intracanal bacterial reduction by EA irrigation technique has not been superiority compared with conventional irrigation with NaOCl on *E*.*faecalis*. Passive ultrasonic irrigation (PUI) was introduced more than 30 years ago as a method for improving the efficacy of irrigation. This method is based on the transmission of ultrasonic waves from a file to an irrigant in root canals. Ultrasonic energy creates growing bubbles. After the collapse of these bubbles, a pressure-vacuum effect is created, resulting in the killing of bacteria and cleaning. The oscillation of the PUI instrument also creates a resonance that agitates the irrigant that is called stable cavitation. The combination of these physical effects produces an acoustic streaming that enhances the cleaning and decontamination efficacy of the irrigant. When NaOCl is used with PUI, its organic tissue-dissolving and antibacterial capacity increase due to the ultrasound. Recently, Ertuğrul et al. has discovered the low level electric current (μE) agitation has accelerated the tissue dissolution activity of NaOCl. In addition, the combination of μE agitation and dynamic movement of solution has also increased the tissue dissolution efficacy of NaOCl than PUI and EA activation methods. However, the antibacterial efficacy of μE activation has not been evaluated against intracanal microorganisms yet. In literature, low-micro amperage is capable to reduce the number of micro-organisms and suppressed Gram- negative bacterial growth. The aim of this study was to evaluate the intracanal bacterial reduction performance of μE-assisted sonic agitation on NaOCl and to compare with different activation techniques. The null hypothesis tested was that μE-assisted sonic agitation does not affect the antibacterial efficiency of NaOCl, and there is no difference in the antibacterial effect of different activation techniques. # Materials and methods ## 1 Teeth collection and specimen preparation The study protocol was approved by the ethics committee of Süleyman Demirel University with the reference number 72867572/050/2469. Eighty extracted canine teeth were selected for this study. Single root canals of teeth were examined using radiographs taken in both the mesiodistal and buccolingual directions for checking obliterations. Following tissue remnants were removed, the coronal parts of the teeth were sectioned horizontally using a diamond disc (NTI^® Diamond Discs, Axis-SybronEndo, TX, USA), and 15 mm-long roots were obtained. A K-file \#10 (Anteos K-files, Lot \# 1109000906, VDW GmbH; Munich, Germany) was placed in the root canal until its tip was visible at the apical foramen through magnification. The working length (WL) was determined as 1 mm short from the tooth length measurement. The roots were biomechanically prepared with rotary instruments up to F3 (Protaper Universal Lot# 1299410, Dentsply Maillefer, Ballaigues, Switzerland) in apical size, under 2 mL of 5.25% NaOCl (Chlorax 5.25, Lot \# 2708151, Cerkamed Medical Company, Stalowa Wola, Poland) irrigation between each file. The smear layer was removed by the sequential use of 5 mL of NaOCl, 5 ml of 15% EDTA (Endo-solution Lot \# 0512131, Cerkamed Medical Company), for 60s, followed by application of 5 mL of distilled water for 60 s. A 30G needle (Max-I-Probe Lot# 291048, Dentsply Int. York, PA, USA) was used for all irrigation procedures. The outer surfaces of root samples were sealed with double-layer discolored nail polish (Catherine Arley Lot \# 012159, Alfar Cosmetic co., İstanbul-Turkey) as a closed-end system to prevent bacterial leakage. Following the nail polish set, the root samples were mounted vertically in sterile multiple-well plates (Costar^® Product \#3524 Corning Incorporated, MA, USA) fixed with silicon impression material (Zetaplus, Zhermack SpA Lot \# 199554, Badia Polesine, Italy). The multiple-well plates containing the root samples were packaged and sterilized using ethylene oxide gas (Etomari ETO C 1445, Ankara, Turkey). ## 2 Experimental contamination with *E.faecalis* A previously described method was used for root canal contamination. A suspension was prepared by adding 1 mL of a pure culture of *E*. *faecalis* (ATCC 29212) that was grown in brain hearth infusion (BHI) broth (Merck, Darmstadt, Germany) at 37°C for 24h to 10 mL of fresh BHI. Root canals were infected by inoculating them up to the orifices with 1.2 x 10⁸ cfu mL^-1of *E*. *faecalis* ATCC29212 diluted in 10 mL of BHI broth. Each root canal was filled with a monospecies suspension by using a sterile 1-mL micropipette (Interlab, 10–1000μL, Interlab Co., Istanbul, Turkey), without overflowing. Sterile \#15 K-type files were used to carry the bacterial suspension to the entire root canal length. Fresh culture medium was added to the canal every 48h after the initial inoculum. The samples were kept at 37 C for 21 days in 100% humidity. After *E*. *faecalis* contamination, three root samples were randomly selected and fixed in ethanol using SEM to allow visualization of the pattern of colonization (JEOL, JSM-5800LV, Tokyo, Japan). ## 3 Description of the prototype μE-assisted sonic irrigation device We developed a prototype device (PD) that generates μE energy and acoustic streaming to agitate the irrigation solution in Electric, Electronic, and Telecommunication Engineering Laboratories of Süleyman Demirel University. The scheme of PD is illustrated in. The technical characteristics of the device are: Dimension of board (cm) (40x29x6); Weight (kg) 1.8kg; Working voltage 220 V; Output voltage 0–18 (V); Intensity current (0–25 mA); Power (0–2.5W); Inversion Polarity (0–999 s); and Frequency (50 Hz). In this study, DC output was adjusted at the 10 mA level in Groups 6 and 7. The handpiece part consists of two-electrode-mounted 5 mL liquid container and custom-made adapter part consists of a sonic motor and a cannula. The shape of cannula was fitted for the adapter of a standard disposable irrigation needle. When the PD is launched, μE agitated irrigant is actively released into the root canal via the handpiece, and the amount of irrigant can be controlled by the practitioner. ## 4 Decontamination procedures Seventy-seven roots were distributed into seven groups of 11 root samples each, according to final decontamination procedures and a negative-control group (n = 11). The groups were as follows: - Group 1 (Control), no decontamination procedure of the root canal dentin. - Group 2 (SS), biofilm decontamination with 5 mL of SS irrigation for 60 s using a 30G needle inserted up to 2 mm short of the WL. - Group 3 (NaOCl), biofilm decontamination with 5 mL of 5.25% NaOCl irrigation for 60 s using a 30G needle inserted up to 2 mm short of the WL. - Group 4 (PUI), biofilm decontamination with PUI used with continuous flush technique with 5 mL of 5.25% NaOCl. The tip (DT-007, EMS SA, Nyon, Switzerland) attached to the PUI device (SybronEndo miniEndo, EMS SA) and was inserted up to 2 mm short of the WL. PUI was performed at a power setting of 4, with up and down continuous motion for 60 s. - Group 5 (EA), biofilm decontamination with EA used with a continuous flush technique with 5 mL of 5.25% NaOCl. The noncutting polymer tip (medium size 25/0.04) inserted up to 2 mm short of the WL, at a frequency level of 167 s^-1 for 60 s. - Group 6 (μE), biofilm decontamination with the μE with no sonic agitation by the PD. The tip inserted up to 2 mm short of the WL was activated, and 5 mL of 5.25% NaOCl was released into the root canal for 60 s. - Group 7 (μE+A), biofilm decontamination with the μE-assisted sonic agitation mode of the PD. The tip inserted up to 2 mm short of the WL was activated, and 5 mL of 5.25% NaOCl was released into the root canal for 60 s. All groups were then rinsed with 5 mL of SS for 60 s and aspirated using flexible capillary tips (Ultradent Products, South Jordan, USA). After the decontamination protocols were completed, the root canals were rinsed with 1 ml of 5% sodium thiosulfate for 30 s to neutralize chlorine of NaOCl and with 1 ml of SS for 30 s. ## 5 Microbial sampling of the root canals Each root canal was sampled according to the method used by Brito, et al.. Following the decontamination procedures, the root canal was filled with SS, and the first sample was taken immediately using the ISO \#30 sterile paper point (DiaPaper, Lot \# 011013, DiaDent Group Int.; Seoul, Korea) placed at the WL level and remained in the root canal for 60 s. The paper point was then transferred to a sterile Eppendorf tube. The root canal was then filled with SS again, and a sterile ISO \#30 Hedstroem (VDW GmbH, Lot \# 081953) file was used for shaving the dentinal walls. The second ISO \#30 sterile paper point was placed at the WL level for 60 s. The paper points and the file were transferred together into the same Eppendorf tubes containing 1 mL of SS (0.85%, w/v) and vortexed for 60 s. Then, 10-fold serial dilutions of the samples were prepared with SS (0.85%, w/v) and 0.1 mL of the dilutions was inoculated onto the BHI agar using the spread plate technique. Following incubation for 24 h at 37°C, bacterial colonies were counted. The counts of the bacterial survivors in the root canals were determined using the direct plating method. ## 6 Statistical analyses Statistical analyses were performed with commercially available software (SigmaPlot 12; Systat Software Inc; Chicago, IL, USA). After the colonies were counted, the normality was tested using the Shapiro-Wilk test for both control and activation groups. All groups were compared with one-way ANOVA. Multiple comparisons were analyzed with Tukey’s test. The significance level was set at 5%. # Results Biofilm formation of *E*. *faecalis* was confirmed at the apical, middle, and coronal levels on the dentinal walls in SEM images. The mean 1log₁₀ cfu mL^-1 values and standard deviations were given in. Statistical analyses indicated that 5.25% NaOCl groups were more reduced cfu values than the control and saline groups (p\< 0.001) whereas, 5.25% NaOCl did not eradicate the biofilms with any activation methods. Data regarding the μE+A revealed the lowest cfu values (2.51 log₁₀ cfu mL^-1) (p\<0.05), whereas no significant difference was observed among EA, PUI and μE groups on *E*.*faecalis* (p\>0.05). Device-assisted irrigation groups were significantly more effecting in reducing *E*. *faecalis* populations than non-activated 5.25%NaOCl irrigation (p\<0.05). # Discussion This study investigated the intracanal bacterial reduction performance of μE- assisted sonic agitation on NaOCl. The null hypothesis was rejected, as μE- assisted sonic agitation of NaOCl significantly reduced the cfu values of *E*. *faecalis* than PUI, EA and μE alone (p\<0.05). Root canals were contaminated by *E*. *faecalis* ATCC 29212 for 21 days in our study. A bacterial plaque formation was similar to previous studies. The *E*. *faecalis* has been found into dentin tubules after 3–4 weeks after incubation.. A recent study has reported that the apical flora of secondary infections has been similar to primary infections which involved multiple species. Although a monospecies biofilm of *E*. *faecalis* might not be simulated *in vivo* apical flora, this *in vitro* model has been used in previous studies. Our results showed that cfu values were reduced in all device-assisted agitation techniques of 5.25% NaOCl. However, there was unable to eradicate *E*. *faecalis* from the root canal system in any sample even under uncomplicated anatomical conditions. Similarly, previous reports have emphasized that using any irrigation technique is not able to ensure complete decontamination in single and uncomplicated canal systems against biofilm structure. *E*. *faecalis* resists the decontamination methods to completely kill the bacteria could be attributed to biofilm state of the microorganism and Gram-negative specific cell wall structure. A previous report showed that *E*.*faecalis* can invade up to 653 μm the depth of the dentin. However, chemical disinfectants could only penetrate 100 μ into the dentin, which could be the reason for the inability to eradicate *E*. *faecalis*. The aim of the chemomechanical preparation is to weaken mechanically biofilm structure with instrumentation and allows to increase the efficiency of irrigation agent together. But in present study did not include the instrumentation process to eliminate biofilm. Therefore, this might be contributed *E*.*faecalis* growth after decontamination. Previous studies have shown that the combined use of PUI and NaOCl is effective in eliminating *E*. *faecalis*. The antibacterial efficiency of PUI could be explained by the following mechanism; an acoustic streaming by ultrasound produces a disagglomeration of bacteria biofilms in the root canal. The deconstruction of bacterial biofilms gives rise to planktonic bacteria that are more susceptible to the bactericidal activity of NaOCl. The cavitation effect of PUI tips might temporarily weaken the cell membrane, making more permeable to NaOCl, which could be the reason for the reduction of bacteria within the root canals in the present study. In literature, EA-assisted NaOCl has increased the antibacterial activitiy of NaOCl on *E*. *faecalis*. The EA device is a form of sonic agitation that generates subsonic micro acoustic streaming in an irrigant and cavitation. When cavitation bubbles are produced by acoustic waves, they eventually collapse and the energy released is transferred to the root canal, providing effective biofilm dislodgement, which could be the reason for the reduction of bacteria after using EA in this study. The apices of canine teeth were enlarged up to F3 size due to increase the irrigation efficacy. Mathew et al. has reported that there has not been antibacterial superiority between conventional needle irrigation and EA against *E*. *faecalis* biofilm in narrow root canals. When EA or PUI tips are used in small or curved canals, their free vibratory movement is restricted and consequently their cleaning efficacy could decrease. We used the closed-end root canal model owing to prevent bacterial leakage in this study. However, air bubbles could be locked into the acoustically activated irrigation solution when the closed-end model used is called “vapour lock effect”. Therefore, device-assisted irrigation might be negatively influenced by the effect in this study due to microstreaming and cavitation are only possible in a liquid phase. Ertuğrul et al. has been reported the first time that μE agitation causes acceleration of tissue dissolution capacity of NaOCl. According to the dynamic balance theory, anions and cations tend to change continuously in NaOCl in water solutions. External factors such as any agitation method or heating are capable to change dynamic balance of these charged ions. Therefore, μE agitation could change the dynamic balance of the electrolytic liquid as NaOCl and tissue dissolution capacity has been increased. Furthermore, the dynamic movement or mixing of irrigation solution has also increased to the μE agitation effect on the tissue dissolution property of NaOCl. The antimicrobial activity of NaOCl occurs by these mechanisms: Hypochlorous acid (HOCl) which is divided into hydrochloric acid and reactive oxygen, is formed in NaOCl solutions. The reactive oxygen ion is a very strong oxidator for microorganisms. We speculate that electrical potential energy might be accelerated the polarity of the positive charged reactive oxygen ions by an excess of electrons exists in NaOCl; therefore, antibacterial efficiency of NaOCl increases. A Gram-negative specific cell wall structure of *E*.*faecalis* consists of peptidoglycan, teichoic acid and polysaccharide. Another Gram-negative strains such as *Escherichia coli* and *Salmonella typhimurium* have inhibited and killed by low microamperage. Davis et al. have proved that the effectiveness of electric current on inhibition of growth and mortality is directly related to increasing microamperage. Furthermore, lipopolysaccharides capsule and surface proteins of Gram-negative bacteria have been influenced by an applied electric field which can cause mortality. Beside of the antibacterial efficiency of NaOCl, the μE agitation might be influenced directly to the cell wall and surface proteins of *E*.*faecalis* hence, this may also contribute to reduce cfu values. PUI is recommended to use between 30 s and 3 min in literature. However, there is no defined consensus on the exact duration. The cleaning efficiency of PUI for 60 s activation has not been significant difference than 3 min activation. Moreover, EA is suggested between 30 s and 60 s for hydrodynamical agitation of irrigation solution. The duration of irrigation was set at 60 s for standardization in accordingly with previous values of PUI and EA devices in this study. The output power was set at 10 mA. This microamperage is tolerable for human beings according to the standard. However, platinum electrodes were located into the container part of the PD. When voltage is applied, electrons move through a circuit between anode to cathode. Thus, electric current principally could not transfer to the human body such as electronic apex locators. Consequently, further studies can be designed to evaluate the electric-current effects of NaOCl at different concentrations and higher output energy levels against endodontic pathogens. # Conclusions This is the first study in the literature enlightening the antibacterial effects of μE-assisted sonic agitation on NaOCl. μE-assisted sonic agitation increased the antimicrobial activity of NaOCl than passive ultrasonic irrigation and EndoActivator against *E*. *faecalis* biofilm. μE-assisted sonic agitation on 5.25% NaOCl is not capable to eradicate biofilms at 10mA energy level in 60s. # Supporting information This project labelled 3691-D1-13 has been supported by Division of Scientific Research, Suleyman Demirel University. The authors deny any conflicts of interest. [^1]: The authors have declared that no competing interests exist.

# Introduction Male-killing bacteria are ultra-selfish maternally transmitted endosymbionts whose spread within host populations depends on the damage they do to these hosts. Although some male-killers act late in development, here we focus on male-killers that act early in development and whose dynamics do not entail significant levels of horizontal transmission. All known agents of early male- killing are bacterial, with these agents being taxonomically diverse (including *Rickettsia*, Flavobacteria, *Spiroplasma*, *Wolbachia* and γ-Proteobacteria. The diversity of the agents associated with male-killing sets the male-killing strategy apart from the other ultra-selfish manipulations of host reproduction, most of which are caused by *Wolbachia*. Early male-killers have been found in a wide range of insect species, including Coleoptera, Lepidoptera, Diptera, Hemiptera and Hymenoptera. Some groups, such as aphidophagous coccinellids, milkweed bugs and nymphalid butterflies, particularly of the genus *Acraea,* are especially prone to invasion. Three characteristics of aphidophagous ladybirds make them prone to invasion by male-killers. First, they lay eggs in tight clutches, predisposing them to strong interactions among sibling larvae. Second, ladybirds are highly cannibalistic, with neonate larvae habitually consuming any unhatched eggs in their clutch, whether these are viable or not. The potential for sibling egg cannibalism has imposed selection for embryos to develop and hatch rapidly. As a result, neonate larvae are poorly resourced and show high mortality from starvation when they fail to find and subdue their first aphid prey. In egg clutches laid by females infected with male-killing bacteria, male eggs fail to hatch and so are available to be eaten by infected female siblings, which thereby gain significant extra resources before they disperse to find aphid prey. They are, therefore, able to search for longer and subdue larger prey than are larvae from uninfected clutches. Finally, the aphid prey of ladybirds is highly ephemeral due to rapid population increases and crashes. Thus, ladybird larvae are often confronted with local resource scarcity, thus magnifying the benefit of sibling cannibalism. Of the ladybirds possessing these traits – laying eggs in clutches, exhibiting sibling cannibalism, and feeding on aphids – about half of those surveyed (13 of 30) have been found to be infected with male-killers. Conversely, none of 12 surveyed species lacking one or more of these traits has been found to be infected with male-killing endosymbionts. However, it should be noted that Weinert *et al.*, 2007 , assaying 21 species of European coccinellids, did find inherited symbionts of three clades (*Spiroplasma*, *Rickettsia* and *Wolbachia*) from *Chilocorus bipustulatus* and *Halyzia sedecimguttata*, both of which lack one or more of the three characteristics. However, in neither species was the phenotype caused by the bacteria ascertained. In addition to the behavioural and ecological prerequisites for invasion and persistence of male-killers, the environmental conditions must also be conducive. Specifically, high temperature may, by suppressing growth or killing the bacteria, prevent their spread within a host species. For instance, *Drosophila bifasciata* can be cured of male-killing *Wolbachia* by culturing the flies at 26°C instead of 21°C. Similarly, *Drosophila equinoxialis* could be cured of male-killing *Spiroplasma* by exposing eggs to relatively high temperatures (34–40°C). Using, artificially transferred *Spiroplasma* into *Drosophila melanogaster*, Sakaguchi and Poulson showed that high temperature exposure produced a full cure from infection. High temperature impacts on male-killers have been demonstrated in three species of coccinellids: *Adalia bipunctata* infected with *Rickettsia* , *Adalia decempunctata* infected with *Rickettsia*, and *Coleomegilla maculata* infected with *Flavobacterium*. In these cases, egg hatch rates increased, either completely or partially, and sex ratios became less female biased, tending towards 1∶1, following high temperature treatment (25–30°C). Furthermore, molecular investigations confirmed the absence of the bacteria in temperature cured lines. In some cases, curing was partial, as in *A. decempunctata*, where *Rickettsia* was found in some male progeny, leading to the deduction that male death depends on the bacterial density in the host. As a result of these experiments showing that high temperature can cure hosts of male-killer infections, it has been hypothesized that male-killers may be rare in hot climates. In all aphidophagous coccinellid species previously tested, male-killing bacteria have been shown to be sensitive to high temperature. Furthermore, all coccinellid populations found to date to be infected with male-killing bacteria have been from temperate or Mediterranean climates. This may result from research bias, or may be a consequence of the temperature sensitivity of male- killing bacteria, thus restricting the distribution of coccinellid-infecting male-killers to geographical regions lacking very high temperatures. The mode by which coccinellids become infected with male-killing bacteria is not known. However, there is phylogenetic evidence to suggest that horizontal transmission of male-killing bacteria does occur, although probably very rarely. This means that it is possible for one species of coccinellid to be invaded by two or more different types of male-killing bacteria. However, as the intracellular environment in which bacteria in a particular host species live and are vertically transmitted seems to be essentially the same, and as the bacteria employ the same strategy of ultra-selfish manipulation, the competitive exclusion principle would suggest that only a single male-killer should survive in a particular host population. Indeed, models of the invasion dynamics of early male-killers show that two male-killers cannot occupy the same population at equilibrium, unless there is some degree of male-killer suppression. Randerson *et al*. demonstrated that the ‘strongest’ male-killer: i.e. the most efficient in terms of high vertical transmission, low direct cost on infected females and high male-killing efficiency, would out-compete and exclude ‘weaker’ male-killers, unless the host can evolve resistance against the strong male- killer, which then allows the two to coexist in the same host population. Despite these theoretical findings, two or more male-killers have been recorded from some host populations without any indication that suppressor genes have evolved in the host. In the butterfly *Acraea encedon,* two strains of male- killing *Wolbachia* have been reported from Tanzanian populations. In the coccinellid *Adalia bipunctata*, Majerus *et al*. reported that four male- killers (a *Rickettsia*, a *Spiroplasma* and two distinct strains of *Wolbachia*) were found in a single sample collected from a single location in Moscow. Here, we report the discovery of two male-killing strains of *Wolbachia* in a population of a coccinellid, *Coccinella undecimpunctata*, from hot regions of lowland Egypt and Jordan. This finding indicates that climate is less of a limiting factor for the distribution of male-killers in ladybirds. # Materials and Methods ## Sample Collection and Culturing 200 individuals of adult *Coccinella undecimpunctata* L. were collected from Abo-Rawash, Giza, Egypt, in September 2004 and July 2005. An additional 50 individuals were collected in Amman, Jordan, by Professor T. F. Allawi, in October 2004. They were grouped in 10 individuals and housed in 9 cm stock Petri-dishes, and allowed to mate freely. Individual pairs were removed to clean dishes in which to lay eggs for the establishment of individual matrilines. Reproductive adults and larvae were allowed to feed on pea aphids (*Acrythosiphon pisum*), following methods described by Majerus and Kearns. Eggs were harvested daily by removing adults to clean dishes, leaving the eggs *in situ*. Immature stages were reared in a controlled temperature room at 21°C and 16L: 8D. An artificial diet was used to maintain non-reproductive adults. No specific permits were required for the described field studies. The samples location is not privately-owned or protected in any way. The field studies did not involve endangered or protected species. ## Phenotypic Indicators of Male-killing Two phenotypic indicators of male-killing were assessed for each matriline: the egg hatch rate and the sex ratio of progeny. Once neonate larvae had dispersed, the eggs within each clutch were categorized as hatched, unhatched and yellow (which is indicative of there having been no embryonic development), or unhatched and grey (indicative of some embryonic development). Progeny sex ratios (given as proportion male) were obtained for each family by determining the sex of all progeny, using sex-specific ventral abdominal sternite traits, visualized microscopically under CO₂ anaesthesia. Families were split into four categories: 1. Sex ratio (SR) = low egg hatch rate (\<0.5) and all progeny female. 2. Incomplete sex ratio (iSR) = low egg hatch rate (\<0.5) and a statistically significant female bias amongst progeny. 3. Normal (N) = high egg hatch rate (\>0.5) and progeny sex ratio not significantly different from 1∶1. 4. Not ascertained (?N) = high egg hatch rate (\>0.5) and progeny sex ratio significantly different from a 1∶1 sex ratio. In some instances, matrilines initially designed as SR or iSR subsequently showed increased hatch rates and sex ratio (proportion males), and these were described as revertant families. In general, these lines were removed from the study stocks, unless otherwise stated. No obvious case of progressive sex ratio was observed in the study stocks. ## Inheritance of the SR Trait Female offspring of SR and iSR lines were crossed with males from N families, to test the inheritance of the sex ratio trait. Pairs and their offspring were treated as before. ## Susceptibility of the SR Trait to Antibiotics The effect of tetracycline, a broad-spectrum antibiotic, on females showing the SR trait was investigated. If the causative agent of the sex distortion was a bacterium, tetracycline treatment should produce a partial or full, cure of the trait. Eggs were collected from three females from an SR line for up to three weeks. The egg hatch rates of these females were recorded. Once it had been confirmed from egg hatch rates that these females remained less than 50%, females were fed for two hours per day on a diet of golden syrup containing 10% w/v tetracycline, being otherwise fed on aphids. The time of exposure to tetracycline was varied from 1–4 weeks for different females. Five females from normal (N), presumably uninfected lines were treated in the same way as controls. The egg hatch rates after treatment and the progeny sex ratios from the eggs laid before and after treatments were recorded. Subsequently, molecular assays on progeny from treated females were used to assess the presence or absence of bacteria. For two of the treated SR families, one female resulting from an egg laid after tetracycline treatment was mated to a male from an N line, and her progeny reared, with the sex ratio of progeny being recorded. ## Identification of the Male-killing Agent Genomic DNA was extracted from adult ladybirds, following a fast protocol modified by Majerus *et al.* from Walsh *et al.*. Successes of extractions were tested using general insect primers (*COI* gene) C1-J-1751f and C1-N-2191r, to check for the presence of ladybird DNA. Modified versions of the standard PCR protocols for male-killers, were used to assay two females from matriline B (SR) for bacteria. *16S rDNA* general eubacterial primers 27f and 1495r, were used. Product bands were sliced from the agarose gel and purified using Qiaquick PCR purification kit (Qiagen), and directly sequenced using dye-labelled terminators in a cycle-sequencing reaction (Applied Biosystems). The products were then visualised on an ABI 377 automated sequencing machine. Initial identification of the sequence was established through a BLAST search. ## Relation between the SR Trait and the Putative Causative Agent Following identification of *Wolbachia* as a candidate male-killer, samples from all matrilines, irrespective of their sex ratio status, were assayed for *Wolbachia* using primers *wsp*81F and *wsp*691R. From the first Egyptian sample (2004) 39 SR or iSR females from three SR lines, 16 N or ?N females and five N males from three N lines were assayed. In addition, ten females produced by an SR line female before administration of tetracycline and eight females and two males produced by females from this SR line after tetracycline treatment were assayed. From the second Egyptian sample (2005) and the Jordanian sample, 28 SR or iSR line females, ten N females and five N line males were assayed. In addition, five females produced by SR line females before administration of tetracycline and six females and two males produced by SR line females after tetracycline treatment were assayed. DNA samples extracted from two previously known *Wolbachia*-infected *Adalia bipunctata* were used as positive controls, while negative controls were prepared using the same PCR premix, except that sterile distilled water replaced the genomic DNA. PCR reactions were carried out in a Techne Progene PCR machine with a heated lid. All PCR cycling conditions were as described by Majerus. PCR products were then run on a horizontal 1% agarose gel, beside an appropriate marker containing DNA fragments of known size (1 kb DNA ladder). Following electrophoresis, the gels were visualised and photographed using a Biogene UV trans-illuminator. ## Phylogenetic Analysis DNA from SR lines was amplified with *Wolbachia* specific primers *wsp*81F and *wsp*691R for the *wsp* gene. Products were sequenced as described above for the *16S rDNA* gene and the resultant partial sequence was subjected to a BLAST search. The sequences were then manually aligned with a selected group of *wsp* sequences (see for details), including those known to cause male-killing (including the only known two male-killing *Wolbachia strains* in ladybirds, which are in *Adalia bipunctata*), at least one strain causing each of the other reproductive manipulations of hosts, and those showing closest homology to the sequence under investigation. Phylogenetic analysis was performed using Mega version 3.1. Phylogenetic trees were constructed using maximum parsimony and support was obtained from 500 bootstrap replicates; seed 64238 (program Mega 3.1). Gaps or missing data were dealt with by complete deletion. ## Statistical Analyses Using StatXact 8 software, Fisher’s exact test was used to test the significance of deviation from 1∶1 sex ratios among family progeny. Bonferroni correction was used to correct for multiple comparisons among the different families. The association between the F1 and F2 sex ratios and the F1 sex ratio and hatch rate were evaluated using a simple linear regression model to report slope, associated error of estimate and P-value, as well as the correlation coefficient. The heterogeneity between lines was investigated using a Chi-square test to test the differences in proportions of sex ratio and hatch rates. Sex ratio changes subsequent to antibiotic treatments were also evaluated with a simple linear model which looked at information that we combined over different observations periods. # Results ## Phenotypic Results Sixteen matrilines were obtained from the Egyptian and the Jordanian collections, one of which failed to produce sufficient offspring for analysis. From these lines, nine lines were designated as SR lines, having low egg hatch rates and absence of male progeny (3 from Egyptian 2004 collection, 3 from the Egyptian 2005 collection, and 3 from the Jordanian collection). The rest of were designated as normal lines due to their normal sex ratio. Lines that had produced all or almost all female progeny in the F1 also did so in the F2, and lines with normal F1 sex ratios remained normal in the F2. The strong correlation between the F1 and F2 sex ratios (*r²* = 0.86, slope  = 0.93±0.11; *P*\<0.0001) demonstrates essentially perfect maternal inheritance of the male-killing effect. The F1 progeny sex ratios of all matrilines showed significant heterogeneity (χ²₁₁ = 65, *P*\<0.0001)Furthermore, the progeny sex ratio was significantly correlated with egg hatch rate (*r*² = 0.68, slope  = 0.91±0.11, *P*\<0.001), such that clutches with hatch rates less than 50% were made up almost entirely of females. ## Effect of Tetracycline Treatment The sex ratios of progeny produced before and after administration of tetracycline to females from an SR line are given in. Families reared from F1 females produced by SR mothers after tetracycline treatment all produced high egg hatch rates and normal progeny sex ratios. The treatment resulted in an initial decrease in egg hatch rates, with hatch rates close to zero for two to five days (characteristic of ladybirds when treated with tetracycline), followed by an increase in egg hatch rates to greater than 0.5. The sex ratio before antibiotic treatment was significantly different from that after antibiotic treatment in females 2 (p\<0.05), 3 (p\<0.01) and 4 (p\<0.05), but not in females 1 (p  = 0.27) or 5 (p  = 0.91). The sex ratio of progeny from SR females produced after tetracycline treatment increased rapidly, and converged upon 1∶1. Females from N lines fed on tetracycline showed no significant alteration of egg hatch rate or progeny sex ratio. ## Relation between *Wolbachia* Presence and the SR Trait All SR females that had not been treated with tetracycline were found to be positive for the presence of *Wolbachia*. None of the females from N lines, or progeny of females from SR lines after tetracycline treatment, males from either these lines or N lines, or either of the negative controls were found to have *Wolbachia* present. Moreover, while females produced by SR females before being treated with antibiotic tested positive for *Wolbachia,* none of these females and two males produced by these same SR females after tetracycline treatment did. Both negative control samples also tested negative. ## Phylogenetic Analysis The analysis of the *16S rDNA* gene from females from the Egyptian 2004 SR matrilines showed that the male-killing causative agent may be a B-type *Wolbachia* (Accession number DQ993358). Phylogenetic analysis of this *wsp* sequence (Accession number EF502046) shows closest affinity to the two male- killing *Wolbachia* sequences from *A. bipunctata* reported by Hurst *et al.*. This conclusion should be viewed with some caution because the *wsp* can undergo recombination between different *Wolbachia* strains, thus affecting phylogenetic patterns. Analysis of the *wsp* gene revealed that our samples included two distinct strains of *Wolbachia* strains in our samples. One strain (*wsp* Accession number EF502046) was found in three lines from Egypt in 2004 (A, B and F) and one line from Egypt in 2005 (line 1). A second *Wolbachia* strain (*wsp* Accession number EF608161) was found in two lines from Egypt in 2005 (lines 2 & 3) and three lines from Jordan (Lines 2, 3 &4). The sequence of this second strain differed from that of the first strain by 20.97% (of 515 bases). The *wsp* sequence from the second strain is most similar to that of the U strain of *Acraea encedon*, with divergence of 15.13% (of 522 bp), including a central identical region of 387 bp. On the basis of the limited data in this study, the prevalence of the first *Wolbachia* strain, which was only recorded from the Egyptian samples (sample size  = 50), is 0.4 (3 of 6 in 2004 and 1 of 4 in 2005). The mean fidelity (±se) of vertical transmission (calculated as: 1 - (number of males/number of females) is estimated as 0.987±0.012. The second *Wolbachia* strain, which was recorded from both Egypt and Jordan, had an overall prevalence of 0.33 (combining 2 of 10 in Egypt and 3 of 5 in Jordan). For this second strain, the vertical transmission is estimated as \>0.99 from Egypt and 0.85 from Jordan, giving a mean vertical transmission for this strain of 0.94±0.109. Overall, 60% (9 of 15) of the lines were infected with one of these male-killing *Wolbachia* strains. # Discussion The results show that *C. undecimpunctata* is a host to at least 2 different strains of male-killing B-type *Wolbachia*. The association between the *Wolbachia* infection and the SR trait was perfect. The *Wolbachia* was detected in females from all strains with significantly female-biased sex ratios and low hatch rates (SR and iSR), but from none of the strains with normal sex rations (N and ?N). One line (G) from the 2004 Egyptian sample had a significantly female biased sex ration among the F1, but had a high egg hatch rate (0.86). However, the F2 exhibited both a high hatch rate and normal sex ratio, and this line was found to be negative for *Wolbachia* infection. Thus, the F1 sex ratio bias may have been due to chance or to a factor other than a male killing endosymbiont (e.g., meiotic drive). Both male-killing strains of *Wolbachia* identified here are antibiotic sensitive, as antibiotic treatment increased hatch rates of eggs laid by treated females, and restored a 1∶1 progeny sex ratio, and resulted in the production of offspring uninfected with *Wolbachia*. Moreover, the offspring of tetracycline- treated females produced normal progeny sex ratios. Although the two *Wolbachia* strains found in *C. undecimpunctata* had identical 16S sequences, their *wsp* sequences were 20% divergent. Using the rapidly evolving *wsp* gene to infer relationships to other closely related strains, we find that one of the two strains in *C. undecimpunctata* is closely related to the two different but closely related male-killing *Wolbachia* strains that occur in the ladybird *A. bipunctata*. Such a pattern could result from each beetle species having been invaded by closely related strains of *Wolbachia* that subsequently evolved male-killing within their new ladybird hosts. Alternatively, which is more likely, a male-killing strain may have jumped from one ladybird species to another via lateral transfer. The other *Wolbachia* strain found in *C. undecimpunctata* is most similar to the male-killing *Wolbachia* in *A. encedon*. This result supports previous hypotheses that horizontal transmission of male-killers may occur occasionally between hosts from different genera or even orders. Dyson *et al*. suggested that some *Wolbachia* strains might specialise on particular host sex determination systems. The apparently close relationship between the second *C. undecimpunctata* strain and the U strain of *A. encedon* is thus striking, since most ladybirds, including *Coccinella*, have an XY sex determination system, while butterflies have a ZW sex determination system. If the same *Wolbachia* strain can cause male-killing in XY male beetles and ZZ male butterflies, this would suggest that some male-killing *Wolbachia* are generalists when it comes to sex determination systems. This possibility could be tested by transferring *Wolbachia* between *C. undecimpunctata* and *A. encedon*. This is the first report of a male-killing endosymbiont in any coccinellid species from a hot climate. The daily high temperatures in Amman, Jordan and Giza, Egypt average ∼32°C and ∼34°C, respectively, during the hottest summer months, well above the temperatures shown to kill or debilitate male-killing endosymbionts from more temperate regions. Thus, high temperature is not an insurmountable barrier to infection of coccinellids by male-killing bacteria. Further work on aphidophagous coccinellids from hot climates is likely to extend the number of cases of male-killing infection in this family of beetles. In some coccinellids, such as *Harmonia axyridis* infected with *Spiroplasma*, the temperature required to kill male-killing infections is close to the critical temperature at which coccinellids can survive, making cure of the SR trait by temperature difficult. It would be interesting to determine whether the *Wolbachia* that infects Egyptian *C. undecimpunctata* can be cured by heat treatment without adversely affecting the beetles. Whether or not this is the case, the *Wolbachia* strains found in *C. undecimpunctata* from Egypt and Jordan provide an opportunity to explore the evolution of high-temperature tolerance in *Wolbachia*. The present findings also indicate that the ecology and evolution of ladybird beetles from hot climates may be as likely to be affected by male-killing endosymbionts as those from more temperate regions. In previous assays of *C. undecimpunctata* samples from Britain (n  = 47), male- killers have not been found. This may be because of low prevalence and incomplete ascertainment, or it may be because male-killers are absent from British populations of this species. If the latter is the case, this might be due to lack of invasion opportunity, or the result of inherent resistance, making British *C. undecimpunctata* an unsuitable host for male-killers. Microinjection of one of the male-killing *Wolbachia* that occurs in *C. undecimpunctata* in the Middle East, into British individuals could be used to test this latter possibility. This is the first confirmed case of a heritable male-killing bacterium having been reported from the genus *Coccinella*. This is noteworthy because the majority of species in this genus are aphidophagous and have the ecological characteristics that should make them liable to infection by such bacteria. Despite this, in previous assays of 19 collections from seven species of *Coccinella* from temperate or Mediterranean climates (including two previous samples of *C. undecimpunctata* from England) (M. Majerus, unpubl. data), no male-killers have been detected. The results herein suggest that either further testing of members of this genus would be worthwhile or might reveal that *Wolbachia* in this genus may not only tolerate, but actually require high temperature for expression and transmission. The results to date do not allow resolution of whether *C. undecimpunctata* was invaded by a single strain of *Wolbachia* that has subsequently diverged, or was invaded by two already different strains of *Wolbachia*. This question might be resolved by analysis of mtDNA variability associated with hosts of each *Wolbachia* strain. This is the third finding of more than one male-killing bacterial strain occurring sympatrically in the same host. This instance is similar to that in *A. encedon* in Tanzania, where two different strains of *Wolbachia* coexist. The similarity to *A. bipunctata*, is less clear, for here, in addition to two strains of *Wolbachia* coexisting, two other phylogenetically disparate male-killers occur. Randerson *et al*. have tried in their model to explain the observed co- existence of multiple male-killing strains in the same host populations. They proposed the evolution and spread of a costly resistance gene in the host that should weaken the resident (the stronger) male-killer, although this may mean that both the resistance gene and the male-killer may be lost form the population. Then the weaker male-killer may spread at the expenses of the stronger male-killer, since it is tolerant to the effect of the evolved host resistance gene. As a result, the frequency of the stronger male-killer will decrease and in turn the frequency for the resistance gene will decrease as well. Thereafter, the frequency of the stronger male-killer may increase again, in a cyclical manner in response to the resistance gene. Such frequency dependent selection should allow stable coexistence of multiple male-killers in the same host population. However, the cause of male-killers coexistence in a population seems unconfirmed, especially no resistance genes have been reported in *C. undecimpunctata*. Three scenarios seem possible. First, the cases of multiple male-killer existences in a host population may be the result of independent allopatric invasions by different strains of male-killer, followed by migration such that the different male-killers become sympatric. While this is tenable for the two coccinellid cases, for coccinellids can be highly dispersive, it seems to be less tenable for *A. encedon*, which is a highly colonial species showing low female dispersal. Moreover, in *A. bipunctata*, Tinsley has shown that prevalence of two of the male-killers in this species (*Rickettsia* and *Spiroplasma*) show correlations with environmental factors in Scandinavia, suggesting long-term persistence. Second, the Randerson *et al*. model may be essentially correct, but the approach to male-killer prevalence equilibrium frequencies may be extremely slow, such that in those host populations that harbour two or more male-killers, equilibrium frequencies have yet to be reached, and the competitive advantage of the ‘strongest’ male-killer has yet to be manifest in the decline and elimination of ‘weaker’ male-killers. While this may be the case, if it is, it makes the prediction of competitive exclusion in the case of male-killers of little value in the field. Finally, it is possible that biotic and abiotic factors affect the level of the three principle parameters that control male-killer invasion and prevalence (vertical transmission efficiency, direct effect of infection on females, fitness compensation) in coccinellids (1) in such a way that there is no consistently ‘strongest’ male-killer. Here, if the relative fitnesses of different male-killers oscillate such that the ‘strongest’ is first one strain and then the other, coexistence may be maintained for very long periods of time, particularly if there was a mechanism by which male-killer fitness was inversely related to prevalence. We are grateful to Ian Wright who provided technical assistance, in particular by maintaining aphid supplies and to S.S.M. Hassan, R.L. Ware and L.J. Michie who helped to care for cultures. Special thanks for Prof. John Jaenike for his great help and advice through this manuscript. [^1]: The authors hereby declare that they received funding from a commercial source (BP Egypt) to conduct this study and they do confirm that this does not alter their adherence to all the PLOS ONE policies on sharing data and materials. [^2]: Conceived and designed the experiments: SME MEM. Performed the experiments: SME. Analyzed the data: SME. Contributed reagents/materials/analysis tools: SME MEM. Wrote the paper: SME SM MEM.

# Introduction Adaptive memory refers to cognitive adaptation based on survival that prioritizes the storage and retrieval of important information in human memory to deal with recurring challenges related to the Pleistocene, 2,5 millions of years ago—that is, ancestral priority in the functioning of memory. This may reflect the importance of ancestral environments in the evolution of most human psychological mechanisms, enabling better mnemonic performance to occur when induced by problems present in ancestral settings, such as escaping predators and treating infections. This adaptive memory bias leads to greater information retention, regardless of the specific proximate mechanisms that may be involved. In this sense, a mnemonic adaptation refers to the selective pressures that shaped a memory that helped early human beings to solve adaptive problems and, therefore, increased the chances of survival and reproduction. For example, it was observed that situations involving the search for medicinal plants for treating an infection in a savanna ancestral environment promoted survival advantage in memory processing compared to the modern environment. Other studies analyzing this mnemonic adaptation also obtained similar results, including studies among people living in different environmental contexts. In this sense, the book published by Schwartz et al. gathers several theoretical and empirical works about the adaptive nature of memory. In addition, remembering the location of food, the habitat of a predator or a partner available to mate facilitated the survival and reproductive success of hominids. These mental adaptations were shaped by the way hunter-gatherer hominids solved recurrent adaptive problems in ancestral environments. However, recent evidence shows that human beings have undergone a multiregional evolution across the African continent and in other regions of the planet, with little evidence on the implications of these origins on adaptive memory. For instance, a study by Yang et al., showed that important information for survival is remembered by people both in ancestral survival scenarios (savanna-like pastures) and in non- ancestral/modern environments (mountains). Moreover, a study by Silva et al. provided evidence that the previous experience of people with common illnesses and the frequency of these illnesses intensifies the memorization and retrieval of important information to deal with these illnesses. This suggests that the human mind can adaptively respond well to recurrent threats in the current environment. Therefore, it is reasonable to consider that adaptive memory can operate in recent environments and in other evolutionary environments besides the savanna, such as in humid and dense tropical environments; however, this has not been tested so far. During evolutionary history, the first human beings had to constantly interact with the environment, using natural resources to obtain nutrients or avoiding dangers and threats from predators. This indicates that their interactions may have affected our cognitive apparatus such that the human mind has adapted and developed psychological mechanisms responding to a wide range of challenging situations. Recurrent selection pressures in the *environment of evolutionary adaptedness*, acting over evolutionary time on a regular basis, boosted the construction of mental adaptations in the first humans to solve lasting adaptive problems. Environment of evolutionary adaptedness does not refer to a specific environment or time; it is a set of statistical factors of recurrent selection pressures or cause and effect relationships that drove the increase in the frequency of certain alleles underlying an adaptation, until they became typical species or promoted a frequency-dependent balance. According to Tooby and Cosmides, evolutionary processes are slow and require thousands of years to develop complex cognitive adaptations; therefore, the human mind remains adapted to the Pleistocene savanna. However, this concept does not consider two important aspects of the evolutionary history of *Homo sapiens*: i) human evolution and development may have occurred in various environments during the Pleistocene; and ii) human activities can accelerate biological evolution, thus modifying genetically inherited predispositions. These aspects may even reflect flexible mental adaptations, which integrate the mechanisms molded in paleoenvironments with mechanisms built during the individual’s ontogenetic development. For example, an idea proposed by Orians assume that exist the predisposition of human beings to prefer savanna environments, due to the importance of this environment during evolutionary history, which, being an environment with sparse trees that allowed a panoramic view, facilitated, among other things, the observation of the approach of predators, which helped in the survival and reproduction of hominids. However, this preference is not observed in some cultures, which may result from the establishment of human beings in different modern environments. In addition, the study by Soderstrom and McCabe brought evidence that the good performance of recall occurs in modern (city) and ancestral (pasture) scenarios, suggesting that in some situations there is no ancestral priority in the survival processing advantage in memory. In this sense, based on recent paleoanthropological evidence from the fossils of the first human beings, we consider the rainforest, deciduous forests, and the savanna asancestral environments. For example, recent fossil evidence suggests that *H*. *sapiens* may have originated and evolved in distinct regions throughout the African continent. The first hominids adapted to various environments during the Pleistocene over a wide latitudinal range, such as the temperate north (which includes deciduous forests) the subtropical region of China, and in the tropical regions of Southeast Asia. At the end of the Pleistocene, some *H*. *sapiens* had already occupied the cold temperate environments of the Arctic region. Richerson and Boyd even argue that the long and slow evolutionary history of human beings in different environments may have driven the growth and development of the brain as we know it today, because the human construction of subsistence systems adapted to spatial variation. In this study, we analyzed the performance of human memory in situations of survival in the ancestral savanna, rainforest, and deciduous forest environments, and in the modern coniferous forest, desert, tundra, and urban environments, to investigate whether an ancestral priority of adaptive memory really exists. The modern environments considered here are regions in which there has been no paleoanthropological evidence about hominid origin and evolution until now. This study can help to elucidate the controversial results in literature that may reflect the conception of evolutionary psychology suggesting that we evolved in a specific environment (savanna) and that is why we adaptively respond better to challenges related to this biome, not considering, for example, the ancestral environment of the rainforest. For this, we carried out an experiment adapted from Nairne et al. in which we simulated both ancestral and modern survival scenarios. In the typical survival processing paradigm, participants imagined that they were trapped in a savanna without any survival materials and must face potential dangers, and that they needed to evaluate a list of words and their relevance to survive in the relevant scenario. In this study, we asked volunteers to assess the relevance of information to solve the recurring challenges imposed in these scenarios and later subjected them to a surprise memory test. The memory test allowed us to analyze whether the relevant information was remembered when related to ancestral environments. If human evolution was in fact multiregional, it is reasonable to consider that adaptive memory is a reflection of an evolutionary history that has trodden several paths, with hominids venturing and adapting to threats that exist in hot, dry, and sparse environments as well as in humid and dense environments. Therefore, these evolutionary paths must have some effect on adaptive memory. We suggest that the understanding of how adaptive memory operates can be more complete with the expansion of its analysis in a manner that encompasses rainforests as well as modern environments. Thus, we tested the hypothesis that there is an ancestral priority in the cognitive strategies mobilized in the human mind to find medicinal plants to treat diseases. The challenge related to the search for medicinal plants for treating diseases is an interesting model for a broader understanding of adaptive memory, as this was a basic recurrent challenge essential for the survival and evolution of hominids in ancestral environments. In our experiment, we predicted that, on average, people would significantly remember the words classified as most relevant to finding and using medicinal plants in the simulated ancestral survival scenarios (savanna, rainforest, and deciduous forest) when compared to the modern scenarios. There is evidence that *Homo sapiens* and *H*. *neanderthalensis* have used several parts of different plants for medicinal and technological purposes. Situations of parasite infections or invasion by predators during foraging activities may have stimulated the first humans to look for medicinal plants. In addition, *Homo habilis* used medicinal plants to treat diseases based on observation of plant-based self-medication by other animals. The use of medicinal plants by non-human primates may also involve different plant parts. For instance, some orangutans (*Pongo pygmaeus*) in Indonesia ingest *Dracaena cantleyi* leaves for the treatment of parasites. Although the therapeutic use of plants is understood to be essential for the survival and evolution of hominids, the events that led to this behavior are still uncertain, with only few hypotheses about its origin (for a complete argument see Albuquerque et al.). In this sense, we conducted an empirical experiment that analyzed how human memory stores and retrieves the relevant information to challenges related to the use of medicinal plants in different environments. We used the same challenge related to medicinal plants in all simulated environments because using qualitatively different survival threats (for example, escape from a predator in the savannah *versus* escape from a murderer in the city) could lead to different levels of threat meaning, that is, one threat can be perceived as more dangerous than the other. Therefore, our aim was to obtain evidence about how human memory may have evolved to be versatile in solving the problems imposed by ancestral and recent environments. # Materials and methods ## Experiment and word selection We designed an experiment, adapted from the work of Nairne et al., analyzing the survival processing paradigm, in which we simulated different survival scenarios wherein each volunteer evaluated the relevance of information to perform the task of finding and using medicinal plants to treat a disease. We do not use a control scenario in our experiment. Although, we chose to only use scenarios related to a survival situation encompassed ancestral environments, considering that the task of looking for medicinal plants to treat diseases activates the advantage of survival processing in memory. The scenarios encompassed ancestral environments—rainforests, deciduous forests, and savanna—and modern environments—coniferous forests, deserts, tundra, and urban. We maintained the same survival conditions in all environments. The objective was to analyze whether the most relevant information would be quantitatively more recalled in a free recall surprise test. We predicted that relevant information would be best remembered when the challenge was present in ancestral environments. The "information" in our experiment is represented by words (concrete nouns) that act as stimuli in the tests of memorization. The relevance of this information was decided by the volunteers themselves, as even seemingly irrelevant stimuli for fitness, such as a pencil, could become relevant depending on the situation and a jacket could be relevant depending on the location. Relevance was measured using a Likert scale that ranged from 5 for extremely relevant to 1 for extremely irrelevant. In the recall test, our objective was not to analyze the word itself, but rather if the relevance given to each word would influence its recall depending on the environment, regardless of the meaning of the information. Thus, we randomly selected 32 words that were unrelated concrete nouns, based on the free association norms for Brazilian Portuguese words in memory studies, which include 1004 standardized words. The selection criterion was based on two aspects: i) concrete nouns, as they generate less confusion in free recall tests; and ii) unrelated nouns that have little or no semantic relationship, as this relationship can be a confusing factor in the analysis. The 32 selected words that we used as a stimulus, followed by their translation into English, are described in. The same set of 32 words was used for all simulated survival scenarios. The words were randomized before being presented to each participant. ## Survival scenarios To test whether there is an ancestral priority in the cognitive strategies mobilized in the human mind, we use simulated survival scenarios that are dangerous situations, but made some adaptations. In adaptive memory studies, survival scenarios usually cover only the ancestral savanna environment and the modern city and mountains, among others. However, to test our hypothesis, we used the other five great terrestrial biomes, considering the rainforest and deciduous forest as ancestral environments, according to the classification by Odum, to represent the scenarios of survival, maintaining the same situation of danger in all the scenarios. In addition, we displayed an image of the landscape corresponding to the respective environment to each participant so that the situation had another associated stimulus. The use of images is common in psychology studies that analyze, for example, the preference for landscapes by human beings, as carefully selected images promote a good representation of the actual environment. In this study, we used seven images used by Moura et al.. We describe all seven simulated scenarios used in. Below is an example of one of the simulated scenarios used in the experimente: 1. Scenario: “*Imagine that you are alone and sick in a rainforest*, *without basic materials for survival*. *In the coming days*, *you will need to find and use medicinal plants to treat this disease*. *We will show you a list of words*, *and would like you to assess the relevance of each of these words in your attempt to treat the disease and to survive in this environment*”. ## Participants We recruited 210 volunteers (age mean = 22.4 years; *SD* = 4.07), all students from the Universidade Federal Rural de Pernambuco, Brazil, including 51% women and 49% men, with the objective of analyzing how adaptive memory operates in different simulated survival environments with challenges related to the use of medicinal plants. Volunteers were recruited through direct contact with each volunteer. Before the experiment, the following question was asked: "*Do you know what medicinal plants are*?", with answer options as "yes" or "no". Volunteers who did not know what medicinal plants were were excluded from the study, as not knowing the model used as a challenge (medicinal plants) could bias the experiment. New recruitments were made to replace these participants. The Research Ethics Committee involving human beings at the Universidade de Pernambuco approved this study (decision number 2.944.271). All participants read and signed the Free and Informed Consent Form, which explained the procedures and objectives of the research. ## Procedure Participants were tested individually in an isolated room without any external interruption. The volunteers were divided into seven groups (n = 30 in each group) that differed by the type of simulated environment. These environments were considered ancestral—rainforest, deciduous forest, and savanna—and modern—coniferous forest, desert, tundra, and urban. The objective was to observe how information processing in memory varies depending on the type of environment. In the first part of the experiment, a simulated survival scenario was presented to the participants, and each participant was instructed to assess the relevance of 32 words to survive in the presented danger situation. Each word was exposed on a computer screen for 5 seconds. Immediately after that, there was a 2-minute distraction interval, which is the time necessary to avoid the tendency to remember the first elements of a list (primacy effect) and the last elements (recency effect), during which the participants filled out a form about demographics. After that, a surprise test of free recall was made, in which the participants were instructed to write the maximum number of evaluated words that they could remember on a sheet of paper, regardless of the order, within 10 minutes. Each experiment had a total duration of 30 minutes. We expected that the most relevant words were retrieved in the memory test when related to the ancestral environment. ## Data analysis To test our hypothesis that there is an ancestral priority in the cognitive strategies mobilized in the human mind to find medicinal plants to treat diseases, we developed mixed generalized linear models (GLMs) with binomial distribution, using the *R lme4* package. As the environment and relevance of the information are factors that influence memory in the recall test, the dependent variable was the *recall* of the words evaluated by the participants. Therefore, the independent variables were the types of simulated *environments*, the words’ *relevance*, and the *interaction* between these variables. As each participant has a different memorization capacity, we included the participants as a random factor to discard the effect of the difference in memorization among the three models developed. The effect size of the variables was measured using the R broom.mixed package. All analyses were performed in the R version 3.6.1 environment. We adjusted the model to obtain the lowest AIC value, as proposed by Agresti. Although the variable relevance when isolated, was not significant (assuming a significance level of p \< 0.001), we included it in the adjusted model (AIC = 9013.2), as analysis of the more complex model showed significant interactions between the variables (relevance + environment) that influenced the recall. For Agresti, when the variables interact significantly, the variables that make up this interaction should not be removed from the model. The resulting model explains about 4.5% of the variation observed in the recall (R2 marginal = 0.044). We also created a graph to observe the relationship between the variables using the ggplot2 package. # Results The relevance of the words influenced their recall by the participants (n = 210, which was 30 per scenario) in the modern urban, tundra, and desert environments, in addition to the ancestral environments of the savanna and rainforests (*p* \< 0.001). The analysis of our most complex model, which had a better fit (AIC = 9013.2), showed that the interactions between the relevance and environment variables had a significant effect on the model. We observed that the following interactions were significant: relevance + desert (*z* = 3.893; *p* = 9.90e-05), relevance + savanna (*z* = 3.008; *p* = 0.00263); relevance + rainforest (*z* = 4.694; *p* = 2.67e-06); relevance + tundra (*z* = 5.042; *p* = 4.61e-07); relevance + urban (*z* = 4.447; *p* = 8.72e-06). This means that the greater the relevance of information, which ranged from 1 to 5 on the Likert scale—to use medicinal plants to treat a disease in a given environment—in this case, the desert, savanna, rainforest, tundra, and urban environments—the more it is recovered in the memory test. However, the relevance + deciduous (*z* = 0.932; *p* = 0.35126) and relevance + coniferous (*z* = 0.784; *p* = 0.43313) interactions were not significant. The interactions show that for each unit of increased relevance, the chances of a word being remembered in the tundra environment are increased by approximately 35% (OR = 1.35), in the rainforest by 32% (OR = 1.32), in the urban environment by 30% (OR = 1.30), in the desert by 25% (OR = 1.25), and in the savanna by 19% (OR = 1.19). Each of the 32 words used in the experiment is presented in, along with descriptive information on the average classification of each word and the recall ratio (95% confidence interval) when processed in each environment. ## Comparison of word evaluations and the proportion of recall between environments We performed an additional Kruskal–Wallis test (non-parametric test) to compare the word evaluation and the proportion of recall between all survival scenarios. This test was chosen because our data had a non-normal and heteroscedastic distribution. The comparison of word evaluation values showed that there were no significant differences in the relevance given to the words in each environment (*H* = 1.12; *p =* 0.98). The descriptive analysis results of the words assessment are shown in. Regarding comparison of the proportion of words remembered in each environment, there were also no significant differences (*H* = 1.42; *p =* 0.96). The descriptive analysis results of the proportion correct recall are shown in. Thus, both the evaluation and the proportion of word recall were similar between the environments. However, as noted in the GLMM analysis, the greater the relevance of information in the tundra, rainforest, urban, desert, and savanna environments, the more are its chances of recovery. # Discussion This study comprised an analysis of the functioning of adaptive memory in different simulated environments rejected the hypothesis that there is only ancestral priority in the way it operates, i.e., it is not restricted only to the savanna, rainforests, and deciduous forests. We provided evidence that relevant information to search for medicinal plants in the desert, rainforest, and tundra environments was also recovered in memory. Based on our evidence, human cognition may have been selected slowly over evolutionary history in response to recurring challenges, present in diverse environments explored by hominids, such as the use of medicinal plants to treat diseases. According to Nairne and Pandeirada, the effects of survival processing on memory may be acting through a general survival optimization system that deals with varied challenges, selected for having helped human beings to face recurrent threats both in ancestral as well as modern environments. For example, the effects of survival processing were observed in different simulated scenarios, such as “city” and even in “outer space” and in situations involving zombie threats. In this sense, the human mind adaptations exist because they helped to solve recurrent and relevant adaptive problems to survival or reproduction of the hominids during evolutionary history. Since in our study in the deciduous forest and coniferous forest environments there was no significant interaction with the relevance of the information, this may reflect the absence or less occurrence of certain challenges in these environments. However, this needs to be analyzed in future studies. In fact, the environments explored by hominids are not restricted to the savanna of the African continent and may include other biomes (see). For example, there is evidence that the oldest hominid, *Graecopithecus freybergi*, inhabited a savanna environment in the region of Greece, between 7.37 and 7.11 million years (Ma). In addition, according to Zhu et al., fossil evidence indicates that hominids dispersed from the African continent towards East Asia at least 2.1 Ma. This suggests that even at the beginning of the Pleistocene, hominids entered Asia, adapting to different geographic and climatic configurations. For a long time, there was a consensus that *H*. *sapiens* originated in South Africa, but fossil evidence involving recent discoveries in the Gruta da Aroeira in Portugal, added to the evidence that *H*. *sapiens* inhabiting the Morocco region had a mixture of *H*. *sapiens* fossil characteristics from other parts of Africa, suggest a multicentric genesis for our species. Therefore, this evidence suggests that several regions may have been important during the evolution of humans. Although the savanna is still considered the main environment of human evolution, fossil evidence indicates that the origin and large divisions of hominids may have occurred outside Africa. There is even evidence of hominid adaptations in humid and dense environments. Among these adaptations, we can include the use of fire, and traces of foraging activities in the interior of the rainforest. The idea of hominid origin and evolution in rainforests is not new. Andrews had already proposed the possibility of the first hominids to have evolved in both closed and open forests. This proposition has gained strength as new paleoanthropological evidence of human evolution in the rainforests has emerged. In this sense, given that cognitive mechanisms result from responses to relevant and recurring selective pressures from ancestral environments, a challenge for evolutionary psychologists would be to expand their analysis in a way that covers the adaptive problems of different environments in which we evolve and the influence of the evolved psychological mechanisms in solving problems different from those found only in the African savanna. For example, dealing with infections was a challenge in several environments during the Pleistocene, and success in this, regardless of the environment in which such an adaptive problem was present, may have been crucial for the survival and reproduction of hominids. This would help explain, for example, why Yang et al. found that important information for survival is recovered both in ancient survival scenarios as well as in modern environments. The work by Silva et al. also showed that the challenges encountered in current environments are important for our cognition. We do not deny the important role of the ancestral past in the construction of the human mind, but it is essential to abandon the idea that we are still mentally predisposed to the ancestral environments of the Pleistocene. We provide evidence that human respond adaptively to challenges related to the use of medicinal plants present in ancient as well as modern scenarios. In this case, we assume that human beings have inherited adaptations from their hominid ancestors; however, human activities carried out in different environments over time can accelerate biological evolution and may modify genetically inherited predispositions. Barrett argues that certain adaptations can provide plasticity to the human mind. According to the author, it is possible that the human mind integrates both general and specific mechanisms, which were shaped during evolutionary history and by the individual’s ontogenetic development. As per Barrett, the cognitive processes of human beings can operate through mechanisms of heterogeneous origin, with new structures evolving from older structures and ancestral characteristics combined with relatively recent characteristics. This aspect of human cognition may have been important in reducing the incompatibility between ancestral and modern environments. According to Tooby and Cosmides, evolutionary processes are slow and require several generations to develop complex psychological mechanisms. This argument supports the idea that we still act in the face of imposed environmental problems in a similar manner to our hominid ancestors during the Pleistocene. A classic example is the preference for fatty and sweet foods, which is behavior adapted to ancestral environments with little fat availability, but poorly adapted in the current environment, thus increasing the incidence of cardiovascular illness. However, there is recent robust evidence of significant genetic changes in human populations. Although previous examples do not involve changes in cognitive aspects, it is probable that psychological mechanisms evolved in the Pleistocene may be subject to evolutionary processes that promote rapid changes in a few generations, being able to generate new cognitive structures that used “model” older mechanisms to adapt to today’s environments challenges. In this sense, evidence that people remember important information in modern environments, may indicate that there is no incompatibility, and that genetically inherited psychological mechanisms can be modified and adjusted to the current environment as well. Our study provides evidence that adaptive memory can operate through a flexible and adaptive cognitive mechanism, as argued by some other scientists as well. However, the absence of a control scenario in our study limits the interpretation of our results, making it important to analyze the mnemonic performance in other survival scenarios compared to a control scenario. From an evolutionary point of view, memorizing information related to a wide range of environments and recurring danger situations over the generations may have given the first hominids greater ability to survive. If this is true, cognitive adaptations do not just respond to situations related to a specific environment. The first hominids may have evolved psychological mechanisms, reusing pre-existing mechanisms, which adjusted to different environments during their evolution, not just being restricted to the Pleistocene era. Thus, future studies should investigate the factors leading to certain environments and not just ancestral environments, to intensify the effect of survival processing advantage on memory and to determine the cognitive mechanisms that act as stimuli for information prioritization in memory. # Conclusion The cognitive apparatus selected throughout evolution allows humans to survive and create survival strategies to face recurring challenges in various environments. Retrieving important information for survival related to the use of medicinal plants can be one of these strategies. We suggest that this is only possible if the human mind operates through an evolved versatility that gives it flexibility. This flexibility can reflect, for example, the different environments that the first hominids inhabited and the different and recurring situations of danger that they had to face. ## Limitations and future research The main limitation of this study was that the performance of memory was not compared to a control condition (for example, “moving” scenario commonly used in studies that analyze the survival paradigm). Therefore, this makes it difficult to understand whether the mnemonic performance was really good or bad between environments. In this sense, the absence of control limits our interpretation, which leads us to conclude that ancient and modern survival scenarios produce similar recall performance. Furthermore, we analyzed the effect of survival processing advantage on the use of medicinal plants to treat diseases in people who do not have continuous contact with this type of challenge or a dependence on this natural resource. Recruiting participants who do not experience the challenges present in simulated survival scenarios in the real world is common in experimental psychology studies (see studies). However, we argue this based on evidence that familiarity and previous experience with a recurring challenge in the environment influences how people perceive this risk. It was also observed in an experiment that medicinal plants previously known by the participants were prioritized in memory and were more easily remembered. Thus, we suggest that future studies use adjusted protocols to test adaptive memory in real systems, in which the population has already faced a challenging event at some point or that deals with the challenge presented by the researcher on a relatively daily basis. In addition, we suggest that future studies examine if prior knowledge about the use of medicinal plants to treat disease, or of another natural resource, is a critical factor in the performance of the survival processing advantage effect on memory. # Supporting information The authors would like to thank Dr. Marco Varella (Universidade de São Paulo, BR), Dr. Gustavo Taboada Soldati (Universidade Federal de Juiz de Fora, BR), and Dr. Josefa Pandeirada (Universidade de Aveiro, PT) for their insightful comments on the previous version of this paper; and to thank the Laboratory of Ecology and Evolution of Socioecological Systems at the Universidade Federal de Pernambuco for their physical and intellectual support. [^1]: The authors have declared that no competing interests exist.

# 1. Introduction Mineral resources are the necessities of human and economic development. With the continuous progress of society, the demand for mineral resources around the world continuously increases. The shallow resources are gradually exhausted, and open-pit mining gradually transits to underground mining. Underground mining will inevitably lead to a large number of goaf in the strata, which is the main cause of rock instability and safety accidents. In addition, the decline of groundwater level, surface subsidence and tailings reservoir pollution are also troublesome problems in underground mining. In order to solve the above problems, the filling mining method has become the preferred method for underground resources mining. At present, Portland cement is commonly used as the cementing material for tailings cementation. A great amount of Portland cement is filled underground, leading to high pH value and pollution of groundwater. Furthermore, carbon emission is inevitable in the process of cement production. The cement industry has become one of the main greenhouse gas emission industries. Therefore, finding a tailings cementing material that can replace Portland cement has become a hot scientific research direction. The use of environmentally friendly tailings cementitious materials is of great significance for reducing carbon dioxide emissions and achieving carbon neutrality as soon as possible. The microbial induced calcite precipitation (MICP) technique is a typical biomineralized consolidation method. In a certain physical and chemical environment, the free calcium ions in the solution are transformed into solid minerals dominated by calcium carbonate, so as to realize solidification of granular particles by controlling the biological organic matter. There are many kinds of solidification methods: urea hydrolysis, dentrification, ferric salt reduction, sulfate reduction, and photosynthesis etc., among which urea hydrolysis is widely used. Urea hydrolysis refers to the continuous production of highly active urease by specific microorganisms (urease bacteria) in the process of metabolism. Urease will promote the hydrolysis of urea to generate carbon dioxide and ammonia, which combine with water to generate carbonate ions and ammonium ions, and increase the alkalinity of the environment. When calcium ions are added externally, calcium carbonate precipitates are formed in the environment. The reaction equation is shown in (1–6). The mechanism of the MICP technique based on urea hydrolysis is relatively simple and a large amount of calcium carbonate is precipitated within a short period. It is highly efficient, sustainable and pollution-free, and hence has attracted attention from many researchers around the world. In 2004, Whiffin first proposed cementation of sand particles by the MICP technique, so as to improve the strength and stiffness of sand. With properly selected bacillus, Dick et al. observed calcium carbonate precipitation on the surface of weathered limestone by using the surface coating method, which greatly reduced the capillary water absorption coefficient of limestone surface. Tittelboom et al. repaired concrete cracks by using the MICP technique. With silica as the protective carrier of bacteria, 0.3 mm wide cracks can be completely repaired. By using porous aggregates as the carrier for bacteria and lactic acid bacteria, Wicktor et al. repaired cracks in concrete under immersion curing for 100 days. A great amount of white calcite precipitated in the cracks and the repair effect was good. Chu et al. found a proportional relationship between the uniaxial compressive strength of the sand column cemented by the MICP technique and the amount of calcium carbonate precipitation, and proposed a linear empirical equation between them. Van Paassen et al. conducted consolidation tests on a large volume of sand (100m³) by using the MICP grouting. After 12 days of cyclic grouting, about half volume of sand was cemented, indicating the the MICP technique can enhance the integrity of sand. Deng et al. mixed *Sporosarcina pasteurii* (*S*. *pasteurii*) and metal ore tailings directly and cured the mixture in a sealed environment. The effect of tailings cementation by the MICP was studied. Many scholars have conducted studies on repair and reinforcement of various materials, including concrete-based materials, soil and sand, by the MICP technique. Surface brushing or grouting was adopted in most of the previous studies, which cannot effectively reinforce the interior of sand or soil mass. <img src="info:doi/10.1371/journal.pone.0272281.e001" id="pone.0272281.e001g" /> C O N H 2 2 \+ H 2 O → C O 2 \+ 2 N H 3 <img src="info:doi/10.1371/journal.pone.0272281.e002" id="pone.0272281.e002g" /> C O 2 \+ H 2 O ↔ H 2 C O 3 <img src="info:doi/10.1371/journal.pone.0272281.e003" id="pone.0272281.e003g" /> 2 N H 3 \+ 2 H 2 O ↔ 2 N H 4 \+ \+ 2 O H − <img src="info:doi/10.1371/journal.pone.0272281.e004" id="pone.0272281.e004g" /> 2 O H − \+ H 2 C O 3 ↔ C O 3 2 − \+ 2 H 2 O <img src="info:doi/10.1371/journal.pone.0272281.e005" id="pone.0272281.e005g" /> C a 2 \+ \+ C a l l → C a l l − C a 2 \+ <img src="info:doi/10.1371/journal.pone.0272281.e006" id="pone.0272281.e006g" /> C a l l − C a 2 \+ \+ C O 3 2 − → C a C O 3 ↓ To sum up, a tailings cementation method by mixed facultative anaerobes (*Castellaniella denitrificans* or *C*. *denitrificans*) and *Sporosarcina pasteurii* (*S*. *pasteurii*) is proposed in this paper. The influences of different mixing modes on the cementation effect are investigated so as to improve the cementation effect. Based on the research in this paper, it is found that in the process of soaking and curing, the cementation effect is better when facultative anaerobic bacteria and aerobic bacteria are used to act together on the tailings. # 2. Test materials and device ## 2.1 Test materials ### (1) Tailings The tailings under study were collected from the tailings reservoir of an iron mine in Anshan, Liaoning, Northeast China. The main chemical composition of the tailings was determined by X-ray fluorescence analysis (XRF), which mainly include SiO₂ (77.57wt%), Al₂O₃ (13.59wt%), CaO (2.89wt%), Fe₂O₃ (2.73wt%) and MgO (2.40wt%), as well as some minor chemical components, as listed in. The crystal phase of tailings was determined by X-ray diffraction analysis (XRD). As shown in, the main minerals are quartz, merlinoite and coquimbite. In order to investigate the effect of grain size on tailings cementation, the original tailings were ground by a vertical planetary ball mill (XQM-4L) for 30 min, 60 min and 90 min, respectively, with the composition of tailings remaining unchanged. The original tailings were mixed with the three types of ground tailings in a mass ratio of 1:1:1:1 to prepare tailings specimens with a new particle size distribution. A laser particle size analyzer (Mastersizer3000 manufactured by Malvern Panalytical) was employed to analyze the particle size distribution of five tailings specimens and the results are presented in. ### (2) Microbes The bacteria used in this study were *S*. *pasteurii* and *C*. *denitrificans* obtained from China General Microbiological Culture Collection Center with a strain number of CGMCC 1.3687 and 1.10720, respectively. *S*. *pasteurii* is a kind of Gram-positive aerobic bacteria, while *C*. *denitrificans* is kind of Gram-negative facultative anaerobe. The microbes were cultured in nutrient medium containing 10 g/L peptone, 5 g/L NaCl, and 3 g/L beef extract. The pH value of the culture medium was adjusted to 7.0. The culture medium was sterilized in an autoclave before it was taken to the sterile bench for tests. The activated *C*. *denitrificans* and *S*. *pasteurii* were inoculated to the sterile culture medium, respectively, and put into a vibration incubator at 30˚C with an agitation speed of 140r/min. ### (3) Cementing solution The cementing solution mainly provides urea and Ca²⁺ for microbial induced calcite precipitation during the immersion process, meanwhile, it provides necessary nutrients for the growth and reproduction of *S*. *pasteurii*. According to the previous studies on the MICP technique based on the immersion method, the main composition of the cementing solution includes urea, CaCl₂, NH₄Cl, NaHCO₃, nutrient broth and deionized water. Based on the reaction mechanism of calcium carbonate precipitation induced by urea hydrolysis of *S*. *pasteurii*, the concentration of urea and CaCl₂ remained unchanged for studies on the influence of cementing solution concentration on the cementation effect. The concentration of NH₄Cl, NaHCO₃ and nutrient broth were 0.28 mol/L, 0.025 mol/L and 5 g/L, respectively. As nitrate ions and acetate ions are required to participate in dentrification by *C*. *denitrificans*, an approximate amount of Ca(NO₃)₂ and Ca(CH₃COO)₂ was added into the cementing solution in addition to the above-mentioned substances when mixed *S*. *pasteurii* and *C*. *denitrificans* were used. ## 2.2 Test device The device for immersion curing is shown in, which mainly consists of a plexiglass box with a size of 450 mm×400 mm×300 mm for containing cementing solution and tailings specimen; a magnetic stirrer, which mixes the cementing solution in the plexigalss box evenly by driving the magnetic rotor; a bracket with circular holes at equal intervals to facilitate full contact between the lower part of the specimen and the cementing solution; a thermostatic heating rod to maintain a constant temperature of the cementing solution; bubble stones to provide oxygen for the growth and reproduction of *S*. *pasteurii* in the plexiglass box; a thermometer for real-time display of the temperature and monitoring the temperature changes of the cementing solution. # 3. Test method ## 3.1 Cementation method Firstly, dry tailings and cultured *S*. *pasteurii* solution were mixed. Each tailings specimen was mixed with 70mL bacterial solution. The mixture was put into a steel mold (diameter 50 mm, height 100 mm, as shown) for shaping and sample preparation. In order to avoid loss of tailings particles during the contact process between the cementing solution and the tailing specimen, the specimen was wrapped with 1 mm thick geotextile and laterally confined by a 3D printed mold (as shown in), so as to prevent specimen deformation during immersion curing. Finally, the upper and lower ends of the specimen were wrapped by geotextile. 30 L cementing solution was injected into the plexiglass box, and the power supply of the magnetic stirrer and the thermostatic heating rod was connected. When a preset constant temperature was reached, the tailings specimen was put into the cementing solution and the air pump was turned on. After being immersed and cured under a constant temperature in an aerobic environment for 28 d, the specimen was taken out and dried for demolding. Tests were then conducted on the tailings specimen for analysis of the cementation effect. According to previous studies, the spray repair method based on MICP technology is limited to repairing and coating the surface cracks of the specimen. The injection and percolation repair methods based on MICP technology are limited to the application in earthwork construction, and the process is relatively complicated and the cost is high. If it is applied to the curing of tailings, the cost is too high and the economic benefit is poor. In this paper, the tailings are cured by immersion curing. Compared with spraying, injection and percolation repair methods, immersion curing has the advantages of good effect, simple technology and easy practical application. To study the combined effect of *S*. *pasteurii* and *C*. *denitrificans*, two mixing modes were adopted to mix the bacterial solution and tailings. For the first mode, the cultured *C*. *denitrificans* solution was centrifuged and the supernatant was taken out. The same volume of cultured *S*. *pasteurii* solution was poured in and dissolved to obtain a mixed solution containing two types of bacteria. The solution was then mixed evenly with the tailings and filled into the mold for specimen preparation. For the second mode, the two cultured bacterial solutions were mixed with the tailings respectively. The end of a 50 ml syringe with an inner diameter of 25 mm was cut off to make a cylindrical container and the length was marked. The mixture of tailings and *C*. *denitrificans* solution was filled in the syringe and compacted. The syringe was then put at the center of the steel mold and surrounded by the mixture of tailings and *S*. *pasteurii* solution. The tailings inside the syringe was slowly injected into the steel mold and compacted. Finally, the specimen with tailings mixed with *C*. *denitrificans* in the center and *S*. *pasteurii* in the surrounding was obtained. After shaping, the specimen was wrapped and cured by the method as the same as that for the tailings specimen mixed with *S*. *pasteurii* solution. ## 3.2 Test on mechanical strength Uniaxial compression tests and direct shear tests were conducted on the cemented tailings specimens. The device for uniaxial compression tests was Humboldt HM-5030 loading test machine manufactured in US. The precision of load sensor is 0.01N, the precision of displacement transducer is 0.001 mm and the maximum loading capacity is 50kN. The direct shear tests were carried out by YAW-3000 microcomputer-controlled electro-hydraulic servo rock mass direct shear test machine in Rock Mechanics Laboratory of Northeast University, China. A normal stress of 0.25 MPa was adopted. Cylindrical specimens with a dimension of ϕ50 mm\*100 mm were tested. For each group of tests, the same tests were conducted on five specimens and the average values were taken for analysis. The mechanical parameters of tailings specimens under different conditions were obtained and the influences of various factors on the cementation effect were investigated. ## 3.3 SEM analysis The SEM (Scanning Electron Microscope) analysis was conducted by the field emission scanning electron microscope in Analysis and Testing Center, Institute of Science and Technology, Northeast University, China. Model: Ultra Plus; Manufacturer: Zeiss, Germany; Resolution: 0.8 nm/15kv, 1.6nm/1kv; Accelerating voltage: 20V\~30kV; Magnification: 12\~1 million times; Secondary electron imaging; Fully digitized computer and manual parallel control system. After sample preparation and conductive coating, the material structure and morphology formed in the MICP process were scanned and analyzed by SEM. ## 3.4 Test on calcium carbonate content The CaCO₃ content in the cemented tailings specimen directly influences the cementation effect and the mechanical properties of the specimen. With appropriate adjustment based on the previous studies, the cemented tailings specimen was divided into 4 regions in order to analyze the uniformity of carbon carbonate precipitation, as shown in. The different regions of the specimen were separated and pulverized with a mortar. The specimen powder was then dried in an oven. Its weight was recorded as W₁. The dry specimen powder was soaked and washed with deionized water and dried in an oven. The dry weight at this moment was recorded as W₂. The dry powder was then soaked and washed with 0.1mol/ml HCl solution until no bubbles were observed. The specimen powder was then washed with deionized water, filtered and dried. The weight at this moment was measured as W₃. The difference between W₂ and W₃ is the weight of CaCO₃ in the cemented specimen. The CaCO₃ content can then be calculated according to the original weight W₁, by using. For each group of tests, 5 specimens were tested and the average values were taken. <img src="info:doi/10.1371/journal.pone.0272281.e007" id="pone.0272281.e007g" /> w CaCO 3 = W 2 − W 3 W 1 % # 4. Results and discussions ## 4.1 Influence of bacterial concentration on cementation effect As the optical density of bacterial solution is directly proportional to its concentration, the optical density corresponding to a wavelength of 600 μm (i.e., OD₆₀₀) was adopted to characterize the concentration of *S*. *pasteurii* solution. After being activated and cultured for 24h, the *S*. *pasteurii* solution was centrifuged and the supernatant was poured out. Based on the calculation results, different volumes of culture media were added to prepare bacterial solutions with OD₆₀₀ of 0.4, 0.8, 1.2, 1.6 and 2.0, respectively. The tailings were cemented by bacterial solutions with five different concentrations. In order to control the uniqueness of variables, the urea concentration in the cementing solution remained 0.5mol/L, and the original tailings without grinding were used. The tailings specimens were under immersion curing at a temperature of 30°C for 28 days. The dried specimens were tested for uniaxial compressive strength and direct shear strength. The average values were taken, as shown in. The CaCO₃ content was determined for each region and the average values were taken, as shown in. The SEM image of tailings specimen corresponding to an OD₆₀₀ value of 1.6 is presented in. As can be seen from, when OD₆₀₀ of the bacterial solution increases from 0.4 to 1.6, both the uniaxial compressive strength and the direct shear strength of the cemented tailings specimen increase significantly. The uniaxial compressive strength increases by 343.8% from 0.32MPa to 1.10MPa. The mechanical properties have been remarkably enhanced and can satisfy the strength requirement on mine backfill (the uniaxial strength of backfill has to be 0.7-2MPa for underground mining. However, when OD₆₀₀ of the bacterial solution increases from 1.6 to 2.0, both the uniaxial compressive strength and the direct shear strength of the cemented tailings specimen vary slightly. It can be seen from that, except slight change in Region III, the CaCO₃ content in the other three regions increases significantly when OD₆₀₀ of the bacterial solution increases from 0.4 to 1.6, and changes slightly when OD₆₀₀ increases from 1.6 to 2.0. The cementation effect of *S*. *pasteurii* gradually increases with increasing bacterial solution concentration. However, when OD₆₀₀ exceeds 1.6, the growth and reproduction of *S*. *pasteurii* reach a certain balance and the increase of bacterial solution concentration has little impact on the number of bacteria in the specimen. As can be seen from, a large amount of CaCO₃ is formed and closely bound in the cemented specimen, which greatly enhances the specimen strength. Therefore, when OD₆₀₀ exceeds 1.6, further increase of bacterial solution concentration cannot improve the concentration effect of microbes on the tailings. This part of the research proves that the MICP technology based on *S*. *pasteurii* can solidify the tailings into specimens with certain strength under the condition of immersion curing. It is confirmed that *S*. *pasteurii* has the potential to replace Portland cement as the cementing material of tailings. However, the urea hydrolysis of *S*. *pasteurii* will increase the emission of ammonia, which will also have a certain impact on the environment. In the follow-up, it can be considered to solve this problem by adjusting the concentration of reactants and adding ammonia gas absorbent. ## 4.2 Influence of cementing solution concentration on cementation effect In order to investigate the influence of cementing solution concentration on the cementation effect of the MICP technique based on immersion curing, the cementing solutions with urea concentration of 0.25 mol/L, 0.50 mol/L, 0.75 mol/L, 1.00 mol/L and 1.25 mol/L, respectively, were prepared for tests. The *S*. *pasteurii* solution with an OD₆₀₀ value of 1.6 and the original tailings were mixed for sample preparation. The specimens were immersed in the cementing solutions with various urea concentration and cured at 30°C for 28d. The uniaxial compressive strength and direct shear strength were measured for dry specimens and the average values were taken, as shown in. The CaCO₃ content in different regions were measured and calculated and the average values were taken, as shown in. presents the SEM image of the specimen cemented in a cementing solution with urea concentration of 0.25mol/L. As can be seen from Figs and, with the increase of cementing solution concentration, the uniaxial compressive strength, the direct shear strength and the CaCO₃ content in each region of the cemented tailings specimen have similar trends. When the urea concentration in the cementing solution increases from 0.25mol/L to 0.75mol/L, the mechanical properties and the CaCO₃ content in each region of the cemented tailings specimen are enhanced gradually. The function of urea in the cementing solution is mainly to provide energy and raw materials for urea hydrolysis by *S*. *pasteurii*. When the concentration is below a certain level, the increase of urea concentration can promote the urea hydrolysis by *S*. *pasteurii*. As sufficient calcium ions are available in the cementing solution, the stronger the urea hydrolysis is, the more calcium carbonate precipitation is, and the stronger the mechanical properties of the cemented tailings specimen are. However, when the urea concentration in the cementing solution increases from 0.75 mol/L to 1.25 mol/L, the CaCO₃ content in each region decreases slightly, and the uniaxial compressive strength and the direct shear strength of specimen decreases rapidly. It can be seen from that, when the urea concentration is 0.25 mol/L, although a few CaCO₃ particles are formed on the tailings particles, they are not bound as a whole and cannot fill up the gaps between tailings particles. Hence, the specimen strength is relatively low. When the urea concentration in the cementing solution is higher than 0.75 mol/L, the increase of urea concentration can no longer promote, instead restrains the urea hydrolysis by *S*. *pasteurii*. The main reason is that too high urea concentration leads to an environment unsuitable for the growth and reproduction of *S*. *pasteurii*. Consequently, the urea hydrolysis process is slowed down, less CaCO₃ is formed in the specimen, and the mechanical properties are weakened. ## 4.3 Influence of tailings particle size on cementation effect In order to investigate the influence of tailings particle size on the cementation effect, the original tailings were ground by a vertical planetary ball mill for 30 min, 60 min and 90 min, respectively, and the original tailings were mixed with the three types of ground tailings in a mass ratio of 1:1:1:1 to prepare a tailings material with a new particle size distribution. Five types of specimens, i.e., the original tailings (ground for 0min), the tailings ground for 30 min, 60 min, 90 min, and the mixture in a mass ratio of 1:1:1:1, with the same composition and different particle size distributions were obtained. The *S*. *pasteurii* solution with an OD₆₀₀ value of 1.6 was mixed with the tailings having various particle size distributions to prepare the tailings specimens. The specimens were immersed in the cementing solution with a urea concentration of 0.5mol/L at 30°C for 28 d. Tests were conducted on the dry specimens for uniaxial compressive strength and direct shear strength. The average values were taken, as shown in. The CaCO₃ content in each region was measured and calculated. The average values were taken and shown in Figs and presents the SEM image of the cemented tailings specimen which was ground for 90 min. As can be seen from Figs and, among the tailings specimens with five types of particle size distributions, the cementation effect is the best for the original tailings. The compressive strength and the direct shear strength of the original tailings after cementation are much higher than those of ground specimens. The reasons are: for immersion curing, the cementing solution can easily pass larger tailings particles and enter the interior of the specimen, and participate in the urea hydrolysis process of *S*. *pasteurii*, leading to precipitation of CaCO₃. When the grinding time is longer, the proportion of small particles in the tailings is higher, it is harder for the cementing solution to enter the specimen and participate in the urea hydrolysis process. In addition, the smaller the particle size is, the fewer the pores between the particles are. *S*. *pasteurii* is a kind of aerobic bacteria, which can only survive in the presence of oxygen. When there are fewer pores, most *S*. *pasteurii* would die and cannot hydrolyze urea due to hypoxia. Hence, the cementation effect is poor. It can be seen from, when the proportion of small or medium-size particles is too high, little CaCO₃ is formed in the pores between small particles and hence cannot bind fine particles. Therefore, the strength of the cemented tailings which was ground for 90 min is relatively low. Among the tailings specimens with different particle size distributions (except the original tailings), the cementation effect for the specimen with a mass ratio of 1:1:1:1 is the best, as the existence of larger particles in the specimen with a mass ratio of 1:1:1:1 provides support and enough space for the survival and growth of *S*. *pasteurii*. Furthermore, the small particles act as fine aggregates between pores. Hence, the coarse and fine particles can still be cemented when the CaCO₃ crystals formed by the MICP process are small and the cementation effect is good. ## 4.4 Influence of curing temperature on cementation effect When studying the influence of temperature on the cementation effect, the *S*. *pasteurii* solution with OD₆₀₀ of 1.6 was mixed with the original tailings and the specimens were cultured in a cementing solution with urea concentration of 0.5 mol/L. The specimens were under immersion curing at 20°C, 25°C, 30°C, 35°C and 40°C, respectively, for 28 d. The uniaxial compressive strength and the direct shear strength were tested for dried specimens and the results are shown in. The CaCO₃ content in each region was measured and calculated, as presented in. According to the curves shown in, the uniaxial compressive strength and the shear strength of the cemented specimen reach the peak values at a curing temperature of 30°C, indicating that 30°C is the optimal temperature for urea hydrolysis by *S*. *pasteurii* and precipitation of CaCO₃. 30°C is also the most suitable temperature for culturing of *S*. *pasteurii*, which has high activity and can grow and reproduce rapidly at this temperature. When the temperature exceeds 30°C, the mechanical properties of the cemented specimen weaken rapidly with increasing temperature. High temperature is unfavorable for the survival of *S*. *pasteurii*. Most *S*. *pasteurii* would die and cannot hydrolyze urea any more. According to the curves shown in, the variation trend of CaCO₃ content in Regions I, II and III is similar to that of mechanical properties of the cemented specimen, i.e., reaching the peak at a temperature of 30°C, being less at a temperature higher or lower than 30°C. However, the temperature has little impact on the CaCO₃ content in Region III, which is in the center of the specimen and less affected by the curing temperature. In Region III, before the temperature decreases or increases to an value unfavorable for *S*. *pasteurii*, a certain amount of CaCO₃ has already formed in the center. As can be seen from, a great amount of CaCO₃ is formed in the interior of the specimen cured at 30°C. The pores between tailings particles are well filled and the cementation effect is good. Hence, the strength of the specimen cured at 30°C is relatively high. ## 4.5 Distribution of products According to the above results, the variation trends of the uniaxial compressive strength and the direct shear strength of cemented tailings specimens are similar to that of CaCO₃ content in each region. However, the CaCO₃ content is obviously different in various regions. The CaCO₃ precipitation in each region of the cemented specimen under various test conditions is summarized and presented in. It can be seen that the CaCO₃ content in the top (Region I) and side (Region II) of the cemented tailings specimens subjected to immersion curing is higher, followed by the bottom (Region IV) and the interior (Region II). As the *S*. *pasteurii* is a kind of aerobic bacteria, sufficient amount of oxygen is required for its urea hydrolysis process and the pH value of the environment is increased, leading to binding of calcium ions to carbon dioxide and precipitation of calcium carbonate precipitation. Although an air pump is connected to the test device, oxygen can hardly enter the interior of the specimen due to few pores between tailings particles. Therefore, the microbes in the top and side of the specimen can contact with more oxygen, and more calcium carbonate is formed in these regions. Some microbes in the interior of the specimen die due to hypoxia and can no longer hydrolyze urea. Although the bracket is perforated, the holes are gradually blocked with the accumulation of calcium carbonate. Hence, the CaCO₃ content at the bottom of the specimen is lower compared to that at the top. In addition, it is difficult for the cementing solution to enter the interior through tailings particles, resulting in the lack of urea and calcium ions and thus less calcium carbonate precipitation in the interior. ## 4.6 Comparison of cementation effect under different bacteria mixing modes In order to solve the uneven distribution of CaCO₃ in each region of the specimen cemented by the MICP technique with aerobic bacteria, the urea hydrolysis by *S*. *pasteurii* and the denitrification by *C*. *denitrificans* were combined. Two mixing modes, i.e., direct mixing and mixing by using a syringe, were adopted to mix the bacterial solution and the tailings. The mixed specimens were immersed and cured under the same conditions. The uniaxial compressive strength, the direct shear strength and the CaCO₃ content in each region were measured for dried specimens, as shown in. As can be seen from, under the same condition, the average CaCO₃ content in Region III is 2.9% when only *S*. *pasteurii* is used. The CaCO₃ content in Region III of the specimen prepared by the direct mixing mode (Mode I) is 5.1%, which is increased by 75.9%. The CaCO₃ content in Region III of the specimen mixed by a syringe (Mode II) is 6.5%, which is increased by 124.1%. When both *S*. *pasteurii* and *C*. *denitrificans* are used for tailings cementation, the CaCO₃ content in Region III is greatly enhanced by either of the two mixing modes. It indicates that denitrification of facultative anaerobes is dominant in the anoxic environment inside the specimen. The CaCO₃ distribution in various regions is relatively uniform and more CaCO₃ is precipitated inside the specimen mixed by Mode II. As both *S*. *pasteurii* and *C*. *denitrificans* exist in the interior of the specimen mixed by Mode I, *S*. *pasteurii* cannot hydrolyze urea due to hypoxia, but it competes for living space and nutrient with *C*. *denitrificans*. As a result, some *C*. *denitrificans* cannot carry out the denitrification process normally. However, in the specimen mixed by Mode II, only *C*. *denitrificans* exist in the interior, hence more CaCO₃ is formed. Although the CaCO₃ content increases in the specimen with mixed bacteria, the formation rate of CaCO₃ by denitrification is slower compared to that by urea hydrolysis. The amount of CaCO₃ due to denitrification is only 62% of urea hydrolysis, and cannot reach the CaCO₃ content at the top and side of the specimen. As the CaCO₃ content increases slightly in the specimen with mixed bacteria, it has little impact on the overall mechanical properties of the specimen. The uniaxial compressive strength and the direct shear strength of the specimen cemented by mixed bacteria increases by 9.3% and 14.3%, compared with the specimen cemented by *S*. *pasteurii* only. # 5. Conclusions In this paper, experimental studies were carried out on the influencing factors for cementing tailings specimens by the MICP technique with *S*. *pasteurii* under immersion curing. The single variable method was adopted to investigate the influences of bacterial solution concentration, cementing solution concentration, tailings particle size and temperature on the cementation effect. The following conclusions can be drawn: 1. With the increase of bacterial solution concentration, the uniaxial compressive strength and the direct shear strength increased first and then gradually decreased. When OD₆₀₀ of the *S*. *pasteurii* solution was 1.6, the cementation effect was no longer enhanced rapidly with increasing bacterial solution concentration. High urea concentration constrained the growth of *S*. *pasteurii* to a certain extent. When the urea concentration was 0.75 mol/L, the cementation effect was optimal. With the increase of temperature, the uniaxial compressive strength and the direct shear strength tended to increase first and then decrease. The cementation effect was optimal at a temperature of 30°C. 2. The CaCO₃ content in each region of the specimen varied in a trend similar to that of uniaxial compressive strength and direct shear strength. The CaCO₃ content was positively correlated with the mechanical properties of the specimen. 3. Two mixing modes were adopted to mix *S*. *pasteurii* and *C*. *denitrificans* with the tailings. The test results indicated that combined action of aerobic bacteria and facultative anaerobes can improve the amount of CaCO₃ precipitation and result in more uniform distribution of CaCO₃ in the interior of the specimen. In addition, the cementation effect in the specimen mixed by a syringe was improved more obviously. # Supporting information 10.1371/journal.pone.0272281.r001 Decision Letter 0 Pimraksa Kedsarin Academic Editor 2022 Kedsarin Pimraksa 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 May 2022 PONE-D-21-40689Experimental study on tailings cementation by MICP technique with immersion curingPLOS ONE Dear Dr. Changyu Jin, 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. ============================== ACADEMIC EDITOR: Please carefully check the writing before re- submission.============================== Please submit your revised manuscript by June 13, 2022. 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, Kedsarin Pimraksa, 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\. We note that the grant information you provided in the ‘Funding Information’ and ‘Financial Disclosure’ sections do not match. When you resubmit, please ensure that you provide the correct grant numbers for the awards you received for your study in the ‘Funding Information’ section. 3\. Thank you for stating the following in the Funding Section of your manuscript: “This work was supported by the Fundamental Research Funds for Central Universities, China (N2101041).” We note that you have provided funding information that is not currently declared in your Funding Statement. However, funding information should not appear in the Acknowledgments section or other areas of your manuscript. We will only publish funding information present in the Funding Statement section of the online submission form. Please remove any funding-related text from the manuscript and let us know how you would like to update your Funding Statement. Currently, your Funding Statement reads as follows: “The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.” Please include your amended statements within your cover letter; we will change the online submission form on your behalf. 4\. We note that 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. 5\. 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>. \[Note: HTML markup is below. Please do not edit.\] **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: 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: 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: In the manuscript titled „Experimental study on tailings cementation by MICP technique with immersion curing” experimental study on MICP performance in the cementation of tailing under different condition are presented. Some experimental findings regarding the influences of bacterial concentration, concentration of urea solution, particle size, temperature and mixture mode of bacteria were addressed. Although the reviewer think the most results are reasonable and straightforward, as many qualitative as well as quantitative investigations on the different influence factors on MICP treatment are already available in the current literature, compared to other researches the study in this manuscript seem to be not deep enough. There is lack of quantification and comprehensive analysis of the experimental observations. Thus, it’s hard to get more insight into MICP cementation of tailing based on this study. Further, the reviewer questions also the novelty and scientific contributions of the present study. On one side, usually, one of main contributions of the experimental study is that the experiments can provide data base which can be used as reference for other experimental study as well as for validation the numerical study. However, in the manuscript only few data points are available. On the other side, only some qualitative conclusions are addressed in the manuscript, and the tests were carried out on specific tailing material in laboratory scale. Therefore the experimental findings cannot be applied to other further studies. Except for the main concerns above, the design of some tests is so far not clear to me (please see the specific comments below). Moreover, the language aspect and format should be checked (some suggestions with regard to the language and format aspects can also be found in the specific comments below). Specific comments: • Line 15: what does “ore energy” mean? Does it mean ore and energy? • Line 36: There should be always a space between worlds and bracket. e.g. should be “underground mining (Li et al. 2021; Wang et al. 2018)” instead of “underground mining(Li et al. 2021; Wang et al. 2018)” • Lines 45-47: the expression sounds a bit weird. It should be rephrased • Lines 54-57: The description of urea hydrolysis is improper. Urea hydrolysis actually refer to the reaction process, in which urea is consumed, and carbonate ion and NH4 are produced. The authors should also mention that in the MICP treatment expect for urea and bacteria, Ca2+ should be additionally provided in the system. For better explanation of the chemical reactions by MICP, one should give the chemical equation here. • Section 4.1: Any explanation for the experimental observation that after excess 1.6 OD, no significant increase of calcite production was observed? Actually this observation is quite common. With the increase of bacterial amount, the urease activity is improved. However, after reaching a certain bacterial mass competition among the bacteria increases due to the insufficient nutrients which on the contrary leads to the increase of bacterial decay thus decreases the activity. This can usually be seen in the growth rate of biomass. By means of changing the ambient environment of bacteria e.g. changing the nutrition receipts, or providing more nutrition) the bacteria activity can be further improved. Considering this, it is more interesting to investigate the optimal ratio of the bacterial concentration with respect to the urea/calcium concentration rather than to study these two concentrations separately. • Section 4.2: not sure if the conclusions drawing here make sense. Firstly, this observation is of no general applicability. Second, if one want to establish the relationship between urea concentration and bacterial urease activity, it is better to analyze the urease activity of bacteria in cases of different combination of bacteria concentration and urea concentration based on monitoring the decrease rate of urea mass during the test (e.g see. Xiao et al., 2020). • Section 4.3: Only uniform and homogenous distribution of different particle sizes is investigated. However, in the in-situ MICP cementation of tailing, the heterogeneous particle size distribution in a large ´field is expected. • Section 4.4: To analyze the relationship between temperature and spatial distribution of produced calcite, one should look into the urease activity of bacteria at different temperature (see e.g. Xiao et al., 2020). Reference: Xiao, Y., Wang, Y., Wang, S., Evans, T. M., Stuedlein, A. W., Chu, J., … Liu, H. (2021). Homogeneity and mechanical behaviors of sands improved by a temperature- controlled one-phase MICP method. Acta Geotechnica, 4. Reviewer \#2: The authors have presented a work performed to stabilize the tailing by MICP. The work is original and can be recommended for publication IN PLOS ONE. However, the following comments must be carefully addressed. 1\. The authors should be careful when writing the scientific names of bacteria species. The species name should be in italic - revise throughout the manuscript. 2\. There should be a space between number and units. For example, 70mm should be written as 70 mm. 3\. I wonder, what is the new finding in the manuscript, comparing to the published article? If it is optimization, that has already been published in previous works. Authors should highlight this in the latter part of the introduction section. 4\. When writing OD600, 600 should be subscripted. 5\. What is the advantage in immersion curing, compared to the spraying/ injection/ percolation? How the proposed method is suitable for tailing applications? The discussion should be expanded in a way to answer these questions. 5\. Similar to Ordinary Portland cement, the use of MICP also has adverse effect related to the ammonia emission/ release. The authors should briefly discuss the negative side of MICP technique, and should propose the possible methods to overcome this issue. Several recently published works confirmed that the ammonium by-products can be managed by struvite precipitation (<https://doi.org/10.1007/s13762-021-03138-z>), zeolite use (<https://doi.org/10.1520/GTJ20170353>), and etc. The authors are recommended to refer those papers and expand the discussion. 6\. I still confused about one thing. Can this be practical to implement MICP to tailing? In the MICP, the bacteria as well as resources should reach the depths to got treated at those levels. However, the tailing materials are more likely to be very fine (in the range of nano scale). Will it be possible to penetrate the bacteria? Moreover, tailing used to have organic content. A recent study showed only a limited improvement is possible in fine and organic soil materials. Refer the following paper (<https://doi.org/10.3389/fenvs.2021.690376>) and enhance the discussion, which helps to improve the manuscript standard. 7\. Regarding the concentration of cementation solutions, 0.25mol/L, 0.50mol/L, 0.75mol/L, 1.00mol/L and 1.25mol/L, what is the demerit in using 1 mol/L or 1.25 mol/L? Few previous works recommend the use of 1 mol/L. Scientific explanation may be needed. 8\. In some places, 3 in CaCO3 is not subscripted. Revision needed throughout. 9\. Is this the first article using Castellaniella denitrificans? \*\*\*\*\*\*\*\*\*\* 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.0272281.r002 Author response to Decision Letter 0 29 Jun 2022 Responses to reviewer comments have been submitted in file format. 10.1371/journal.pone.0272281.r003 Decision Letter 1 Pimraksa Kedsarin Academic Editor 2022 Kedsarin Pimraksa 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 Jul 2022 Experimental study on tailings cementation by MICP technique with immersion curing PONE-D-21-40689R1 Dear Dr. Changyu Jin 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, Kedsarin Pimraksa, PhD Academic Editor PLOS ONE Additional Editor Comments (optional): Reviewers' comments: 10.1371/journal.pone.0272281.r004 Acceptance letter Pimraksa Kedsarin Academic Editor 2022 Kedsarin Pimraksa 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 2022 PONE-D-21-40689R1 Experimental study on tailings cementation by MICP technique with immersion curing Dear Dr. Jin: 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. Kedsarin Pimraksa Academic Editor PLOS ONE [^1]: The authors have declared that no competing interests exist.

# Introduction Bacille Calmette-Guérin (BCG) is the only available vaccine for the prevention of tuberculosis and has been given to over 3 billion individuals, making it the most widely administered vaccine to date. BCG is typically administered soon after birth, and while it is effective at preventing TB during childhood, its effectiveness wanes over time. To that end, the efficacy of BCG against adult pulmonary TB is highly variable, ranging from 0–80%. Due to the success of BCG in reducing childhood TB, and its proven safety record, strategies to develop a more effective TB vaccine have focused on improving the efficacy of BCG, either through the development of recombinant BCG strains and attenuated *Mycobacterium tuberculosis* (*Mtb*) vaccine strains or through the development of subunit and virus-vectored vaccines that can be used as a boost for BCG. In that regard, most of the novel TB vaccines currently in the vaccine pipeline are designed to incorporate BCG or attenuated *Mtb*. Optimizing the delivery of this live bacterial vaccine is a further way in which the efficacy of BCG could be improved. Oral delivery of BCG has many advantages over the standard intradermal method of BCG vaccination, including reduced cost, ease of administration, avoidance of needles and the associated risk of disease transfer. More importantly, it has been shown that oral delivery more effectively targets the mucosal immune response than intradermal vaccination. This is critical, given that the primary site of TB infection is the lungs. BCG is most effective when delivered as a live vaccine. We have previously reported that oral delivery of BCG in a lipid formulation protects the bacilli from degradation in the stomach and provides immunity against an aerosol *Mtb* or *Mycobacterium bovis* challenge in mice and guinea pigs. Moreover, oral BCG vaccination has been shown to boost antigen-specific immune responses in human volunteers, and reduce the incidence of virulent *M. bovis* in livestock and wildlife. Immunity to TB is highly dependent upon CD4⁺ T cells and the acquisition of a T helper cell type 1 (Th1) immune response. Control of a primary TB infection is reliant on the production of IFNγ and TNFα in the lungs by CD4⁺ effector T cells. These cytokines activate infected macrophages, enabling them to kill or restrict the growth of the invading mycobacteria. The requirements for a protective memory response to TB are less clear. Lung-resident CD4⁺ T cells appear to be the principal mediators of protection, since following BCG vaccination lung-resident memory T cells have been shown to be sufficient for protection against a mycobacterial challenge. However, the level of IFNγ produced by CD4⁺ T cells is not a reliable predictor of vaccine efficacy. More recently, efforts have focused on measuring the quality of the vaccine-induced immune response by assessing the proportion of multifunctional Th1 CD4⁺ T cells, which are capable of simultaneously producing high levels of IFNγ, TNFα and IL-2. The increased presence of these cells in the spleens, and perhaps more importantly the lungs, of vaccinated mice has been shown to correlate with protection against TB. Although the vaccine-elicited immune response to parenteral BCG vaccination is well characterized, the CD4⁺ T cell immune response to lipid- formulated oral BCG vaccination is unknown. In this study, we compare the magnitude and quality of the CD4⁺ T cell response in the spleen and lungs of mice induced by orally delivered lipid-formulated BCG to the response induced by subcutaneous (s.c.) BCG vaccination. We report that lipid-formulated oral BCG vaccination (Liporale™-BCG) induced a long-lived CD4⁺ T cell response, demonstrated by the persistence of activated, antigen-specific CD4⁺ T cells in the spleens and a greater number of multifunctional CD4⁺ T cells in the lungs than s.c. vaccinated mice at 30 weeks post immunization. These findings suggest that Liporale™-BCG vaccination could present an effective means of delivering novel BCG-based vaccines for the prevention of TB. # Materials and Methods ## Ethics Statement This project was undertaken within the provisions of the Animal Welfare Act (1999) of New Zealand and was approved by the Victoria University of Wellington Animal Ethics Committee. ## Mice Inbred C57BL/6 mice were purchased from The Jackson Laboratory and bred and housed under SPF conditions at the Malaghan Institute of Medical Research Biomedical Research Unit in Wellington, New Zealand. Groups of 5 age- and sex- matched mice were used at each time point for each experimental group. ## BCG Preparation and Vaccination Mice were vaccinated per oral or subcutaneous route, as indicated, with *M. bovis* BCG Danish strain 1331. BCG was grown to mid log-phase in 175 ml flasks (Falcon, NJ, USA) containing Middlebrook 7H9 medium (Difco, Detroit, MI, USA) supplemented with albumin-dextrose-catalase (BBL, Becton Dickinson, MD, USA) and 0.01% Tween 80. BCG in vaccine preparations was enumerated by plating onto modified Middlebrook 7H11 agar (Difco, Detroit, MI, USA) containing oleic acid- albumin-dextrose-catalase ((BBL, Becton Dickinson, MD, USA) and glycerol and counting retrospectively after incubation for 2–3 weeks. For formulating the oral vaccine, broth-grown BCG bacilli were pelleted by centrifugation and encapsulated into Liporale™ as previously described (13). For subcutaneous vaccination, 50 uL of 7H9 medium, containing approximately 1×10⁶ CFU BCG, was injected into the right flank. For oral BCG administration, mice were temporarily separated into individual cages and offered 0.3mL chocolate-flavored Liporale™ containing 1–2×10⁷ CFU BCG. After 12 hours, the oral vaccine had been entirely consumed and mice were placed back in their original cages. ## Tissue Preparation At times indicated, mice were culled by cervical dislocation. Single lymphocyte suspensions were prepared from spleens of mice by passing them through a 70 µm cell strainer and subjecting them to red blood cell lysis (Red Blood Cell Lysing Buffer, Sigma, St. Lois, MO). Lung lymphocytes were isolated by enzymatic digestion of lung tissue (2.4 mg/mL Collagenase Type I (Invitrogen, Carlsbad, CA), 0.12 mg/mL DNase 1 (Roche, Mannheim, Germany) in Iscove’s Modified Dulbecco’s Medium (IMDM) without additives), and CD45⁺ cells were isolated using magnetic bead enrichment with CD45 MicroBeads (MACS, Miltenyi Biotec, Germany). Total cell counts per organ were determined using a haemocytometer following red blood cell lysis of spleens or following CD45 MicroBead isolation from lungs. ## In Vitro Restimulation Single cell lymphocyte suspensions from spleens and lungs were plated at a density of 4×10⁶/mL in a 24 well plate and incubated for 6 hours at 37°C in IMDM (supplemented with 5% FCS, 1% penicillin/streptomycin, 1% L-glutamine (GlutaMAX, Gibco, Invitrogen, Auckland, New Zealand), 0.1% 2-mercaptoethanol (Gibco, Invitrogen)) containing 2 µg/mL anti-CD3 (clone 2C11) and 2 µg/mL anti-CD28 (clone 37.51) (both prepared in house). 3 µg/mL Brefeldin A (eBioscience, San Diego, CA) and 2 µM monensin (Sigma) were added for the last 4 hours of incubation. ## Identification of Tetramer-specific Cells Single cell suspensions from spleens were stained with I-A(b) *Mtb* antigen 85B precursor 280–294 (FQDAYNAAGGHNAVF) tetramer-APC or with I-A(b) human class II- associated invariant-chain peptide (PVSKMR MARPLLMQA) tetramer-APC as a negative control (NIH MHC Tetramer Core Facility at Emory University, Atlanta, GA), and enriched by positive magnetic bead isolation using the AutoMACS cell sorter (Miltenyi Biotec) following staining with anti-APC MicroBeads (MACS, Miltenyi Biotec) and were identified by flow cytometry. For phenotype analysis we used a minimum number of 50 events in the tetramer gate, with a mean of 180 events. ## Flow Cytometry Lymphocytes were labeled with anti-CD4-Pac Blue, (BD Biosciences, San Diego, CA), anti-CD44-PE-Cy7, anti-CD62L-APC-AlexaFluor 750 (both from eBioscience) for cell surface staining, and anti-IFNγ-PE-Cy7, anti-IL2-APC (both from eBioscience), and anti-TNFα-PE (BD Biosciences) for intracellular staining. Dead cells were excluded following staining with the viability dye LIVE/DEAD® Fixable Blue Dead Cell Stain (Invitrogen). All samples were collected on a BD LSRII SORP (Becton Dickinson, San Jose, CA) and FlowJo software version 9.4 was used for data analysis. ## Statistical Analysis Statistical significance was determined by one-way ANOVA followed by the Tukey post-test, two way ANOVA followed by the Bonferroni post test, or by the Mann Whitney test, as indicated in figure legends, using Prism software. # Results ## Liporale™-BCG Significantly Increases the Number of Ag85B-specific CD4⁺ T Cells in the Spleen To compare the immune response elicited by Liporale™-BCG to the response induced by s.c. BCG vaccination, we isolated lymphocytes from the spleens of vaccinated C57Bl/6 mice at 4, 8 and 30 weeks post immunization. Naïve mice served as unvaccinated controls. Splenic CD4⁺ T cells specific for the immunogenic BCG and *Mtb* antigen Ag85B were identified using an MHCII-Ag85B tetramer. Importantly, only mice that received Liporale™-BCG had a significant increase in the number of Ag85B-specific cells in the spleen compared to naïve controls at 8 and 30 weeks following immunization. ## Liporale™-BCG Induces an Ag85B-specific CD4⁺ Effector T Cell Phenotype in the Spleen To distinguish between naïve, effector (T_EFF)/effector-memory (T_EM) and central memory (T_CM) CD4⁺ T cell populations, total CD4⁺ lymphocytes from naïve mice or Ag85B-specific CD4⁺ T cells from vaccinated mice, were stained for the expression of CD62L, a lymphoid homing receptor that is downregulated shortly following T cell activation, and the activation marker CD44. CD4⁺ T cell subsets were categorized as follows: naïve cells as CD62L^hi, CD44^lo; T_CM as CD62L^hi, CD44^hi and T_EM/T_EFF as CD62L^lo, CD44^hi. At 4 weeks post immunization, the Ag85B-specific CD4⁺ T cells from the spleens of mice vaccinated with Liporale™-BCG or s.c. BCG displayed a predominantly T_EM/T_EFF phenotype ( a and b). The antigen- specific CD4⁺ effector immune response was maintained up to 8 weeks post immunization, illustrating that Liporale™-BCG can induce an antigen- specific CD4⁺ effector T cell response in the spleen that is equivalent to s.c. BCG vaccination. By 30 weeks post immunization, \>50% of the Ag85B-specific CD4⁺ T cells in the spleens of vaccinated mice were T_CM, consistent with reports demonstrating that central memory cells are capable of long term survival. Interestingly, there was a significant increase in the proportion of T_CM cells in mice that received BCG orally compared to mice that received a s.c. BCG immunization (p\<0.05), suggesting that Liporale™-BCG vaccination can induce a long-lived antigen specific memory immune response. ## Liporale™-BCG Induces a Long-lived CD4⁺ Effector T Cell Phenotype in the Lungs To investigate the immune response to lipid-formulated oral BCG vaccination in the lung, we isolated lymphocytes from the lungs of mice 4, 8 and 30 weeks post oral or subcutaneous BCG vaccination. We determined the phenotype of the total CD4⁺ T cell population using antibodies against CD62L and CD44 to distinguish between naïve, T_EM/T_EFF and T_CM subsets. Similar to the antigen-specific response in the spleen, at 4 and 8 weeks post immunization, Liporale™-BCG induced a T_EM/T_EFF CD4⁺ T cell phenotype in the lungs that was equivalent to the immune response observed when BCG was administered subcutaneously. Importantly, by 30 weeks following immunization, only mice that received Liporale™-BCG maintained a CD4⁺ T_EM/T_EFF phenotype relative to naïve controls. The maintenance of a CD4⁺ T_EM/T_EFF cell population in the lungs of orally vaccinated mice 7 months after immunization suggests that oral BCG vaccination induces a prolonged effector response in the lungs compared to subcutaneous vaccination. ## Cytokine Production by CD4⁺ T Cells from the Lungs of Liporale™-BCG Vaccinated Mice To assess the quality of the immune response in the lung following Liporale™-BCG vaccination, lymphocytes were isolated from the lungs of naïve or BCG-vaccinated mice and restimulated for 6 hours *in vitro* in the presence of Brefeldin A and monensin. Cytokine producing CD4⁺ T cells were identified by intracellular staining and flow cytometry. At 4 and 8 weeks after vaccination, there was a significant increase in the proportion of CD4⁺ T cells producing IFNγ or TNFα, but not IL-2, in the lungs of mice that received either Liporale™-BCG or s.c. BCG vaccines compared to naïve controls ( a and b). Of note, by 30 weeks post immunization, there was a significant increase in the proportion of IFNγ-producing CD4⁺ T cells from the lungs of mice that received Liporale™-BCG vaccination relative to cells from s.c. BCG vaccinated mice and naïve controls ( a and b). The frequency of multifunctional CD4⁺ T cells in the lungs of vaccinated and naïve mice was determined using Boolean gating. At 4 and 8 weeks after vaccination with either Liporale™-BCG or s.c. BCG there was a significant increase in the percentage and number of multifunctional CD4⁺ T cells in the lungs of mice compared to cells from naïve controls ( a, b and c). The proportion of IFNγ⁺TNFα⁺IL-2⁻ double positive CD4⁺ T cells also was also significantly increased in the lungs of Liporale™-BCG or s.c. BCG vaccinated mice compared to naïve controls ( b and c). Interestingly, at 30 weeks after immunization, there was a significant increase in the frequency of multifunctional CD4⁺ T cells in the lungs of mice that received Liporale™-BCG compared to s.c. vaccinated or naïve controls. # Discussion Oral delivery of live BCG using a lipid formulation has been shown to be effective at protecting animals against a virulent mycobacterial challenge, ; however, the immune response elicited by this vaccine had not been fully investigated. Therefore we compared the CD4⁺ T cell response in mice vaccinated with Liporale™-BCG to mice vaccinated with BCG through the conventional s.c. route. We observed a significantly increased number of Ag85B tetramer-specific CD4⁺ T cells in the spleens of mice vaccinated with Liporale™-BCG and found that oral vaccination induced T_EFF/T_EM and T_CM CD4⁺ T cell populations in the spleen that were similar to s.c. BCG vaccinated mice. Moreover, following polyclonal stimulation, we found that mice vaccinated with Liporale™-BCG had significantly more IFNγ-producing, and multifunctional CD4⁺ T cells in the lungs, the primary site of TB infection, than s.c. BCG vaccinated or control mice \>6 months after vaccination. The earlier control of bacterial growth observed following pulmonary mycobacterial challenge of memory immune mice has been shown to coincide with the early arrival of antigen-specific Th1 CD4⁺ T cells in the lungs. Supporting this, it has been demonstrated that adoptively transferred activated, transgenic Th1 ESAT-6-specific cells traffic to the lungs and protect from an *Mtb* challenge in a frequency dependent manner. Moreover, using FTY720 to block egress of lymphocytes from the lymph nodes, it has been shown that T cells in the lungs of BCG vaccinated mice are sufficient to protect against a mycobacterial challenge. Together, these studies suggest that the presence of Th1 CD4⁺ T cells in the lungs is critical for the immune protection afforded by vaccination. We found that both Liporale™-BCG vaccination and s.c. BCG vaccination led to a statistically significant increase in the frequency of IFNγ-producing CD4⁺ T cells in the lungs at 4 and 8 weeks after immunization, which coincided with the presence of a predominantly effector phenotype of the CD4⁺ T cells in the lungs. By 30 weeks post vaccination, a significant population of T_EFF/T_EM CD4⁺ T cells capable of producing IFNγ was identified only in the lungs of mice that were vaccinated with Liporale™-BCG, suggesting that the oral route of vaccination produces a more sustained immune response in the lungs than traditional s.c. vaccination. Although IFNγ is necessary to control an *Mtb* infection, the IFNγ response induced by TB vaccination is an unreliable correlate of vaccine-elicited protection. For this reason we also assessed the frequency of multifunctional cells in the spleens and lungs of vaccinated mice, since these cells have been shown to correlate with vaccine-elicited protection from *Mtb* in mice. We found an increase in the proportion and frequency of triple cytokine producing cells in the lung at 4 and 8 weeks in mice vaccinated with BCG orally or s.c., but interestingly, only Liporale™-BCG maintained a significant population of multifunctional CD4⁺ T cells in the lungs of mice by 30 weeks post vaccination. Therefore, the oral, mucosal vaccination route maintains the multifunctional CD4⁺ T cell population in the lung for longer than the traditional s.c. route of immunization. By contrast, Kaveh *et al*. recently reported that intradermal BCG vaccination induces a long-lived population of multifunctional CD4⁺ T cells in the lungs. In this study an increase in the proportion of multifunctional cells in vaccinated mice compared to naïve controls was detected at 6 weeks post immunization, however, it is unclear whether the percentage of multifunctional cells in BCG vaccinated mice were above that found in naïve mice at 6, 12 and 18 months after vaccination because unvaccinated controls were not included at later time points. The phenotype of the multifunctional cells in the spleens of intradermally BCG vaccinated mice was reported as T_EM–like (CD44^hi, CD62L^lo); however, it is important to note that this phenotype was assessed after an 18 hour *in vitro* restimulation, and it is likely that the phenotype of these cells was altered by the restimulation. In our study we found that Ag85B tetramer-specific CD4⁺ T cells in the spleen maintained an effector phenotype up to 8 weeks following either s.c. or Liporale™-BCG vaccination. By contrast, at 30 weeks post vaccination, over 50% of the antigen-specific CD4⁺ T cells in the spleens of vaccinated mice had a T_CM phenotype, with a significantly higher proportion of T_CM in the orally vaccinated mice. Similar to the early response in the spleen, we observed a pool of CD4⁺ T_EFF/T_EM cells at 4 and 8 weeks in the lungs of mice that received BCG either orally or subcutaneously. This population of CD4⁺ T_EFF/T_EM cells was maintained up to 30 weeks post Liporale™-BCG vaccination. It is possible that the low level of persisting antigen in the lungs following oral BCG vaccination contributed to the maintenance of T_EFF/T_EM cells in the lung up to 6 months after vaccination; however it should be noted that an earlier study demonstrated that viable bacteria could not be recovered from the lungs by 8 weeks following oral BCG vaccination. A further possible mechanism for the extended presence of T_EFF/T_EM lymphocytes following oral BCG vaccination is tissue-specific homing, in which T cells preferentially migrate to the tissues in which they were primed. There is evidence that T cells primed in mucosal lymphoid sites, such as the mesenteric lymph nodes and Peyer’s patches, express homing markers and chemokines specific for the major mucosal sites in the body, the lung and the intestine. Following Liporale™-BCG vaccination, live mycobacteria are predominantly found in mucosal lymphatic tissues, such as the mesenteric lymph nodes, cervical lymph nodes and the Peyer’s patches, and is therefore thought to be where T cell priming occurs. Interestingly, lymphocytes isolated from the spleen of mice vaccinated with Liporale™-BCG did not express the mucosal homing molecules CD103 or α4β7, but differentially expressed β1 integrin, which has been shown to be involved in T cell homing to the lung epithelium. Whether the antigen-specific CD4⁺ T cells recovered from the spleens of Liporale™-BCG vaccinated mice express β1 integrin remains to be determined. We have shown previously that Liporale™-BCG vaccination effectively protects against *Mtb* infection and in this present study we have provided evidence that orally delivered, Liporale™-BCG vaccination induces a strong antigen-specific CD4⁺ T_EFF/T_EM and T_CM response, which appears superior to s.c. BCG vaccination. Due to the ability of BCG to protect against childhood TB, most TB vaccines currently in clinical trial incorporate either recombinant BCG, attenuated *Mtb* strains or boosting regimes to maintain this protection. Given the long-lived immune response we see in the lungs of mice following Liporale™-BCG vaccination, we speculate that the delivery of novel live TB vaccines via the oral route may more effectively target the mucosal immune response than traditional immunization routes, enhancing protection against aerosol *Mtb* infection. We would like to thank Cornelia Walker for her technical assistance. [^1]: FEA is an Inventor on a patent related to the Liporale technology and is a Director on a company (Immune Solutions Ltd) which has a vested interest in commercializing Liporale as an oral delivery platform. Liporale is covered by the following pending patents and applications: “Antigenic Compositions” WIPO WO/2003/009868 (PCT/NZ2002/00132, NZ 546141, AU 2002/326233, US 2004/0234533, EP 02760915.5, CA 2454920, ZA 2004/1211, CN 02817408.9, IN 00302/DELNP/2004, JP 2003-515260, HK 04109263.7. This study was partly commercially funded by Otago Innovations Ltd. This does not alter the authors’ adherence to all the PLOS ONE policies on sharing data and materials. [^2]: Conceived and designed the experiments: LRA FEA JRK. Performed the experiments: LRA FJR. Analyzed the data: LRA JRK. Contributed reagents/materials/analysis tools: FEA. Wrote the paper: LRA JRK.

# Background Policy makers and program managers, particularly those in low- and middle-income countries (LMIC), often face prioritization decisions with limited resources. Economic evaluation, such as cost-effectiveness analyses (CEA), can provide insight into the comparative value of various health interventions and therefore help inform priority setting. Since the original Panel on Cost‐Effectiveness in Health and Medicine proposed the use of a reference case as a benchmark of quality and methodological rigor, various guidelines for conducting economic evaluation have been proposed. The Consolidated Health Economic Evaluation Reporting Standards (CHEERS) Checklist, a widely cited reporting guideline, is used to ensure study results are reported with clarity and accuracy, yet does not provide methodological guidelines for how analyses should be conducted. Many countries, particularly high-income ones, have also developed their own reference cases to inform decision-making in their health care systems. In contrast, most low- and middle-income countries (LMICs) have not developed such guidelines, possibly due to their limited capacity to do so. In fact, only 12 LMICs currently have economic evaluation guidelines specific to their country. Although the general principles of guidelines for high-income countries can still be applied to LMICs, variations in both approaches and methods used limit their usefulness. For example, most high-income country guidelines suggest health outcomes be measured using quality-adjusted life-years (QALYs). The estimation of QALYs requires a preference weight for different health states, called health-related quality of life, on which LMICs often have limited data. To address the need for a reference case that could broadly apply to different contexts, particularly in LMICs, the Bill and Melinda Gates Foundation (BMGF) supported the development of the Gates Reference Case to ensure high quality and transparent CEA in global health priority setting. One of the key recommendations is to support the use of disability-adjusted life years (DALY), as disability weights are more readily available and more easily transferable across different countries. The first version was published in 2014 as the Gates Reference Case and, later in 2016, was renamed the International Decision Support Initiative (iDSI) Reference Case to convey the breadth of its intended applicability. The iDSI reference case fills a major gap in global health economics, as it is the only resource of economic evaluation best practices for many policymakers in LMICs looking for guidance on resource prioritization. However, no study has assessed whether the development of the guideline is associated with an improvement in research practice for CEAs employing DALYs. This paper aims to quantify the methodological and reporting quality of cost-per-DALY averted studies over time, as measured by adherence to best practices enumerated by the iDSI reference case. # Methods ## Data ### The iDSI reference case The iDSI reference case includes 11 principles: transparency, comparator, evidence, measures of health outcome, costs, time horizon/discount rate, perspective, heterogeneity, uncertainty, budget impact, and equity considerations. Each principle has a number of corresponding methodological specifications and reporting standards. By using this tiered structure, the reference case aims to serve as a framework that provides best practice guidance while allowing for flexibility depending on context, and thus is the most appropriate economic evaluation guideline for LMICs without their own national guidelines. ### Global health CEA registry We analyzed data from the Tufts Medical Center Global Health CEA Registry, a continually updated database of English-language economic evaluations in the form of cost-per-DALYs averted. Among 620 cost-per-DALY averted studies in the database, we selected all articles published three years before and after the initial release of the iDSI reference case (2011–2017) to examine the impact of its publication on the literature (N = 398). We focused particularly on economic evaluations using the DALY metric because it is recommended as a main outcome metric by the iDSI reference case and it is used more often as a health outcome measure in LMICs than equivalent metrics such as the QALY. To ensure a comprehensive assessment of adherence to the reference case, two independent readers (JE and AP) extracted additional information from each study in our sample using REDCap, an online data collection platform, including data on: currency reported; subgroup analyses conducted; limitations reported; structural sensitivity analyses conducted; budget impact conducted; justification of alternative methodology; and comparator setting. ## Adherence score We first translated all 30 methodological specifications and 38 reporting standards (across 11 principles) listed in the reference case into questions with discrete binary outcomes (standard satisfied or standard not satisfied). We then designated reference case elements as “required” or “optional” based on our interpretation of the language in the report (Table A). We deemed 19 methodological specifications and 21 reporting standards “required”. Our base-case analysis examined adherence scores consisted only of “required” elements. We evaluated each published cost-per-DALY averted study’s adherence to methodological specifications (0–19 items) and reporting standards (0–21 items). We then separately calculated reporting and methods raw scores as a percentage of total possible points, measured as normalized adherence score (0% = no adherence, i.e., no requirements adhered to; 100% = full adherence, all requirements adhered to). ## Analysis ### Descriptive analysis We examined the association between adherence score and certain study characteristics, including whether the study cited the reference case, the study funder characteristics, and journal attributes. We categorized study funders into the following groups (not mutually exclusive): academic, government, healthcare organization, industry, intergovernmental organization, BMGF, non- BMGF, and other. We also stratified selected articles into clinical versus non- clinical journals using SCImago Journal Rank’s subject categorization (medicine vs. health policy, public health, non-health). Finally, we recorded 2016 journal impact factor quartiles and categorized studies as high impact (first quartile), medium impact (second quartile), or low-impact (third and fourth quartiles). ### Statistical analysis To examine whether the iDSI guideline has, since its release in 2014, improved the methodological and reporting practices of cost-per-DALY averted studies, we calculated mean adherence scores by year from 2011 to 2017. We conducted a pre- post analysis of improvement in methodological and reporting adherence scores. As the reference case was first released in January of 2014, we considered that year to be the reference case’s dissemination period, and hence did not include articles published during that year in our pre-post analysis. We also compared the overall methodological and reporting adherence scores, stratified by the 11 principles. ### Sensitivity analysis We conducted three sensitivity analyses. First, we included the “optional” criteria in the calculation of adherence scores for a random 10% subset of the articles to explore the impact of including optional items in the adherence score. Second, given that efforts to increase awareness of new guidelines may take longer than one year, and subsequent development and publication of adherent CEAs can span more time, we conducted a sensitivity analysis to explore alternate dissemination period lengths. Primarily, we expanded the dissemination period from 2014 to 2014–2015 to examine the influence of a longer dissemination period on adherence. Third, we used an alternative classification to determine adherence to the comparator principle’s corresponding methodological specification. In our base case analysis, we designated an article adherent to the iDSI’s comparator methodological specification only if the article explicitly reported their comparator as the “standard of care”. In this sensitivity analysis, we classified an article as adherent so long as it specified a comparator other than “do nothing” or some other non-action. To be consistent with the iDSI reference case principle that the standard of care must include at least “minimal supportive care”, we designated “do-nothing” interventions as non-adherent. # Results ## Descriptive statistics Among 398 cost-per-DALY averted studies published from 2011–2017, 215 (54%) focused on LMICs and 263 (68%) targeted communicable diseases, such as diarrhea, HIV/AIDs, tuberculosis, and malaria. Articles averaged 60% adherence to the reference case’s methodological specifications and 74% adherence to reporting standards. summarized iDSI Reference case normalized adherence scores by year, sponsor, and journal aspects (The raw scores are available from Table B). No article achieved full adherence to either the methodological specifications or the reporting standards. Of the 213 articles published after 2014 (i.e. 2015–2017), only 9 (4%) cited the iDSI reference case. For articles that did so, adherence to reporting standards averaged 79%, five percentage points higher than mean adherence for the full sample, while adherence to methodological specifications did not differ from adherence for the full sample. Funding source (BMGF vs. non-BMGF) was not significantly associated with a change in adherence scores for either methodological (mean score of 60% vs. 60%) or reporting (mean score of 75% vs. 74%). Studies published in clinical journals had marginally higher adherence (60% methodological adherence, 74% reporting adherence) than studies in non-clinical journals (57% methodological adherence, 73% reporting adherence). On average, methodological adherence scores for articles published in high-impact journals exceeded the corresponding scores for studies published in low-impact journals (61% vs. 50%); for reporting adherence, the corresponding difference was 74% vs. 71%. Methodological adherence did not improve after publication of the reference case compared to the pre-2014 period (59% adherence pre-2014 vs. 60% post-2014, p = 0.53). Reporting standard adherence slightly increased (72% adherence pre-2014 vs. 75% post-2014, p\<0.01) ( and Table C). ## Methodological adherence versus reporting adherence scores Across the 11 principles, reporting standard adherence exceeded methodological specification adherence by 14 percentage points (74% vs. 60%). Reporting adherence score were highest for the following principles: uncertainty (mean of 100%), comparator (97%), and evidence (95%). Methodological adherence scores were highest for the outcome measure (100%), transparency (89%), and evidence (74%) principles. Methodological adherence scores were higher than reporting adherence scores for the following principles: transparency (89% methodological adherence score vs. 86% reporting adherence score), outcome (100% vs. 54%), and costs (65% vs. 54%). Reporting adherence scores exceeded methodological adherence scores for the following principles: comparator (36% methodological adherence score vs. 97% reporting adherence score), evidence (74% vs. 95%), time horizon/discounting (57% vs. 82%), perspective (64% vs. 85%), and uncertainty (57% vs. 100%). Articles seldom addressed the principles of budget impact (10% methodological adherence score, 9% reporting) or equity (7%, 7%). ## Sensitivity analyses Inclusion of optional criteria in our adherence score calculation decreased mean methodological adherence by 14 percentage points (60% to 46%) and mean reporting adherence by 22 percentage points (from 74% to 52%). When we increased the dissemination period to 2014–2015 (base case: 2014), we found no change in our results. Using an alternate comparator principle classification (base case: comparator must be standard of care; alternative: comparator can be any intervention other than “do-nothing”) also had little impact. # Discussion Since its introduction in 2014, adherence to the iDSI reference case among published cost-per-DALY averted studies has improved for reporting standards, but not for methodological specifications. Adherence to the reference case’s reporting standards exceeds adherence to its methodological specifications, perhaps reflecting the relative ease of revising the way information is presented and greater effort needed to conform to analytic requirements. Moreover, other reporting guidelines, such as CHEERS or country-specific recommendations, may have independently promoted more rigorous reporting, with the unintended effect of boosting adherence to the iDSI reference case. However, methodological and reporting adherence scores varied substantially across reference case principles, demonstrating ways in which articles are falling short of guidelines. For example, articles almost always report their comparator clearly (as recommended by reporting standards), but do not necessarily specify whether the comparator is considered standard of care (as recommended by methodological specifications). Similarly, all articles reported findings from sensitivity analyses, but did not always conduct comprehensive structural, probabilistic, and deterministic sensitivity analyses. In some cases, methodological adherence exceeded reporting adherence. For example, articles often included implementation costs (as recommended by methodological specifications), but did not as frequently report these costs in both US dollars and local currency (as recommended by reporting standards). Because the methodological specifications and reporting standards address distinct issues, future guidelines should continue to include recommendations for both types. It is important to consider what level of adherence should be seen as satisfactory. Although articles in our sample were more adherent to reporting guidelines, they adhered to just over half of methodological specifications. Adherence scores were notably lower for particular principles—heterogeneity, budget impact, and equity—indicating overall neglect of these issues in cost- per-DALY averted studies. The adherence scores are perhaps best thought of as a baseline against which to measure improvement, and as a call to action to promote higher quality and comparability. The lack of adherence to the iDSI reference case might reflect the competing influences of other guidelines, as authors may prioritize adherence to local guidelines that are more relevant to their context. For example, the South African pharmacoeconomic guidelines recommend a base case 5% discount rate, which differs from the 3% value recommended by the reference case. Although the iDSI reference case supports the use of alternative discount rates where appropriate to the decision problem and constituency, published CEAs that adhere to the local guidelines may be scored as non-adherent to the methodological specifications in this analysis. Another possible explanation for relatively low adherence for certain items is that authors may not be aware of the guidelines. We found that only 4% of the identified studies published after 2014 directly cited the iDSI reference case. The BMGF and iDSI have focused educational campaigns on national payers and health technology assessment (HTA) agencies in LMICs, rather than on researchers, who are primary authors of published studies. Future studies should examine whether the reference case has influenced country-specific guidelines, such as Thailand’s HTA assessment guideline. ## Limitations The primary limitation of our study is that the post-evaluation period (2015–2017) may not have been sufficiently long to detect the impact of the reference case. Though it was initially released in early 2014, as noted, the iDSI reference case was not officially published in an academic journal until 2016. However, dissemination efforts began in 2013 at a BMGF-hosted workshop for multi-sectoral stakeholders, which was later considered “a major part of the Gates reference case development”. Although more time may be needed for the field to adopt these guidelines, as new CEAs can take years to conduct and publish, we believe our results on adherence to the iDSI reference case can serve as a baseline estimate. Adherence should be re-analyzed in the future as the field continues to grow. Furthermore, our use of dichotomous (i.e., “yes/no”) questions to score adherence may be inconsistent with the more nuanced goals of the iDSI reference case. Because the reference case is designed to be applicable in a range of different country-specific contexts, it must balance the goals of study comparability and quality against the goal of local applicability. To address this limitation, we omitted “optional” standards from our adherence calculation for the base case. That is, we assumed that the “optional” elements represent conditional requirements intended by the reference case authors to allow for local adaptability. Our sensitivity analysis that included all elements in our calculation of the adherence score (i.e. both the “required” and “optional” elements) yielded lower adherence scores. Assessing adherence to the comparator principle posed a particular challenge because this assessment requires subjective judgment on whether the specified comparator constitutes the “standard of care.” Although our sensitivity analysis of altering the definition of appropriate comparator had little impact on our findings, a “do nothing” intervention, which is deemed inappropriate by the iDSI’s comparator methodological specification, can be regarded as “standard of care” for some conditions in some settings, such as a population screening program for tuberculosis. Also, our findings cannot be generalized to the rest of the economic evaluation literature as the Tufts Medical Center Global Health CEA Registry catalogs only published cost-per-DALY averted studies. For example, our analysis excluded gray literature (i.e., material not disseminated in regularly published, indexed journals). Gray literature may be more prevalent in some countries, especially those without local guidelines. Finally, our approach for scoring articles inherently involves reviewer judgment to determine author intent and to resolve ambiguities (e.g., determining whether the comparator is “clearly” stated). We attempted to mitigate this problem by having two reviewers read each article and, in cases of no consensus, we appealed to a third reviewer. ## Policy implications As posited by Nugent and Briggs, future research on the subject should ask, “what specific help does the iDSI reference case offer the analyst, who, while attempting to conform to the principles, nevertheless has to choose and implement the methods?” It is possible that the methodological guidelines impose an excessive burden on researchers, raising “issues about the resources and data requirements to meet the principles”. Future qualitative research can focus on researcher consideration of best practice guidelines in study design and reporting, and on how to increase guideline acceptance among authors. Studies could also further evaluate the methods and reporting adherence for articles that strongly adhere to the iDSI reference case, as these analyses may serve as useful examples for other CEA authors. Analysis of the impact of the reference case on perceived quality and usefulness of economic evaluations by decision makers would be useful. Moving from guideline development to implementation is a vital step towards improving the quality of economic evidence in global health. Future efforts could include additional educational workshops for researchers, students, and policymakers. Policymakers and major funders of economic evaluations, such as the BMGF, could require that researchers adhere to reference case recommendations in grant applications. Journals and reviewers should also impose high-quality standards for economic evaluations. Moving forward, journals may require reviewers to fill out a rubric similar to the instrument in our study that measures the adherence of economic evaluations to the iDSI reference case guidelines. # Conclusion Since its initial launch in 2014, our study indicates that the development of the iDSI reference case is associated with improving reporting standards for economic evaluation focused on global health, but no improvement in methodological practice. Although the reference case has substantial potential to serve as a resource for researchers and policy makers in global health and economics, more effort to promote adherence and awareness may be needed. # Supporting information [^1]: The authors have declared that no competing interests exist.

# Introduction The growth of a tumor beyond a certain small size is highly dependent upon the development of an independent vasculature. This “angiogenic switch”, which may not be activated until years after the initial neoplasm has developed, has until recently generally been viewed as having two origins. The first, termed “sprouting” angiogenesis, or simply angiogenesis, is a consequence of the re- directing of pre-existing neighboring blood vessels into the tumor bed, whereas the second, termed vasculogenesis, stems from the recruitment of bone marrow- derived endothelial cell (EC) precursors. Both processes may operate independently and concurrently and are regulated by specific inductive signals originating from the tumor and/or its associated inflammatory cell infiltrate. Once developed, the neo-vasculature becomes the tumor’s main source of oxygen and nutrients, whose delivery is no longer diffusion-dependent. It also serves as a major portal for tumor cell invasion of the systemic circulation, thus abetting metastasis. Given the neovasculature’s importance in supporting many of the tumor’s most critical metabolic needs and functions, much effort has been devoted to the design of anti-angiogenesis therapies. One attraction of this approach is that the cells comprising the vasculature are presumed to be genomically stable and thus less predisposed to the development of the therapy resistance commonly encountered in tumor cells. Despite this major theoretical advantage, clinical trials with these approaches have extended patient survival only modestly and the eventual development of resistance is actually quite common, although its cause remains largely unexplored. We and others have described a third origin for the EC population of several tumor types in the form of “tumor-derived ECs” (TDECs), which arise stochastically from tumor epitlelium. TDECs express EC-specific markers, contain Weibel-Palade bodies, form tube-like structures in semi-solid media, and incorporate into functional tumor vasculature. These properties distinguish TDECs from tumor cells undergoing vasculogenic mimicry, a reversible process involving a partial epithelial-mesenchymal transition (EMT). As expected for actual tumor cell descendants, TDEC karyotypes are distinctly abnormal and contain tumor-specific chromosomal markers. These findings suggest that, like their ancestral tumor cells, TDECs might possess inherent genomic instability (GI) and thus be prone to develop resistance to anti-angiogenic therapies. Consistent with this idea, we show here that upon continued in vivo serial passage, the TDEC population acquires a more pronounced EC-like phenotype. Whole genome single nucleotide polymorphism (SNP) array analysis indicates that serially-passaged TDECs also acquire a small number of genomic amplifications and deletions. The known genes contained within these affected regions are disproportionately involved in functions pertaining to tumor suppression, vasculogenesis, and EMT. These findings support the idea that the long-term propagation of TDECs leads to the emergence of a more functionally robust clonal population that is facilitated by underlying GI. # Results ## TDEC Recovery Increases with Serial Passage The H460 human lung cancer cell line was originally tagged with lentiviral vectors separately encoding green fluorescent protein (GFP) and resistance to G418. Tumor xenografts from these cells were initially grown in immuno- compromised nude mice and the total human and murine EC populations were isolated with anti-CD31-coupled magnetic beads as previously described. Approximately 40% of the total “tumor-associated ECs” (TAECs) expressed GFP, were G418-resistant and contained only human chromosomes, thus indicating that they were true TDECs. These “serial passage 0” (SP0) TDECs were then mixed with a 20-fold excess of untagged H460 cells and re-propagated with tumor xenografts. Upon reaching maximal allowable size, each tumor was excised and its its total TAEC population, now referred to as SP1, was again isolated. The GFP+ TDECs, which comprised approximately 5% of the entire TAEC population, were then purified and expanded for several days in G418-containing medium, combined again with a 20-fold excess of untagged H460 tumor cells and propagated as tumor xenografts. Several such serial *in vivo* passages were performed with the percent of retrieved GFP+ TDECs in the total TAEC population being determined periodically. While the earliest passage TDECs (SP1 and SP2) consistently yielded about 5% GFP+ TDEC recovery, later passages (SP4–SP6) provided for recovery rates in excess of 30%. This was consistent with the observation that these later passage TDECs were more frequently associated with tumor blood vessel endothelium. To establish further that the late passage cells could out-compete their early passage counterparts, we derived SP0 TDECs from an H460 tumor tagged with a red fluorescent protein (dsRed) rather than GFP. The SP0 (dsRed+) TDECs and SP6 (GFP+) TDECs were then mixed in equal numbers, combined with untagged H460 cells at a 1∶20 ratio as before, and grown as tumor xenografts. Just prior to total TAEC harvesting, fluorescence imaging of the mouse was performed using a Xenogen IVIS-200 *in vivo* imaging system. As seen in, the dsRed signal localized to discrete regions, that recapitulated the distribution previously seen for tumors containing early passage TDECs, while the GFP signal was more evenly distributed throughout the tumor. Following the isolation of TAECs, the red and green TDEC populations were quantified by direct UV microscopy. In three independent tumor xenografts, there was an average of \>6-fold greater abundance of GFP+ TDECs than of dsRed+ TDECs. Taken together, these observations indicate that late passage TDECs outcompete early passage TDECs, even when the two populations co- exist under identical *in vivo* conditions. ## Late Passage TDECs Express Higher Levels of some EC-specific Markers We previously reported that early passage H460-derived TDECs, as well as those originating from several other tumor cell types, stably express numerous EC- specific markers. Having demonstrated that late passage TDECs outcompete early passage TDECs and incorporate into the tumor vasculature more efficiently, we next asked whether they do so in association with higher level expression of these markers. Thus, we used flow cytometry to follow the cell surface expression of CD31 on TDECs during the course of serial *in vivo* passage and observed that it increased by nearly 30-fold. This increase was confirmed by immunoblotting. Similarly, several other EC-specific markers, including uptake of acetylated LDL (AcLDL), expression of E-lectin receptor, von Willebrand’s factor (vWF) and EC-selective adhesion molecule (ESAM), also increased over the course of serial passage, rivaling the levels observed in human umbilical vein endothelial cells (HUVECs) and coinciding with the acquisition of a more EC-like morphology. Finally, an examination of EC-specific transcripts, which had previously shown 2 to100-fold increases in SP0 TDECs relative to parental H460 tumor cells, showed additional 1.6 to 5.1-fold increases in expression between SP1 and SP7 TDECs. Taken together, these studies demonstrate further up- regulation of numerous EC specific markers during the course of TDEC serial passage. ## Late Passage TDECs form more Complete Tube-like Structures and Become Hypoxia-responsive One hallmark of ECs is their ability to form 3-dimensional tube-like structures resembling capillaries in semisolid media. When cultured under standard normoxic conditions for six days in Matrigel, H460 tumor cells formed only acini, while both early and late passage H460-derived TDECs generated abortive, network-like structures only partially resembling those formed by HUVECs ( and reference 16). In contrast, when TDECs were cultured for 6 days under moderately hypoxic conditions (1% O₂) followed by a 4 day normoxic recovery period, a more complete HUVEC-like tube pattern was observed with late, but not early, passage TDECs. TDEC serial passage thus is associated with an enhanced tube- forming ability, which requires hypoxic stress for optimal expression. To explore further the *in vitro* EC-like properties of TDECs, we co-cultured HUVECs with GFP-tagged H460 cells or with early or late passage H460-derived GFP+ TDECs. We reasoned that, during propagation in Matrigel, HUVECs might provide supportive factors that would enhance the ability of TDECs to form tube- like structures. After 2 d of isolated culture, H460 and early and late passage TDECs cultured alone remained predominantly single, rounded cells and had not formed tubes. However, when co-cultured with HUVECs, we observed a greater tendency of late passage TDECs to elongate and incorporate into (not just associate with) HUVEC tubes compared to either early passage TDECs or H460 cells . These results are consistent with those shown in and indicate that late passage TDECs are more responsive to and cooperate better with neighboring HUVECs than early passage TDECs. ## Enhanced Vascularization Potential of Serially-passaged TDECs We next examined whether the tumors arising from H460 tumor cells containing a 5% contribution of late-passage TDECs might be more highly vascularized than those arising from a similar mix with either early-passage TDECs or GFP-tagged tumor cells (control tumors). Paraffin-embedded sections of tumor xenografts were stained with hematoxylin and eosin to identify tumor blood vessels. We found that tumors originating from the inocula containing late-passage TDECs tended to be supplied by blood vessels that were both denser and larger than those seen in the other two tumor groups. Thus, continued serial passage of TDECs is associated with an enhanced ability to establish a functional tumor vasculature. ## Late Passage TDECs have Defined Genomic Gains/deletions that Affect Gene Expression As late passage TDECs have more pronounced EC-like phenotypes than their early passage counterparts, it seemed possible that serial propagation might be selecting for a subpopulation of cells inherently better at contributing to tumor blood vessel formation. This process could be accelerated if TDECs, like the tumor cells from which they originate, were more genomically unstable than host-derived ECs. To test this hypothesis, we compared genome-wide SNP profiles of H460 cells and early and late passage TDECs. No differences were observed in the SNP profiles of early-passage TDECs (SP0 and SP1) and H460 cells, whereas late-passage TDECs (SP5 and SP6) showed identical genomic alterations (, , and S7). Aside from the loss of one copy of chromosome 14, these consisted of a relatively small number of deletions and gains (8 and 2, respectively) ranging in size from approximately 56 kb to 435 kb (, and). Seven of the ten genomic gains/deletions in late-passage TDECs were found to be intragenic, whereas nine additional genes were located within the immediate vicinity of the affected loci. Interestingly, five of the genes with intragenic alterations and 6 of the adjacent genes have been previously linked to tumorigenesis, vasculogenesis or the EMT and several have also been implicated in neuronal development. Taken together, these results indicate that late passage TDECs appear to arise through a process of clonal selection and possess discrete regions of genomic gain/deletion commonly involving genes with functions potentially relevant to the transition from tumor epithelium to ECs. Our SNP findings suggested that the expression of genes contained within or adjacent to the amplified/deleted regions of late passage TDECs might differ from that of early passage TDECs and/or parental H460 cells. To investigate this, we performed quantitative real-time polymerase chain reaction (qRT-PCR) on 16 of these transcripts (primer sets are listed in). Relative to H460 cells, whose expression levels for all transcripts were arbitrarily set at one, we found changes of <u>\></u>1.5-fold in 7 transcripts from early TDECs (*LSAMP*, *DEPDC1B*, *ODZ2*, *FAM155A*, *EFNB2*, *LIG4*, and *MACROD2*) and 6 transcripts from late TDECs (*LRP1B*, *LSAMP*, *NLGN1*, *RAB3C*, *FAM155A*, and *MACROD2*). Relative to early-passage TDECs, late-passage TDECs showed statistically significant altered expression with 5 of 16 transcripts, namely those of *LRP1B*, *RAB3C*, *FAM155A*, *EFNB2*, and *MACROD2*. The nature of these changes was complex and could not always be predicted by the type of alteration of the respective genomic loci. The most dramatic example was seen in the case of *LRP1B* transcript expression, which was increased 16-fold in late-passage TDECs, despite being associated with a hemizygous allelic loss. # Discussion Although the tumor vasculature has been classically viewed as arising from extra-tumoral sources, numerous studies have demonstrated that some of the cells comprising tumor blood vessels or its stroma display evidence of GI. Our recent finding that tumor cells can directly transdifferentiate into ECs *in vivo*, and that they stably retain this phenotype, provides one potential explanation for these findings as does the observation that some tumor cells may acquire EC-like properties under particular *in vitro* conditions. These findings are supported by recent observations that a substantial proportion of TAECs from primary human glioblastomas appear to be of tumor cell or tumor stem cell origin. Taken together, these observations suggest that tumor cells may engage in EMT, a potentially intermediate stage in TDEC differentiation. During the EMT, tumor cells lose junctional attachments, become more spindle-like, and acquire more invasive and migratory behaviors. The tumor cell derivation of TDECs thus provides reason to suspect that they may share the attribute of GI. The combination of functional plasticity and GI could provide the means for TDECs to adapt to microenvironmental challenges imposed by the growing tumor and to evade anti-angiogenic therapies despite being initially sensitive. Based upon previously published as well as our current work, H460 tumor cells and early-passage TDECs are polyclonal populations with essentially no identifiable differences based on karyotypes or SNP profiling. In contrast, late passage TDECs comprise a monoclonal or oligoclonal population with an enhanced revascularization potential. This advantage is only seen upon serial *in vivo* passage as we have found that comparable periods of *in vitro* propagation neither increases the growth rate of early passage TDECs nor allows for clonal populations to emerge (unpublished data) These findings provide further evidence that the *in vivo* micro-environment plays a key role in selecting for TDEC populations that are optimized for participating in tumor neovasculature development. Late-passage, serially-passaged TDECs possess a limited number of well-defined chromosomal gains and deletions. That the majority of these alterations affect genes that have been implicated in EMT, vasculogenesis, or tumor suppression provides support to a model for which one or more of these genes play a role(s) in the TDEC developmental program. Future studies will be needed to determine the relative contribution of these genes to the TDEC differentiation process. In addition to the vascular and tumorigenic roles mentioned above, several of the genes affected by TDEC gains/deletions are also active in neurogenesis. This may relate to the tendency of blood vessels and nerves to develop coordinately along similar anatomic pathways and to the fact that nerve-derived signals may also influence vascular patterning. Moreover, certain highly motile and invasive TDECs, which might be expected to possess a functional advantage for the type of tumor repopulating activity seen in tumor xenografts, have properties that are highly reminiscent of the cells comprising axonal growth cones –. The precise role of these genes in TDEC formation and function and their relationship to the EC and neuronal developmental process remain to be determined. Although the EC-like properties and behaviors of TDECs clearly increase with *in vivo* passage number, the cells remain morphologically distinguishable from primary ECs such as HUVECs. The less than complete EC-like appearance of TDECs may reflect the different pathways that drive differentiation of these cells compared to true ECs. Certain key elements of the normal EC differentiation program that are lost or rendered dysfunctional as a consequence of the transformation process and/or the ensuing GI may play independent roles as well. Despite these less than complete phenotypes, the fact that late-passage TDECs are capable of better vascularizing tumor xenografts attests to their increasing ability to function within an *in vivo* context. Together with previous observations, the approaches described here should allow a number of interesting questions pertaining to TDEC development and adaptive selection to be addressed. Among these is whether TDECs of diverse tumor types share similar regions of gains/deletions and whether, within these regions, the genes most responsible for TDEC evolution can be better defined with respect to function. The short time-frame over which TDEC evolution occurs should also allow investigation into whether the development of clinical refractoriness to anti-angiogenesis therapies is associated with the emergence of TDEC clonal subsets related to but distinct from those described here. # Materials and Methods ## Ethics All mouse studies were conducted according to Animal Welfare Act and the Public Health Service Policy and approved by the University of Pittsburgh’s Institutional Animal Care and Use Committee (IACUC) (Permit Number: 0812276). The animals were housed in pathogen-free units at Children’s Hospital of Pittsburgh Rangos Research Building in compliance with IACUC regulations. Age and gender matched Nu/Nu mice were purchased from Harlan Sprague-Dawley Laboratories (Indianapolis, IN). For all surgical procedures, all efforts were made to minimize suffering to the mice. The University of Pittsburgh Institutional Review Board specifically approved this study. ## Tumor Xenograft Growth, Cell Culture and Serial Passage of TDECs NCI-H460 human lung carcinoma (H460) cells were obtained from the American Type Culture Collection (Manassas, VA) and were cultured as previously described. Authentication of H460 cells, as well as the serially-passaged H460-derived TDECs, was obtained by comparing the results obtained from the whole genome SNP array analysis with previously published results obtained by SKY analysis for NCI-H460 cells (see legend to, and S7). Human umbilical vein endothelial cells (HUVECs) and EGM-2 EC-specific growth medium were purchased from Cambrex Bio Science (Walkerville, MD). Lentiviral packaging and infections to produce GFP- tagged and G418-resistant H460 cells were performed as previously described. Initially, tumor xenografts were propagated by inoculating tagged H460 cells (10⁶) subcutaneously into the flanks of nude mice. The tumors were allowed to grow to the maximal allowable diameter of ca. 2 cm (typically 4–6 wks) and were then excised, minced, collagenase-digested, and passed successively through 100 µm and 40 µm mesh filters to obtain single cell suspensions (BD Falcon Cell Strainer, BD Biosciences, Franklin Lakes, NJ). Total murine and human tumor-associated endothelial cells (TAECs) were isolated from the bulk tumor cell population using two rounds of selection with magnetic beads coupled to human- and murine-specific anti-CD31 monoclonal antibodies (mAbs). The following day, cultures were examined using brightfield and ultraviolet microscopy to determine the percentage of GFP+ ECs (i.e. TDECs) and then subsequently selected for at least 7 days in G418-containing medium (500 µg/ml). The resulting human-derived cell population was termed “serial passage 0 (SP0) TDECs” and was routinely maintained in EGM-2 medium. Serial passage was performed by re-inoculating 50,000 SP0 TDECs with 950,000 untagged H460 tumor cells (a 1∶20 ratio) into nude mice. Subsequently arising tumors were then used for repeat rounds of TDEC isolation and the generation of additional serially- passaged TDECs up to SP7. “Early passage” TDECs were generally comprised of SP1 cells and occasionally of SP0 cells, whereas “late passage” TDECs were comprised of SP6 or SP7 cells and occasionally of SP5 cells. Separate fragments of tumors were used for the preparation of frozen and paraffin-embedded sections. ## Evaluation of EC-specific Markers and Functions Following their expansion, those TDECs not being used for serial passage studies *in vivo* were used to assess EC marker expression. For flow cytometry, at least 10,000 cells from short-term cultures of in vitro-selected TDECs were stained with anti-human CD31 and analyzed with a FACSCalibur flow cytometer (BD Biosciences). Data were analyzed and processed using BD FACSDiva software. For procedures requiring fixation, TDECs were grown on glass coverslips, fixed in PBS-4% paraformaldehyde and stained with antibodies against either human vWF or human ESAM as previously described. We also quantified the uptake of rhodamine- tagged *Ulex europeus* lectin (E-lectin: Vector Laboratories, Burlingame, CA) and AlexaFluor 594-tagged acetylated low-density lipoprotein (AcLDL) (Invitrogen Molecular Probes) as previously described. All photographs for fluorescence microscopy were taken with an Olympus Fluoview 1000 confocal microscope at a magnification of 40X–60X, except rhodamine-labeled E-lectin staining which was photographed at 20X. Adobe Photoshop CS2 (version 9.0, Adobe Systems, San Jose, CA) was employed for image analysis. qRT-PCR analysis of EC markers was performed as previously described. ## Tube Formation Assay Characterization of tube formation in Matrigel was performed at 37°C as previously described under normoxic or hypoxic conditions using a Hypoxic Glove Box incubator (Coy Laboratory Products Inc., Grass Lake, MI). Wells from 24-well plates were initially coated with Cultrex BME Growth Factor Reduced PathClear matrix (R&D Systems, Minneapolis, MN) (100 µl). TDECs, the original tumor cells (1000 cells/well) and HUVECs (10,000 cells/well) were then seeded in EGM-2 medium for the times and under the conditions indicated in the figures. Tube formation was typically assessed at 2, 6, and 10 d, at which points tubes were quantified and photographed using a Zeiss Axiovert 135 inverted microscope equipped with a Sony DXC-970MD 3CCD Color Video Camera. In assessing and quantifying TDEC tube formation, a “tube” was arbitrarily defined as the formation of an enclosed space surrounded by at least 10 cells linked end-to-end with a 1–2 cell layer thickness for at least 75% of the circle. ## Evaluation of Tumor Blood Vessels Hematoxylin and eosin staining of paraffin-embedded sections from tumor xenografts was performed by the University of Pittsburgh Histology Core Laboratory. Criteria for tumor blood vessel identification were the presence of an identifiable lumen containing red blood cells. ## Whole Genome SNP Array, Chromosome Analysis, and qRT-PCR SNP analyses were performed at The University of Pittsburgh Genomics and Proteomics Core Facility with using Ilumina 1 M DNA Analysis Bead Chips, which contains approximately 10⁶ SNPs with a mean genome spacing of 2.7 kb (median 1.7 kb). Analysis of the raw data was carried out using GenomeStudio software (version v2009.2). Changes in Log R Ratio and B Allele Frequency were used to identify gains or deletions within individual chromosomes. The software also allowed for identification of genes within the immediate proximity of duplicated or deleted genomic loci. The boundaries of the chromosomal alterations were determined using the identified SNP coordinates with the Ensembl Genome Browser 57 database. For qRT-PCR analysis of transcripts of genes identified by whole genome SNP array, total RNA was first purified using RNeasy mini-columns (Qiagen, Inc., Valencia, CA) (16) and 50 ng of input RNA was used in a SYBR Green-based assay (Quantitect, Qiagen) according to the directions of the supplier. Primers used to assess transcript levels of genes putatively affected by genomic alterations in late passage TDECs were selected from the MGH/Harvard PrimerBank site (<http://pga.mgh.harvard.edu/primerbank>) or were self-designed, with the exception of those for *NLGN1*. These latter primers are described in and were synthesized by International DNA Technologies (Coralville, IA). Confirmation of the predicted PCR product sizes was achieved by agarose gel electrophoresis and is shown in. Data analysis was performed on an Applied Biosystems 7300 instrument (ABI, Foster City, CA). Relative quantification of gene expression data was carried out using a comparative *C*_T method and *GAPDH* was used as an endogenous control. Fold differences in the SP1 and SP7 mRNA levels relative to H460 RNA were then calculated by 2^{−\[Δ*C*T(SP1/7)−} ^{Δ*C*T(H460)\]}. ## Statistical Analysis Data were analyzed using either the one-tailed or two-tailed Student’s *t* test as appropriate. # Supporting Information We thank Deborah Hollingshead and the staff of The University of Pittsburgh Genomic and Proteomics Core Facility for performing SNP arrays, Julie Speicher, Farzad Esni, and George Gittes for assistance with hypoxia experiments, Brett Phillips for help with IVIS live animal imaging, Dr. Brian J. Henson for expertise with qRT-PCR and data analysis, and Kristin Lang for technical assistance. [^1]: Conceived and designed the experiments: TFM GBS RDN EVP. Performed the experiments: TFM GBS JL. Analyzed the data: TFM GBS JL RDN EVP. Contributed reagents/materials/analysis tools: RDN EVP. Wrote the paper: TFN RDN EVP. [^2]: The authors have declared that no competing interests exist.

# Introduction Human and nonhuman primates attend to other individuals to gain valuable social information about them (such as identity and emotions) and their shared surroundings (such as nearby dangers and resources), and even to infer others’ goals and intentions based on their actions. Fundamental characteristics of social attention are similar between human and nonhuman primates. Primates selectively attend to others’ faces, eyes, and targets of ongoing actions. They follow others’ gaze and attend to the same objects and locations that others are manipulating. They anticipatorily attend to the targets of others’ actions before their actions are completed; for example, apes and macaques look at the goal objects while the actor is reaching to the object, before the actor grabs them. Previous studies also suggested that human and nonhuman primates share common neurophysiological mechanisms underlying social information processing and that they process social information mainly through the two distinct pathways in their brains. One pathway, via subcortical routes, rapidly processes crude social information such as others' faces, eyes, and gaze direction; the other pathway, via cortical routes, processes nuanced social information such as others’ social and emotional status and communicative intentions. Another important feature of social attention is its individual variation. On the one hand, such individual variation in social orienting is related to biological, early-developing, temperamental characteristics in human and nonhuman primates. Attention to others’ eyes is evident at a very young age in human and nonhuman infants. Attention to targets of others’ gaze, pointing, and manual actions ("joint attention") also emerges early in the development (although somewhat later than attention to the eyes). However, human children diagnosed with autism spectrum disorder attend to others’ eyes and targets of gaze and pointing less strongly than do typically-developing children; even preverbal human infants later diagnosed with autism attend to eyes less strongly than typically-developing infants during viewing of social movies. A recent study showed that monozygotic-twin infants show more similar levels of attention to the eyes than dizygotic-twin infants during viewing of social movies. It is also known that endocrine systems mediate social attention in human and nonhuman primates: Human yearlings who experienced higher levels of prenatal androgen show lower levels of eye contact with their mothers, and oxytocin administration leads humans and monkeys to increase the levels of eye contact with the conspecific images. On the other hand, individual variation in social orienting is related to late- developing, experience-dependent characteristics in human and nonhuman primates. For example, human (sighted) infants of blind parents attend less to the eyes and gaze direction of parents compared to control infants. “Enculturated” apes, reared by humans in the human cultural environment, respond more than non- enculturated apes to the targets of human experimenters’ gaze, pointing, and manual actions when interacting with the experimenters. In humans, it is known that cultural background biases attention to both social and physical stimuli. People from East Asian countries tend to attend to the central parts of faces (i.e., around the nose), while people from Western countries tend to directly attend to both eyes and mouth. These two cultural groups differ in the same way even when presented with allospecific faces (e.g., a sheep face) and visually homogeneous non-face objects. It is also known that expertise by profession biases attention among both social and physical stimuli. Patterns of social orienting differ not only across individuals within a species but also across closely-related primate species. Previous studies used eye- tracking to compare social orienting between different primate species: macaques and humans, chimpanzees and humans, orangutans, gorillas and humans, and bonobos and chimpanzees. Great apes and macaques, like humans, view the models’ faces and especially the eyes when presented with still pictures and movies \[–, –\]. Also, like humans, apes and macaques view the targets of models’ manual actions in movies—even anticipatorily looking at targets of their manual actions. However, compared to humans, apes and macaques view the targets of models’ actions for a shorter time during viewing of movies. Compared to humans, apes view the models’ eyes for a shorter time and the models’ mouth for a longer time when presented with pictures. It was also reported that when chimpanzees and bonobos view pictures, bonobos look at the model’s eyes for a longer time and the targets of models’ actions for a shorter time than do chimpanzees. Therefore, although all these primate species view the same social features in pictures and movies—other’s faces, eyes, mouths, and action targets—they differ significantly from one another in the relative strength of viewing of each social feature. Such species-typical viewing patterns likely reflect temperamental characteristics unique to each species. These previous studies established a useful paradigm for comparing viewing patterns of social stimuli across individuals and species under the same experimental conditions. However, the obtained results are still fragmentary with regard to the pattern and nature of the individual and species variations, particularly among nonhuman primates. More specifically, most previous studies compared only two species, and thus procedural differences across studies, such as the differences in stimuli, preclude a straightforward generalization. Also, many of these previous studies used still pictures as stimuli, leaving untested how individuals respond to dynamic social stimuli typical of natural environments. Critically, none of these studies have tested how experience- dependent factors can effect viewing patterns across individuals within nonhuman primate species. Thus, it remains unclear the extent to which within- and between-species variations overlap and how factors such as species and experience affect viewing patterns. This study has two complementary goals. First, we examined individual and species differences in social orientating by presenting naturalistic movie stimuli to larger samples of human and nonhuman primates than previously tested (humans, bonobos, chimpanzees, orangutans, rhesus macaques). Experiment 1 aimed to extend the results from the previous studies and examined overall species similarities and differences in social orienting among rhesus macaques, three species of great apes (bonobos, chimpanzees, and orangutans), and humans. The movies depicted natural behaviors of conspecific and allospecific animals. Experiment 2 examined the effects of individual experience on within-species differences in social orienting. We tested three groups of chimpanzees housed at facilities that differed in their early experiences with media and cognitive experiments. We tested movies depicting natural behaviors of chimpanzees. We presented these same movies to three groups of humans differing in their expertise in observing chimpanzees (experts vs. novices) and in their experience with media in general (adults vs. preschoolers). Second, we contrasted two methods for quantifying viewing patterns. One of the most common analytic strategies is quantifying the viewing times for predefined Areas-Of-Interests (AOIs). However, a clear shortcoming of this approach is that it may overlook attention to certain features that are salient to nonhuman participants but not to human researchers. An alternative, novel, data-driven approach consists of directly measuring gaze similarities using distances and correlations among individuals. We contrasted these two approaches in this study. In the AOI viewing-time analysis, we first defined AOIs for the social features that previous studies typically included—faces, eyes, mouths, and action targets—to measure viewing times of these social features. We then used a Principal Component Analysis (PCA) to identify the components that best explained the observed variations across individuals and species. In the data- driven analysis, we estimated gaze similarities between each pair of participants, created a similarity matrix, and then performed Multi-Dimensional Scaling (MDS) to identify the dimensions that best explained the observed variations across individuals and species. Finally, we used a canonical correlation analysis to test the similarities between the components/dimensions derived from the two different analyses. If the major features distinguishing between individuals’ scanpaths were adequately captured by their viewing times for the defined AOIs, the two analyses would correlate with one another. Combining the two approaches allows us to characterize species’ similarities and differences thoroughly, and to confirm whether gaze toward AOI adequately describes the variations detected in the data-driven analysis. # Experiment 1 We examined how bonobos, chimpanzees, orangutans, rhesus macaques, and humans view movies depicting various natural behaviors of these species and of nonprimate animals. Those behaviors included resting (with the individuals’ neutral faces), intense engagements among individuals such as playing and fighting (with the individuals’ emotional expressions), and extractive foraging such as manipulating foods and using tools. We predicted, in accord with previous studies, that species is the primary factor influencing individuals' unique viewing patterns of particular social features: faces, eyes, mouths, and action targets. ## Method ### Participants A total of 47 nonhuman primates (12 bonobos, 21 chimpanzees, 7 orangutans, and 7 rhesus macaques) and 12 humans participated in this study. An additional macaque was tested but not included in the analysis because of a calibration failure. All species lived in social groups. Twenty-eight apes (6 bonobos, 15 chimpanzees, 7 orangutans) lived in Wolfgang Köhler Primate Research Center (WKPRC) and 12 (6 bonobos, 6 chimpanzees) in Kumamoto Sanctuary (KS). Apes in these facilities had visual access to members of the other ape species. Macaques lived in a conspecific group at The Rockefeller University. All nonhuman participants had some experience watching movies (e.g., in the previous experiments or as enrichment), with KS chimpanzees being more experienced than the others (see and Experiment 2 for the effect of such experience). They were reared by their biological mothers or human caregivers in conspecific peer groups (see Table A in for further details about the participants). No ape or monkey participant showed a behavioral indication of vision deficit through our daily observation. Human participants were zoo workers at the WKPRC with extensive experience in interacting with nonhuman primates. All had normal or corrected-to-normal vision. No participant with neurological disorder or developmental delay was included. They were instructed to simply watch the movies as they normally would. ### Ethics statements Apes lived in Wolfgang Köhler Primate Research Center (WKPRC) and Kumamoto Sanctuary (KS). In both facilities, the living areas were large and complex enough for the apes to rest, exercise, and socialize with the group mates. The outdoor playground areas were larger than 200 m2 and were equipped with climbing trees, vegetation and enrichment devices. The indoor areas including sleeping rooms were larger than 100 m2. The apes received fresh fruits, vegetables, nuts and leaves distributed in three main meals and occasional enrichment programs. Water was available ad libitum throughout the day. They voluntarily participated in the study and were not food or water deprived. In KS, apes were tested in one of their sleeping or in a separate routine testing room (\> 9 m2). In WKPRC, all apes were tested in one of their sleeping rooms (9 m2). No medical, toxicological or neurobiological research of any kind is conducted at KS or WKPRC. Ape husbandry and research complied with the international standards in accordance with the recommendation of the Weatherall report “The use of non- human primates in research” and the institutional guidelines which are strictly adhered to the national laws of Japan or Germany \[KS: Primate Research Institute “Guide for the Care and Use of Laboratory Primates 3rd Edition”, Wildlife Research Center “Guide for the Animal Research Ethics”\] \[WKPRC: “EAZA Minimum Standards for the Accommodation and Care of Animals in Zoos and Aquaria”, “WAZA Ethical Guidelines for the Conduct of Research on Animals by Zoos and Aquariums”, “Guidelines for the Treatment of Animals in Behavioral Research and Teaching” of the Association for the Study of Animal Behavior (ASAB)\]. The study protocol was approved by the institutional committee of Wildlife Research Center (No. WRC-2014KS001A) and Max-Planck Institute for Evolutionary Anthropology. All macaque procedures conformed to the NIH Guide for Care and Use of Laboratory Animals of the National Institutes of Health, and were conducted in accord with a local Institutional Animal Care and Use Committee (IACUC) protocol (#12585-H and \#15849-H at The Rockefeller University). Monkeys were housed in a climate- controlled indoor colony in suites comprising 1–4 individuals. Monkey health was monitored daily, and monkeys were provisioned daily with biscuits, fresh fruits and vegetables, and behavioral enrichment including puzzle feeders. Human adult participants were tested in a testing room located at the Max-Planck Institute for Evolutionary Anthropology (MPI-EVA), Leipzig, Germany. All agreed to and signed the written informed consent, which was in accordance of Helsinki Declaration and approved by the internal committee of MPI-EVA. For the preschooler participants, their parents were recruited by telephone from a database of parents who had volunteered to participate in developmental studies. All parents agreed the informed consent upon coming to the institute. They were tested in a testing room located at MPI-EVA. All agreed to and signed the written informed consent, which was in accordance of Helsinki Declaration and approved by the internal committee of MPI-EVA. ### Apparatus Apes at the two facilities, macaques, and humans watched the same movies in an eye-tracking system. The differences in eye-tracking setups were minimized as much as possible between different facilities. Eye movements of apes were recorded using an infrared eye tracker (60 Hz; down-sampled from X120/X300 eye- trackers; Tobii Technology AB, Stockholm, Sweden). This eye-tracker can record the participants’ eye movement without a head restraint device. WKPRC apes and KS bonobos were separated from the experimenter and the eye-tracker by a transparent acrylic panel (this panel does not add noises in the eye-movement recordings). To keep their heads relatively still, we let apes drink dripping grape juice from a nozzle attached to transparent acrylic panels. For the KS chimpanzees, one of the experimenters stayed inside the room, sat beside them, and lightly held their chins during the recording. Another experimenter stayed outside the room, with the eye tracker, and recorded the participants’ eyes through transparent acrylic panels. No explicit training was conducted for apes. Stimuli were presented using Tobii Studio software (version 3.2.1) at a viewing distance of 65–70 cm with a resolution of 1170×720 pixels (approx. 39×25 degree) on a 22-inch LCD monitor (1366×768 pixel). Human participants were tested in a standard office using the same setups of the eye-tracker and the monitor. Eye movement of macaques was recorded using an infrared eye-tracker (60 Hz; ETL-200, ISCAN, MA, USA). They sat in a primate chair, with head position maintained via head prosthesis, and performed a gaze calibration routine. They were trained to fixate simple shapes for calibration in this and the other experiments. In addition, they were trained in the other experiments (but not in this experiment) to fixate at the center point of the monitor. Fluid rewards (water droplets) were delivered during calibration, and during movie viewing at 3-second intervals independent of the macaques’ visual behavior. Stimuli were presented using Presentation software at a viewing distance of 50 cm with a resolution of 1014x624 pixels on a 20-inch LCD monitor (1024x768 pixels; two macaques, Sam and Thor, were tested at a viewing distance of 57 cm with a resolution of 1202x754 pixels on the monitor, 1600x900 pixel; yet we confirmed that such differences did not affect the results; see below and Fig D). These setups were adjusted so that the images occupied about the same visual angles as in the apes’ and humans’ settings. We conducted calibration procedures previously established for apes, humans, and macaques at each facility. For apes, automated calibration was conducted in Tobii Studio by presenting a small object or movie clip on two reference points. Although the number of these reference points was smaller than that used typically for human and monkey participants, we manually checked calibration accuracy after the calibration, by examining the discrepancies between the participant’s gaze and the 9 reference points presented on the screen, and repeated the calibration until those observed discrepancies became smaller than a degree. For human participants, automated calibration was conducted in Tobii Studio by presenting small objects at 5 reference points. Calibration was conducted for macaques in a Presentation software (Neurobehavioral Systems, California, USA) by presenting simple geometric shapes at 9 reference points. These procedures assured comparable accuracy of calibration for each species (typically within a degree), as detailed in the previous studies. It should be noted that our control analysis ensured that the distribution of fixations around each defined Area-Of-Interest, which accommodate any calibration errors, was similar across the participant species (see and Fig B). ### Stimuli and procedure Movies (total 9 minutes, 25 fps) depicted the natural behavior of conspecifics and allospecifics (obtained from [ARKive.org](http://ARKive.org)). We prepared a total of 18 movies (each 30 seconds) featuring bonobos (3 clips), chimpanzees (3 clips), orangutans (3 clips), rhesus macaques (3 clips), and 3 nonprimate species (1 clip each of horses, dogs, and birds). Human movies were omitted in this study because the previous studies have confirmed similar eye movement patterns for human and allospecific images in adult human participants with experience interacting with nonhuman primates. The contents of movies were selected so that they covered a wide range of species-typical behaviors of each primate. The three clips respectively depicted “actions”, “social interaction”, or “resting”. “Actions” included extractive foraging behavior such as manipulating objects (ground digging by bonobos and macaques, food washing by macaques), tool-using (stick-use by chimpanzees and orangutans, nut-cracking by chimpanzees), and eating the extracted foods. “Social interactions” included intense social engagement among individuals such as fighting (by bonobos, chimpanzees, and macaques) featuring threat and fearful facial and bodily expressions, copulating (by bonobos), and playing (by orangutans). "Resting" included calm, relaxed individuals, mainly showing the face. The non-primate movies depicted scenes of the species-typical behavior of those species (e.g., eating, fighting, flying, and galloping). Also, we presented participants with image-scrambled movies (3 clips) to obtain baseline data for the data-driven analysis (to control for the participants' default viewing biases to the screen). However, nonhuman participants only watched those movies for about half of the time that they spent watching the other movies, and human participants showed peculiar patterns of eye movement (i.e., kept looking at the center of the images). We thus did not use these data in the analysis. Instead, we used the time-shuffled eye movements (derived from the same trials presenting the same movies) as an alternative baseline for the data-driven analysis (see below). All movies were silent. Each ape viewed one movie (1 trial) per day (total 18 trials). If an ape became distracted during any given trial, that trial was dropped and the same trial was repeated on the next day (but no more than once). Each human and macaque viewed all movies consecutively in a single day, with a short blank period between movies. Yet, we did not particularly observe fatigue (or an increase in the percentage of off-screen gaze) in macaques (off-screen gaze was less than 30% in all trials) or in humans (less than 10% in all trials). The order of the movie presentation was counterbalanced across participants. See the video showing all stimulus movies and the superimposed gaze patterns in the first author's online repository (<https://youtu.be/JLLW3ophuTc>**).** ### Data analysis Viewing-time analysis. Areas-Of-Interest **(**AOIs) were defined for the eyes and mouth (when the scene focused on the face), head (when the scene focused on the whole body), and action targets as polygons around each feature of interest, frame-by-frame, by a primary coder using custom software. We did not define AOIs smaller than 1% or larger than 25% of the frame size (i.e. not defining the eyes/mouth when the whole bodies were zoomed-out, and not defining the body when the face was zoomed-in). The size of AOIs was approximately 20% larger than the objects of interest to accommodate small offsets which may derive from calibration errors or quick movements of objects in movies. Our control analysis ensured that small variations in the size of AOIs did not affect overall pattern of results (see and Fig B). The "head" AOI was defined when the head appeared along with the whole body (and the eye/mouth AOIs were too small to be defined). The “action targets” included any goal targets of manual actions, including the foods being grabbed by hands (or pecked by a bird’s beak), the ground being dug, and the tools being manipulated. We then calculated viewing times for each category of AOI. We did not exclude off-screen gaze from this analysis (i.e. we used the raw, rather than proportion, viewing times) but excluded off-screen gaze in another, inter-individual distance, analysis and tested the similarities between these analyses (see below). Moreover, when we used such proportion data in this same analysis, we obtained the same pattern of results (Fig C). See the video showing example frames and superimposed AOIs in the first author’s online repository (<https://youtu.be/fb6N8-olJxk>**).** To reveal the components that best explained the observed individual and species variations, we performed Principal Component Analysis (PCA) on the viewing times for the eyes, mouth, head and action targets. To classify the participants by species, we performed a discriminant function analysis using the same data. We used permutation discriminant function analysis (pDFA) to control for the unequal number of participants in each species, using the code provided by R. Mundry. This analysis samples an equal number of participants from each species (7 samples in this study, based on the minimal group size; 100 iterations) and classifies the sampled participants into predicted species based on the dependent variables. The remaining participants (i.e., non-sampled participants of large groups) were then used for external classification. The success rate of classification was compared with the chance level in a permutation test (i.e., the performance of the discriminant analysis on permuted data in which species identity had been randomly reassigned, 1000 iterations). Inter-Individual Distance (IID) analysis. This data-driven analysis directly measures the gaze distances between the participants. Owing to its data-driven nature, this analysis benefits from some noise reduction. We did this in following way. First, we smoothed the series of horizontal and vertical gaze coordinates (60Hz) using a 100 ms (6 sample) moving average window to reduce the recording noises. We then calculated Inter-Individual Distances (IIDs), defined here as Euclidean distance between the gaze coordinates of a given pair, averaged across time-points for each movie clip. We did this for all pairs of participants. We excluded times of off-screen gaze from this analysis. To minimize a possibility that gaze similarities between participants derive from the similarities in default viewing biases (unrelated to the movie contents; e.g., central bias in humans), we corrected the IIDs for baseline similarities. We calculated those baseline IIDs by shuffling the timestamps of a given scanpath 10 times and averaging over the repetitions. We then normalized the raw IIDs by dividing out the time-shuffled IIDs. These procedures created a similarity matrix. Based on this similarity matrix, we performed multi- dimensional scaling (MDS) and inferred the dimensions that best explained gaze similarities among participants. We took the first three dimensions for the analysis, based on the elbow of a scree plot. Finally, we tested the similarities between the two analyses, using a canonical correlation analysis to compare these MDS dimensions with the PCA components derived from AOI viewing times. ## Results In the viewing-time analysis, we measured the participants’ viewing times for the Areas-of-Interest (AOIs) eyes, mouths, heads, and action targets. shows the viewing-time scores for each species summed across all movies. A MANOVA revealed highly distinct patterns across species in viewing times toward AOIs (F(16,156) = 11.1, p \< 0.001, Wilk's Λ = 0.10) and also in viewing times to particular AOI categories (follow-up ANOVAs; eyes: F(4,54) = 9.40, p \< 0.001, *η*² = 0.41; mouth: F(4,54) = 22.21, p \< 0.001, *η*² = 0.62; head: F(4,54) = 9.03, p \< 0.001, *η*² = 0.40; action target: F(4,54) = 57.58, p \< 0.001, *η*² = 0.81; the alpha was set at 0.0125 with Bonferroni correction for number of comparisons). We then used Principal Component Analysis (PCA) to identify the components that best explained the observed variation. We took the first two components in this analysis because they explained the majority (93.5%) of the observed individual differences (97.1% with the first three components). plots all the participants in these components. The largest coefficient of the first principal component corresponded to the viewing time for action targets (0.71), followed by head (0.44) and mouth (0.42). The largest coefficient of the second component corresponded to the viewing times for eyes (0.67) and mouth (-0.63; see the vectors in for these coefficients). In the Inter-Individual Distance (IID) analysis, we measured IIDs between all pairs of participants and then created the gaze-similarity matrices for all movies. shows the gaze-similarity matrix averaged for all viewed movies (see Fig A in for the data for each viewed species’ movies). We then used Multi- Dimensional Scaling (MDS) to identify the dimensions that best explain the observed variation. shows the distribution of all participants in the 3D space based on this MDS analysis. We then tested the similarity between data derived from the PCA analysis (first two components) and from the MDS analysis (first 3 dimensions) using a canonical correlation analysis. We found that the canonical correlation was 0.81 and 0.61 for the first and second canonical dimensions, respectively. Both canonical dimensions were significant (1^st to 2^nd: F(6,108) = 20.6, p \< 0.001, Wilk's Λ = 0.22; 2^nd: F(2,55) = 16.7, p \< 0.001, Wilk's Λ = 0.62). The first canonical dimension was most strongly influenced by the first MDS dimension (standardized canonical coefficient 0.99) and the first PCA component (0.998). The second canonical dimension was most strongly influenced by the second and third MDS dimensions (0.73, 0.68) and the second PCA component (0.999). Note that IID analysis excluded off-screen fixations from the analysis (i.e., used on-screen gaze distances), while the viewing-time analysis did not (i.e., used raw, not proportion, viewing times). Thus, the observed similarity between the two results ensured that the species differences in overall levels of attention to the movies (i.e., on-screen viewing times; bonobos were slightly less attentive than the other species; see Table B) cannot alone explain those in the viewing patterns of specific social features (also see Fig C in for the replication of the same results with a proportion measure excluding off-screen gaze). Moreover, it indicates that the major features distinguishing between individuals’ scanpaths were adequately captured by their viewing times for the defined AOIs. To test the clustering of participants based on their species, we performed a permutation Discriminant Function Analysis (pDFA;) using the viewing times for AOIs (the data for all movies were averaged). The classification based on the participants’ species was highly successful (83.4%; chance-level, 37.9%; p \< 0.001). Most misclassifications occurred between orangutans and chimpanzees/humans and between bonobos and macaques. Fig D in presents the names of all participants in the PCA graph and Table A in details the properties of each participant. The inspection of the remaining misclassified participants did not reveal common properties (including living facility, sex, age class, or whether mother- or human-reared; yet WKPRC and KS chimpanzees somewhat differed from one another in their viewing patterns of the action targets; this group difference was further examined in Experiment 2). Classification based on the participants’ species was successful with the data for any given depicted species’ movie. This indicates that each species viewed social features (eyes, mouth, face, and action targets) of each depicted species similarly across the movies (although bonobos viewed the conspecific movies somewhat for a longer time than the allospecific movies; see Table B in). Finally, there may be a concern that potential differences in calibration error between species (due to the procedural differences between facilities) may affect the pattern of species differences to some degree. Yet, this was not an issue here. First, we generated matching results through two different analyses with different sensitivity to calibration noise: the AOI viewing-time and inter- individual distance analyses. Moreover, we conducted a control viewing-time analysis (Fig B) manipulating the size of AOIs (shrinking or expanding the size up to 20%) and confirmed that such manipulation did not change the pattern of species differences in the viewing-time data. This result indicates that the distribution of fixations (including any calibration errors) around each defined AOI was similar across the participant species. ## Discussion Overall, bonobos, chimpanzees, orangutans, rhesus macaques, and humans exhibited similar yet highly discriminable gaze patterns to the movies. We found a strong correlation between the results from the data-driven IID analysis and those from the AOI viewing-time analysis. The use of two different analytical approaches revealed that the viewing patterns for the models’ face, eyes, mouth, and action targets satisfactorily characterized overall gaze similarities. It also revealed that variations in overall levels of attention to the movies (somewhat lower in bonobos than in the other species) cannot explain variations in viewing patterns across social features, because one of our analyses excluded the off-screen fixations from the analysis, while the other did not. More specifically, we found that humans viewed the action targets for a much longer time than apes and macaques. Bonobos viewed the eyes for a longer time (and the mouth for a shorter time) than chimpanzees and orangutans. Chimpanzees and orangutans viewed the mouth and the action targets for a longer time than bonobos. Macaques’ viewing patterns were somewhat similar to bonobos in the sense that they viewed the eyes for a longer time than chimpanzees and orangutans, although the data revealed clear differences between bonobos and macaques; with the latter viewing the eyes even longer, and the mouth even shorter, than the former. These results are largely consistent with the previous studies, although some results are unexpected (e.g. monkey-ape difference). We will discuss the implications of these results in General Discussion. Consistent with the previous eye-tracking studies, we found that the observed species-typical viewing patterns were relatively independent of whether the presented species was conspecifics or allospecifics. This result suggests that such species-typical patterns likely reflect their general responses to the social features that are commonly present in animate agents (e.g., face-like shapes, contingent motions). It is noteworthy that bonobos viewed the conspecific (and also the chimpanzee) movies for a longer time than the allospecific movies (Table B). This result suggests that bonobos might have a higher interest in conspecific than allospecific movies, although their viewing bias for each social feature (e.g., eyes versus mouth) was highly similar for both types of movies. # Experiment 2 Experiment 1 revealed some differences in the viewing patterns of action targets of two groups of chimpanzees (WKPRC vs. KS1). Several studies have documented that early experiences with the social and physical environment are especially influential in the adulthood behaviors of great apes. Enculturated apes reared by humans in human cultural environment performed particularly well at tasks requiring joint attention with human experimenters. Additionally, deprivation of social and physical experience in early life adversely affects social behaviors in adult chimpanzees. In Experiment 2 we further examined the role of experience on viewing patterns by presenting movies of chimpanzee natural behavior to three groups of chimpanzees differing in their early social and physical experiences (individuals were reared in three different facilities). One group of chimpanzees (WKPRC) had standard experiences with media and cognitive experiments, another group (KS1) had more extensive early experiences with media, cognitive experiments, and tool-using training, and the third group (KS2) had relatively little early experience with media and cognitive experiments, and relatively little social and physical enrichment during their development (prior to arriving at the sanctuary). In line with the findings from Experiment 1, we expected that WKPRC and KS1 groups differ from one another in their viewing patterns for the action targets in the movies. Additionally, we expected that the KS2 group differ from the other two groups in their general viewing patterns for the social features in the movies. We also presented the same movies to three groups of humans differing in their expertise in observing chimpanzees or in their experiences with media in general; expert fieldworkers who had an extensive experience of observing chimpanzees in the wild, novice researchers who did not have an experience of working with chimpanzees, and preschooler (novice) children who likely had fewer experiences of watching movies in general. We expected to observe the effect of expertise between the first two groups and a more general effect of media exposure between the first two and the last groups. ## Method ### Participants A total of 26 chimpanzees and 58 humans participated in this study. An additional human was tested but not included in the analysis because of a recording failure. Chimpanzee participants consisted of three groups differing in their early experiences (“early” defined here as the infancy and the juvenile period, roughly covering the first nine years). Fourteen WKPRC chimpanzees had moderate experience participating in cognitive experiments and some experience watching movies in previous eye-tracking experiments. They were either reared by their biological mothers or human caregivers (and conspecific peers; See Table C in for further details). Six KS chimpanzees (KS1 group) were recently moved to KS from the Great Ape Research Institute, Okayama, Japan. They had extensive experience participating in various cognitive experiments and watching movies in experiments and as enrichment. They were also trained, since youth, to perform complex tool-use behaviors, including nut-cracking behaviors (while WKPRC chimpanzees were not). WKPRC and KS1 chimpanzees were either reared by their biological mothers or human caregivers and conspecific peers (see Table C in for further details). The other six KS chimpanzees (KS2 group) had almost no experience participating in cognitive experiments or watching movies. They had been housed in isolation for biomedical research and reared by human caregivers during the infancy and juvenile periods. They arrived at Kumamoto Sanctuary between 1980 and 2000 to be integrated into a more naturalistic conspecific social group. Note that, after the adoption to the sanctuary, KS2 chimpanzees live in a socially- and physically-enriched environment as do the other participant chimpanzees (see SI for the details about enrichments and the ethical statements). No ape participant showed a behavioral indication of vision deficit through our daily observation. Human participants consisted of three groups differing in their experience watching chimpanzee behavior and movies. Eighteen humans were professional field-worker researchers who had expertise working with chimpanzees in their wild habitats. Twenty humans were researchers who had no experience working with chimpanzees. Thirteen expert humans (of 18) and 10 novice humans (of 20) reported that they have already seen the movie used for our stimuli, and thus knew the basic stories used in the movies, yet we confirmed that this factor did not affect the results (see below). Most had European or North-American origins (4 expert humans were from Japan; yet, they were not different from other experts, as shown below). They were instructed simply to watch the movies as they normally would. Twenty humans were preschoolers aged between 5 and 6 years (mean age 5.6 ± 0.29). Their parents reported that no preschooler participant watched the “chimpanzee” movie but had some experiences of watching movies of nonhuman animals in general, and that all had regulated opportunities of watching TV and cinemas made for juveniles/adults (see SI for the ethical statements and Table C in for further details about participants). All had normal or corrected-to-normal vision. No participant with neurological disorder or developmental delay was included. ### Apparatus WKRPC and KS1 chimpanzees and humans were tested with the same eye-tracking setup as those used in Experiment 1. KS2 chimpanzees were tested with the same eye-tracking setup as those used for WKPRC apes and KS bonobos in Experiment 1 (i.e. with transparent panels between the participant and the eye-tracker/the experimenter). ### Stimuli and procedure Movies (total 6 minutes, 25 fps) depicted the natural behavior of chimpanzees in the wild (taken from *Chimpanzee* by Disney Nature). We prepared a total of 12 movies (each 30 seconds) featuring resting, grooming, eating, tool-using, playing and fighting (2 clips for each). Resting clips depicted calm, relaxed individuals, mostly faces. Grooming and play clips depicted grooming and playing bouts between dyads. Fighting clips depicted agonistic episodes among individuals that included threat and fear facial expressions. Eating clips depicted individuals grabbing and consuming food. Tool-using clips depicted individuals using a probe-stick to extract the insects inside the wood and a hammer (a log) to crack open nuts on an anvil. No sound accompanied the movie images. Each ape viewed one movie (1 trial) per day (total 12 trials). If an ape became distracted during any given trial, that trial was dropped; and the same trial was repeated on the next day (but no more than once). Each human viewed all movies consecutively in a single day, with a short blank period between movies. Yet, we did not particularly observe fatigue (or a strong increase in the percentage of off-screen gaze) in human adults or preschoolers (off-screen gaze was less than 10% in all trials). The order of the movie presentation was randomized for each participant. See the video showing all stimulus movies and the superimposed gaze patterns in the first author's online repository (<https://youtu.be/KfVqWAP-D6Q>**).** ### Data analysis We used the same method as Experiment 1 for the data analysis except that we distinguished between the “in-hand action targets” and “distal action targets” for the definition of AOIs in this study, because the movies included a long sequence of nut-cracking behaviors. The “in-hand action targets” covered any goal targets of manual actions including the foods being grabbed by hands, the ground being dug, the body part being groomed, and the tools being manipulated. The “distal action targets” are the nuts being placed on anvils and cracked open by chimpanzees with hammers. ## Results In the viewing-time analysis, we measured the viewing times for AOIs comprising eyes, mouths, heads, in-hand action targets, and distal action targets. A MANOVA revealed highly distinct patterns across groups (F(25,276) = 17.1, p \< 0.001, Wilk's Λ = 0.03) and also in viewing times to particular AOI categories (follow- up ANOVAs; eyes: F(5,78) = 20.51, p \< 0.001, *η*² = 0.57; mouth: F(5,78) = 11.19, p \< 0.001, *η*² = 0.42; head: F(5,78) = 39.37, p \< 0.001, *η*² = 0.72; action target: F(5,78) = 31.78, p \< 0.001, *η*² = 0.67; distant action target: F(5,78) = 71.63, p \< 0.001, *η*² = 0.82; the alpha was set at 0.01 with Bonferroni correction for a number of comparisons). We then used PCA to identify the components that best explained the observed variation. We selected the first two components in this analysis because they explained the majority (89.8%) of the variation (97.0% with the first three components). plots all the participants as a function of these two components. The largest coefficient of the first principal component corresponded to the viewing time for the head (0.88) followed by the eyes (0.41). The largest coefficient of the second component corresponded to the viewing time for the mouth (0.71) followed by the in-hand action target (0.49), and the distal action target (0.42; see the vectors in for these coefficients). Next, as in Experiment 1, we measured IIDs between all pairs of participants, created the gaze-similarity matrix, and identified the three dimensions (based on an elbow of the scree plot) that explain the observed individual variations using MDS (see Fig E in for the plot). We then tested the similarity between the data from the AOI-PCA analysis and those from the IID-MDS analysis using a canonical correlation analysis based on the first three dimensions of MDS and the first two (i.e., most influential) components of the PCA. We found that the canonical correlation was 0.91 and 0.78 for the first and second canonical dimensions, respectively. All these canonical dimensions were significant (1^st to 2^nd: F(6,158) = 75.5, p \< 0.001, Wilk's Λ = 0.07; 2^nd: F(2,80) = 63.4, p \< 0.001, Wilk's Λ = 0.39). The first canonical dimension was most strongly influenced by the first MDS dimension (standardized canonical coefficient -0.92) and the first PCA component (0.999). The second canonical dimension was most strongly influenced by the second MDS dimensions (-0.96) and the second PCA component (0.999). Finally, we performed a permutation Discriminant Function Analysis (pDFA,) using the viewing times for Area-Of-Interests. The classification based on the participants’ group was highly successful (81.9%; chance-level, 43.1%; p \< 0.001). The majority of misclassifications occurred between novice and expert humans, between novice adults and preschoolers, and between WKPRC chimpanzees and KS1 or KS2 chimpanzees. Misclassifications across species were rarely observed, although some occurred between KS1 chimpanzees and preschoolers. Fig F in presents the names of all participants in the PCA graph and Table C in details the properties of each participant. The inspection of the remaining misclassified participants did not reveal common properties, including sex, age class, whether the chimpanzee was reared by their biological mother or human caregivers/conspecific peers, and whether the human participant had previously seen the movie. Note that all participants were from Western countries except some Japanese experts, HE15-19, who did not differ from the other experts (Fig F). These results thus indicate that experimentally-selected group (or species) was the major factor in this classification. ## Discussion Overall, several groups of chimpanzees and humans exhibited similar yet highly discriminable gaze patterns during viewing of social movies. Consistent with Experiment 1, we found a strong correlation between the data from the data- driven IID analysis and the AOI viewing-time analysis. As in Experiment 1, the use of two different analytical approaches revealed that variations in attention to the models’ face, eyes, mouths, and action targets could satisfactorily characterize overall gaze similarities. Also consistent with Experiment 1, we confirmed that species was the primary factor affecting the observed variations, although there were substantial differences within each species, notably relating to the participants’ rearing and experimental histories. More specifically, KS1 chimpanzees viewed the models' action targets (both in- hand and distal) longer than the other chimpanzee groups. Moreover, KS2 chimpanzees viewed all social features for a shorter time than the other chimpanzees (i.e., they viewed nonsocial features proportionally for a longer time than did the other chimpanzees). Expert humans viewed the faces and eyes of model chimpanzees longer (and the mouth and action targets for a shorter time) than the other humans. Moreover, children viewed the models’ action targets longer (and the models’ faces and eyes for a shorter time) than adults. # General discussion We examined individual and species variation in the viewing patterns of movies depicting the natural behaviors of nonhuman primates in rhesus macaques, three species of great apes (bonobos, chimpanzees, and orangutans), and humans. We found that social orienting was both individually-variable and species-typical across human and nonhuman primates. Also, we found that variation in the viewing of the models’ faces, eyes, mouths, and action targets can distinguish both the species and experiences of the viewer. This result supports the idea that attention to others’ eyes and their manual actions are related to key aspects of social cognition in human and nonhuman primates. ## Gaze toward action targets Why did individuals and species vary in their viewing patterns in the observed ways? Multiple factors likely contribute to shaping such variation. Let us start by discussing observed differences in viewing targets of depicted actions. In this study, human participants viewed the action targets of any model animal for a much longer time than did the other primate species in both Experiments 1 and 2. In general, humans should be regarded as a special class of participants among the tested primates, because our stimulus movies were created in the human cultural environment, e.g., under specific conventions of cinematography. One interpretation is thus that human participants, presumably even preschoolers, were much more accustomed than nonhumans to watching movies, and therefore better understood (and hence more actively viewed) the goals of depicted actions. The action targets in our stimuli were typical goal targets of manual actions by primate models, including foods being grabbed by hands, tools being manipulated, and nuts being placed on an anvil for cracking by a chimpanzee’s hammer. Moreover, some of the movie scenes contained complex configurations (e.g., zoomed-in manual movements). Humans should understand such movie content readily due to their unique experiences with cinematography, or should at least expect movies to provide some interesting and conceptually-related information across scenes. Importantly, in Experiment 2, those chimpanzees with extensive early experiences with media (KS1) also viewed the models' actions for a longer time than the other chimpanzee groups. As in humans, their experiences with visual media may have enhanced their understandings and expectations about movie contents. Also, their early experience with cognitive experiments and training in tool-use, including nut-cracking, could have enhanced their understandings of movies and their attention to the distal goal targets (i.e., nuts). These results may be related to previous reports that "enculturated" chimpanzees are particularly attentive to human experimenters' action targets. On the contrary, those chimpanzees who experienced relatively impoverished social and physical environment during their youth (KS2) viewed the depicted actions (and eyes) for a shorter time than other chimpanzee groups. Therefore, one candidate factor affecting the observed variation in the viewing of action targets may be related to our participants’ unique experiences with the human environment, including media viewing, tool use and cognitive testing. Then, why did nonhuman species (with similar experiences) differ from one another? In Experiment 1, we observed that chimpanzees and orangutans viewed the models’ targets of manual actions for a longer time than did bonobos and macaques. One possibility is that bonobos and macaques were much more attentive to the models’ faces and eyes than actions, and thus could not spend much time in viewing the other features because of a time trade-off. However, this possibility is unlikely because (unlike humans) their on-screen viewing times to the movies did not reach to the ceiling level; this means that bonobos and macaques viewed elsewhere (including backgrounds and off-screen) instead of viewing the models’ manual actions. Our results may be related to the previous observation that bonobos and rhesus macaques, unlike chimpanzees, orangutans, and humans, do not use tools in foraging contexts or show clear evidence of cultural transmission of tool-using in the wild. Thus, another possibility is that, similarly to what we discussed above, bonobos and macaques may have more poorly understood the models’ manual actions depicted in the movies and hence attended them less actively than did chimpanzees and orangutans. However, at least for bonobos, this explanation is inconsistent with the previous evidence. Studies have shown that bonobos can perform tool-using behaviors as dexterously as the other primate species if they have an opportunity to do so in a laboratory. Also, researchers largely agree that motivational factors rather than competence better explain the absence of tool-using behaviors in bonobos living in the wild. Moreover, studies have shown that bonobos follow a model’s gaze more sensitively than the other ape species and consequently attend more to the target objects in such situations. Studies also have shown that bonobos are comparable to the other ape species in their performances of anticipatory looking to the agent’s manual reaching. Therefore, a more plausible explanation for our results with bonobos is that they were simply less interested than chimpanzees and orangutans in others’ manual actions due to their unique motivation and temperament. It is also possible that their unique experiences during development, such as more limited opportunities for observing conspecifics’ manual actions than the other ape species, may have further discouraged them from gazing toward complex actions. However, it should be noted that this same explanation may not to apply to macaques. It is certainly likely that macaques understood the models' manual actions less well than did apes. Even so, it is unlikely they failed to understand simple actions such as macaques handling food, especially given previous studies showing that macaques can learn from conspecifics’ actions in natural experiments and that their mirror-neuron system responds to both their own actions and actions performed by others. Overall, motivational factors rather than competences likely explain the observed variations in the viewing of the models’ action targets across species. ## Gaze toward faces Next, why did individuals and species differ in their viewing patterns of the model’s eyes and mouth? Regarding the (within-species) individual differences, in Experiment 2, we observed that attention to the eyes and mouth varied to a larger extent among humans than chimpanzees. Specifically, expert field-workers of chimpanzees viewed the face and eyes of model chimpanzees for a longer time than did novice researchers and preschoolers (and the mouth and action targets for a shorter time presumably due to a time trade-off). One interpretation of this result is that experts habitually attend to chimpanzees' faces and eyes to individuate chimpanzee faces. Specialization for processing and individuating particular faces or exemplars of inanimate objects (e.g., cars) is one of the well-known effects of expertise. Our expert participants may be trained to individuate chimpanzee faces, or at least be more motivated than novices to individuate chimpanzees faces, and therefore may have attended to their faces more strongly than novices in the movies. In contrast, our preschooler participants’ inexperience with allospecific movies, or movies in general, may have discouraged them from attempting to identify individuals. Their inexperience may have instead motivated them to watch unfamiliar models’ performing certain actions. The observed adult-child differences in humans may be also related to certain developmental changes in social attention, in that adults may have a stronger tendency of looking at face and eyes of both conspecifics and allospecifics. This aspect cannot be fully examined in nonhuman primates in our study because most of our nonhuman participants were adults; the few juvenile participants did not obviously differ from the adult participants (Figs D and F and Table A and C). In Experiment 2, we also observed that the chimpanzees who underwent relatively impoverished social and physical environment during their youth (KS2) showed a decreased level of attention to all social features including face and eyes (i.e., they viewed nonsocial features proportionally for a longer time than the other groups). This pattern could derive from their lack of experience in watching movies or participating in cognitive experiments more generally. Given that early social deprivation adversely impacts social behaviors in chimpanzee adults, it is also likely that their reduced experience in communicating with conspecifics (and social agents in general) during their early lives discouraged them from attending to chimpanzees in the movies. Overall, therefore, individuals’ unique experiences likely affected their patterns of gaze toward eyes and mouths. Why, then, did nonhuman species with similar early experiences differ from one another in the viewing of eyes and mouth? We observed that in Experiment 1, bonobos and macaques viewed the eyes for a longer time than the mouth, while chimpanzees and orangutans showed an opposite pattern. Given that, in a previous study, orangutans (the same participants as in this study) showed a similar viewing pattern for the face and eyes as gorillas, it is likely that bonobos are exceptional among great apes in their viewing patterns of the face and the eyes. Importantly, in Experiment 2, while the time spent viewing the eyes and the mouth varied to a large extent within the human species, it varied only to a small degree among the chimpanzee participants. Therefore, at least in chimpanzees, the observed viewing bias should reflect some inherent species- typical characteristic. Several previous studies may help to identify the nature of this trait. First, in humans, increased motivation to affiliate with particular others can lead to an increased level of eye contact with them. Bonobos live in a more egalitarian society and exhibit more frequent and diverse affiliative behaviors towards social partners than do chimpanzees. Thus, their general affiliative attitudes toward others may have led them to attend to others’ eyes than chimpanzees (and orangutans). Second, previous studies reported that bonobos and chimpanzees differ in brain areas implicated in social interaction, which were activated, in humans, when engaging eye contact. Third, previous studies reported bonobos and chimpanzees differ in their endocrine systems. Bonobos have a lower level of prenatal androgens than do chimpanzees, which is known to cause an increased level of eye contact in humans. Bonobos and chimpanzees are also known to differ in their oxytocin- and vasopressin- receptor genes; in humans and macaques, a higher level of oxytocin is reported to cause an increased level of eye contact. Therefore, bonobos may differ from chimpanzees (and possibly also from orangutans) in their psychobiological characteristics affecting the pursuit and tolerance of eye contact with others. Interestingly, macaques viewed the models’ face and eyes for a longer time and the mouth for a shorter time than any other species, including bonobos. Our macaques viewed the models’ mouth and action targets very little; thus, overall, they almost exclusively viewed the models’ eyes among all social features in the presented movies. Such strong viewing bias to eyes (versus mouth) is consistent with previous studies \[–, –\]. However, it was somewhat surprising that they did so even more than great apes in this study, because some researchers believe that prolonged eye contact is more commonly observed in great apes than in rhesus macaques. There are several possibilities that could explain this result. First, our macaques, unlike our apes, had previously received fixation training. Thus, one possibility is that such different prior training may have encouraged them to search for certain salient stimuli (e.g., faces or eyes) as cues that could produce rewards. However, note that we did not reward the macaques for their viewing of any particular social features in this study. Also, the eye viewing patterns exhibited by the macaques in the previous studies with different training histories (or no reported prior training) were very similar to those exhibited by our macaques in this study. Therefore, overall, it is unlikely that their viewing patterns derive solely from their training histories. The second possibility is that, like bonobos, a high level of social tolerance led them to focus on the models’ eyes more than the other species. However, this is unlikely because rhesus macaques live in a relatively despotic society and make eye contact with conspecific adults in affiliative contexts less frequently than other macaque species ( but see). The third possibility is that, rather than tolerance, vigilance led our macaques attend to the models’ eyes more than the other species. In general, attention to eyes is enhanced in both affiliative and threating situations. Although it is reported that tolerance enhances attention to the eyes of others in macaques, it is also reported that vigilance enhances their attention to the eyes (or the attentional status) of others such as when an experimenter maintains eye contact with them at a close distance. Therefore, our macaques may have been more vigilant than apes to our movie stimuli and hence monitored the eyes of potentially threatening models exclusively in the movies. Finally, it is likely that the differences in rearing experience and the level of understanding of movie contents complicate a direct comparison between monkeys and great apes. Future studies should address this question by testing multiple species of monkeys using eye-tracking. It would be especially interesting to examine how the tolerance levels of social systems in closely-related macaque species (despotic versus egalitarian societies) affect their distinct viewing of the eyes and the mouth. ## Conclusion Lastly, from an animal welfare perspective, it is important to highlight that the patterns exhibited by chimpanzees who had poor experiences with media, cognitive experiments, and social and physical enrichments in youth. These chimpanzees had been isolated from their mothers and conspecifics and reared by human caregivers at a biomedical laboratory during their infant and juvenile periods, and only later they were transferred to more naturalistic groups in sanctuaries. Previous studies found that impoverished early social experiences negatively affect social behaviors of chimpanzees in general, but importantly, not all chimpanzees reared were affected in similar ways. Thus, one possibility raised by our results is that the tests with eye movements can be used as a diagnostic tool to assess psychological differences across chimpanzee individuals to offer individualized care for those animals; for example, when they are integrated into more naturalistic social groups. In summary, we found that although great apes, humans, and macaques view social movies overall similarly, individuals and species also have unique viewing patterns for several key social features (i.e., eyes, mouths and action targets). Also, we found that individual experiences and species-typical motivation and temperament explain some of the observed individual and species differences. This suggests that the underlying mechanisms affecting variation in social attention are similar across species. From an evolutionary perspective, our results suggest that closely-related primate species can acquire particular attentional biases relatively rapidly through ontogeny and evolution based on shared mechanisms. Such attentional biases might help them to learn effectively from the social environment and enhance their chances of survival and reproductive success. # Supporting information We thank the staffs at Kumamoto Sanctuary, Kyoto University, Japan, and Wolfgang Köhler Primate Research Center, Germany, for the help of data collection. We also thank R. Mundry for the help of statistical analysis. [^1]: The authors have declared that no competing interests exist.

# Introduction Globally, 37.9 million (32.7 million-44.0 million) people were living with HIV in 2018, of which 20.6 million (18.2 million–23.2 million) were from the eastern and southern Africa region. The proportion and risk factors of HBV co-infection vary widely with geographical location: 5–10% in North America, Europe and Australia, and 20–30% in sub-Saharan Africa (SSA) and Asia. HIV positive individuals are more likely to be infected with HBV than HIV- negative individuals, possibly as a result of shared risk factors. Progression of chronic HBV to cirrhosis, end-stage liver disease (ESLD), and hepatocellular carcinoma (HCC) is more rapid in HIV co-infected individuals. HBV replication markers appear to be influenced by HIV infection. HIV co- infection prevented HBsAg secretion with significantly elevated HBsAg quantity present in cell lysates in co-infected hepatic cell lines. In HBV co-infection, HBsAg may hence be too low to be detected using serological tests and anti-HBc may be the only serological evidence of exposure to the virus. In HBsAg negative individuals, anti-HBc positivity may be associated with occult hepatitis infection (OBI), characterized by low but detectable viral replication using PCR based approaches.. HIV infection is also a risk factor for HBsAg negative HBV infection and the development of OBI, since it occurs more frequently in HIV- infected patients. Testing and diagnosis of hepatitis infection is critical to both prevention and treatment services, and provides an opportunity to reduce transmission, through counselling on risk behaviors and vaccination. HBV co-infection status is not commonly tested in ART clinics despite the recommendation for hepatitis B and C routine screening and vice versa. Lack of HBV screening may lead to treatment without the use of the recommended tenofovir (TDF) in the regimen, which may be further associated with flares of hepatitis B due to ART-associated immune reconstitution. In addition, the use of lamivudine (3TC) as the single agent in an ART regimen with activity against hepatitis B is contraindicated due to high likelihood of resistance development to YMDD (tyrosine-methionine-aspartate- aspartate), the highly conserved motif in HBV. In Ethiopia, reports indicate that HBsAg prevalence ranges from 2.7% to 14% among HIV infected individuals. Nearly all reports are based on HBsAg testing with less emphasis on anti-HBc and anti-HBs seromarkers despite their clinical importance. Screening for these seromarkers is also important to identify non- HBV exposed individuals who would benefit from HBV vaccination or to identify those with chronic disease requiring follow-up monitoring or treatment. This study therefore, assessed HBsAg, anti-HBc and anti-HBs seromarkers among HIV infected adults on ART to determine their magnitude, exposure related factors and reveal the unmet need of HBV screening. # Materials and methods ## Study setting and period The study was conducted in ART clinics of three selected public hospitals (Hiwot Fana Specialized University Hospital, Dilchora General Hospital, and Karamara General Hospital located in Harar, Dire Dawa and Jigjiga towns, respectively), in Eastern Ethiopia. The hospitals provide specialist services to both the urban trade hubs and the pastoral and agro-pastoral communities inhabiting the borders with Somalia and Djibouti, each catering to an estimated five million populations. The study was conducted from September 2017 to February 2018. ## Study population Consenting ART experienced HIV infected adults who attended ART clinic during the study period were recruited. The study included participants which were not HBV vaccinated and not screened for HBV seromarkers. ## Sample size determination The sample size was calculated using OPEN Epi 3.1 assuming HBV co-infection of 6.9% among HIV infected individuals on ART, and 3.7% among blood donors, and adding 10% for assumed non-response. The final calculated sample size was 920. ## Sampling procedure A sample size of 385, 325 and 210 was allocated proportionally to Hiwot Fana, Dilchora and Karamara hospitals, respectively. Participants who were not HBV vaccinated and not previously screened for HBV were recruited consecutively at each site until the allocated sample size was attained. ## Data collection Based on the routine follow-up of HIV infected individuals, an ART nurse screened eligible subjects, conducted interview and sent participants to the laboratory for sample collection with a unique identifier. History of opportunistic infection, history of tuberculosis (TB), baseline and current CD4+T cell count, duration on ART, initial and current ART regimen, and if changed, reason for regimen change, WHO clinical stages, medication adherence and other clinical and demographic information were captured from participants’ ART follow-up form using a checklist. A senior Internal Medicine specialist was involved for patient consultations as necessary at each site. ## Sample processing and serology In the laboratory, 10 ml blood was collected in sterile anticoagulated tube, plasma was separated, labelled and stored at -20°C by medical technologists until transferred to Haramaya University. The plasma samples were screened and interpreted for HBsAg, ant-HBc and ant-HBs using ELISA (Monolisa HBsAg ULTRA, Monolisa Anti-HBc PLUS, Monolisa Anti-HBs PLUS, BIORAD, France) in the Medical Laboratory Science Department, College of Health and Medical Sciences, Haramaya University, following the manufacturer’s instruction. ## Quality control To maintain data quality, data collectors were trained and questionnaire was pre-tested in different sites other than those selected for the study before the actual data collection. Samples were kept at -20°C until processed. Standard operating procedures (SOP) and pre-analytical, analytical and post analytical quality control measures were applied. ELISA test results were determined based on the cut-off values following the manufacturer’s instruction. ## Data management and analysis Data were cleaned, coded and entered to EPI Data version 3.1 and analysed using Stata version 11(Stata Corp, USA). Descriptive analysis was used to calculate prevalence, and summarize sociodemographic and associated factors. A binomial logistic regression model was used and associated variables with p≤0.25 were entered to multiple regression analysis to control for potential confounders. Adjusted odds ratio (AOR) and 95% confidence interval (CI) were used to assess the strength of association between dependent and independent variables. Finally, statistical significance was decided at p\<0.05. ## Ethical considerations This study was reviewed and approved by the Institutional Health Research Ethics Review Committee (IHRERC) of the College of Health and Medical Sciences, Haramaya University (Ref. No. IHRERC/137/2017) and AHRI/ALERT Ethics Review Committee, Addis Ababa (Ref. No. P019/17). Written signed informed consent was obtained before data collection. Laboratory results of HBV seromarkers were reported to the respective attending clinician for the necessary intervention. To maintain confidentiality, participants’ information was coded and names and personal identifiers were not used. # Results ## Socio-demographic characteristics of study participants A total of 901 (98%) HIV infected individuals on ART were included in this study. Of the total participants, 817 (90.7%) were urban residents, and 622 (69%) were female of which 539 (86.6%) were within the reproductive age group. The median age of the respondents was 40 years (IQR 32, 45) with an average family size of 3.12. Among the study participants, 291 (32.3%) did not attend formal education, and 71 (7.9%), had attended tertiary education. “” ## Behavioral and health related characteristics of study participants Of the screened HIV infected individuals on ART, 173 (19.2%) reported current alcohol consumption, 118 (13.1%) khat chewing, 544 (60.4%) a history of body piercing, 293 (32.5%) had tattoos and 314 (34.9%) shared sharp tools such as needles and razors. Three hundred sixty-two (40.2%) had history of hospital admission, 59 (6.5%) genital discharge, 186 (20.6%) history of dental extraction, and 119 (13.8%) had multiple sexual contacts. ## Distribution of HBsAg, anti-HBc and anti-HBs seromarkers among study participants Based on the three HBV seromarkers assessed, 334 (38.1%) were negative for all, 176 (20.1%) were positive for both anti-HBc and anti-HBs and 53 (6.0%) were positive only for HBsAg. Those positive for “anti-HBc only” (IAHBc) totaled 184 (21.0%), of which 113 (61.4%) were females and 93% of these females were urban residents. “” ## Clinical characteristics of HBV co-infected and HIV mono-infected individuals Nearly all, 99.8%, of the study participants were on ART, and had been receiving treatment for a median duration of 86.0 months (IQR 51,118). A total of 850 (95.6%) were taking first line and 39 (4.4%) were taking second line ART regimen currently. From a total of 313 (35.2%) who had ART regimen changed, 144 (46.0%) were due to undesirable side effects and 39 (12.5%) were due to apparent treatment failure. Among participants taking second line drugs due to treatment failure, 3 (7.5%) were HBsAg positive, and one was in WHO clinical stage III category. Based on the clinical data record of the study participants, 244 (27.2%) had history of tuberculosis (TB) and 476 (53.1%) had a history of opportunistic infections (OI). The vast majority of the participants, 835 (93.8%) had good ART adherence, and 844 (94.8%) were in WHO stage I category. “” The median baseline and current CD4+T cells counts of the participants were 180 and 519 cells/μl (IQR: 104.5, 274.5; 353.75, 691), respectively. Among the total participants, 481 (58.0%) had baseline CD4+ T cells count ≤200 cells/μL. Among these, 57 (11.9%) were HBsAg positive, of which 23 (40.4%) were taking ART combination with TDF and 3TC. Currently, 63 (7.9%) had CD4+ T cells count ≤200 cells/μL. From HBV co-infected individuals, 31 (29.5%) had a history of tuberculosis, 81 (77.1%) had history of opportunistic infections and 101 (96.2%) had good adherence and were in the WHO clinical stage I category. A total of 567 (63.8%) participants were currently taking ART containing TDF and 3TC combinations. From those who were both HBsAg and anti-HBc positive, 29 (58%) were currently taking an ART containing both TDF and 3TC, however, 21 (42%) were taking ART regimens comprising 3TC only. ## Prevalence of Hepatitis B virus surface antigen and associated factors Among the study participants, 105 \[11.7%, 95%CI (10–14)\] were positive for HBsAg. The distribution was 11.9% among females and 11.1% among males. The majority, 92 (87.6%), of the HBsAg positives were urban residents and 74 (70.5%) were female. Nearly half, 50 (47.6%) of the HBsAg positives were also positive for anti-HBc, among which, 42 (84%) were urban residents, and 31 (62.0%) were females, of which 27 (87.1%) were within the reproductive age group. Binary logistic regression analysis was conducted including sociodemographic, behavioral, health related and clinical variables. Marital status, tattooing, hospital admission, sharing sharp tools, discharge from genitalia, genital mutilation, history of opportunistic infections and presence of TDF and 3TC in the ART regimen were considered for multivariable logistic regression analysis (p≤0.25). In the final model, marital status, history of genital discharge and ART with TDF and 3TC combination remained statistically significant (p\<0.05). Besides, sharing sharp tools and genital mutilation remained strong predictors though they were marginally significant. HIV infected individuals who were single were 2 times more likely to be HBsAg positive compared to married ones \[AOR 2.10; 95%CI (1.03, 4.28) p = 0.041\]. History of genital discharge increased the chance of being HBsAg positive by 2.9 times \[AOR 2.90; 95%CI (1.18, 7.09) p = 0.020\]. Those who were currently treated with ART without TDF were 1.9 times more likely to be HBsAg positive compared to those treated with TDF combinations \[AOR 1.89; 95%CI (1.10, 3.23) p = 0.020\]. Likewise, those who shared sharp tools were 1.9 times more likely and females who had genital mutilation were 1.8 times more likely to be HBsAg positive compared to their counterparts. However, the p value was marginally significant \[AOR 1.97; 95%CI (0.99, 3.93) p = 0.051; AOR 1.81; 95%CI (0.99, 3.28) p = 0.052\]. “” ## Prevalence of Anti-HBc and associated factors among study participants A total of 419 \[(46.5%, 95%CI (43, 50)\] HIV infected individuals on ART were positive for anti-HBc seromarkers, among which the majority, 272 (64.9%) were female and 383 (91.4%) were urban residents. The distribution was 43.7% among female and 52.7% among male. To assess independent predictors of anti-HBc positivity, factors that were significant in bivariate analysis (p≤0.25) were entered into multivariable logistic regression analysis. In the final model gender, level of education, occupation and multiple sexual contact remained significant predictors (p≤0.05). Males were 1.6 times more likely to be anti-HBc positive \[AOR; 1.59 95%CI (1.11, 2.26) p = 0.010\]. Unemployed participants were 2 times and daily laborers were 1.9 times more likely to be anti-HBc positive compared to government employees \[AOR; 2.17 95%CI (1.28, 3.67) p = 0.004; AOR; 1.90 95%CI (1.11, 3.26) p = 0.020\]. Those who had multiple sexual contacts were 2 times more likely to be anti-HBc positive compared to their counterparts \[AOR; 2.21 95%CI (1.44, 3.38) p = 0.001\]. High school students were 63% less likely of being anti-HBc positive compared to college or university graduates \[AOR; 0.37 95%CI (0.19, 0.69) p = 0.002\]. “” # Discussion The overall prevalence of HBsAg and anti-HBc among HIV infected individuals on ART was 11.7% and 46.5%, respectively. Among the HBsAg positives, 47.6% were also positive for anti-HBc and (29/50) 58% of these were currently on an ART regimen containing TDF and 3TC. Among the total participants screened for the three seromarkers, 38.1% were negative for all, and 21% were positive only for anti-HBc (IAHBc), and 96% of the participants were on ART for more than 6 months. Being single, history of genital discharge and taking ART with TDF were significant predictors of HBV co-infection. Male gender, unemployment, daily laborer, history of multiple sexual contacts and level of education were significant predictors of HBV exposure. The HBV co-infection prevalence in this study is higher compared to HBV prevalence of 4.7% in Addis Ababa, 5.9% in Mekelle hospital, 5.5% in University of Gondar Hospital, 6.3% in Southern Ethiopia, 6.9% in Hawassa Referral Hospital, a pooled prevalence of 5.2% in meta-analysis regardless of ART status and a 7.4% global prevalence report. However, it is less than the 14% HBV prevalence reported from Shashemene town in Southern Ethiopia and the overall prevalence of 15% in sub-Saharan Africa. A similar finding of 11.7% was reported among HIV infected HAART naïve individuals in north-west Gondar. A recent population-based HIV impact assessment (EPHIA 2017–2018) report in Ethiopia indicated a 4.8% HBV co-infection among adults of 15–64 years of age (3.6% in women to 7.4% in men) in urban Ethiopia using rapid diagnostic test. The variations in prevalence reports may be attributed to ART exposure that could significantly reduce the level of HBV DNA, geographical variation of HBsAg carriage that range from 1.9% to over 40% and mutations in the S region of HBV. The difference in sample size and the diagnostic tools may also affect to the prevalence reports in Ethiopia. The high HIV/HBV co-infection in this report is an evidence for the serious health burden that demands immediate intervention considering the recent HIV increase from 1.14% in 2014 to 3% in 2018 in urban population in Ethiopia due to the shared risk factors. Additionally, 87.1% of those females who were positive for both HBsAg and anti-HBc were within the reproductive age group. This increases the risk of vertical and horizontal transmission, as well as progression to chronic liver disease unless the necessary precautions are implemented. Studies reported that one-third of the world’s population has serologic evidence of past or present HBV infection. Though anti-HBc is widely used in HBV screening as an epidemiological marker, data on the magnitude of anti-HBc is limited in Ethiopia. The anti-HBc seropositivity in this study is higher than the 22.5% anti-HBc prevalence reported in southern Ethiopia, and slightly lower compared to the 52.4% reported among ART experienced patients in Addis Ababa and the 55.1% reported from sub-Saharan Africa. About 70–90% of all HIV patients show evidence of past or active HBV infection in Kenya. The possible explanation for such variations could be the simultaneous infection with HIV that reduces immune control of previous HBV infection facilitating HBV reactivation and HBV DNA replication without presence of detectable HBsAg. Moreover, immune status of the study participants and stages of HBV disease may affect likelihood of HBsAg detection. The higher occurrence of occult HBV infection in HIV-positive people may also be attributed to the lower rate of HBsAg. The anti-HBc carriage among HIV infected individuals on ART may indicate a high rate of HBV transmission as anti-HBc typically persists for life, regardless of whether the infection resolves or remains chronic. However, the anti-HBc test should be supported by testing for HBsAg and anti-HBs in order to decide whether it indicates HBV immunity through natural infection, chronic HBV infection, IAHBc or OBI. In our study, based on the seromarkers assessed, 37.3% of the study participants were negative for HBsAg, anti-HBc and anti-HBs, and therefore susceptible to HBV infection. These group could be protected from HBV infection through vaccination if pre-ART HBV screening had been implemented in Ethiopia. The unmet need of pre-ART HBV screening and subsequent appropriate vaccination hampered the HBV infection control effort, and as well as contributing to progression of untreated chronic HBV to cirrhosis, ESLD, and HCC. Similarly, 20.1% were positive for both anti-HBc and anti-HBs indicating protection due to natural infection, and 4.8% were positive for both HBsAg and anti-HBc indicating the requirement for further follow up, to prevent adverse consequences. Though it was not defined in the advisory committee recommendations, we found 6.1% of HBsAg only positive cases. This might have occurred due to low anti-HBc levels depending on the immune tolerant and inactive phases of HBV infection which may affect its detection. Furthermore, 21% of the total screened were positive for anti-HBc and negative for both HBsAg and anti-HBs, hence categorized as “isolated anti-HBc” (IAHBc), or “anti-HBc alone”. In Ethiopia, we have not observed any data reporting IAHBc to date and therefore, the burden is unknown despite its clinical and public health implications. IAHBc may represent several clinical entities including the window phase of acute HBV when anti-HBs is not yet detected, the late stage of prior infection after anti-HBs has fallen to undetectable levels, OBI, or false positive anti-HBc. HIV co-infection was demonstrated to be a risk factor for HBsAg negative infection, and pre-S1 and ‘a’ determinant mutations also prevent HBsAg secretion, ultimately affecting virus detection. High anti-HBc levels in the immune active and immune reactivation phases of chronic HBV infection and the higher occurrence of OBI (2% to 10%) in HIV-positive people may also affect detection rate of HBsAg, leading to IAHBc. Patients who have undergone viral clearance may also lose the ability to produce anti-HBs after long periods of time due to a waning T-cell response. OBI has much impact on different clinical and public health aspects, including transmission, risk of reactivation and enhancing liver disease progression that can lead to HCC. This further emphasizes the critical need of HBV screening among HIV infected individuals before ART initiation to prevent the potential risk of HBV transmission. In our study, nearly all study participants were on ART with different combinations for a median of 86 months. For HBV co-infected individuals if treatment is indicated, the national guideline recommends TDF + 3TC (or FTC) + EFV as a preferred regimen. We found that 70% of the participants had been treated with an ART regimen containing TDF compared to a the 51.2% report among HBsAg and anti-HBc positive individuals on ART. The rest were on ART with 3TC as the only anti-HBV-active agent. This may cause unnecessary consequences on the patients due to continued HBV viremia and progression, resulting from high rates of drug resistance mutations (DRMs) and virologic breakthrough. Furthermore, 3TC resistance confers partial or complete cross-resistance to other HBV inhibitors such as emtricitabine (FTC), telbivudine and entecavir (ETV), thus limiting treatment options. In Ethiopia, the unmet need of HBV screening, might have affected the clinical outcome of the patients who were eligible for TDF but remain taking 3TC only leading to circulation of drug resistant strains among the community. Being single, history of genital discharge, and ART without TDF were statistically significant predictors of HBsAg positivity. This study, did not find a statistically significant association between HBsAg positivity and many of the sociodemographic characteristics such age, sex, residence and level of education like other similar studies. History of hospital admission, surgery, dental extraction and sharing of sharp materials were not significantly associated with HBsAg. Male gender, occupation (being unemployed and daily laborer), level of education and multiple sexual partner were significant predictors for HBV exposure. The risk difference by gender may be due to differences in risk behavior over years of life. However, exposure to HBV infection was not significantly associated with body piercing and tattooing, and age and history of genital discharge. No significance difference was observed by residence, marital status and history of dental extraction as a risk for HBV exposure. Those who shared sharp tools were 1.9 times more likely and females who had genital mutilation were 1.8 times more likely to be HBsAg positive compared to their counterparts. However, the p value remained statistically insignificant (p\>0.05). ## Limitations This study did not conduct HBeAg and anti-HBc IgM tests to differentiate between active viral replication and acute infection, respectively among the study participants. HIV viral load, liver function tests and related data were incomplete to assess other important clinical parameters. # Conclusion and recommendations This study showed a relatively high prevalence of HIV/HBV coinfection, HBV co- infected participants taking 3TC as the only anti-HBV agent, susceptible HIV infected adults that require HBV vaccine and IAHBc group. In general, the unmet need for HBV screening prior to ART initiation may affect the quality of care given to the patients, increase the risk of life-threatening complications among the co-infected, but untreated ones and hamper the prevention effort towards HIV and HBV as they share transmission routes. In addition, it brings into attention the importance of integrating anti-HBc screening due to its public health implication in HBV transmission as well as drug resistant mutations. We would like to express our gratitude to HIV infected individuals on ART in the study areas. We also would like to thank ART clinic and Medical Laboratory staff, and administration of Dilchora Hospital in Dire Dawa, Hiwot Fana Specialized University Hospital in Harar and Karamara Hospital in Jigjiga, for their unreserved support and collaboration during data collection. [^1]: The authors have declared that no competing interests exist.

# Introduction The identification and treatment of sepsis continues to be a major health issue. The incidence of sepsis is particularly high in the neonatal population, where low birth weight and other compromising factors make it a primary cause of morbidity and death. Early identification and treatment are critically important to healthy patient outcomes given the inconsistent presentation of sepsis in terms of body temperature, which may be either above or below normal. The most reliable diagnostic of neonatal sepsis, often referred to as the *gold standard*, is a blood culture test for bacteria. While this test is the most reliable available, it can take hours to obtain the results. As a result, treatment must often begin before the results are known. An additional complication is the fact that the blood culture test can be negative for one in five subjects with sepsis. Thus, it is of critical importance to identify new biomarkers that will enable fast and reliable hematological scoring systems for sepsis in its earliest stages. The current hematological scoring system was first proposed by Rodwell, et al. in 1988 and is based on the following seven quantities: total leukocyte (or White Blood Cell, WBC) count, mature neutrophil count (also named Segs, Absolute Neutrophil Count, or ANC), immature neutrophil count (also named Bands, Absolute Band Count, or ABC), Immature to Total neutrophil count ratio (IT-ratio), Platelet count (Plt), and adverse changes in the total neutrophil count. Another scoring system was proposed in Ref. that characterizes a patient as septic if any two of the following four criteria are satisfied: - or - - \- -. These hematological scores are supplemented by other observational evidence and measurements collected by physicians including body temperature, blood pressure, and clinical presentation in determining the course of treatment before the blood culture results are available. Additional diagnostic hematological biomarkers have been studied such as C-reactive protein (CRP), and procalcitonin. While these biomarkers have shown to be correlated with sepsis, they are considered to have limited diagnostic information. More recently, the blood biomarker neutrophil CD64 has proved to be particularly promising for early detection of sepsis. Neutrophil surface CD64 expression is a high affinity Fc receptor for immunoglobulin G (IgG) expressed on neutrophils (and other white blood cells). Quantities of CD64 increase markedly when neutrophils are activated by the human body's response to infection, and in particular, to sepsis. The challenge of biomarker identification is reflected by the fact that over sepsis biomarker studies have been published with almost candidate biomarkers evaluated. Nonetheless, clinicians are unsatisfied with the diagnostic tools currently available for making accurate and timely sepsis diagnoses that would also support appropriate therapies. The challenge is not to identify single biomarkers that pass a univariate test for diagnostic efficacy, but to determine which sets of biomarkers, when considered as a group, yield the most accurate prognosticator. In this investigation, we integrate two tools for discovering information in large data sets. Embedded feature selection using a sparse support vector machine classifier, and canonical correlation analysis, a tool for identifying relationships between two sets of variables. This two-pronged analysis provides a powerful general tool for the identification of biomarkers useful for multivariate scoring systems. In this manuscript, we present a systematic study of the multivariate diagnostic capacity of a set of ten hematological biomarkers. Our goal is to establish a general approach that can be used effectively on potentially much larger sets of biomarkers. We develop an approach to identify a minimum set of predictive biomarkers with the ultimate goal of improving the early detection of sepsis. We verify the results by conducting an exhaustive evaluation of all possible combinations of biomarkers. We envision that the algorithms proposed here will be helpful tools as advances in biomedicine produce additional candidate biomarkers arising from new proteomic and metabolomic tests. # Results A total of sepsis evaluations were performed on neonates during the study period. Blood cultures, complete blood counts (CBC), and neutrophil CD64 data were obtained for of the sepsis evaluations. One evaluation was excluded due to the high neutrophil CD64 value that skewed the results. Evaluations were partitioned into three groups: (1) blood culture positive septic group, (2) clinically probable septic group, and (3) nonseptic group. In this study, we combined groups and and labeled these subjects as having sepsis. Our analysis is based on the comparison between this combined septic group and nonseptic group. See Materials and Methods for details. Data for ten hematological biomarkers were analyzed in this study including: (1) Age, (2) WBC count, (3) Hemoglobin count (Hgb), (4) Hematocrit percentage (Hct), (5) Plt, (6) Segs, (7) Bands, (8) Lymphocyte (Lymph) count as a percentage of WBC, (9) Monocyte (Mono) count as a percentage of WBC, and (10) neutrophil CD64 expression. Following Ref., P-values were computed for the biomarker data and all ten biomarkers were determined to have predictive capacity. ## Optimal Subsets of Biomarkers Multivariate correlation analysis is a general tool for exploring how variables are inter-related. Canonical correlation analysis provides a powerful tool for discovering relationships between two sets of variables. Given two sets of variables, CCA can identify subsets of each set, which when combined as latent variables, produce the maximum correlation between the two sets. In this study, we choose one set of variables to be the sepsis score, and the second set is taken from all possible subsets of the ten biomarkers. CCA can thus generate an ordered list of biomarkers that are most correlated with the sepsis score. See Materials and Methods for details. Here we discuss the results of applying CCA to select the best combinations of sepsis biomarkers. We first consider the single biomarker with highest correlation to the sepsis score. As shown in, this biomarker is Bands. If we consider all pairs of biomarkers, Bands and Plt possess the highest correlation with sepsis score. We note that CD64 has the second highest correlation with sepsis score, in the univariate sense, but improves the correlation of Bands to sepsis score less than Plt, which has a lower univariate correlation with sepsis score. This is due to the fact that Bands and CD64 are more correlated than Bands and Plt, and so less information is provided by adding CD64. Hgb enters at even though it has a very weak pairwise correlation with the sepsis score given it also has very weak pairwise correlation with Bands and Plt. The correlation saturates at with the following combination set of biomarkers: Bands, CD64, Segs, WBC, and Plt. The rest of the biomarkers do not provide significant additional information about the sepsis score. The above analysis suggests that these five variables should be included in our sepsis scoring system. ### Comparison with Forward Selection Method Forward Selection (FS) is a well known data-driven selection method, where additional variables are added in one-by-one to improve the model. The FS method selects the single variable out of the remaining set that gives the highest absolute correlation with the residual vector. The results from FS on the sepsis data set are compared with those from CCA in. Both methods involve linear correlations, but FS is a greedy algorithm, which only produces a locally optimal solution. However, we find that up to, CCA and FS select the same subset of biomarkers. At, CCA and FS differ. Since FS can only select one feature at a time, at, FS selects Segs, while CCA selects Segs and WBC and replaces Hgb. The manner in which we implemented CCA ensures a globally optimal solution for each. In the next section, we validate this result using a classifier to predict the sepsis score in terms of these biomarkers. ## The Diagnostic Classifier We seek to construct a decision function from the biomarker data that serves as a hematological scoring system, *i.e.* a function that maps a sample vector of biomarkers to a positive or negative sepsis diagnosis. Using the biomarkers identified by CCA above, WBC, Plt, Segs, Bands, and CD64, we propose the linear decision function: From the sparse support vector machine approach described in Materials and Methods, we determined the optimal decision function to be See for the weights, and their errors, and means and standard deviations of the biomarkers. With this decision function, if the Score is greater than or equal to zero the diagnosis is positive for sepsis, whereas if the Score is less than zero, the diagnosis is healthy or aseptic disease. We note that since the range of values of the biomarkers varies widely, all values of the biomarkers are normalized by subtracting the mean over all cases and then dividing by the standard deviation. The results of applying the classifier) to the full sepsis dataset are shown in. We calculated the true positive rate (TPR), true negative rate (TNR), positive predictive value (PPV), negative predictive value (NPV), and accuracy (ACC) (defined in Materials and Methods) for these five biomarkers. We emphasize that there are two remaining questions of interest. How good is the classifier? Did we identify the most predictive biomarkers from the original set of ten? We focus on the validation of these biomarkers in the next section. ## Biomarker Validation In this section, we have two goals. First, we will verify that the number of biomarkers suggested by CCA, is optimal. Secondly, we seek to provide evidence that the CCA-selected biomarkers are optimal. To do this, we will perform an exhaustive analysis of all possible scoring systems for the ten biomarkers. Clearly this approach is not feasible for large sets of biomarkers, but we exploit the fact that we only have ten to illustrate the power of CCA biomarker selection by constructing all possible SSVM classifiers. We used the accuracy of the resulting decision functions for our validation. ### Validation of the Classifier For each, we select the -combination set of biomarkers as identified by CCA and shown in. We construct a decision function for each from to and evaluate several measures of the quality of the scoring system in. We find that each measure begins to saturate near, although one could argue that some slight improvement could be obtained by adding one or two more biomarkers for the given model. (We note that this particular model was not optimized over variations in the parameter.) Receiver operating characteristic (ROC) curves for true positive versus false positive rate provide additional insight into the determination of the minimal number of biomarkers that provide predictive information about sepsis infection. In, we show that the ROC curves become independent of for, and thus is indeed the appropriate number of biomarkers. In the inset to, we show the ROC curve for averaged over SSVM models. ### Validation of the CCA Selected Biomarkers We provide further evidence that our CCA biomarker selection was in fact the optimal one by applying SSVM to all possible combinations of biomarkers for each. We show the TPR, TNR, PPV, NPV, and ACC for the top of all possible combinations in. It is clear that the CCA-selected biomarkers possess the best statistical measures for each. ### Comparison with Logistic Regression Logistic Regression (LR) is widely used for classification problems. A LR model can predict the outcome variable, such as the disease state ( *i.e.* sick or healthy), by the new predictor inputs. The LASSO (Least Absolute Shrinkage and Selection Operator) algorithm is a -norm regularized logistic regression, which is extensively used for feature selection. By the -norm penalty, LASSO Logistic Regression (LLR) can achieve a sparse solution and exhibit a significantly high tolerance to the presence of many irrelevant features. Here we also construct a LLR based classifier for each from to and plot the same statistical measures of the performance of the diagnostic system in as for SSVM. The sets of biomarkers determined by CCA are used for each -combination. We observe a similar saturation for all of the measures near. On our test data set we observe that LLR has a superior true negative rate while its true positive rate is inferior. The variability of the measures is substantially wider for LLR than SSVM (see for the LLR and SSVM performance of the classifier at k = 5). In practice physicians may be concerned with a specific measure, e.g., negative predictive value. In this case, either of these methods could be used to optimize the negative predictive value. Please see Supporting Information for more details. # Materials and Methods The data sets were obtained from a prospective study conducted in the Neonatal Intensive Care Unit at Yale-New Haven Hospital.This study was approved by the Yale University School of Medicine Human Investigation Committee. Consecutive patients, who underwent a sepsis work-up as deemed necessary by the attending neonatologist during the time period 1/2008-6/2009, were enrolled in the study. ## Sepsis Evaluations The clinical and historical features used to identify patients at risk for sepsis include one or more of the following, as determined by the attending neonatologist, : (1) respiratory compromise (*e.g.* tachypnea, increase in frequency or severity of apnea, or increased ventilator support); (2) cardiovascular compromise (*e.g.* increased frequency or severity of bradycardic episodes, pallor, decreased perfusion, or hypotension); (3) metabolic changes (*e.g.* temperature instability, feeding intolerance, glucose instability, or metabolic acidosis); (4) neurological changes (*e.g.* lethargy, hypotonia, or irritability); and (5) antenatal risk factors (*e.g.* maternal Group B Streptococcus (GBS) colonization without adequate intrapartum prophylaxis, unknown maternal GBS status, maternal temperature, chorioamnionitis, preterm labor, or prolonged rupture of membranes). After the sepsis evaluation was performed, we utilized the following values derived from CBC to assign a sepsis score, : (1) Absolute Neutrophil Count (ANC) or ; (2) Absolute Band Count (ABC) ; (3) Immature to Total neutrophil ratio (IT-ratio) ; and (4) Platelet (Plt) count. Infants who met 2 or more of these laboratory criteria were categorized as having a positive sepsis score. Hemoglobin was measured in the clinical hematology laboratory using a calorimetric method. The hematocrit was calculated after measuring the total red blood cell count (RBC) and the mean corpuscular volume (MCV) of the RBCs. All blood cultures were collected using standard sterile techniques. As per unit protocol, we attempt to obtain 2 blood cultures with a minimum of ml. The BACTEC (Becton Dickinson and Co., Sparks, MD) microbial detection system was used to detect positive blood cultures. Neutrophil CD64 expression was measured using l of whole blood incubated for minutes at room temperature with a saturating amount of fluorescein isothiocyanate (FITC)-conjugated anti-CD64 monoclonal antibody or isotype control (Leuko64 kit, Trillium Diagnostics, Scarborough, ME), followed by ammonium chloride-based red cell lysis. Samples were washed once and re- suspended in ml of phosphate-buffered saline with 0.1% bovine serum albumin. Flow cytometric analysis was accomplished using a Becton-Dickinson FACScan (Mountainview, CA) to collect log FITC fluorescence, log right-angle side scatter and forward scatter on a minimum of 50,000 leukocytes. Interassay standardization and neutrophil CD64 quantification were performed using FITC calibration beads (Leuko64 kit). Data analysis was performed using light scatter gating to define the neutrophil population, and the neutrophil CD64 Index was quantified as mean equivalent soluble fluorescence units using QuickCal for Winlist (Verity Software House, Topsham, ME) with a correction for nonspecific antibody binding by subtracting values for the isotype control. This was expressed as an absolute value. Investigators checking and confirming the neutrophil CD64 results were blinded to the clinical data, including the blood culture results. Clinicians did not have access to the neutrophil CD64 values and these were not used to decide initiation or duration of antibiotic therapy. ## Evaluation Studied Evaluations were obtained by accessing the electronic medical record from January 2008 through June 2009. Each evaluation typically included a CBC, two peripheral blood cultures, and other optional cultures. A patient could undergo multiple sepsis evaluations during admission. Since a single evaluation represented a separate episode of suspected sepsis and could be treated independently, we therefore treated all evaluations equivalently in this manuscript. Evaluations were excluded if the CBC, neutrophil CD64, or blood culture tests were not provided in the patient record. A total of sepsis evaluations with complete hematologic, neutrophil CD64, and blood culture data were used for the analyses. (One evaluation, which was positive for was excluded due to the high neutrophil CD64 value that skewed the results.) Information about each sepsis evaluation included (1) sepsis diagnosis type, (2) day of life that the evaluation was performed, and (3) CBC data and neutrophil CD64 expression. Ten biomarkers were included in the analysis: Age, WBC, Hgb, Hct, Plt, Segs, Bands, Lymph, Mono, and CD64. Additional details about the laboratory and clinical data were recently published. ## Defining Sepsis Outcome Individual sepsis evaluations with positive blood cultures were diagnosed as culture-proven sepsis according to the current National Healthcare Safety Network definitions for laboratory-confirmed bloodstream infections. Individual sepsis evaluations with positive sepsis scores were categorized as clinical sepsis. This might include infants with other infectious diagnoses that were not accompanied by a positive blood culture, such as pneumonia, urinary tract infection, and necrotizing enterocolitis. Three groups of evaluations were defined, and each evaluation was assigned to only one of the three groups. Group consisted of evaluations with a positive blood culture. Group with “suspected sepsis” consisted of evaluations, where the patients lacked a definitive positive blood culture, but the clinical diagnosis was unable to rule out bacterial infection. Group consisted of evaluations, for which either the blood culture or clinical diagnosis showed no evidence of infection. ## Data Preprocessing First, each evaluation, with data, is categorized as septic (groups and above) and nonseptic (group), where is a real-valued vector with components (biomarkers) and is the total number of evaluations. For convenience, we labeled each evaluation using the variable, where is the label for the septic group and is given to each in the nonseptic group. To standardize the range of independent biomarkers, we normalized the real-valued data, a matrix, to have zero mean and unit standard deviation for each biomarker : where and are matrices, and are the mean value and standard deviation of for each biomarker. ## Canonical Correlation Analysis CCA is a multivariate statistical tool that facilitates the study of interrelationships among multiple variables. A linear combination of variables can be chosen by CCA such that the correlation between two sets of data is maximized. In our studies, the two sets of data are the sepsis score and each distinct -combination of the biomarkers in the data matrix. In this investigation, CCA is used to identify the set of -biomarkers most correlated as a group to the sepsis score. Please see Supplementary for more details. To explore the redundancy among the biomarkers, we calculated the correlations between each possible -combination of and using CCA. By varying from (single biomarker) to (all biomarkers), we selected the specific set of biomarkers that possessed the highest correlation with for each. The sets of biomarkers that had the largest correlation with the sepsis score and their corresponding correlation coefficients are shown in. ## Sparse Support Vector Machines (SSVM) We applied the SSVM ensemble method to build a classifier for each of the CCA- selected -combination of biomarkers selected by CCA. A linear support vector machine (SVM) is a widely used classifier, which finds the hyperplane that separates high-dimensional data with maximum margin by categories. The search of this hyperplane can be translated into the following optimization problem: where is the -norm of a vector, which induces the sparsity in the weight vector. We refer to the solution of) as a sparse support vector machine (SSVM) following Ref.. Note that splitting the classes in the objective function allows for unbalanced sample sizes. Due to the limited size and noise of our data, a bootstrap aggregation method was applied to build an ensemble of SSVM classifiers using the following procedure, : 1. The data set is randomly divided into a learning set and a test set. is one third of the data. 2. Based on the bootstrap aggregation method, a bootstrap training set is randomly selected from the original learning set with replacement. That is, has the same number of samples as the original training set, but with several training samples appearing multiple times. Each bootstrap set contains unique samples of the original training set. By repeating this process times, an ensemble of classifiers, with, is built by the SSVM. To have the same total cost for both false positives and false negatives, the parameters and of the SSVM are chosen according to with since the results are not sensitive to the overall scale of. 3. The final classification is obtained by calculating the mean of the ensemble of classifiers. 4. The random division of the data into and is repeated times, after which we calculate the mean and standard deviation. We used the same random divisions of the training and test sets for each. ## Calculation of Statistical Measures The statistical measures of the performance of a classifier are measured using ACC, TPR, TNR, NPV, and PPV. For the sake of completeness, we include their definitions: Here, these statistical measures are calculated for each one of the random divisions of test sets by the classifier built on the bootstrap aggregation method. Their mean and standard deviation are calculated from the groups obtained from the random divisions. # Discussion ## Clinical Issues in Sepsis Diagnosis The problems associated with confirming sepsis with positive blood cultures, as mentioned earlier, has led clinicians to investigate alternate approaches for confirmation of diagnosis of blood-cluster negative or clinical sepsis and prevention of missed or under treatment of neonates with antibiotics. Clinical parameters are notorious for their non-specific nature in detecting infection, especially in premature neonates; however, a scoring system based on a 7-item weighted clinical score has been suggested. In real world settings, most clinicians rely on clinical judgment, in concert with specific hematological criteria, to identify infants with sepsis. The hematological criteria have usually included ANC, ABC, IT-ratio, and platelet counts, as was done in the present study,. Unfortunately, this approach has not proven very reliable due to the inherent subjective nature of the clinical assessment and the variability of the hematological parameters secondary to physiological derangements and non- infectious medical conditions, and has led to over-treatment of neonates. It has been suggested that the addition of specific molecular markers might improve diagnostic accuracy of neonatal sepsis. Among the acute-phase reactants, CRP is probably the most well studied, but its value for diagnostic accuracy has had mixed results. Among the newer ones, procalcitonin – and neutrophil CD64, have shown promise. Neutrophil CD64 values have been reported to be sustained for at least 24 hours in neonates with sepsis. Several studies have investigated the usefulness of the CD64 Index in the NICU population, albeit in much smaller cohorts, but with promising results in both the preterm and term populations, as well as in cases of both early-onset sepsis and late-onset sepsis,. Recently, studies have suggested that the diagnostic accuracy of neutrophil CD64 is superior to the IT-ratio and CRP for the early detection of neonatal sepsis. Secondly, testing can be done on the same sample sent for a CBC evaluation as it requires only 50 l of blood. Thirdly, the CD64 results can be made available within hours of the CBC, since most clinical laboratories in the developed and some developing countries have flow cytometry technology. Furthermore, standard cell counters which use flow cytometry have the potential to incorporate anti-CD64 antibodies and software to provide an even more rapid enumeration of CD64 indices nearly simultaneous with CBC results. Hence, we believe that a scoring system that can incorporate the common CBC parameters with the neutrophil CD64, as was done with our analyses, would provide objective criteria for recognizing neonatal sepsis and guidance for initiation and/or early termination of antibiotic therapy. Additional independent validation of our results is needed before incorporation of our diagnostic sepsis score can be recommended for routine clinical use. ## Identification of the Optimal Subset of Biomarkers We have proposed a new approach for biomarker identification based on the integrated use of CCA and SSVM on a labeled data set. We found that for the neonatal sepsis data our approach produced the optimal set of hematological biomarkers for all possible. We validated our results by conducting an exhaustive search of all combinations of biomarkers and ranking them based on their classification accuracy. These results showed that our approach produced either the absolute top combination, or a combination of biomarkers with statistically indistinguishable performance. Although this study explored a relatively small set of biomarkers, the CCA approach can be applied to potentially much larger sets by exploiting the relative weighting of the features (see Equations S1 and S2) and selecting only the most important features. From Equation S3, CCA requires finding the pseudo-inverse of a matrix, where is the number of biomarkers. Even without invoking sparse methods, one can easily investigate systems with of order. Our approach identifies Bands as the most significant biomarker in our set for detecting neonatal sepsis. The next four most sigificant hematological biomarkers, in order of importance, appear to be Plt, neutrophil CD64, WBC, and Segs. We note that, as illustrated in, the reason for the significance of Bands may be attributed to the fact that it is highly correlated with subjects with a negative diagnosis while much less correlated with subjects with a positive diagnosis. This could be related to the significant variation in Bands for sick individuals as evidenced by. We explore LLR in addition to SSVM to corroborate our results. In each case we see that an exhaustive combinatorial evaluation of the classifiers determines that the CCA selected biomarkers were indeed optimal. See and for a graphical summary of these numerical experiments. Additionally we found that the Forward Selection results were very similar to the biomarkers identified by CCA on this data set, i.e., Bands, Plt, Hgb, CD64 and Segs. CCA selected WBC and not Hgb for the best 5-combination. The classifiers using these two sets of biomarkers perform very similarly with a very slight edge to the CCA biomarkers. However, in general, forward selection is a greedy algorithm and it is possible that the sets of biomarkers identified by CCA and FS could be quite different. Classification algorithms such as SSVM or LLR can then assist in comparing and evaluating the selected biomarkers. We propose that the results found in this investigation, in particular, the new sepsis scoring system, sets the stage for independent investigators to clinically validate these results using alternative sepsis databases. In particular, it will be interesting to ascertain whether this scoring system is also relevant for adults. It is also possible to envision modifying the scoring system based on new data related to alternative scenarios, e.g., septic adults infected by Gram negative microorganisms. Although we propose CCA in conjunction with SSVM as an approach for biomarker identification, the strength of the methodology lies in the exploitation of multivariate relationships within the data and other methods that do this also merit further exploration. # Supporting Information [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: VB GH CSO MDS MK. Performed the experiments: KW VB SC MK. Analyzed the data: KW VB SO CSO MK. Contributed reagents/materials/analysis tools: KW SC SO MK. Wrote the paper: KW VB SO CSO MK.

# Introduction Animal metabolic rate is variable and may be influenced by both endogenous factors (e.g. circadian rhythm, individual physiological traits) and exogenous factors (e.g. oxygen availability). A surge of research interest continues to uncover the mechanistic basis of variability in metabolic rate, and metabolic rate is now one of the most widely measured physiological traits in animals. In many aquatic animals, measurements of oxygen consumption rate (*M*O₂) provide a robust proxy for aerobic metabolic rates. Under static conditions, measurements of *M*O₂ are typically repeatable in individual animals, suggesting that metabolic rate may be an organismal trait, although the repeatability tends to decline over time. Circadian rhythms in physiology and behaviour have evolved to allow animals to anticipate changes in the light-dark environment that are tied to the rotation of Earth. Circadian rhythms reflect endogenous rhythms that are self-sustained, unlike exogenous rhythms that depend on external factors, including changing light levels. Circadian rhythms play a tremendous role in most organisms; ranging from decentralized regulation of the daily timing of mitosis to influencing the migration of animals. Circadian rhythms have been described in details in several teleost fishes. For example, circadian rhythms influencing metabolic rate and behaviour have been documented in Nile tilapia *Oreochromis niloticus* and puffer fish *Takifugu obscurus*. In contrast, in many primitive fishes, the influence of circadian rhythms on metabolism and behaviour remains largely unknown. Standard metabolic rate (SMR) is a basic maintenance requirement measured as the minimum rate of oxygen consumption of postprandial unstressed animals at rest. Long-term energy demands for swimming, food acquisition and treatment, regulation owing to environmental perturbation, and reproduction are additional to standard metabolism. These demands are met within the range set by the maximum metabolic rate (MMR). Animal metabolic physiology is often influenced by exogenous factors, including environmental hypoxia. Hypoxia occurs in a wide range of aquatic systems, and the severity, frequency of occurrence, and spatial scale of hypoxia have increased in the last few decades, primarily due to anthropogenic activity. There are two distinct metabolic responses to environmental hypoxia: 1) oxygen independent respiration in which the metabolic rate remains constant in spite of changing oxygen availability; and 2) oxygen dependent respiration in which the metabolic rate varies with oxygen availability. The two responses are commonly termed oxygen regulation and oxygen conformity, respectively. The vast majority of literature suggests that most teleost fish are oxygen regulators, capable of maintaining both MMR and SMR down to certain oxygen thresholds. In contrast, it remains controversial if oxygen regulation or conformity occurs in a number of primitive fishes exposed to hypoxia. For example, among members of the family Acipenseridae (sturgeons), previous studies have reported conflicting results stating that the metabolic rate remains constant or tends to increase – (i.e. oxygen regulator) or decrease – (i.e. oxygen conformer) when Acipenserids are exposed to environmental hypoxia. Using Adriatic sturgeon *Acipenser naccarii*, McKenzie et al. suggested that swimming *A. naccarii* are oxygen regulators, whereas immobile *A. naccarii* are oxygen conformers. Knowing whether species are oxygen regulators or conformers is important to understand the capacity of fish to respond to environmental changes and to assess assumptions for disparate metabolic theories in ecology. Intraspecific variation in animal metabolic rate may correlate with endogenous factors, including behavioural or life history traits. For example, Niitepõld and Hanski found positive correlations between MMR and life span in a species of butterfly. In fish, MMR is typically measured in the laboratory using either a critical swimming protocol or a chase protocol. Using the latter protocol, Norin and Malte reported that MMR is repeatable over several weeks. Assuming repeatability and heritability, MMR may represent a measure of organism performance, and it is possible that the trait is subjected to natural selection and could evolve over time. Little is known, however, about potential correlations between forced MMR (MMR_F; e.g. measured using the chase protocol) and spontaneous MMR (MMR_S) measured in volitionally performing fish. For example, is there a positive relationship between MMR_F and MMR_S such that an individual fish with an unexpectedly high MMR_F also has an unexpectedly high MMR_S? Clarifying potential correlations between MMR_F and MMR_S is important, because from an evolutionary point of view, selection regimes may not always operate on a trait's maximal value, but rather on the spontaneous use of the trait. If MMR_F and MMR_S are correlated, measurements of MMR_F could function as a predictor of MMR_S in individual fish. Using juvenile lake sturgeon (*Acipenser fulvescens*), we employed intermittent flow respirometry and video analysis to test four hypotheses: 1) *A. fulvescens* exhibit a circadian rhythm influencing metabolic rate and behavior; 2) *A. fulvescens* has the capacity to regulate metabolic rate when exposed to environmental hypoxia; 3) measurements of MMR_F are repeatable in individual fish, and 4) MMR_F is positively correlated with MMR_S. Our results reveal that the metabolic rate of *A. fulvescens* is influenced by a circadian rhythm, and *A. fulvescens* has the capacity to regulate SMR when exposed to environmental hypoxia, demonstrating oxygen regulation. In contrast, MMR_F tends to decrease with increasing levels of hypoxia. Measurements of residual body mass corrected MMR_F are repeatable in individual *A. fulvescens*; and residual body mass corrected MMR_F and MMR_S are correlated positively, but only in *A. fulvescens* exposed to an environmental stressor including hypoxia or 24 h of light. # Materials and Methods ## Ethics statement All procedures were reviewed and approved by the Animal Care Committee at the University of Manitoba, Canada (Approval ID: AUP-F11-004) under the guidelines of the Canadian Council of Animal Care. No animals were sacrificed, all efforts were taken to ameliorate animal suffering and undue stress, and there was no mortality during any of the tests. ## Experimental animals A total of 70 juvenile *A. fulvescens* (body mass: 30.51±1.21 g (mean ± S.E.); age: 1+; sex: unknown) obtained from Grand Rapids Fish Hatchery (Grand Rapids, MB, Canada) were kept at 17±1°C in flow-through holding tanks at the University of Manitoba, Canada. The light regime was 12 h light: 12 h dark (12L∶12D). *A. fulvescens* were fed daily using a mixture of bloodworm (San Francisco Bay Brand, Newark, CA, USA) and sinking trout pellet (Martin Mills Ltd., Elmira, ON, Canada). ## Respirometry Four static respirometers (each 0.83 l) and a mixing pump were submerged in a 100 l opaque tank, filled with freshwater maintained at 17±0.1°C. Oxygen content (% air saturation; O_2sat) of the water in the tank was controlled using two air stones combined with a stream of nitrogen bubbles. Depending on the experiment, water in the tank was maintained at an oxygenation level between 100% and 30% O_2sat. Respirometers were made of transparent glass tubing and were designed to allow a degree of spontaneous activity of *A. fulvescens*, including body undulations with tail excursions\>90° relative to the body axis. Respirometers were situated in a sound isolated room with no other ongoing experiments to minimize any disturbance of the fish. Measurements of *M*O₂ (mg O₂ h⁻¹) were carried out every 9 min using computerized intermittent flow respirometry allowing long term (\>48 h) repeated measurements. Each respirometer was fitted with two outlet and two inlet ports as described previously. The repeated respirometric loops consisted of a 4 min flushing phase during which a pump flushed the respirometer with ambient water through one set of ports. The second set of ports and a pump secured re-circulation of water in the respirometer in a closed circuit phase for 5 min, divided into a waiting phase (2 min) and a measurement phase (3 min). Oxygen partial pressure was measured at 1 Hz by a fiber optic sensor (Fibox 3 connected to a dipping probe; PreSens, Regensburg, Germany) located in the re- circulated loop. The flush pump was controlled by AutoResp software (version 2.1.3; Loligo Systems, Tjele, Denmark) that also calculated the *M*O₂ in the measurement phase using the oxygen partial pressure and standard equations. Preliminary testing demonstrated that the duration of the measurement phase (3 min) ensured that the coefficient of determination (r²) associated with each *M*O₂ measurement was always\>0.95, similar to previous studies. Corrections of background respiration (i.e. microbial respiration) followed Jones et al.. ## Experimental protocols *A. fulvescens* were selected randomly and fasted for 48 h to ensure a post absorptive state prior to experimentation. Subsequently, *A. fulvescens* were introduced to the respirometers and acclimated for 20 h. The light regime during the fasting and acclimation periods was 12L∶12D, which included 0.5 h of gradually shifting light intensity from light to darkness and *vice versa*. Light intensities were 3.0 and 0.0 μmol s⁻¹ m⁻² in daylight and darkness, respectively. Starting at 16:00 h on the next day, *M*O₂ data were collected for the following 24 h. Measurements of *M*O₂ over 24 h comprised three test groups: 1) control (100% O_2sat; 12L∶12D); 2) treatment A (30% O_2sat; 12L∶12D); and 3) treatment B (100% O_2sat; 24L). The oxygen content in treatment A (30% O_2sat) corresponded to approximately 6.2 kPa. Data collection for the three test groups was carried out in a random fashion and each test group included 10–12 individuals. After each 24 h trial, MMR_F was measured as described below. ## Standard metabolic rate (SMR) and maximum metabolic rates (MMR_F and MMR_S) For each test group, SMR in individual fish was estimated as the average of the lowest 10 *M*O₂ values collected over 24 h. This method to estimate SMR was employed because it provides measurements that are repeatable in individual fish. MMR_F was measured immediately after each 24 h trial at the corresponding O_2sat level (i.e. 100% or 30% O_2sat) inside the respirometer. MMR_F was elicited using a standard chase protocol. Briefly, individual *A. fulvescens* were transferred to a circular trough and chased to exhaustion, similar to previous studies on Atlantic sturgeon (*Acipenser oxyrhynchus*) and shortnose sturgeon (*Acipenser brevirostrum*). Upon exhaustion, identified by no further response after 5 min of manual stimulation, *A. fulvescens* were transferred (\<20 s) to the respirometer where *M*O₂ measurements started immediately. MMR_F was the highest of three consecutive *M*O₂ measurements. In addition, following the same chase protocol, MMR_F was measured in 36 *A. fulvescens* exposed to 100%, 90%, 80% or 70% O_2sat inside the respirometer. A total of 8–12 *A. fulvescens* were tested at each of the four O_2sat levels. Measurements of MMR_F in 100% O_2sat were repeated after 4.5 h to examine the short term repeatability of MMR_F in individual fish. These two measurements were termed initial and final MMR_F. Finally, for each test group (i.e. control and treatments A and B), MMR_S was estimated as the single highest measurement of *M*O₂ (i.e. one respirometric loop) in volitionally performing individual fish during the complete 24 h trial (i.e. after acclimation). These data were used to test for correlations between MMR_F and MMR_S in individual fish (see Data analysis). ## Behaviour *A. fulvescens* in the respirometers were recorded (25 frames s⁻¹) dorsally using a UEye camera (model UI-1640SE-C-GL; IDS, Woburn, MA, USA) equipped with a CCTV lens (model HF6M-2; Spacecom, Whittier, CA, USA). The software UEye Cockpit (version 3.90; IDS, Woburn, MA, USA) was used to download recordings to a PC. Two Scene illuminators (model S8030-30-C-IR; Guangdong, China) provided infra-red light for nocturnal recordings. All recordings were synchronized with the respirometric loops (to the nearest 1 s). For each *A. fulvescens*, behavioural data were collected over a 45 s time interval during the measurement phase of the respirometric loop (i.e. once every 9 min.). Behavioural data included total activity (i.e. % of time moving), and the number of body undulations with tail excursions\<90° or \>90° relative to the body axis (i.e. body undulations min⁻¹). For each test group, behavioural data were collected over a 1 h time interval (i.e. 6–7 respirometric loops) at 16, 20, 21, 22 and 23 h. These hourly measurements were selected to record simultaneous metabolic and behavioural changes during the light-dark transition at 21 h. ## Data analysis *M*O₂ data were body mass adjusted following previous studies. Metabolic rates from the three test groups were calculated over 1 h intervals, with two exceptions, because the light intensity was gradually changing over 0.5 h periods at 21 h and 9 h. Therefore, the two 1 h intervals associated with 21 h and 9 h were each divided into two: 0.5 h with changing light intensities and 0.5 h with constant light intensity. The compiled data were used to compare metabolic rates over 24 h within the three test groups (i.e. control and treatments A and B). Behavioural data were compiled in the same fashion. Metabolic and behavioural variables were compared within each test group across the time interval from 16:00 to 23:00 h using a repeated measure (RM) one way ANOVA. Relationships between behaviour and metabolic rates were investigated using least squares linear regression. To test for metabolic differences, SMR, MMR_F and MMR_S measurements were compared between the three test groups using a one way ANOVA. MMR_F data from the four oxygen treatments (100 – 70% O_2sat) were analyzed using least square linear regression to examine the effect of decreasing oxygen levels on MMR_F. The method recommended by Norin and Malte was used to examine repeatability of the MMR_F measurements. All values of MMR_F and body mass were log₁₀-transformed prior to the analysis. Mass-independent data of MMR_F were expressed as residual values using the relationship between body mass and MMR_F. Fish with higher than expected MMR_F have positive residuals and fish with lower than expected MMR_F have negative residuals. Repeatability of the two sets of residuals (initial and final) was estimated using Spearman's rank correlation coefficient (ρ). Using metabolic rate data from the three test groups, MMR_S of each individual fish was extracted to test for correlations between individual MMR_F and MMR_S. The comparison of individual MMR_F and MMR_S was carried out in the same fashion as the repeatability analysis described above. Log₁₀ transformations of data prior to statistical analysis were employed to meet assumptions of normal distribution of data and homogeneity of variance. If the assumptions were met, ANOVA or RM ANOVA were employed depending on design as described above. If significant, the tests were followed by pairwise multiple comparisons using the Holm-Sidak method. If data transformations did not permit the use of parametric testing, ANOVA on ranks or RM ANOVA on ranks (Friedman) were employed depending on design as described above. The tests were followed by pairwise multiple comparisons using Dunn's method to take unequal sample sizes into account. Tests were carried out using SigmaStat 3.01 (Systat Software, San Jose, CA, USA) and SPSS 20.0 (IBM, Armonk, NY, USA). Results were considered significant if α\<0.05. All values are reported as means ± S.E. unless noted otherwise. # Results For all the experiments, there were no indications that the health status of the test animals changed during any of the tests. ## Body mass adjustments There were no differences between test groups (i.e. control, treatments A and B) in terms of body mass and SMR measured as mg O₂ h⁻¹ (both P\>0.05). Consequently, SMR data were pooled, and the relationship between log₁₀ SMR and log₁₀ body mass was described using a linear equation. The slope of the relationship was 1.00±0.12 indicating that a 1.0 body mass scaling coefficient was appropriate for the SMR data. A 1.0 body mass scaling coefficient is consistent with two previous studies on green sturgeon *Acipenser medirostris*. Because the 1.0 body mass scaling coefficient was appropriate for the SMR data, the same coefficient was used for the *M*O₂ data collected over time. MMR_F measured as mg O₂ h⁻¹ did not differ between the control group and treatment B (P\>0.05), but MMR_F from treatment A was lower than both the control group and treatment B (P\<0.001). To examine the relationship between body mass and MMR_F (mg O₂ h⁻¹), data collected in normoxia were combined and the relationship between log₁₀ MMR_F and log₁₀ body mass was described using a linear equation. The slope of the relationship was 0.90±0.05 indicating that a 0.9 body mass scaling coefficient was appropriate for the MMR_F data. Consequently, all MMR_F data were standardized to the mean body mass of 30.5 g using 0.9 as the body mass scaling coefficient. In the following, MMR_F standardized to 30.5 g is denoted MMR_F30.5. ## Metabolic rates over 24 h Metabolic rates varied substantially over the 24 h periods. In the control group, metabolic rate increased significantly (P\<0.001) from 112 mg O₂ kg⁻¹ h⁻¹ at 20:00 h to reach a maximum of 237 mg O₂ kg⁻¹ h⁻¹ when the light went off (control), indicating a dusk metabolic peak. Thereafter, metabolic rate decreased and reached 157 mg O₂ kg⁻¹ h⁻¹ shortly before daylight. The metabolic rate decreased further in daylight and reached 112 mg O₂ kg⁻¹ h⁻¹ after 3 h. In treatment A, metabolic rate increased significantly (P\<0.05) from 116 to 150 mg O₂ kg⁻¹ h⁻¹ when the light went off (treatment A). Although truncated, this metabolic peak corresponded to the dusk metabolic peak observed in the control test group. Thereafter, metabolic rate decreased to 128 mg O₂ kg⁻¹ h⁻¹ at 03:00 h, and then increased to reach 139 mg O₂ kg⁻¹ h⁻¹ during the period with increasing light intensity (09:00 h). Thus, treatment A indicated two metabolic peaks; one associated with dusk and one associated with dawn. After the light went on, the metabolic rate changed little for 1 h and then decreased to 118 mg O₂ kg⁻¹ h⁻¹ (treatment A). In treatment B, the metabolic rate remained below 119 mg O₂ kg⁻¹ h⁻¹ until 02:00 h (treatment B). Data showed that the dusk metabolic peak, observed in the control group and in treatment A, was eliminated by the constant light (P = 0.64). In contrast, in treatment B, the metabolic rate tended to increase at 02:00 h and continued doing so until it reached 178 mg O₂ kg⁻¹ h⁻¹ at 08:00 h (treatment B). These data indicated the presence of a darkness independent increase in the metabolic rate. The increasing metabolic rate peaked around dawn, just before the light would normally come on. Collectively, data indicated the presence of two metabolic peaks occurring over 24 h. The first metabolic peak occurred around dusk and was noticeable in the control group and treatment A. The second metabolic peak occurred around dawn and was noticeable in treatments A and B. ## Behaviour across the light-dark transition Behavioural recordings from the control group indicated that the total activity increased in darkness (control), but no statistically significant differences were identified over time (P\>0.05). Similarly, the frequency of body undulations with tail excursions\<90° did not change significantly over time (P\>0.05). In contrast, body undulations with tail excursions\>90° increased significantly over time (P\<0.001) (control). Data from treatment A revealed no significant changes in the total activity over time or in the frequency of body undulations with tail excursions\<90° (both P\>0.05) (treatment A). In contrast, the frequency of body undulations with tail excursions\>90° increased significantly over time (P\<0.001). Data from treatment B revealed no significant changes over time in the total activity or in the frequencies of body undulations with tail excursions\<90° or \>90° (all P\>0.05;, treatment B). ## Correlations between behaviour and metabolic rate The behavioural data suggested that the frequency of body undulations with tail excursions\>90° could be a major driver of the increase in metabolic rate associated with dusk. Regression analysis revealed highly significant (P\<0.001 in all cases) linear relationships between the frequency of body undulations with tail excursions\>90° and metabolic rate. The coefficients of determination (r²) for the relationships varied between test groups and were 0.68, 0.64 and 0.15 for control and treatments A and B, respectively. These data suggest that metabolic variation was coupled with behavioural variation. ## Environmental effects on standard metabolic rate (SMR) and forced maximum metabolic rate (MMR_F) SMR was unaffected by hypoxia (treatment A) and constant light (treatment B) (P\>0.05;), and the pooled average was 92.39±2.00 mg O₂ kg⁻¹ h⁻¹. Corresponding analyses of MMR_F30.5 revealed no differences between the control and treatment B (P\>0.05), whereas MMR_F30.5 from treatment A was lower than both the control and treatment B (P\<0.001;). These findings showed that 30% O_2sat reduced MMR_F30.5. MMR_F30.5 was quantified in four separate groups of *A. fulvescens* exposed to 100%, 90%, 80% or 70% O_2sat to estimate the effect of hypoxia on MMR_F30.5. Body mass did not differ between the four treatments (P = 0.95). MMR_F30.5 was affected by hypoxia and decreased across the range from 100% O_2sat to 70% O_2sat as revealed by the linear regression analysis (P\<0.03; r²\>0.94). These findings indicated that the maximum metabolic rate of *A. fulvescens* is sensitive to increasing levels of hypoxia. ## Repeatability of forced maximum metabolic rates (MMR_F) Analysis of repeatability followed a previous study and showed that measurements of residual body mass corrected MMR_F are repeatable in individual fish. Spearman's rank correlation coefficient (ρ) for the relationship between the initial and final residual MMR_F was 0.76, and the relationship was highly significant (P\<0.006). ## Spontaneous maximum metabolic rate (MMR_S) MMR_S was extracted from each 24 h trial for comparisons between test groups. MMR_S differed significantly between all three test groups (P\<0.05). These findings showed that MMR_S was suppressed in treatments A and B, with the most pronounced effect in treatment A. Standardizing MMR_S to a 30.5 g fish using a 0.9 body mass scaling coefficient (i.e. equivalent to MMR_F30.5) changed MMR_S values by \< 1% and had no impact on the conclusions. ## Correlations between forced (MMR_F) and spontaneous (MMR_S) maximum metabolic rates MMR_F and MMR_S were compared to test the hypothesis that they would correlate positively. Data showed that residual MMR_F and residual MMR_S were not correlated in the control group (P = 0.40; ρ = 0.27) (control). In contrast, residual MMR_F and residual MMR_S were positively correlated in both treatments A (P\<0.05; ρ = 0.69) and B (P\<0.05; ρ = 0.66) (, treatments A and B). These data indicated that an individual with an unexpectedly high MMR_F also has an unexpectedly high MMR_S, at least when the individual is exposed to an environmental stressor, such as hypoxia (treatment A) or constant light (treatment B). # Discussion This study provides evidence that the organismal physiology of *A. fulvescens* is influenced by a circadian rhythm and strongly indicates that *A. fulvescens* is an oxygen regulator. Using residual (i.e. body mass corrected) values, the study suggests that MMR_F is repeatable in individual *A. fulvescens*, and MMR_F can be positively correlated with MMR_S. The relationship between MMR_F and MMR_S appears, however, to depend on the presence of an environmental stressor such as hypoxia or constant light. Our data indicated the presence of two metabolic peaks in *A. fulvescens* occurring over 24 h. The first metabolic peak occurred around dusk (control group and treatment A), whereas the second metabolic peak occurred around dawn (treatments A and B). The dusk metabolic peak was eliminated by the constant light in treatment B, suggesting that the dusk metabolic peak reflected an exogenous rhythm, depending on exogenous stimuli (i.e. decreasing light levels). In contrast, the dawn metabolic peak occurred regardless of constant light, suggesting that a circadian rhythm, including an endogenous mechanistic basis, control the metabolic rate of *A.* fulvescens. As far as is known, our study provides the first evidence of a circadian rhythm in Acipenserids. It is not clear why the dawn metabolic peak was not distinct in the control group. We suggest that the relatively high metabolic rates masked the dawn metabolic peak in the control group. In the hypoxic treatment, metabolic rates were suppressed, but not to an extent where the dawn metabolic peak was eliminated. Therefore, the metabolic suppression in hypoxia helped revealing the underlying presence of two metabolic peaks. In a recent field study, Forsythe et al. reported that adult *A. fulvescens* initiate upstream migration around dusk and dawn. The authors suggested that the observations could ultimately be explained by reduced risk of predation and harvest by humans at dusk and dawn. While the present study used juvenile *A. fulvescens*, our data indicate that the migratory peaks at dusk and dawn observed by Forsythe et al. could reflect proximate mechanisms that include an exogenous rhythm at dusk and a circadian rhythm at dawn. This study tested the hypothesis that *A. fulvescens* is an oxygen regulator. Our data provide two lines of evidence that *A. fulvescens* is an oxygen regulator, capable of regulating metabolic rate and maintaining metabolic rhythms in environmental hypoxia. Firstly, we found no evidence that SMR differed between normoxia and hypoxia (30% O_2sat). Thus, *A. fulvescens* maintained SMR regardless of fluctuating environmental oxygen levels. Secondly, *A. fulvescens* exposed to hypoxia (30% O_2sat) exhibited a similar metabolic rate rhythm over the time interval from 16 h to 23 h as *A. fulvescens* exposed to normoxia and was capable of increasing the metabolic rate around dusk in the hypoxic environment (treatment A). The metabolic increase had a strong behavioural component in both hypoxia and normoxia, and correlated positively with the frequency of body undulations with tail excursions\>90°. These data show that *A. fulvescens* is capable of regulating metabolic rate (SMR) and maintaining metabolic rhythms in hypoxia. Thus, *A. fulvescens* is an oxygen regulator, like most teleost fishes. In contrast to SMR, data indicated that MMR_F30.5 is sensitive to increasing levels of hypoxia in *A. fulvescens*. Physiologically, the result is expected because if a fish is exercising at MMR before the hypoxic exposure, compensatory mechanisms (e.g. increasing gill ventilation and cardiac output) are already utilized to support the elevated oxygen requirements and are unavailable to compensate for environmental hypoxia. The result is not, however, consistent with previous studies on teleost fish. Most previous studies have reported that the maximum metabolic rate in normoxia is maintained in low levels of hypoxia, typically down to approximately 80% O_2sat. The reason for the discrepancy between the present and previous studies remains unknown, but is it possible the maximum metabolic rate of *A. fulvescens* is more sensitive to low levels of hypoxia than in most teleost fishes. Further tests comparing Acipenserids and teleost fishes using identical equipment and experimental approaches are required to examine the discrepancy. Previous studies have demonstrated that SMR and MMR are repeatable physiological traits in a wide range of taxa. Repeatability (or temporal consistency) is important when ascribing certain properties to an individual animal on the basis of a single physiological measurement. Repeatability indexes the reliability of the protocol used to measure a trait and further sets a general upper limit to the intensity of selection that can be applied to the trait. If a trait is not repeatable over time, a single measure of the trait may not be representative of future physiological performance and it becomes unlikely that natural selection can act on the trait, i.e. separate the favoured from disfavoured individuals. Little is known about repeatability of traits in Acipenserids, but a recent behavioural study demonstrated that spawning times and locations are highly repeatable in mature *A. fulvescens*. To our knowledge, the present study provides the first estimate of physiological repeatability in Acipenserids. Our data suggest that body mass corrected measurements of MMR_F are repeatable in *A. fulvescens*, at least over short time intervals (4.5 h) and set the stage for studies examining repeatability over longer time intervals. Recently, it has been shown that not only SMR and MMR, but also routine metabolic rate (RMR) can be a repeatable trait in fish. Repeatability of RMR suggests that the spontaneous activity within a respirometer is not simply random bouts of movement over time, but rather, that individual fish exhibit consistent behavioural patterns when evaluated at different times. The present study tested whether body mass corrected values of MMR_F and MMR_S are positively correlated to examine whether an unexpectedly high value of MMR_F would indicate an unexpectedly high value of MMR_S. By demonstrating positive relationships between MMR_F and MMR_S in *A. fulvescens* exposed to an environmental stressor, the present study adds to the growing body of evidence indicating that variation in metabolism, as determined over time in a respirometer, is not random, but may reflect physiological or behavioural traits in individual animals. Measurements of physiological performance, including MMR_F and critical swimming speed (*U*_crit), are widely used whole-organism indicators of maximal performance, examined to better understand evolutionary and physiological ecology,. While maximal performance is crucial for a wide range of behaviours tightly connected to fitness, animals may not exercise at maximal intensity very often. Therefore, measurements of maximal performance could have more pronounced functional importance if maximal performance correlated positively with spontaneous performance, which is used more frequently. In particular, this is important because selection regimes may not only operate on a trait's maximal value, but alternatively on the spontaneous use of the trait (i.e. ecological performance). In the present study, we examined maximal forced and spontaneous performances by measuring MMR_F and MMR_S to test whether the two traits are correlated. Considering treatments A and B, data indicated that *A. fulvescens* exhibiting an unexpectedly high MMR_F also exhibit an unexpectedly high MMR_S. These data suggest that MMR_F may be indicative of MMR_S in individual *A. fulvescens*. Nevertheless, we only found relationships between MMR_F and MMR_S when fish were exposed to an environmental stressor (hypoxia or 24 h light), and no relationship when fish were exposed to normoxia and a normal light regime (12L∶12D). It remains unclear why we observed relationships between MMR_F and MMR_S when *A. fulvescens* were exposed to an environmental stressor, and no relationship without an environmental stressor. Our findings are, however, consistent with a recent review by Killen et al.. The authors described how environmental stressors, including hypoxia and light, may either reveal or mask relationships between behaviour and physiology. Because we found evidence of correlations between behaviour and metabolic rate, it is likely that MMR_S not only reflected a physiological trait, but also a behavioural trait. As such, our relationships between MMR_F and MMR_S could be considered relationships between physiology and behaviour that were revealed by environmental stressors, as suggested by Killen et al.. All our measurements of MMR_F were stressful for *A. fulvescens*, whereas the measurements of MMR_S were probably most stressful under hypoxia and constant light. Physiological stress is associated with increased concentrations of plasma cortisol in Acipenserids – with secondary responses involving metabolism. In the present study, MMR_S was suppressed in treatments A and B, and stress experienced by *A. fulvescens* under hypoxia and constant light could have influenced the relative distribution of phenotypes with regard to MMR_S, such that positive correlations between MMR_F and MMR_S were revealed in treatments A and B (see in Killen et al.). This remains speculation, however, and further studies of the coupling between behaviour and physiology in divergent environments are needed to evaluate the hypothesis. We thank Dr. S. Renault at the University of Manitoba for the light sensor. We thank Dr. G. R. Ultsch for advice concerning body mass adjustments of oxygen consumption rates. We thank T. Smith and the animal care staff at the University of Manitoba for assistance in the care and maintenance of experimental animals. We thank an anonymous reviewer and academic editor Dr. E. V. Thuesen for helpful and constructive comments on an earlier version of the manuscript. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: JCS JG WGA JAS DW ECE. Performed the experiments: JCS JG JAS. Analyzed the data: JCS. Contributed reagents/materials/analysis tools: JCS WGA DW ECE. Wrote the paper: JCS. Revised the manuscript critically for important intellectual content: JCS JG WGA JAS DW ECE.

# Introduction Gastroschisis (GS) is a congenital anomaly of the abdominal wall, which results in extrusion of abdominal viscera from the abdominal cavity. The prevalence of GS is increasing; currently reaching 4.9 per 10,000 live births. Although survival in GS exceeds 90%, some babies experience significant morbidity, which is largely determined by the severity of prenatal and postnatal bowel injury. It is still unclear how the structural changes correlate with malabsorption and small bowel dysmotility in GS and therefore with patient outcomes. The extensive injury of intestinal mucosa could be a major source of complications both during the surgery and during the postoperative period and may have a major impact on patient recovery. A biomarker capable of quantifying the condition of intestinal mucosa after closure of the abdominal wall and during the acute post-surgery period is needed in clinical practice. I-FABP (Intestinal fatty acid-binding protein) is present in the cytoplasm of mature enterocytes in the small and large intestine and is released as soon as the cell membrane integrity is compromised, thus reflecting the extent of gut damage. It is used as a biomarker of mucosal injury and other diseases affecting the intestine. Serum I-FABP correlates with the severity of villous atrophy in coeliac disease and with Crohn's disease activity. It is increased after abdominal surgery or infection and can be used as a biomarker of acute gut mucosa damage in necrotizing enterocolitis (NEC) or distinguish it from neonatal sepsis. The aim of this prospective study is to find out if I-FABP could be used as a biomarker for mucosal injury in neonates after the GS surgery and if it can predict the speed of their clinical recovery. # Materials and methods ## Population under study Thirty-two patients with GS were recruited from the patients admitted to the Neonatal Intensive Care Unit, Department of Pediatric Surgery of University Hospital Motol, Prague, between 2012 and 2015. Two patients were excluded from the study because they died within the first 48 hours of life without passing any urine. One died of fulminant septic shock and one died of multiorgan failure resulting from prenatally acquired massive intestinal necrosis. Their samples and follow up were therefore not complete for later analysis. Out of our study group, 26 were born in our center and 4 were born in peripheral hospitals and transferred to our department immediately after delivery. All patients were operated within first 6 hours of life regardless of the place of birth. Categorization of GS as simple or complex was based on the absence or presence of intestinal atresia, stenosis, perforation, necrosis or volvulus at birth, as suggested by Molik et al.. All of our complex GS patients had small intestinal atresia, in 2 patients was intestinal atresia found during second revision, so these two individuals were reclassified as complex GS. All subjects of this study were operated under general anesthesia in the operating theatre. A primary operative abdominal wall repair was attempted in all our patients. Primary closure was performed by an interrupted absorbable suture with attention to preserving the umbilicus. Gore-Tex silo (1mm Dual Gore- Tex patch) was used when primary closure was not possible. The decision was made by attending surgeon when viscera could not be primarily reduced without danger of impeding venous return, increase of peak inspiratory pressures and increase of abdominal pressure. The final closure and silo removal was performed at 18 (12–29) (median (range)) day of life. Total parenteral nutrition was started within 24 hours of life via central venous catheter and performed in compliance with the current recommendation. The nutritional regimen for all patients included mixture of amino acids (Primene 10%, Baxter Czech, Prague, Czech Republic), fat (Smoflipid 20%, Fresenius Kabi AB, Uppsala, Sweden) and dextrose, supplemented with the vitamins and trace- elements (Soluvit ^® N, Vitalipid^® N Infant, Peditrace^®, Fresenius Kabi AB, Uppsala, Sweden). Enteral feeding was introduced once postoperative ileus had been resolved and the decision was made by attending neonatologist. Neonates received mother’s milk or bank milk continuously, via a nasogastric tube. Feeding was increased gradually according to tolerance and measurement of gastric residua. After initiation of feeding we followed minimal enteral feeding protocol as follows. Breast milk was given via nasogastric tube continuously, starting at 0.3–0.5 ml/kg/h. If it was well tolerated, the feeding was increased by 7–12 ml/kg daily. If neonate didn’t tolerate the daily increase, the dose was restored to the last tolerated amount or stopped entirely. To exclude the confounding factor of surgery, we established a different control group for either type of GS—newborns who underwent surgery without intestinal mucosa disruption served as controls for simple GS. All of our complex GS patients had small intestinal atresia, so control group for complex GS sustained from patients operated for small intestinal atresia, who underwent as well as complex GS patients resection of atretic part of the intestine and then end-to end anastomosis. There were 12 newborns in the control group for simple GS and 6 newborns in the control group for complex GS. To analyze the capacity of urinary I-FABP to predict the speed of the postoperative recovery, we used clinical data from patients’ health records and chose time to minimal enteral feeding of 20 ml/kg/day (MEF), time to full enteral feeding (FEF) and length of hospitalization (LOH). The median number of days was a cut off. Therefore, we defined "early" MEF as less than 13 days, "early" FEF as less than 19 days and "short" LOH as less than 31 days. The detailed demographics of studied population is summarized in. The study was approved by the Ethics committee of the University Hospital Motol, and written informed consent was obtained from parents of all children included in this study. ## Sample collection and processing Urine samples were collected for 48 hours in 6 hours’ intervals, starting during the first 6h after the surgery. Samples were collected from urine bag connected to an indwelling catheter. The urinary creatinine was measured in each sample in our hospital biochemistry laboratory immediately after sampling and samples for I-FABP analysis were frozen at -20°C until the analysis. ## Enzyme-Linked Immunosorbent Assay (ELISA) Urinary I‐FABP was measured by Human I-FABP ELISA (Hycult Biotech, Uden, The Netherlands). The assay was performed according to the manufacturer’s instruction. To eliminate fluctuation in urine excretion, the urinary I-FABP was normalized to urinary creatinine and is presented as pg/nmol of creatinine. ## Statistical analysis The continuous variables were tested for normality by D'Agostino-Pearson normality test. The differences between GS and control group were analyzed either by Mann-Whitney test (continuous variables) or by Fisher's exact test (dichotomous variables). I‐FABP levels in complex and simple GS were compared with each other and with these in controls using Kruskal-Wallis test with Dunn's post test. I-FABP was correlated with the time to MEF, FEF and with LOH using Spearman's rank correlation coefficient (r). Continuous variables are presented as median (range), dichotomous variables as the number of cases (percentages), and p values \< 0.05 were considered statistically significant. Correlation matrix was created using Spearman's rank correlation coefficient (r), and p values were divided by the number of analyzed values to prevent the type I error (false positivity). Predictive value of the I-FABP for the MEF, FEF and LOH was analyzed using receiver operating characteristic (ROC) curves. Statistical analyses were performed using GraphPad Prism version 6.0 (GraphPad Software, San Diego, CA, USA). Ordinary least squares (OLS) regression was used to model the relationship between MEF, FEF and LOH and I-FABP levels at different time points. Regression analysis was performed in R (ver. 3.3.2; R Foundation for Statistical Computing, Vienna, Austria) and Akaike information criterion (AIC) was determined in the MASS package (ver. 7.3–45). Next, we performed both backward elimination and forward selection based on AIC to determine the best regression model. # Results Simple and complex GS patients have higher urinary I-FABP after the surgery than control subjects. The I-FABP dynamics in simple and complex GS don’t differ. In both cases, I-FABP levels reach maximum in the first 6h after surgery (9.29 (0.59–58.56) pg/nmol for simple GS, 23.99 (16.59–41.90) pg/nmol for complex GS). Interestingly, I-FABP peaks 36 hours after the surgery in both patients with complex GS (15.57 (28.73–8.03) pg/nmol) and in controls for complex GS (4.20 (0.31–10.89) pg/nmol). The level of urinary I-FABP in controls for simple GS is generally low and without a distinct peak; in first 6 hours reaches 1.15 (0.06–3.41) pg/nmol. In the first 48h after the surgery, the levels of I-FABP in patients with complex GS are higher than in patients with simple GS. We did not find significant differences in I-FABP levels between patients treated with primary closure (PC) and stepwise reconstruction (SR), but I-FABP in both these groups was significantly higher than in controls. Urinary I-FABP during the first 6 hours after surgery is significantly higher in complex GS patients who will be later operated for mechanical ileus than in those operated only for silo removal. There is a clear difference in the outcome between complex and simple GS. Patients with complex GS have received FEF significantly later (median 59 vs. 17 days, p = 0.0055) and have significantly longer LOH than patients with simple GS (median 72 vs. 26 days, p = 0.0120). We used a linear regression model to analyze at which time point the urinary I-FABP has the highest capacity as a predictive biomarker for clinical outcome (MEF, FEF and LOH). Due to the clear differences in I-FABP dynamics, we included the changes in I-FABP as additional variables. None of the I-FABP levels measured at 6-hour intervals was a suitable predictor for the outcomes. However, three changes in I-FABP levels in time (decrease between 12 and 18 hours, decrease between 12 and 24 hours and increase between 24 and 30 hours) were found to be exceptionally good predictors. All these models provide an equivalent fit for LOH and FEF (Δ AIC \< 3), but 12h-18h is equivalent to 12-24h (Δ AIC = 3) and gives slightly better fit than 24-30h (Δ AIC = 6). This was, however, not supported by ROC curve analysis, in which the decrease in I-FABP between 12-18h was not capable of predicting early MEF/FEF or short LOH, as defined for our dataset. The levels of I-FABP at the time of surgery do not predict multiple surgeries, as analyzed by ROC curve analysis. To analyze the correlation of individual values and demographic characteristics of GS patients, we constructed a correlation matrix. We found a clear correlation among the levels of I-FABP at different time points, among all three recovery characteristics and between gestation length and birth weight. Another well-established connection is that patients with multiple surgeries are less likely to be operated with primary closure and have longer LOH. The raw data with appropriate metadata are in. # Discussion Disruption of intestinal mucosa causes major complications in patients with GS. The extent of this injury and its capacity to predict patient’s recovery has not yet been sufficiently analyzed. Our study showed that I-FABP can serve as a biomarker for the gut mucosa damage after the closure of abdominal wall in GS. I-FABP is a small cytoplasmic protein localized in epithelial cells of the small intestine, which is released into the circulation after enterocyte damage and quickly passes into urine. Therefore, urinary I-FABP could be used as a non- invasive biomarker of acute gut mucosa damage in spontaneous and surgery-related necrotizing enterocolitis \[, , –\]. Since gut mucosa damage is a typical pathological feature of GS, we studied if urinary I-FABP could be also used as a biomarker to predict a patient's outcome. We found that I-FABP is higher in patients with complex GS as compared to simple GS, which is consistent with significantly more severe mucosal damage in complex GS. However, traumatization to the gut during intestinal incision leads to a quick increase in plasmatic I-FABP as well, so it is unclear if this is an effect of more extensive damage during complex GS or just an effect of extensive surgery. All complex GS patients in this study had intestinal atresia, so they underwent intestinal wall incision, intestinal resection and intestinal wall suture. To control for the confounding factor of surgery and general anesthesia, we established a different control group for each type of GS—newborns who underwent surgery without intestinal mucosa disruption served as controls for simple GS and those operated for intestinal atresia with the same surgical technique were selected as controls for complex GS. These control groups were well matched for all characteristics except for those inherently associated with GS itself (gestational age and birth weight) and GS severity (hospital stay). We were not able to find any relevant information on the effect of gestational age and birth weight on I-FABP in the literature. None of these factors, however, correlated with I-FABP levels, so they do not seem to be crucial. Nevertheless, overall immaturity may influence the recovery regardless of gut damage, thus presenting a potential confounding factor. We found that while surgical damage to the gut mucosa increases the urinary I-FABP, its levels are significantly higher in both types of GS when compared to the relevant control group. Our results also showed that anesthesia, cold stress, volume therapy during surgery and general postoperative care do not influence I-FABP. Moreover, the I-FABP levels in controls for simple GS are not markedly higher than those in generally healthy newborns. These results clearly show that damage to the gut mucosa in GS is not just a result of mucosal integrity disruption during surgery. In this study, both ways of abdominal wall closure (i.e. primary closure and stepwise reconstruction) led to similar levels of I-FABP, suggesting that neither approach leads to more severe damage to the gut mucosa. In a recent meta-analysis, Kunz et al. compared short term outcomes of primary closure (PC) versus stepwise reconstruction (SR), finding that SR is associated with improved outcomes only if the method is selected randomly. Conversely, when the method is selected by the surgeon, PC is associated with improved outcomes. This is probably caused by the fact that patients receiving silo are also more likely to be prone to worse clinical outcomes. We found that while I-FABP quickly decreases in GS after the surgery, in patients with complex GS and controls with intestinal atresia, it increases again with a distinct peak at 30–36 hours after the surgery. This suggests that this delayed release of I-FABP after the small intestine surgery is either caused by protracted stricture of the circulation, which combines higher intestinal damage and delayed I-FABP release, or that it is a result of re- perfusion damage to the intestine. This is in agreement with studies on animal models of GS, showing that increased intra-abdominal pressure leads to gut mucosa damage, possibly via oxidative stress and an increase in apoptotic activity of enterocytes. Interestingly, I-FABP is significantly higher in neonates with complex GS that were later operated for mechanical ileus compared to those operated just for silo removal, which further supports the hypothesis about continuous gut damage. These data, however, need to be interpreted with caution, because the number of neonates in both subgroups is low. We found that the patient's recovery (MEF, FEF and LOH) is significantly faster in patients with simple GS than in those with complex GS. Several markers have been used to determine the patient outcome after GS surgery. Most of them focus on the sonographic findings on the fetal gut during prenatal examination. Dilated stomach of the fetus with gastroschisis is associated with higher neonatal death rate, volvulus, delayed enteral feeding and longer hospital stay postnatally. Intraabdominal bowel dilation of multiple intestinal loops predicts not only earlier delivery, but also postnatal bowel complications in neonates with gastroschisis. Not only the extent of the intraabdominal bowel dilation, but also its early appearance is associated with poor prognosis. Prenatal bowel dilatation is associated with increased morbidity in patients with simple GS. Measurement of the intraabdominal pressure (IAP) during surgery for gastroschisis may help to select optimal surgical technique and shorten the hospital stay. Since we have just started with routine measurements of IAP during this study, we do not have data to correlate this potential biomarker with I-FABP levels. We hypothesized that disruption of intestinal mucosal layer could be a major cause of complications in patients with GS and that the higher degree of gut mucosa damage predisposes to slower post-operative recovery. Since there are no guidelines for the division into early MEF/FEF and short LOH, we decided to use an unbiased approach and divide our dataset into two halves. Unfortunately, none of the measured levels of I-FABP could distinguish between these categories, although the decrease in I-FABP between 12 and 18h (or 12 and 24h) post-surgery explained well the variation of all three measured outcomes in regression analysis. We found that urinary I-FABP during the first 6 hours after surgery was significantly higher in complex GS patients who would be later operated for mechanical ileus than in those operated only for silo removal. There were, however, not enough patients with complex GS re-operated for ileus in our dataset to allow appropriate statistical analysis. The main advantage of single- center studies of rare diseases is the absence of variation between centers, but it comes at a cost of a small study population. This means that the statistical analysis is underpowered and notable to properly assess some important facets of the disease. This is the main reason why we could not compare individual surgery protocols and why we could define only the broadest disease stages (simple GS vs. complex GS) and basic procedures (PC vs. SR). # Conclusions Urinary I-FABP is a marker for intestinal mucosa damage in GS. Patients with complex GS have significantly higher levels of I-FABP and their recovery takes longer than in patients with simple GS. I-FABP fails to predict early MEF/FEF or shorter LOH, so it is not suitable for prediction of these parameters in clinical settings. Its capacity to predict subsequent operation for ileus in patients with complex GS needs to be interpreted with caution until a larger cohort of these patients is analyzed. # Supporting information We thank Kristina Kverková (Faculty of Science, Charles University in Prague) for regression analysis. GS Gastroschisis I-FABP Intestinal fatty acid-binding protein MEF Minimal enteral feeding FEF Full enteral feeding LOH Length of hospitalization PC Primary closure SR Stepwise reconstruction [^1]: The authors have declared that no competing interests exist.

# Introduction Prostate cancer (PCa) develops in the unique gland of the male reproductive system and becomes the most common malignancy in men, which leads to a detriment to men’s health. In 2015, PCa was ranked the second most frequently diagnosed cancer in males worldwide and the fifth leading cause of cancer deaths in the world. In the United States, PCa alone accounts for almost 1 in 5 new dignoses of the most common cancers expected to occur in men, and accounts for 8% of all cancer deaths in men which is the third leading cause of cancer-related deaths for men. Besides, in China, it was estimated that the incidence of prostate cancer was ranked sixth and the mortality of prostate cancer was ranked seventh in men. Currently, there are several effective treatments including surgery, androgen ablation and radiation therapy for hormone dependent PCa, but the curative effect to hormone independent cases is unsatisfactory. Though widely studied, the precise mechanism of PCa has not yet been fully clarified and further investigation is needed. Hepatoma-derived growth factor (HDGF), an acidic heparin-binding growth factor, was originally purified from the conditioned medium of a human hepatoma cell line, Huh-7. HDGF is widely expressed in several embryonic tissues including brain, kidney, liver, and cardiovascular system, which is closely associated with cellular processes including proliferation, differentiation and migration of these fetal cell types. Previous studies have shown that HDGF is highly expressed in numerous cancer tissues and characterized by a novel prognostic factor, such as oral cancer, hepatoma, lung cancer, gallbladder cancer, pancreatic cancer, endometrial carcinoma, and gastric carcinoma esophageal carcinoma, colorectal cancer, and cholangiocarcinoma. Moreover, increasing investigations revealed that HDGF is significantly correlated with the malignant biological potential of a variety of cancer cells including PCa cells. However, the signaling pathway and molecular mechanisms are largely unknown. Epithelial-mesenchymal transition (EMT), defined as that tumor cells lose the epithelial morphology and acquire mesenchymal characteristics during carcinogenic progression, participates in increased cell migration and invasion, and induces the initial stage of metastatic progression during the development and progression of malignant tumors. Besides, matrix metalloproteinases (MMPs) including MMP2 and MMP9 also contribute to the invasive and metastatic phenotypes of a variety of cancer cells by degrading the extracellular matrix and other barriers. However, it is unknown whether HDGF regulates cell migration and invasion via EMT process or the expression of MMPs in PCa. Therefore, we investigate the role of HDGF knockdown in the regulation of migration and invasion and EMT, as well as MMP2, MMP9 in PCa cells. # Materials and methods ## Cell culture Human PCa cells(DU145, PC3 and LNCaP cells) were obtained from the Shanghai Institutes for Biological Sciences in China. DU145 and LNCaP cells were maintained in RPMI-1640 medium containing 10% fetal bovine serum (FBS) purchased from Hyclone, and PC3 cells were cultured in F12K medium containing 10% FBS. All cells were routinely cultured in 5% CO2 at 37°C. ## Lentivirus-mediated shRNA interference The recombinant lentivirus short hairpin RNA (shRNA) targeting HDGF sequence (shRNA-HDGF) and control shRNA (shRNA-CTR) were purchased from Bio-Link (Shanghai, China). The target sequence of shRNA-HDGF as described by Jun *et al*. was as follows: `5′–AACCGGCAGAAGGAGTACAAA–3′`. The scrambled sequence of shRNA-CTR was as follows: `5′–TTCTCCGAACGTGTCACGT–3′`. Cells treated by shRNA- CTR or PBS were used as controls. In vitro transfection were done following the manufacturer’s protocols. ## Quantitative real time PCR Total RNAs were extracted from the cells using TRIzol (Invitrogen, Carlsbad, CA, USA) according to the manufacturer’s instructions. For the detection of HDGF mRNA levels, reverse transcription and quantitative real time PCR (qRT-PCR) was performed on Prism 7500 (ABI, Foster City, CA, USA) by using standard SYBR green assay protocol. Primers for HDGF is `5′- CTCTTCCCTTACGAGGAATCCA -3′` (forward) and `5′- CCTTGACAGTAGGGTTGTTCTC -3′` (reverse). Primers for β-actin used as normalization control is `5′- CATGTACGTTGCTATCCAGGC -3′` (forward) and `5′- CTCCTTAATGTCACGCACGAT -3′` (reverse). The relative level of HDGF mRNA was calculated and normalized using the 2^-ΔΔCt method relative to β-actin. ## Cell scratch assay All cells were plated in 6-well plates and incubated until 70% - 80% confluent. Vertical scratches were then made using a 200μl plastic filter tip to create a ‘wound’. To eliminate dislodged cells, culture medium was removed and wells were washed with PBS. ‘Wound closure’ was observed at 0, 12, 24 hours and digital images were taken under an inverted microscope. The wound area was measured using Image J software to calculate fold change of area coverage in the wells. ## Cell migration assay Transwell chamber was used to examine cell migration ability. Cells (1.0×10⁵ cells /well) were placed into the upper chambers (BD Bioscience, San Jose, CA, USA) containing 10% FBS. The lower chambers were filled with RPMI-1640 or F12K with 20% FBS. After incubation for 12 h for DU145 cells, 24 h for PC3 cells, and 48 h for LNCaP cells at 37°C, cells on the lower surface of the membranes were fixed and stained with 0.1% crystal violet, and five random fields of vision were selected to be counted under a light microscope (Olympus). Each experiment was performed in triplicate. ## Cell invasion assay The invasion assay was similar with the migration, except for that the upper chamber was precoated with matrigel (BD Bioscience, San Jose, CA, USA). In addition, the incubation time for DU145 cells, PC3 cells and LNCaP cells is 24 h, 48 h, and 72 h, respectively. ## Western blotting Cells were lysed in RIPA buffer (Beyotime, Shanghai, China). Protein concentration was quantified with BCA Protein Quantitative Kit (Beyotime, Shanghai, China). Equal amount of protein (20ug) was separated on 10% SDS-PAGE and transferred to PVDF membranes (Millipore, Billerica, MA, USA), which were blocked with 5% non-fat milk for 1 hour and incubated overnight at 4°C with rabbit anti-human primary antibodies against HDGF, E-cadherin, N-cadherin, Snail, Slug, MMP2, MMP9 and β-actin (all 1:1000 dilution) followed by incubation with HRP-conjugated goat anti-rabbit secondary antibody (1:1000 dilution) for 1 hour and detected by ECL. ## Statistical analysis Data were expressed as the mean ±SD from at least three independent experiments. The Student’s t-test was used to compare the differences between two groups with SPSS 22.0. *P* \< 0.05 was considered statistically significant. # Results ## Relative HDGF mRNA levels were inhibited by lentivirus-mediated shRNA interference in DU145, PC3 and LNCaP cells DU145, PC3 and LNCaP cells were transfected by lentivirus shRNA-HDGF or shRNA- CTR. To investigate the transfection efficiency, primarily we examined the expression of Green Fluorescence Protein (GFP). The transfection efficiency of \>80% was determined by detecting the expression of GFP 24 h after transfection at a MOI of 50 for DU145 and PC3 cells, and 36 h after transfection at a MOI of 20 for LNCaP cells. Then the cells were cultured in their medium with puromycin to select cells transfected stably for later HDGF knockdown experiments. Further, we analyzed the HDGF mRNA levels in the selected cells by quantitative real time PCR analysis. Results of this analysis showed that the relative levels of HDGF mRNA were suppressed in DU145, PC3 and LNCaP cells treated by lentivirus shRNA-HDGF compared with these cells treated by lentivirus shRNA-CTR or PBS. ## Expression of HDGF was inhibited by lentivirus-mediated shRNA interference in DU145, PC3 and LNCaP cells Protein is significantly correlated with the function of cells. To investigate whether the expression of HDGF was knocked down successfully, we examined the protein level of HDGF by western blotting analysis. Results of this analysis suggested that the relative protein expression levels of HDGF were significantly downregulated in DU145, PC3 and LNCaP cells treated by lentivirus shRNA-HDGF compared with these cells treated by lentivirus shRNA-CTR or PBS. ## HDGF downregulation reduces PCa cell invasion and migration in vitro Since tumorigenesis is involved of multiple oncogenic processes including migration and invasion, we examined whether modulation of HDGF expression affected these processes. The effects of HDGF on the migration of PCa cells were assessed by scratch assay and migration assay, while the effect of HDGF on the invasion of PCa cells was assessed by Matrigel invasion assay. The results of scratch assay demonstrated that compared with lentivirus shRNA-CTR transfected cells, HDGF knockdown significantly reduced the area of migration of DU145 and PC3 cells respectively. Besides, The results of migration assay demonstrated that HDGF knockdown significantly reduced the number of migrated DU145, PC3 and LNCaP cells compared with lentivirus shRNA-CTR transfected cells or PBS treated cells. Meanwhile, The results of the Matrigel invasion assay indicated that inhibition of HDGF reduced the number of invaded DU145, PC3 and LNCaP cells compared with lentivirus shRNA-CTR transfected cells or PBS treated cells. These findings provided evidence that inhibited HDGF expression levels were involved in suppressing the migratory and invasive ability of PCa cells. ## Epithelial-mesenchymal transition (EMT)-associated proteins in PCa cells To further investigate the mechanism by which HDGF regulates cell invasion and migration, the protein expression levels of EMT-associated genes in PCa cells was analyzed. It was demonstrated that HDGF knockdown decreased the expression of the mesenchymal markers Vimentin and N-cadherin, and the EMT central regulators Snail and Slug compared with shRNA-CTR transfected cells or PBS- treated cells, while the expression of epithelial marker E-cadherin was upregulated in PCa cell lines DU145, PC3 and LNCaP following HDGF knockdown. Thus, the results of the current study demonstrated that knockdown of HDGF altered the expression levels of EMT-associated proteins, therefore, suppressing the invasive and migratory phenotype of PCa cells. ## Effects of HDGF knockdown on the expression of MMP2 and MMP9 As the invasive ability of tumor cells is often correlated with the production of secretory proteases, the effect of HDGF knockdown on the expression of tumor invasion-associated secretory proteases MMPs including MMP2 and MMP9 were determined. Western blotting analysis indicated that downregulation of HDGF reduced the protein expression levels of MMP2 and MMP9 in PCa cells DU145, PC3 and LNCaP. # Discussion HDGF, a member of the hepatoma-derived growth factor family, is located on chromosome 1, region q21-q23 and encodes 240 amino acids, which contains two putative bipartite nuclear localization sequences (NLSs) for nuclear localization and an important PWWP domain in the N-terminal region for binding DNA. HDGF is a multifunctional protein which involves in the many cellular events, such as RNA processing, DNA damage repair, ribosome biogenesis, and transcriptional regulation. Previous studies reveals that HDGF promotes the proliferation of several types of normal cells such as fibroblast, hepatocyte, endothelial cells, vascular smooth muscle cells, and takes part in many cellular physiological processes, such as lung remodeling, sensitization to irradiation, neurotrophic effects, vascular injury, and cardiovascular differentiation. As a mitogen, HDGF is highly expressed in the early developmental stage of several organs and decreases rapidly after birth, however, several findings recently suggested that HDGF is highly expressed in a variety of cancers, which seems to suggest that HDGF plays an important role in the tumorigenesis and cancer progression. It has shown that HDGF stimulates the growth of several types of cancer cells such as HuH-7 hepatoma cells, gastric cancer cells and melanoma. Moreover, HDGF involves in the aggressive biological properties, including invasiveness, angiogenesis and metastasis of cancer cells. In PCa, HDGF is overexpressed in DU145, PC3 and LNCaP cells compared with the normal prostate cell RWPE-1 and down-regulation of HDGF expression inhibits the proliferation of PCa DU145, PC3 and LNCaP cells, and the migration and invasion of DU145 cells. Moreover, up-regulation of HDGF expression in normal RWPE-1 prostate cell facilitates the migration and invasion of RWPE-1 cells. However, the underlying molecular mechanism associated with HDGF-mediated cell migration and invasion in PCa remains elusive. Since tumorigenesis is related with multiple oncogenic processes including proliferation, anchorage-independent growth, migration and invasion, and our recent study has demonstrated that down-regulation of HDGF expression inhibits proliferation of DU145, PC3 and LNCaP cells, and migration and invasion of DU145 cells, we investigated the effects of HDGF knockdown on the migration and invasion of DU145, PC3 and LNCaP cells. As a result, the scratch assay demonstrated HDGF knockdown significantly reduced the area of migration of DU145 and PC3 cells, and the migration assay demonstrated that HDGF knockdown significantly reduced the number of migrated DU145, PC3 and LNCaP cells, and the Matrigel invasion assay indicated that inhibition of HDGF reduced the number of invaded DU145, PC3 and LNCaP cells. These findings revealed that inhibited HDGF expression levels were involved in suppressing the migratory and invasive ability of PCa cells. However, it is unknown how HDGF affect cell migration and invasion in PCa. Our recent study demonstrated that cancer cell migration and invasion is partly dependent on EMT process and matrix MMPs such as MMP2 and MMP9. Moreover, HDGF, a secreted growth factor, is correlated with EMT process in breast cancer cells. Usually, EMT process is regulated through the transcription repressors Snail, Slug, Twist and ZEB1, and initiated by the down-regulation of epithelial genes such as E-cadherin, and the up-regulation of mesenchymal genes such as Vimentin and N-cadherin. Previous study shown that EMT is involved in increased cancer cell invasion and facilitates the initial stage of metastatic progression, because they lose the epithelial morphology characterized by down-regulation of E-cadherin expression and acquire the mesenchymal characteristics characterized by up-regulation of Vimentin and N-cadherin expression during carcinogenic progression. In addition, extracellular matrix degradation is also known to be a major step during cancer progression. Previous studies have shown that MMPs including MMP2 and MMP9, which are families of zinc and calcium dependent endopeptidases, contribute to extracellular matrix remodeling. It has revealed that MMP2 and MMP9 are correlated with the invasive and metastatic phenotypes of prostate cancer cells. However, whether HDGF participates in the progression of PCa mediated by EMT and MMPs is largely unknown.Previous studies have shown that the nuclear targeting of HDGF is essential for its translocation to the nucleus and functioning as a direct DNA binding protein to regulate gene transcriptions. In the present study, after examining the inhibiting effect of HDGF knockdown on migration and invasion of Pca cells, we then investigated the molecular events associated with EMT and the expression level of MMPs, mainly including MMP2 and MMP9 after stable HDGF knockdown in PCa cells DU145, PC3 and LNCaP. Consequently, Western blot showed that the expression of transcription repressors Snail and Slug, as well as the mesenchymal markers Vimentin and N-cadherin, were all downregulated (P \< 0.05), and the epithelial cell marker E-cadherin were upregulated (P \< 0.05). This EMT process was partly verified in breast cancer cells with counterevidence, in which HDGF overexpression stimulates the cell invasion and metastasis by decreasing E-cadherin expression and increasing Vimentin expression. In addition, the expression level of MMP2 and MMP9 in PCa cells were both downregulated (P \< 0.05) after stable HDGF knockdown. All these evidence suggested that HDGF downregulated stably by targeting shRNA could inhibit the migration, invasion and EMT process, as well as the expression level of MMP9 and MMP2 in PCa cells. In EMT process, HDGF knockdown may suppressed the expression of transcription repressors Snail and Slug, and further stimulates the expression of E-cadherin and inhibits the expression of Vimentin and N-cadherin, which may result in the inhibition of PCa cell migration and invasion. Moreover, the inhibition of PCa cell migration and invasion may be also partly caused by the down-expression of MMP9 and MMP2 after HDGF knockdown stably. In conclusion, the downregulation of HDGF significantly is associated with the inhibition of the migration and invasion in prostate cancer cells, at least in part, by suppressing EMT process and the expression of MMP9 and MMP2, which play an important role in the tumorigenesis of PCa. Therefore, HDGF and its downstream molecular events, such as EMT process and the expression of MMP2 and MMP9 may serve as a novel therapeutic target for PCa. # Supporting information This study was supported by financial grants from the National Natural Science Foundation of China (grant Nos: 81502213 and 81372335) (<https://isisn.nsfc.gov.cn/egrantweb/>), the Natural Science Foundation of Shandong Province (grant No: 2015ZRE27283) (<http://jihlx.sdstc.gov.cn/STDPMS/ZR/Default.aspx>), and the Focused Research and Development Program of Shandong Province (grant Nos: 2016GSF201171 and 2017GSF18130) (<http://jihlx.sdstc.gov.cn/STDPMS/GG/Default.aspx>). [^1]: The authors have declared that no competing interests exist.

# Introduction Detection and monitoring of retinal ischemia have the potential to play an important role in the management and prognosis of patients with diabetic retinopathy and a number of other vision-threatening pathologies, including retinal vein occlusion and sickle cell retinopathy. Prior to any visible vascular changes, diabetic retinopathy patients develop thinning of specific retinal layers as seen on spectral domain optical coherence tomography (OCT) images. Later, vascular changes are evident and histopathologic studies show microvascular occlusions in retinal capillaries. Further, irreversible damage and occlusion of macular capillaries has been shown to cause their dropout, leading to ischemia. Ischemia in the macula is especially concerning for vision loss. A cross-sectional study in diabetic patients found higher rates of diabetic macular ischemia (DMI) in more severe stages of diabetic retinopathy. Reduced visual acuity has been seen in eyes with moderate to severe DMI. Two imaging modalities are currently used to visualize blood flow in retinal vessels: fluorescein angiography (FA) and OCT angiography (OCT-A). Quantifying retinal vascular changes using these modalities is important to reproducibly compare longitudinal changes in the retinal vasculature of patients, to evaluate the impact of treatments on retinal vasculature, and to improve our understanding of the relationship between retinal vasculature and clinical outcomes, such as visual acuity and macular edema. Studies using quantitative vasculature data from OCT-A have found a positive correlation between macular vessel density and visual acuity in diabetic patients. Studies are equivocal on the correlation between vessel density and macular thickness. A study of eyes with diabetic macular edema, found that lower macular vessel density correlated with greater macular thickness. Another study found that in diabetic patients with no diabetic retinopathy (DR) or mild non-proliferative diabetic retinopathy (NPDR), the loss of vessel density was associated with macular ganglion cell layer thinning on OCT. Similarly, another study in diabetic patients with and without retinopathy found a positive correlation between vessel density and central retinal thickness (CRT). In clinical applications, as compared to OCT images, ultra-widefield fluorescein angiography (UWFFA) images are currently assessed qualitatively. For research studies, the quantification of UWFFA has been focused on manually measuring the absence of retinal vessels or retinal non-perfusion and calculating an ischemic or nonperfusion index. However, limitations of this labor-intensive approach include the time required for manual measurement and the impact of contrast on accurate vessel detection. Despite these limitations, one study using this approach has found ischemic index in the mid-periphery was negatively associated with CRT and macular volume in patients with diabetic macular edema (DME). Currently, no widely accepted automated techniques exist to quantify retinal vascular changes in UWFFA similar to those employed in OCT-A. Such an automated technique for UWFFA would provide the ability to study longitudinal changes in vessels and relationships between vessel density and clinical endpoints, such as visual acuity and CRT. Our group has worked to address this unmet need by developing an automated algorithm that can detect retinal vasculature on UWFFA using a deep learning approach. In this study, we utilized our automated algorithm to examine the relationship between vessel density, estimated via automated analysis, and best corrected visual acuity (BCVA) and CRT at the baseline visit of patients enrolled in the RECOVERY trial. This trial is studying the effect of intravitreal aflibercept on retinal capillary non-perfusion in patients with proliferative diabetic retinopathy without macular edema. Our overall goal is to investigate how quantifiable vascular changes from UWFFA images can be used to predict clinically meaningful endpoints. For this specific study, we focused on evaluating only the macular vessel density to determine if we can reproduce relationships found in OCT-A studies between macular vessel density and BCVA and CRT, using our automated tools for UWFFA. # Methods ## Study subjects and data Patients with proliferative diabetic retinopathy without center-involving macular edema (defined as central subfield thickness \>320 μm) who were enrolled in the RECOVERY trial were included in our study. The RECOVERY trial assessed the safety and efficacy of intravitreal aflibercept injections given either every 4 weeks or every 12 weeks for the treatment of retinal capillary non- perfusion. Primary objectives of the RECOVERY study were to measure treatment- related adverse events and change in non-perfusion. Secondary outcomes included assessments of changes in visual acuity, changes in area of macular capillary non-perfusion as seen on UWFFA images, and changes in area of retinal capillary non-perfusion outside of the macula. For the current study, patients with complete baseline UWFFA images and visual acuities were included (n = 42). Sterling Institutional Review Board/Ethics Committee approval was obtained for the enrollment, study, and treatment of patients in this Health Insurance Portability and Accountability Act (HIPAA)-compliant trial adhering to the tenets of the Declaration of Helsinki. Data were collected at Retina Consultants of Houston (Houston, Katy, and Woodlands, Texas). De-identified images and corresponding clinical data were transmitted to the University of Rochester for the analysis described in the current study. The University of Rochester Research Subject Review Board approved the current study and determined that the secondary analysis of de-identified images conducted is not research involving human subjects as defined by Department of Health and Human Services and Federal Drug Administration regulations. UWFFA images for each patient were acquired using the Optos California and 200Tx cameras (Optos, Marlborough, MA, USA). Using software available from the manufacturer, UWFFA images were then transformed into stereographic projection images to create standard montage images for our analysis. Visual acuity was tested using Early Treatment of Diabetic Retinopathy Study (ETDRS) letters. The Heidelberg Spectralis OCT (Heidelberg Engineering, Inc., Heidelberg, Germany) was used to determine the CRT. ## Vessel detection and quantification The proposed framework for correlating vessel density with BCVA and CRT is depicted in. As shown in the lower branch in, retinal blood vessels are detected using a deep neural network organized in the U-Net architecture that has been extensively utilized for medical image segmentation. The network was trained on a dataset consisting of 8 UWFFA images and the corresponding binary ground truth vessel maps using a novel human-in-the-loop procedure that reduces the burden of annotation. The trained network achieved the area under the Precision-Recall curve of 0.930. Additional, technical details can be found in an independent publication, parts of which have been reported in preliminary form. We apply the trained deep neural network to the (baseline) UWFFA images used in this study. The output from the deep neural network is a vessel map where pixel intensity, ranging from 0 to 1, indicates the probability of a corresponding pixel being a vessel. The probabilistic vessel map is then converted into the binary representation by setting a threshold of 0.5, which is shown to be close to the optimal threshold. The top branch in shows the selection region over which the analysis is performed. For each UWFFA image, we manually annotated the center of the fovea. We chose the center of the fovea because it provides a consistent landmark that allows us to select the same region of interest for analysis across different participants. Circular regions centered at the fovea are created for computing the vessel density, which is estimated as the ratio of the number of vessel pixels to the total number of pixels within the circular area. The vessel density was estimated for circular regions centered on the fovea. pixels. The step size is 1 pixel, which translates to 540 circular regions with radii ranging from 10 to 550 pixels. The physical dimensions corresponding to the pixel measurements were determined by using the OptosAdvance software. To measure the reliability of the automated vessel detection technique, we chose, for each patient, two consecutive UWFFA images that were captured after the vessels were fully perfused. Vessel density within a common circular region containing the macula was estimated for each pair of UWFFA images using the automated technique. For the reliability test, we chose the circular region around the fovea with the radius of 550 pixels, which is the largest region of analysis for the study thus far. The reliability test is currently limited to this circular region and future studies will aim to analyze the peripheral regions of the UWFFA. Note that the reliability analysis requires more than one FA image for each patient, which were selected from the original FA frames without the stereographic projection into the standard montage. ## Statistical analysis For each radius, the Spearman rank correlation coefficient was computed between the estimated vessel density and the BCVA reported as per the ETDRS letters. The correlation coefficient takes values from -1 to +1. Values of +1 and -1 indicate perfect positive and perfect negative monotonic correlations between the ranks, respectively, and a value of 0 represents no correlation. A statistical significance test was performed; the null hypothesis is that no correlation exists between the vessel density and BCVA. A p-value less than 0.05 is considered statistically significant. The radius at which the computed correlation coefficients showed minimum variability (assessed by repeatedly computing correlation coefficients leaving out data for one patient and computing the standard deviation of the resulting 42 values) was used for further analysis. Correlation between the estimated vessel density and the CRT was similarly computed and analyzed for statistical significance. The reliability of estimating vessel density was assessed by visualizing the inter- sample agreement using Bland-Altman plots and quantified by computing the intraclass correlation coefficient (ICC). # Results Forty-two patients from the RECOVERY trial were included in our study. Statistics of the baseline demographics and clinical data of these patients are reported in. Examples of vessel maps detected from UWFFA images using the automated technique are shown in. These examples highlight the challenge of contrast variations in FA images and the advantage that automated detection provides. The highlighted rectangular regions in the leftmost image in contain many fine vessels that are difficult to identify in the original UWFFA images. Our automated detection algorithm, however, is not affected by changes in image contrast and is able to detect these vessels. Careful contrast enhancement of these regions validates that the fine vessels are actually present. The automated vessel detection can be performed quickly to allow it to be integrated into the clinical workflow; mean time for automated vessel detection from a UWFFA image is 22.05 seconds (standard deviation 0.21 seconds). shows the Bland-Altman plot that illustrates the agreement between vessel density estimates from the two different captures. The mean difference between vessel density was 0.002 (standard deviation 0.008). The ICC is computed by a 2-way mixed-effects model, average measures, and absolute agreement. A high ICC of 0.980 (95% confidence interval: 0.963 to 0.990) was obtained that indicates excellent reliability. shows the correlation coefficients and mean vessel density as a function of the circular radius, in pixels, used for the analysis. We found significant positive correlations between vessel density and BCVA for radii larger than 53 pixels (p\<0.05), indicating that higher vessel density is associated with better vision. No significant correlation was found between vessel density and CRT. The correlation coefficients between the vessel density and BCVA showed minimum variability for radius of 315 pixels, which was chosen for presentation of further results and analysis. shows the circular mask with a radius of 315 pixels overlaid on a sample UWFFA image and the corresponding detected vessel map. Scatter plots between vessel density and BCVA, and between vessel density and CRT, are shown in, respectively. For this mask, the correlation coefficients of BCVA and CRT are 0.4071 (p = 0.0075) and 0.0533 (p = 0.7376), respectively. Using the Optos *Advance* software we found that masks with a radius of 315 pixels corresponded to a diameter of approximately 11.05 mm (standard deviation 0.12 mm). Detected vessel maps from UWFFA regions of two patients who have high BCVA and two patients who have low BCVA are shown in. Patients with worse BCVA exhibited lower vessel densities. # Discussion Current literature on OCT-A image analysis has shown positive correlations between macular vessel density and visual acuity. The objective of our study was to demonstrate that quantitative estimates of vessel density obtained from automated vessel extraction in UWFFA, a modality that has only been studied qualitatively in the past, similarly exhibits a positive correlation with visual acuity and is consistent with the OCT-A based finding. Replication of the finding based on automated UWFFA image analysis potentially opens the path for greater use of quantitative analysis based on UWFFA images. UWFFA is the current method of choice for assessing far peripheral retina diabetic ischemia, and, to our knowledge, we are the first group to offer an algorithm capable of quantifying vessel density in an automated fashion from UWFFA images, allowing us to work toward our goal of automating peripheral retinal vessel evaluation. Additional strengths of our algorithm include its high reliability, and its ability to automatically detect vessels in image regions in the presence of significant variations in the contrast between the background and the vasculature, variations which are ubiquitous in UWFFA imagery. While the detected vessels can be seen manually via a process of repeatedly adjusting contrast and viewing different regions, the tedium and time requirement for doing this are prohibitive in clinical practice. As a way to limit the amount of manual adjustments, a number of prior studies have focused on only the foveal avascular zone (FAZ) as a potential metric to screen for and track ischemic eye diseases. Studies using both qualitative and manual quantification methods have shown that the FAZ can enlarge and become irregular as ischemic diseases, such as DR, advance. One study examining FA images for a link between FAZ and visual function found no relationship between FAZ area and visual acuity in mild grades of ischemia, but found worse vision in patients with moderate to severe ischemia who had larger FAZ areas. A few studies, however, indicate that FAZ interpretation has diagnostic limitations—the FAZ can be highly variable within control groups and among healthy subjects. It is also suggested that parafoveal retinal tissue is able to function normally without direct retinal blood supply. Vessel density in the entire macula is being explored as a promising new metric to measure the extent of disease in eyes affected by DR and other conditions that can alter retinal vasculature. Current commercially available OCT-A machines are generally used to image the macula. Studies have shown, using OCT-A, that patients with eye diseases, such as DR, have lower macular vessel densities than healthy controls. Further, macular vessel density has been correlated with important clinical outcomes, such as visual acuity and macular thickness. Though OCT-A has been the imaging modality of choice in these studies due to its non-invasive nature, methods gauging vessel density using published sets of OCT-A images have differed significantly in their results. The algorithms used to calculate vessel density in OCT-A images vary; they may accompany instruments as proprietary software, or they may be made by research groups or companies to allow for post-image processing of exported images. As a result, the ensuing vessel density measurement of the same image sets or patients differ among the various devices, algorithms, scan locations, and scanned area sizes. Moreover, studies comparing a variety of OCT-A-based algorithms to detect vessel density have found no single algorithm that outperforms others in all retinal plexuses examined. Significant motion and segmentation artifacts have been noted when using the manufacturer’s OCT-A algorithms. Overall, it has been found that methods that incorporate manual annotation of OCT-A images offer the best discrimination of retinal vessels across plexuses. Manual annotation, however, is highly subjective, highly time-consuming, and further, requires trained graders. However, macular vessel density has not been studied using FA due to the limited ability to quantify macular vessels using FA prior to this work. Using the approach described in this paper, we successfully demonstrated a relationship between visual acuity and macular vessel density, similar to that found in prior OCT-A studies. While OCT-A is becoming more widely adopted, FA, especially UWFFA, continues to play an important role in the assessment and management of diabetic retinopathy. Significant peripheral retinal pathology can be seen in diabetic patients, outside of the area examined by the ETDRS 7 standard fields. With advances in retinal imaging technology, UWF images can capture more than 80% of the retina for examination, leading to identification of additional diabetic lesions in the retina periphery. UWF has advantages over the ETDRS 7 fields assessment, in that it increases the frequency of identifying diabetic retinopathy, and decreases image acquisition and evaluation time. A computerized approach to detecting retinal vasculature, like ours applied to UWFFA images, provides the ability to quantify retinal vasculature more peripherally than swept-source OCT-A devices. Importantly, our algorithm is not affected by changes in image contrast, which enables it to detect vessels that are not able to be visualized by manual vessel detection without extensive attention to adjusting the contrast in order to detect all the vessels present as highlighted in. Future applications of the algorithm can potentially provide a quantitative assessment of retinal vasculature density in both the posterior pole and far periphery without montaging, providing additional data compared to current OCT-A machines. Furthermore, the algorithm can be applied to study longitudinal vascular density changes in vascular diseases, and to study how local and systemic treatments impact retinal vascular disease progression. Quantification of retinal vasculature can facilitate studying the relationship between vascular changes and important clinical endpoints, including visual acuity, in ischemic disease such as diabetic retinopathy, as shown in this study. We have highlighted the clinical significance of our work by establishing a correlation between vessel density and visual acuity in PDR patients without significant center-involving macular edema (CRT \>320 μm). Our finding of a correlation between vessel density and visual acuity is in agreement with that noted in a number of OCT-A studies, mentioned previously. Although previous studies have found that decreased vessel density correlates to increased macular thickness in patients with diabetic macular edema as opposed to healthy controls. Vessel density does not correlate with CRT in our study population, likely because all of the patients in our study did not have significant center- involving macular edema due to the exclusion criteria for the RECOVERY study (CRT \>320 μm). One limitation of our study is that we have used a small sample size to train our algorithm and to report our outcomes. Despite this limitation, we have developed a highly reproducible algorithm for automated vessel detection. Although we developed an algorithm that was capable of detecting far peripheral vessels, we focused on the posterior pole in this study to ensure that the algorithm produced findings consistent with previously published reports using OCT-A data. Our future directions include studying the relationship with visual acuity and vessel density beyond the posterior pole at baseline as well as longitudinally during the course of the study. This will also enable us to assess the impact of intravitreal aflibercept on peripheral non-perfusion in a quantitative fashion. As a number of prior studies examining retinal vessel density in various disease states have concluded, though vessel density may offer a future surrogate endpoint for clinical trials, research is still needed to uncover its full efficacy in predicting clinical outcomes before it can be routinely incorporated into clinical practice. # Supporting information We would like to thank the Center for Integrated Research Computing, University of Rochester, for providing access to computational resources. We also thank Shaun Lampen, Alex Rusakevich, and Brenda Zhou for collecting UWFFA images from the RECOVERY trial. BCVA best corrected visual acuity CRT central retinal thickness DME diabetic macular edema DMI diabetic macular ischemia DR diabetic retinopathy ETDRS Early Treatment of Diabetic Retinopathy Study FA fluorescein angiography FAZ foveal avascular zone ICC intraclass correlation NPDR non-proliferative diabetic retinopathy OCT optical coherence tomography OCT-A optical coherence tomography angiography PDR proliferative diabetic retinopathy UWFFA ultra-widefield fluorescein angiography [^1]: I have read the journal's policy and the authors of this manuscript have the following competing interests: The research presented in this manuscript made use of images from the RECOVERY study. While the RECOVERY study was funded by Regeneron, the work presented here was not supported by any funding from Regeneron or any other commercial entity. Charles C. Wykoff has received consulting fees from Adverum, Bayer, PolyPhotonix, RecensMedical, Kodiak, Alimera Sciences, Allegro, DORC, ONL Therapeutics, Allergan, Apellis, Clearside Biomedical, EyePoint (formerly pSivida), Genentech/Roche, Novartis, Regenxbio, Santen, Regeneron, Aerpio, Fosun, Iveric Bio (formerly Ophthotech), and Takeda. Charles C. Wykoff also has grants from Neurotech, Opthea, Samsung, and Chengdu Kanghong. Ajay E. Kuriyan receives consulting fees from Genentech/Roche, Regeneron, Alimera Sciences, Allergan, Bausch Health. AEK also receives grants from Second Sight and Genentech/Roche. No other authors have any financial interests to disclose in relation to this study. No author has any conflict of interest in any material presented in the paper and this does not alter our adherence to PLOS ONE policies on sharing data and materials.

# Introduction Ruptured cerebral aneurysm is the most common cause of subarachnoid hemorrhage (SAH), causing significant morbidity and mortality. The incidence of SAH in western populations is about 9 to 15 per 100,000 persons per year. The epidemiology of cerebral aneurysm in western populations is well reported in the literature. Some studies attempted to determine whether there is a critical size at which an aneurysm is likely to rupture and thus warrant treatment. It has also been reported that there is a higher incidence of rupture of cerebral aneurysms in patients in Japan. An annual risk of cerebral aneurysm rupture of 2.7% has been reported in Japan, and this is relatively high compared to results from Europe and North America. Rinkel *et al* reported an annual risk of rupture of 1.9% in Western populations. Unlike Caucasian and Japanese populations, few studies describing the epidemiology of cerebral aneurysm in the Chinese population have been published. The purpose of this study is to review the epidemiology of sporadic ruptured cerebral aneurysm, in terms of size, location, the prevalence of multiple cerebral aneurysms, and cerebral aneurysm's gender difference in Chinese population. Reliable knowledge about the risks of cerebral aneurysm will help in planning, screening and prevention strategies and in predicting the prognosis of individual patients. # Patients and Methods This is a retrospective study of consecutive 1256 Chinese patients with ruptured cerebral aneurysm between January 2006 and January 2013, who were admitted to the Second Hospital of Hebei Medical University which is a tertiary referral center of neurological diseases in Northern China, for spontaneous SAH due to a rupture of cerebral artery aneurysm. The data used in this study was retreated from the medical records of the hospital. The institutional review board of the Second Hospital of Hebei Medical University approved this retrospective analysis, and informed consent was waived. The patients included 472 males and 784 females, with mean age of 53.85 yrs (SD: 10.64 yrs, range: 14–88 yr). SAH was initially diagnosed by brain computed tomography, and digital subtraction angiography (DSA) was followed and SAH was confirmed to be due to cerebral aneurysm. In 288 males and 478 females (group 1), the size of the aneurysms was measured by a neuroradiologist at the time of diagnosis, and we used the measurement the neuroradiologist reported. The measurement was done on DSA with the largest diameter measured through the long axis of the aneurysm. In 123 males and 184 females, the size of the ruptured aneurysm was not measured though the location of the ruptured intracranial aneurysms was recorded (group 2). The remaining patients, with 61 males and 122 females, had multiple aneurysms present (group 3), and our medical record could not reliably determine the specific aneurysm responsible for the rupture. All statistical analyses were performed with SPSS 14.0 for Windows (SPSS, Inc., Chicago, IL). Group comparisons were performed with Student's t-test for normally distributed continuous variables or Mann-Whitney U test for other continuous variables. Contingency tables were analysed with Fisher's exact test for dichotomized variables or χ² statistics. *P*\<0.05 was considered statistically significant. # Results ## 1. Gender characteristics in ruptured cerebral aneurysm patients The total 1256 patients (inclusive of groups 1–3) had 784 females and 472 males, with a female/male ratio of 1.66, indicating overall females had a higher cerebral aneurysm rupture incidence than males (*p*\<0.0001). The female/male ratio was 0.72 for patients younger than 40 yrs, and down to 0.50 for patients younger than 35 yrs (*p*\<0.05), indicating in younger patients males had a higher cerebral aneurysm rupture incidence than females. ## 2. Age characteristics in ruptured cerebral aneurysm patients The age distribution of the total 1256 patients (inclusive of group 1–3) is shown in. For both males and females, aneurysm rupture was most common during the age of 50–59 yrs. There were subjects who suffered ruptured aneurysm before the age of 30 yrs (n = 19) and before the age of 20 yrs (n = 3). The mean age of male patients was significantly lower than that of female patients (51.6±11.0 yrs vs 55.2±10.2 yrs, *p*\<0.001). ## 3. Size characteristics in ruptured single cerebral aneurysm patients The size distribution of ruptured single cerebral aneurysm patients (group 1, total = 766) was: \>25 mm: 4 aneurysms (0.52%); \>20 mm: 5 aneurysms (0.65%); \>10 mm: 63 aneurysms (8.22%); \>5 mm: 304 aneurysms (39.69%); \>2 mm: 361 aneurysms (47.13%); and ≤2 mm: 29 aneurysms (3.79%). The mean ruptured aneurysm size was 6.01 mm, with 6.17 mm (median 5.45 mm) for males, and 5.91 mm (median 5.00 mm) for females. There was no significant difference in ruptured aneurysm size between males and females (*p*\>0.05). Ruptured aneurysms were most likely in the region of 2 mm–5 mm (47.13%), followed by 5 mm–10 mm (39.69%). Twenty nine patients (3.79%) had ruptured cerebral aneurysm sized less than 2 mm. The trend that males had a larger aneurysm size than females was seen in younger subjects. In female subjects, a trend of smaller ruptured aneurysm in younger subjects was seen, while this trend was not observed in males. ## 4. Location of ruptured single cerebral aneurysms The analysis of location of ruptured cerebral aneurysms included group 1 and group 2 patients (n = 1073). Ruptured single cerebral aneurysm occurred in anterior circulation in 95.0% of the cases, while 5.0% occurred in posterior circulation. Ruptured aneurysms most commonly occurred in posterior communicating artery (PcoA, 34.86%) and anterior communicating artery (AcoA, 29.45%). There were more cases of AcoA aneurysm rupture before the age of 50 than PcoA aneurysm rupture (149∶80), while there were more cases of PcoA aneurysm rupture after the age of 50 than AcoA aneurysm rupture (294∶167). The size of ruptured single cerebral aneurysm at different arteries is shown in. The aneurysms at anterior choroidal artery and pericallosal artery tended to have a smaller size, followed by aneurysms at anterior cerebral artery. ## 5. Characteristics of multiple cerebral aneurysms In the total 1256 patients, 183 cases (14.57%) had multiple aneurysms, with 61 (61/472, 12.92%) in male patients and 122 (122/784, 15.56%) in female patients. There was no difference of multiple aneurysms prevalence in males and females (*p* = 0.2). In these 183 patients, 159 patients (86.89%) had two aneurysms; the remaining 24 patients (13.11%) had more than two aneurysms. The mean age of single aneurysm patients (mean age = 53.50±10.72) was slightly younger than those with multiple aneurysms (mean age = 55.92±9.87, *p* = 0.003). The site information of multiple aneurysms is shown in. Multiple aneurysms occurred in anterior circulation in 91.8% cases, and in posterior circulation in 8.2% cases. Multiple aneurysm most commonly occurred in PcoA (38.62%), followed by internal carotid artery (23.02%) and medial cerebral artery (13.81%). Of the multiple aneurysm cases, 59 cases had mirrored aneurysms (32.24% out of the 183 cases with multiple aneurysms, and 4.7% out of the total 1256 cases), where aneurysms distributed both on the right side and left side in a mirrored manner. With the cases of mirrored aneurysms, 44 cases (74.6%) occurred in PcoA. There was no significant difference of incidence rate for mirrored aneurysm between males and females (*p*\>0.05). # Discussion To our knowledge, this is largest retrospective analysis on the epidemiology of ruptured cerebral aneurysms in the Chinese population with SAH. One comparable study in Chinese population is the Hong Kong study of 267 Chinese patients with SAH from ruptured cerebral aneurysms. The patients in our series ranged in age from 14 to 88 yrs with a mean age of 53.9 yrs, which is slightly younger than the mean age of 59 yrs reported by the Hong Kong study, while older than the mean age of ruptured cerebral aneurysms for Caucasian patients. Weir et al reported that in a database of 945 patients, the average age of patients with ruptured aneurysms was 46 yrs. In Aarhus et al's study, the median patient age was 50.9 yrs. The same as the Hong Kong study, males presented with ruptured cerebral aneurysms at a younger mean age (51.6 yrs) than females (55.2 yrs). This observation has also been reported in Western literature. Aarhus et al reported male patients were younger than female patients \[48.2 yrs vs. 53.8 yrs\]. A female predominance of patients with ruptured cerebral aneurysms has been reported in studies from the West, Japan, and Taiwan. With the total 1256 patients in our series, the female to male ratio was 1.7∶1. However, our results showed with younger patients (≤39 yrs), there was a male predominance, and the trend was more apparent when even younger patients (≤34 yrs) are considered. The Hong Kong study demonstrated a trend of larger ruptured aneurysms in men (mean size, 6.3 mm) than in women (mean size, 5.6 mm), however, statistical significance was not achieved. This study demonstrated that males also had a slight larger aneurysm size (mean size, 6.17 mm) than females (mean size, 5.91 mm), again statistical significance was not achieved. However, males had a larger aneurysm size than females in younger subjects with statistical significance (*p*\<0.05 for subjects less than 35 year old). Many series, including this study, demonstrated that the majority of ruptured aneurysms are less than 10 mm in diameter. In this series, 90.6% (694/766) of the patients had ruptured aneurysms sized ≤10 mm. 50.9% (390/766) of the patients in our series had ruptured aneurysms sized ≤5 mm. Previous study in the Chinese population demonstrated a proportion of 64% had aneurysms of size 5 mm or less. This is different from the results in western and Japanese populations, where it was reported a lower proportion of ruptured cerebral aneurysms had a size of 5 mm or less,. Kassel and Torner analyzed 1092 patients with SAH and reported that 71% of the aneurysms were less than 10 mm in diameter and 13% were less than 5 mm in diameter, respectively. For female subjects, trend of smaller ruptured aneurysms in younger subjects is seen in this study. While some studies classified PcoA as part of the posterior circulation , the PcoA connects the posterior and anterior cerebral circulations and almost all of the aneurysms that affect it arise at the anterior circulation end and such is usually considered part of the anterior circulation. In our study, we classified PcoA as part of the anterior circulation. A high proportion of ruptured aneurysms located in the PcoA and AcoA, which is similar to the pattern reported in western and Japanese populations. Such findings are also consistent with previous literatures from Hong Kong and Taiwan. Our data showed there were more cases of AcoA aneurysm rupture before the age of 50 than PcoA aneurysm rupture, while there were more cases of PcoA aneurysm rupture after the age of 50 than AcoA aneurysm rupture. The prevalence of multiple aneurysms in our series was 14.57%. This is broadly similar to the previous report of 17% in Hong Kong population and 15% in Japanese population, whilst some literature describing western populations reported this figure to be 30–40%,. Some reports listed the risk factors of multiple aneurysms, including smoking, hypertension, and family history of cerebrovascular diseases. In the Hong Kong series, there was no significant difference in the incidence of multiple aneurysms between men and women, which is contrary to the general finding that female gender is a risk factor for multiple aneurysms. The current study also showed more females had multiple aneurysms than males, however this was due to overall females had a higher incidence of ruptured aneurysms, and there was no statistical difference in incidence of multiple aneurysms between men and women in this study. ‘Mirror- like’ aneurysm, which are located bilaterally on corresponding arteries, has been reported to constitute less than 5% of overall aneurysm. In this series of ruptured cerebral aneurysms, 4.7% (59/1256) had mirrored aneurysms and dominantly occurred in PcoA (44/59). This result is different from Meissner et al's report that the most common distribution for mirror aneurysms was the middle cerebral artery followed by noncavernous internal carotid artery. It has been reported that the presence of a mirror aneurysm is not an independent predictor of future SAHs. The prevalence of giant aneurysms (sized \>25 mm) in our series (0.52%) was slightly lower than previous reports on Hong Kong Chinese (1%) and the Japanese population (1%), and substantially less than the figure of approximately 4% published in literature from the West. There are many limitations with our study. This is a retrospective analysis of the medical record of a single hospital. In 39.01% of patients the size of rupture aneurysms was not recorded. Also in multiple aneurysms the specific aneurysm for the rupture could not be asserted. Screening for asymptomatic cerebral aneurysms is not routinely undertaken in China. Our study simply examined ruptured aneurysms in a population with an unknown number of unruptured aneurysms. Furthermore, we do not know how many patients with ruptured aneurysms did not seek medical attention. Hence, the risk of rupture of those unruptured aneurysms cannot be extrapolated from the data of patients with ruptured aneurysms. However, despite these limitations, our data provided interesting and important data of the epidemiology of ruptured cerebral aneurysms in Chinese population. # Conclusions This study indicates in younger subjects males has a higher cerebral aneurysm rupture incidence than females. Aneurysm rupture is most common during the age of 50–59 yr, and the mean age of male patients is younger than that of female patients. A high proportion of the ruptured aneurysms in our series have a size less than 5 mm. Ruptured aneurysms most commonly occur in AcoA and PcoA. About fifteen percent of patients in our series have multiple aneurysms. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: YXJW L. Zhang L. Zhao. Performed the experiments: L. Zhao L. Zhang ZL. Analyzed the data: YXJW L. Zhang XZ LT. Wrote the paper: YXJW L. Zhang XZ LT. Read and approved the final manuscript: L. Zhao L. Zhang XZ ZL LT YXJW.

# Introduction Serial blood sampling through a vascular catheter is used for pharmacokinetics (PK) and other studies that require multiple small blood volumes from mice. A novel computer assisted system for serial automated blood sampling (ABS) minimizes handling stress and potential difficulties associated with blood collection from mice. However, few studies have evaluated the physiological effects of long-term catheterization and ABS in mice, and these have focused on stress parameters. Even with careful aseptic surgical technique, fibrin forms at the catheter tip, promoting the formation of microthrombi that may occlude the catheter, or be released to the circulation, with possible complications such as local vessel inflammation, sepsis or major organ damage. Fonseca *et al*. and found kidney infection and inflammation in catheterized rats seven days after surgery. In the ABS system, regular infusion of small aliquots of heparinized saline is employed to maintain catheter patency. While this measure reduces or prevents development of large thrombi, we hypothesized that it increases the detachment of microthrombi that embolize to distant sites. Tissue damage, arising from catheterization, initiates an inflammatory response with secretion of cytokines and acute phase proteins (APPs) to the circulation, promoting tissue inflammation and repair. The catheter may mechanically injure the vessel endothelium, additionally contributing to the secretion of cytokines and APPs. The elevation may be sustained further by stressors, such as the continuous connection to the ABS system and associated single housing. The catheterized mouse undergoing ABS may be affected both systemically, by inflammatory mediators and stress, and in multiple tissues by damage arising from microthrombi. Even if clinical effects are not evident, such sequela may affect animal welfare and confound experimental results. In the light of the increasing use of ABS in mice, e.g. in pharmacological and toxicological studies, it is important to understand the pathophysiologic effects that may influence study outcomes. The present study compared the inflammatory responses in catheterized mice, that had been subjected to ABS, with non-catheterized control mice by quantifying pro- inflammatory (IL-1β, TNF-α, IL-6) and immune regulatory (IL-10, IFN-γ, GM-CSF, IL-2, IL-17A) cytokines. Additional assessments included body weight measurements and histological evaluation of the surgical sites, salivary glands, heart, brain, spleen, liver, kidneys, adrenals and lungs. The hypothesis was that catheterized mice would express elevated concentrations of plasma cytokines, weigh less and demonstrate pathologic changes of highly vascularized organs, such as the heart, lung, liver and kidneys, in contrast to control mice. # Materials and Methods This study was conducted in a fully AAALAC accredited facility and approved by The Animal Experiments Council under the Danish Ministry of Environment and Food (license number: 2012/561-169). The facility followed the FELASA guidelines for routine health monitoring of mice and rats. The animals had tested positive for *Helicobacter sp*. but none of the other pathogens on the FELASA list. ## Animals and housing Fifteen male BomTac:NMRI mice (Taconic, Ry, Denmark), eight control mice and seven catheterized mice, were submitted for pathology at 9–11 weeks of age following a preceding study on stress associated effects in relation to catheterization and ABS. The number of mice although limited by the preceding study, was estimated to be adequate by means of a sample size calculation based on data from previous studies with α and β levels set at 0.05 and 0.80, respectively; an estimated mean difference in the concentrations of cytokines exceeding 30% was considered biologically relevant. The mice were single-housed in either Macrolon type II cages (Techniplast, Buguggiate, Italy) (control mice) or cages associated with an ABS system (Dilab Accusamplerμ, VeruTech AB, Lund, Sweden) (catheterized mice). The mice were provided with aspen chip bedding (Tapvet, Kortteinen, Finland), shredded paper nesting material (Lilico, Horley, UK), bite bricks (Tapvey) and cardboard houses (Brogaarden, Gentofte, Denmark) for environmental enrichment. Feed (Altromin 1314; Altromin Gmbh & Co KG, Lage, Germany) and acidified tap water were provided ad libitum. A diurnal rhytm was maintained with a 12:12 hour light-dark cycle, starting with lights on at 6 a.m. Cage temperature was kept at 22°C, relative humidity at 55% and the air was exchanged approximately 75 times h^-1. The catheterized mice had in the preceding study been subjected to surgery with catheterization of the right common carotid artery and tunneling of the catheter subcutaneously to the nape of the neck. An analgesic regimen, consisting of 1 mg/kg body weight (BW) buprenorphine (Temgesic; Schering-Plough Europe, Brussels, Belgium) was given in nut paste (Nutella®, Ferrero, Pino Torinese, Italy) for voluntary ingestion prior to surgery and then once daily for two days post-surgery, as described previously. To ensure adequate pre-emptive analgesia, the mice were injected with 0.1 mg buprenorphine/ kg BW before being brought out of anesthesia. After surgery, the mice were monitored every other hour for six hours and then twice daily for three days post-surgery. In the ABS system, catheter patency was maintained through infusion of 65 μl of 25 IU/ml heparinized saline every 30 minutes. On the third day post-surgery, 25 μl blood samples were drawn automatically every third hour during the following 24 hours. The mice had been subjected to 15 minutes of video recording in a behavioral test, consisting of a combined open field, elevated plus maze and light-dark box. Control mice were subjected only to 15 minutes of video recording in the same behavioral test, and had not been subjected to surgery or in life blood sampling. All animals were submitted for pathology immediately after termination of the preceding study, at which time all catheters were functional. For a full description of the preceding study, please refer to Teilmann *et al*. ## Procedure Upon termination of the preceding study, all mice were weighed and then placed in an induction chamber and anesthetized with 5% isoflurane in oxygen. When the hind paw withdrawal reflex was absent, the mice were exsanguinated by closed cardiocentesis. The mice were returned to the induction chamber, where the isoflurane concentration was maintained at 5% until cessation of respiration. The blood was collected in heparin-coated tubes (BD Microtainer; BD Inc., Franklin Lakes, USA) and centrifuged at 1000 g in a microcentrifuge (Labnet International Inc., Edison, NJ, USA) for 15 minutes to isolate plasma and a second time at 10,000 g for 10 minutes. The supernatant was transferred to clean microcentrifuge tubes and stored at—80°C until analysis for plasma cytokine concentrations. The cytokines; interleukin-1 beta (IL-1β), interleukin-2 (IL-2), interleukin-6 (IL-6), interleukin-10 (IL-10), interleukin-17A (IL-17A), granulocyte macrophage colony-stimulating factor (GM-CSF), interferon gamma (IFN-γ) and tumor necrosis factor alpha (TNF-α) were quantified in duplicate using Bio-Plex Pro assays (Bio-Rad Laboratories, Copenhagen, Denmark) on a Luminex®100™ system (Ramcon, Birkerød, Denmark). A nine point standard dilution series was prepared and all assay reagents were prepared according to the manufacturer’s instructions. The lower limit of quantification (LLOQ) (calculated by the software) was defined as two standard deviations above the levels measured in the zero samples on the standard curve. The upper limit of quantification (ULOQ) was defined as the highest point on the standard curve with an intra-assay coefficient of variation (%CV) of less than 20% and with a recovery value between 70%-130%. The assay detection range was thus bounded by LLOQ and ULOQ. The assay principle has been described previously After euthanasia, the mice were subjected to necropsy and histopathologic sampling. All euthanasia, bleeding and necropsy procedures were performed between 9 am and noon. The heart, brain, spleen, liver, kidneys, adrenals and lungs, and surgical area at the ventral neck including the salivary glands were collected. The specimens were immersed in a 4% aqueous formaldehyde solution (Gurr® formaldehyde, VWR, Vienna, Austria). After approximately one month’s fixation, the surgical area including salivary glands was collected in one tissue block after the catheter was carefully removed. The area was dissected from side to side to include the midline surgery site intact. The heart was bisected longitudinally for evaluation of both ventricles and base of aorta. The brain was cross-sectioned, coronally, from dorsal to ventral using a razor blade at the levels of the olfactory bulb, cerebral cortex (neocortex), hippocampus and cerebellum, as described elsewhere. The spleen was bisected longitudinally. Of the liver, two sections of the median lobe, including the gall bladder, were collected and the total number of microgranulomas was quantified in both cross sections of the median liver lobe for each mouse. Microgranulomas were scored only in the presence of focal granulocyte, sometimes mixed granulocyte- mononuclear cell, accumulations associated with hepatocyte necrosis, and were thus differentiated from extramedullary hematopoiesis (EMH), which is typically not associated with cellular degeneration or necrosis. The left kidney was bisected longitudinally through the midline and the right kidney was transected near the midline. The lungs were inflated with formalin in situ before collection. Trimmed tissue was embedded in paraffin. A minimum of three sections for each organ, stained with hematoxylin and eosin, were examined using brightfield microscopy and scored in random order by two blinded pathologists. Furthermore, gram stains of liver sections from the mice with the highest numbers of microgranulomas were retrospectively evaluated (Mouse A-D). ## Statistics Data were analyzed in SPSS Statistics 20 (IBM, Armonk, NY, USA) and analyzed for normality using Shapiro-Wilk’s tests. Normally distributed data sets were analyzed with an independent samples t-test for comparing the overall difference between groups. Levene’s test of equality of error variances was conducted to test that the variances were equal across groups. Statistics are presented as a t-value, t(df), where df are the degrees of freedom. The data points of the cytokines IL-6 and IFN-γ did not conform to a Gaussian distribution and were thus subjected to a Mann-Whitney U test. Statistics are presented as a U value, as well as the asymptotic significance (2-tailed) *p*-value. For weight data p-values \< 0.05 were considered significant. To account for multiple comparisons, a narrower p-value of \< 0.01 was considered significant for cytokine concentrations. # Results ## Cytokines Catheterized mice did not express significantly different concentrations of the cytokines IL-1β (*t*(7.263) = -0.162, *p* = 0.876); IL-2 (*t*(6.000) = 1.549, *p* = 0.172); IL-10 (*t*(6.617) = -0.510, *p* = 0.626); IL-17A (*t*(10) = -1.550, *p* = 0.152); GM-CSF (*t*(13) = -1.294, *p* = 0.218); TNF-α (*t*(7.229) = -0.303, *p* = 0.771) and IFN-γ (*U* = 20, *p* = 0.319) compared to control mice. The concentrations of IL-6 were significantly elevated in catheterized mice compared to control mice (*U* = 8, *p* = 0.006). Raw data are given in. ## Body weights BWs of control mice, 39.2 ± 3.6 (mean ± SD), and catheterized mice, 35.2 ± 3.4, did not differ significantly (t(12) = 2.109, *p* = 0.057) at the time of euthanasia, although catheterized mice on average weighed less. Likewise, the BW change from the start of the preceding study until euthanasia did not differ between groups (*U* = 14.000, *p* = 0.103). Raw data are given in. ## Pathology In, histopathology changes in the examined tissues are summarized. No gross pathological changes were identified in any mice. Surgical sites in all catheterized mice had mild to moderate granulocytic and lymphocytic infiltration of adjacent connective tissue, occasionally also involving the surrounding skeletal muscle. These changes were mostly evident in the midline, but extended laterally in few mice, with fibroblast and myofibroblast proliferation consistent with healing. In two catheterized mice mild arteritis and thrombi were identified near the catheter sites with angiogenesis of granulation tissue in this area of one mouse. Submandibular salivary glands in five catheterized mice had degenerative to necrotic changes with inflammation in four of these mice. These changes were predominantly unilateral, near the catheter tunnel. The ipsilateral right sublingual salivary gland in five catheterized mice had chronic and active inflammation, with degenerative acinar changes in two of these mice. The parotid gland in five catheterized mice had degenerative to necrotic acinar changes, and was unilaterally infiltrated with chronic and active inflammation in three catheterized mice. One control mouse had mild, multifocal, chronic lymphoplasmacytic inflammation involving the subcutis and adjacent submandibular gland with no source or cause identified. The hearts of two catheterized mice had multifocal, mild cardiomyocyte degeneration, loss and replacement (fibrosis) in the left ventricular free wall (LVFW), the interventricular septum (IVS) and the right ventricular free wall (RVFW) (cardiomyopathy). One of these mice had mild associated granulocytic and lymphocytic inflammation. Heart lesions were not identified in the control mice. Kidneys of five catheterized mice had multifocal, mild cortical degenerative and regenerative changes and mild chronic inflammation with cystic tubule dilatation up to 1 mm in diameter (cystic degeneration) in three of these mice. One catheterized mouse had inflammation around arcuate arteries and veins (perivasculitis) with necrosis of two adjacent glomeruli. One catheterized mouse had a fibrinocellular arterial thrombus in adjacent perirenal muscle. In three control mice, focal, lymphocytic inflammation was noted at the renal pelvis. The livers of four catheterized mice had multifocal, randomly distributed, focal lymphohistiocytic inflammatory infiltrates, sometimes with intrahistiocytic pigment or degenerate cells, in association to hepatocyte necrosis (microgranulomas), where up to 37 microgranulomas were identified in sections from one of these mice. One catheterized and one control mouse had one microgranuloma in the studied sections. Four catheterized mice had prominent single cell necrosis, often in close proximity to microgranulomas. Gram stains were negative for microorganisms. No changes could be identified in the spleens, adrenals, brains or lungs of the mice in either group. # Discussion We have previously shown that ABS through a vascular catheter may be applied in rodents with minimal effects on multiple physiological parameters and behaviors. However, a vascular catheter provides direct access to the circulation and should constantly be regarded as a potential source of infection. Fonseca *et al*. demonstrated kidney infection and inflammation in catheterized rats seven days post-surgery, which impacts the utility of chronically catheterized rodents in studies where kidney function is important for the study outcomes, and contributes to our concern that other tissues also may be impacted by chronic catheterization. Fonseca *et al*. studied the isolated effect of the implantation and presence of a vascular catheter, without interfering with the catheters until the end of study. Many automated systems (ABS) are sterilized before connecting a new animal to the machine, but cannot be kept completely sterile during the study period, which may lead to bacterial colonization of tubing over time. Thus, regular flushing and blood sampling through the catheter, as performed automatically in an ABS system, may enhance the risk of introducing bacteria, which may embolize to various tissues. Assessment of the inflammatory and tissue level effects of chronic catheterization with automated blood sampling is important in order to optimize the use of ABS in rodent research, drug discovery and pre-clinical testing. The pro-inflammatory cytokines IL-6, IL-1 and TNF-α act on the central nervous system to induce a complex set of behaviors and physiological responses, collectively termed the *illness responses*, which promote immune defense and tissue repair through reducing energy expenditure and increasing body temperature. These responses may be clinically subtle and easily overlooked in the laboratory mouse; fever, increased sleep and decreased activity, decreased food–and water intake and reduced social interaction. Therefore, measuring pro- inflammatory cytokines can offer valuable information regarding an animal’s health and well-being. In the present study, only the plasma concentration of IL-6 differed between groups, where catheterized mice had higher levels than control mice. Surgery is a major inducer of IL-6, where IL-6 is detectable in the circulation within 30–60 minutes after surgery and may persist for up to ten days, depending on the severity of the tissue damage. IL-6 is also secreted at the site of endothelial damage and is involved in thrombogenic activity. Furthermore, IL-6 is secreted in response to other stimuli such as psychological stressors (open field, immobilization). In the present study, increased IL-6 levels in catheterized mice likely reflected a physiologic response to surgery, but local irritation of the endothelium by the catheter, stress responses to the surgery, post-surgical recovery, the ABS system, and the final behavioral test may have contributed as well. While IL-1β and TNF-α are also secreted in response to tissue injury, these cytokines are released immediately as part of the acute phase response, which subsides within 24–48 hours, and are thus only briefly present in the circulation, unless other stimuli, such as chronic-active inflammation or post- surgical pain, maintains elevated cytokine levels in circulation. The concentrations of IL-1β and TNF-α were not significantly elevated. Likely, these cytokines increased following surgery, but normalized before the time of necropsy in the absence of ongoing acute inflammatory stimuli. IL-10, IL-17A, IL-2, IFN-γ and GM-CSF reflect different stages of immune activation, usually related to the presence of pathogens. They were included to assess the degree of post-operative complications such as infection, or other causes of excessive or inappropriate inflammatory responses, which were not identified. Therefore, based on cytokine quantification, it is assumed that the peak of the *illness response* following surgery had passed at the time of necropsy, four days post-surgery, and that the catheterized mice had recovered with no measurable signs of systemic inflammation. Histopathology confirmed tissue damage remote from the surgical site. At the surgical sites, modest inflammation and fibroplasia were consistent with aseptic surgery and the four-day post-operative period. Salivary gland changes at the side of catheter tunneling, sometimes with tissue necrosis, were generally modest but exceeded expected changes and suggest that catheterization may affect salivary gland vascular supply or induce acinar necrosis from local compression by the catheter. Salivation plays an important role in food consumption and digestion in rodents, and have roles in host defense. Thus, salivary gland impairment may potentially impact food intake, body weight and some innate defense mechanisms. Food consumption was found to decline post-surgically in the preceding study but to reach pre-operative levels by three days post-surgery. Likewise, body weights of catheterized mice were found to decrease post-surgically in the preceding study, but were not found to be significantly different from those of the control mice in the present study. We attributed the transitional decline in food consumption and body weights of catheterized mice to the surgery and possibly also the administration of buprenorphine, as reduced food consumption is often seen in the post-surgical recovery period to other surgical procedures and because buprenorphine is known to decrease food consumption. Supporting these theories, the food intake and body weights normalized in step with surgical recovery and termination of analgesic administration. However, the chronic effects of salivary gland necrosis, even unilateral, on the rodent model have to our knowledge not been examined. These effects should be considered in studies where salivary function, food consumption or digestion are relevant. Cardiac changes related to catheterization may have clinical and experimental effects. However, myocardial degeneration may occur spontaneously in older mice and further investigation is necessary to appropriately elucidate the effects of carotid catheterization on cardiac function in mice. A follow up study is currently in progress. Kidney changes in five catheterized mice and the thrombus in one of these suggest ischemia related to disseminating microthrombi as likely causes or contributors to these findings. Both catheterized mice and controls had mild chronic (non-suppurative) inflammatory changes, which are common background findings in laboratory rodents. Conversely, inflammation associated with the degenerative changes in catheterized mice was interpreted as relevant to tissue injury. One control mouse was found to have mild, multifocal, randomly distributed vacuolization of tubular epithelium consistent with lipid vacuolation, which are often seen in mice of heavy breeds such as NMRI, and was considered non-specific and unrelated to the study. Liver microgranulomas were increased in catheterized compared to non- catheterized mice. Few microgranulomas, consisting of small lymphocytic and histiocytic infiltrates, are common incidental findings in mice in all age groups. Increased presence of microgranulomas, however, has been suggested to be caused by disturbance of microvascular dynamics or by chronic infection with *Helicobacter sp*. or murine norovira. Often, these aggregates are associated with increased single hepatocyte necrosis compared to usually inconspicuous apoptosis related to controlled removal of senescent hepatocytes. Although *Helicobacter sp*. are present in mouse facilities in our department, high numbers of microgranulomas and increased single hepatocyte necrosis were identified in four catheterized mice only, and contrasted clearly with the control mice. In conclusion, chronically catheterized ABS mouse models can mitigate stress and facilitate multiple blood collections relevant to diverse research areas but, as demonstrated in the present study, potential confounding effects of visceral damage should be considered and assessed. # Supporting Information The authors would like to thank Dr. Cory Brayton for reading and commenting on the manuscript, and to acknowledge Helle Porsdal, Trine Marie Ahlman Glahder, Bente Merete Stallknecht, Hanne Hadberg and Pernille Froh for their practical and technical assistance. [^1]: The authors have declared that no competing interests exist. [^2]: **Conceptualization:** ACT. **Formal analysis:** ACT BR OK. **Funding acquisition:** JH. **Investigation:** ACT BR OK. **Methodology:** ACT KSPA. **Project administration:** ACT. **Resources:** JH. **Supervision:** KSPA. **Validation:** ACT OK KSPA. **Visualization:** ACT. **Writing – original draft:** ACT. **Writing – review & editing:** BR OK JH KSPA.

# Introduction STriatal-Enriched protein tyrosine Phosphatase (STEP) is a phosphatase that is specifically expressed in the central nervous system (CNS). The *STEP* gene (*PTPN5*) produces alternatively spliced isoforms that include a 46 kDa cytosolic form (STEP₄₆) and a 61 kDa membrane-associated form (STEP₆₁). Both STEP₄₆ and STEP₆₁ have a wide distribution in the CNS, although STEP₆₁ is enriched in striatum and to a lesser extent in lateral amygdala, hippocampus and cortex, while STEP₄₆ is expressed in striatum and central nucleus of the amygdala. STEP dephosphorylates and inactivates key neuronal signaling molecules including extracellular signal-regulated kinase1/2 (ERK1/2), stress-activated protein kinase p38 (p38), proline-rich tyrosine kinase 2 (Pyk2), Fyn kinase and the GluN2B subunit of the N-methyl-D-aspartate (NMDA) receptor. As a consequence, STEP opposes the development of synaptic strengthening. STEP is an important regulator of spatial and fear conditioning learning processes, as well as motor skills learning and memory. STEP rapidly inhibits p38 signaling after activation by NMDA receptors during learning processes and thereby prevents sustained neuronal excitation and functions as an important neuroprotector. These studies demonstrate that STEP normally regulates several critical neurophysiological functions. In contrast, alterations of STEP expression and/or function contribute to several neurodegenerative diseases and psychiatric disorders that include Alzheimer’s disease (AD), Huntington’s chorea, and schizophrenia. STEP was shown to be associated with physiological responses induced by cocaine, amphetamine or ethanol, and STEP activity or expression is reduced after repeated and intermittent exposure to either stress or ethanol. We recently showed that the intermittent consumption of large amounts of ethanol induces a robust and long-lasting increase in the phosphorylation of STEP₆₁ on a specific inhibitory site in the dorsomedial striatum (DMS) of mice, but not in other striatal regions. Furthermore, we showed that knockdown of STEP₆₁ specifically in the DMS increased ethanol intake and preference. The development of ethanol drinking behaviors relies in part on the balance between the rewarding and aversive properties of ethanol. As our recent data suggested that STEP₆₁ inhibition was required for the development of ethanol consumption, here, we tested the hypothesis that STEP may modulate the intake of rewarding and/or aversive solutions. Therefore, we determined the consequences of global deletion of STEP on voluntary drinking of ethanol compared to voluntary consumption of sweet and bitter solutions. # Materials and Methods ## Materials Saccharin and quinine hemisulfate were purchased from Sigma (St Louis, MO). Denatonium benzoate was purchased from Alfa Aesar (Ward Hill, MA). Lithium chloride was purchased from Santa Cruz Biotechnology (Santa Cruz, CA). ## Ethics statement All animal procedures in this report were approved by the University of California San Francisco (UCSF) Institutional Animal Care and Use Committee (AN091738-02G), and were conducted in agreement with the Guide for the Care and Use of Laboratory Animals and the Association for Assessment and Accreditation of Laboratory Animal Care (AAALAC, UCSF). ## Animals Male and female STEP heterozygote mice were obtained from Jackson Laboratories. Pairs of male and female STEP heterozygote mice (C57BL/6 background) were mated in-house to generate STEP WT and KO littermates. Genotypes were determined by RT-PCR analysis of products derived from tail mRNA as described in.Male STEP WT and KO mice (2–4 months at the time of the experiments) were individually housed in a temperature- and humidity-controlled room under a 12-hr light/dark cycle, with food and water available *ad libitum*. ## Drugs and treatments Ethanol solutions for the drinking experiments were prepared from absolute anhydrous ethanol (190 proof) diluted to 3–20% ethanol (*v/v*) in tap water. Saccharin, quinine hemisulfate, and denatonium benzoate were dissolved in tap water. For systemic administrations, lithium chloride was dissolved in saline and absolute anhydrous ethanol was diluted to 20% ethanol (*v/v*) in saline. ## Ethanol consumption Oral ethanol intake was determined using continuous access to ethanol in a two- bottle choice drinking paradigm as previously described. Briefly, drinking sessions were conducted 24 hrs a day, 7 days a week with one bottle containing tap water while the other contained an increasing concentration of ethanol (3, 6, 10 and 20%) with 7 days of access to each concentration. Fresh fluids were provided each time the concentration was changed. The bottles were weighed on days 2, 4, and 7 of each week and the mice were weighed once a week. The position (left or right) of each solution was alternated as a control for side preference. ## Saccharin, quinine and denatonium consumption STEP WT and KO mice were tested for saccharin, quinine and denatonium intake using a continuous access two-bottle choice drinking paradigm. Drinking sessions were conducted 24 hrs a day, 7 days a week with one bottle containing tap water while the other contained an increasing concentration of saccharin (0.005, 0.015, 0.033 and 0.066%), quinine hemisulfate (0.01, 0.03, 0.06, 0.12 and 0.24mM), or denatonium benzoate (0.03, 0.06, 0.12 and 0.24 mM) with 4 days of access to each concentration. Fresh fluids were provided each time the solution was changed. The bottles were weighed every day and the mice were weighed once a week. The position (left or right) of each solution was alternated as a control for side preference. ## Conditioned place aversion The conditioned place aversion procedure was performed according to. The place conditioning boxes (Columbus Instrument) consist of two distinct compartments that differ in color and floor texture. After 5 days of handling and habituation to subcutaneous (s.c.) injections, the initial aversion of STEP WT and KO mice was assessed (preconditioning test). To do so, mice were allowed to freely explore both compartments for 20 min, and the time spent in each compartment was recorded. Three animals that spent more than 70% of the time in either one of the compartments during the preconditioning test were excluded from the study. Treatments were then further counterbalanced between compartments to use an unbiased procedure. The next day, the conditioning training started with two conditioning trials per day for 3 days. Specifically, mice were injected (s.c.) morning (9:00 am) and afternoon (4:00 pm) with either saline (vehicle-paired session) or 130 mg/kg lithium chloride (drug-paired session) and confined in the corresponding-paired compartment for 45 min. Control animals received saline injections mornings and afternoons followed by a 45 min confinement. On the fifth day (postconditioning test), mice had free access to both compartments for 20 min, and the time spent in each compartment was measured. ## Conditioned place preference The conditioned place preference procedure was performed according to. The place conditioning boxes were the same as used for the CPA experiment described above. After 5 days of handling and habituation to intraperitoneal (i.p.) injections, the initial preference of STEP WT and KO mice was assessed (preconditioning test). To do so, mice were allowed to freely explore both compartments for 30 min, and the time spent in each compartment was recorded. One animal that spent more than 70% of the time in either one of the compartments during the preconditioning test was excluded from the study. Treatments were counterbalanced between compartments to use an unbiased procedure. The next day, the conditioning training started with one conditioning trial per day for 8 days. Specifically, mice were administered (i.p.) saline solution and confined in the vehicle-paired compartment for 5 min. The next day, mice were administered (i.p.) ethanol solution (2.0 g/kg) and confined in the ethanol- paired compartment for 5 min. Control animals received saline injections instead of ethanol injections. This schedule was repeated three more times (i.e., for 4 saline- and 4 ethanol-conditioning trials). On the tenth day (postconditioning test), mice had free access to both compartments for 30 min, and the time spent in each compartment was measured. ## Locomotor activity Spontaneous locomotor activity of mice was measured in activity monitoring chambers (43 cm × 43 cm) with horizontal photo beams (Med Associates, St Albans, VT). Horizontal locomotor activity was monitored and the distance traveled (cm) by the mice was recorded for 30 min. ## Statistical analysis Data was analyzed with two-way analysis of variance (ANOVA) or two-way repeated measures-ANOVA (RM-ANOVA). Significant main effects and interactions of the ANOVAs were further investigated with the Student-Newman-Keuls (SNK) *post hoc* test or method of contrast analysis. Statistical significance was set at *p* \< 0.05. # Results ## STEP controls the consumption of ethanol, quinine and denatonium, but not the consumption of saccharin We recently showed that the inhibition of STEP₆₁ in mice DMS is required for the development of ethanol-drinking behaviors. Specifically, we showed that the voluntary consumption of ethanol induces a robust inhibition of STEP₆₁ in the DMS of mice and that knockdown of STEP₆₁ in the DMS increased ethanol intake. Consumption is strongly correlated with the rewarding properties of ethanol. However, ethanol intake in both rodents and humans is also tempered by their sensitivity to the aversive bitter taste of ethanol. Therefore, we tested whether global deletion of the *STEP* gene in mice leads to changes in the consumption of ethanol (rewarding and bitter), saccharin (rewarding) and quinine and denatonium (aversive) solutions. To do so, STEP WT and KO mice underwent a continuous access to ethanol in a two-bottle choice procedure, during which ethanol concentration was increased every week (from 3% to 20%). Similar to knockdown of STEP₆₁ in the DMS, STEP KO mice consumed more ethanol compared to their WT littermates (Fig), whereas total fluid intake remained unchanged, suggesting that STEP controls ethanol consumption. Next, we tested the consumption of saccharin and quinine solutions in STEP WT and KO mice in a continuous access two-bottle choice procedure, with the concentration of saccharin (0.005% to 0.066%) or quinine (0.01 mM to 0.24 mM) increasing every four days. As shown in Fig and, saccharin intake, as well as total fluid intake, was similar in both genotypes at all saccharin concentrations. On the other hand, we found that deletion of the STEP gene disrupted quinine consumption. Specifically, quinine intake was significantly increased at three out of four of quinine concentrations (i.e. 0.01, 0.03 and 0.06 mM) in STEP KO mice compared to WT littermate mice. Importantly, total fluid intake was similar between both genotypes. We next tested the drinking of another bitter substance with an unrelated structure, denatonium, in STEP WT and KO mice using a continuous access two-bottle choice procedure, with the concentration of denatonium increased every four days (0.03 mM to 0.24 mM). We found that STEP KO mice drank more denatonium than their WT littermate mice at the denatonium concentrations of 0.03 mM and 0.06 mM, whereas total fluid intake was unaltered. We next determined whether the increase in ethanol, quinine and denatonium intake upon deletion of the *STEP* gene was due to alteration in spontaneous locomotor activity. As shown in, the distance traveled in an open field was unaltered in STEP KO mice compared to WT mice, indicating that deletion of STEP does not change spontaneous locomotion. Thus, the observed increase in the ingestion of aversive tasting agents such as quinine, denatonium and ethanol is not due to a general increase of spontaneous locomotion. Together, these results suggest that STEP controls the ingestion of aversive tasting agents such as quinine, denatonium and ethanol. ## STEP contributes to conditioned place aversion to lithium chloride Our data thus far suggest that the increase in ethanol intake upon *STEP* gene deletion arises, at least in part, from an increase in the threshold of bitter taste rejection. As bitter taste is an aversive stimulus that often signals harmful poisonous substances and is generally avoided by mammals, we thought to assess whether STEP is involved in the avoidance toward aversive stimuli. To do so, we submitted STEP WT and KO mice to a conditioned place aversion (CPA) paradigm using lithium chloride a substance that when administered systemically causes gastric malaise similar to food poisoning. The procedure consisted of three conditioning sessions to saline and three conditioning sessions to 130 mg/kg lithium chloride as previously described. Whereas WT mice developed a strong aversion to the lithium-paired compartment, STEP KO mice failed to express any avoidance to the environment associated with the aversive stimulus. These results suggest that STEP contributes to the avoidance to aversive stimuli in mice. ## Conditioned place preference to ethanol is not altered in STEP KO mice As ethanol has a bitter taste component, we thought that this mechanism could explain, at least in part, the increase of ethanol consumption in STEP KO mice. We therefore tested whether STEP signaling also controls the sensitivity to the pharmacological rewarding properties of ethanol. We used an ethanol-induced conditioned place preference (CPP), which measures the rewarding properties of drugs of abuse, in STEP WT and KO mice. The procedure consisted of four conditioning sessions to saline and four conditioning sessions to 2.0 g/kg ethanol as previously described. STEP WT and KO mice expressed a similar CPP to ethanol, suggesting that, under these conditions, STEP does not control the response to ethanol’s pharmacological rewarding properties. # Discussion We report here that global STEP₆₁/STEP₄₆ deletion increased ethanol consumption in mice. We also showed that quinine and denatonium consumption was increased in STEP KO mice compared to WT littermates, whereas saccharin and total fluid intake as well as spontaneous locomotion were unaltered. Our results strongly suggest that STEP controls the consumption of solutions with a bitter taste component such as ethanol, quinine and denatonium. We further showed that mice with global deletion of the *STEP* gene did not develop aversion to the gastric malaise-inducer lithium chloride, although they expressed similar levels of conditioned place preference for ethanol compared to their WT littermates, indicating that STEP plays an important role in the avoidance to aversive stimuli in mice. Altogether, our results suggest a novel mechanism by which STEP participates to the protection against the ingestion of aversive tasting agents like ethanol. Using a continuous access two-bottle choice of increasing concentration of ethanol, we showed that STEP KO mice drank significantly more ethanol than WT mice, in line with our recent findings. Importantly, it has been shown that ethanol clearance is similar between STEP WT and KO mice, ruling out the possibility that the increase in ethanol drinking upon *STEP* gene deletion could be attributable to an enhanced ethanol metabolism, nor was it due to locomotor changes, as spontaneous locomotor activity was similar between both genotypes. Besides the sensitivity to the rewarding properties of ethanol, the perception of its aversive bitter flavor also plays an important role in the levels of voluntary oral consumption of ethanol in both rodents and humans. We therefore hypothesized that STEP may promote the avoidance response to solutions with an aversive bitter taste. As STEP KO were aversion-resistant compared to WT littermate mice, our results suggest that STEP is specifically involved in the avoidance to the bitter tastant ethanol. Consistent with our hypothesis, a relationship between the responsiveness to quinine and ethanol has been suggested, and a blunted sensitivity to bitter taste may be an important predictor of high ethanol consumption in both rodents and humans. To further confirm that STEP contributes to the avoidance to bitter substances, we tested the drinking of another structurally unrelated bitter substance, denatonium, in STEP WT and KO mice. Even though taste receptor cells discriminate between different bitter stimuli, rats fail to distinguish quinine from denatonium. Here we showed that STEP KO mice drank more denatonium than WT mice, confirming the involvement of STEP in the avoidance to bitter tastants. Bitter taste often signals harmful substances and sensitivity to bitter tastants is thus acknowledged to have evolved as a protective mechanism in mammals. Therefore, a bitter taste is considered as an aversive stimulus in rodents. To test whether STEP contributes to the avoidance to aversive stimuli in mice, we used a lithium chloride-induced CPA paradigm. We choose lithium chloride as the aversive stimulus in the CPA experiment because the gastric malaise induced by lithium chloride administration is comparable to food poisoning, which can be the result of the consumption of bitter substances. We found that WT mice developed a robust CPA whereas STEP KO mice were resistant to CPA. The fact that comparable ethanol-induced CPP levels were observed between STEP WT and KO mice allowed us to rule out spatial or associative learning deficits as confounding factors in the CPA experiment. These results indicate that STEP is necessary for the development of conditioned aversion and further suggest that inactivation of STEP abolishes the avoidance to aversive stimuli and hence promotes the consumption of aversive tasting agents like ethanol. We recently demonstrated that inactive STEP₆₁ in the DMS increases the drinking of ethanol without altering the palatability of sweet or bitter solutions, in apparent contrast with the present report. First, we cannot exclude the occurrence of developmental compensatory mechanism in the mutant mice used herein compared to the gene knock down acutely performed in our previous study. Such compensation may account, at least in part, for the behavioral changes observed in the STEP KO mutant mice. Moreover, it is important to note that herein, both STEP₆₁ and STEP₄₆ isoforms are deleted in the whole brain, versus specific downregulation of the 61 kDa isoform in the DMS in our previous study. Thus, ethanol-drinking behaviors are driven by different STEP-related mechanisms in several discrete brain regions, some of them regulating the appetence toward the pharmacological effects of ethanol in the DMS and others controlling the aversion to the bitter taste of ethanol. For instance, STEP is expressed in brain regions such as central amygdala and bed nucleus of the stria terminalis, both of which are involved in negative affective learning and responsive to bitter tastants. Further studies will thus be necessary to dissect the contribution of STEP to aversive stimuli in such brain regions. Recent preclinical reports indicate that STEP inactivation improves cognitive deficits in animal models of AD, Hence, STEP inhibitors are currently being developed as potential treatments for humans suffering from AD. Our previous study and the present report indicate that both specific downregulation of STEP₆₁ expression in the DMS and global deletion of the *STEP* gene lead to the increase of ethanol drinking. Therefore, future use of STEP inhibitors in clinical trials should be closely monitored regarding the consumption of alcoholic beverages and the response to aversive, undesirable signals. The authors thank Eric Zhao for technical assistance. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: RL ED DR. Performed the experiments: RL SAW. Analyzed the data: RL ED. Contributed reagents/materials/analysis tools: PJL. Wrote the paper: RL ED DR. [^3]: Current address: R. L. INSERM ERi24, Université de Picardie, Amiens, France [^4]: Current address: E. D. Douglas Research Center, McGill University, Montreal, Canada

# Introduction Metabolites are sometimes referred to as the “canaries” of the genome. Just as canaries for coalminers served as sensitive indicators of problems in coal mines, metabolites can be exquisitely sensitive indicators of problems in the genome (as well as the transcriptome or proteome). Metabolites are effectively the end products of complex interactions occurring inside the cell (the genome) and events, exposures or phenomena occurring outside the cell or organism (the environment). As a result, the comprehensive measurement of metabolites (via metabolomics) allows one to determine interactions between genes and the environment. In other words, metabolomics allows researchers to obtain a highly sensitive and more complete description of the phenotype. This metabolic readout of the phenotype is often called the “metabotype”. Recent advances in both analytical chemistry and metabolite data analysis techniques are now making metabolomics far more accessible to a wider range of research disciplines. Indeed, metabolomics is now routinely used in biomedical research (for biomarker discovery and disease mechanism research), food and nutritional analysis, crop characterization and environmental monitoring. As a result, the field of metabolomics has experienced very rapid growth with just two papers published on the subject in 1999 to more than 2400 in 2015. However, unlike in other areas of agriculture research where metabolomics is widely used in crop trait selection, pesticide monitoring, crop breeding or crop evaluation, the application of metabolomics to livestock research is somewhat less widely used or appreciated. This is surprising given the potential of metabolomics to address many important questions in livestock and animal science. In particular, the power of metabolomics to non-invasively detect subtle phenotypic changes, innate phenotypic propensities and dietary responses makes it an ideal tool for livestock research, breeding and assessment. Recently, there have been a number of papers in livestock metabolomics that have generated compelling results showing how metabolomics and metabolite-based phenotyping (metabotyping) can help farmers, veterinarians, livestock researchers and the livestock industry. These include papers demonstrating how metabolomics can be used to predict feed efficiency and residual feed intake (RFI), ascertain disease propensity, evaluate dietary responses to different feeds, assess carcass merit, fertility, milk quality, determine bioproduct content and ascertain other important economic or breeding traits associated with livestock. Fast, effective, and quantitative phenotyping is critical for farm trials dealing with animal selection and breeding. Many traditional phenotypic measurements such as those related to animal feed consumption and RFI are expensive, time consuming and require specific recording equipment. Others, such as carcass trait evaluation, may require animal slaughter, which obviously eliminates the potential breeding value of the animal. Similarly for reproductive traits, animals have to reach a stage of maturity and sexual activity to allow measurement of related traits. Metabolomics allows many of these trait measurements to be conducted earlier, more routinely, non-invasively and often at a lower cost than current techniques. However, metabolomics is not without its challenges. Metabolomic experiments must be carefully designed as diet and other variables such as sex, diurnal variations and sampling time can profoundly affect results. Likewise, metabolomic technologies, such as gas chromatography (GC), mass spectrometry (MS) and nuclear magnetic resonance (NMR) spectroscopy are not yet widely available in many livestock research facilities. Furthermore, there continues to be a significant shortage of data resources that could facilitate the interpretation of livestock metabolomic data. Given the many applications of metabolomics in both the livestock industry and livestock research as well as the diversity of journals in which livestock metabolomics is often published, we felt it was important to conduct a thorough, systematic review of the field. By consolidating the results from diverse journals and different studies into a single review paper, we believed this content would provide a more complete picture of both the strengths and the weaknesses of livestock metabolomics. In conducting this review we sought answers to 4 key questions: 1) What are the most common applications of metabolomics in animal science and where are they trending?, 2) What are the preferred metabolomics technologies in livestock metabolomics and how are they evolving?, 3) What are the most obvious gaps or weaknesses in livestock metabolomics relative to other fields of metabolomics research? and 4) What are the known or measured metabolites for the 5 major livestock species (i.e., bovine, ovine, caprine, equine, and porcine) in different tissues and biofluids? This metabolite compilation, which we have called the livestock metabolome database or LMDB (available at <http://www.lmdb.ca>), is intended to help lay a more solid foundation in terms of data resources that would make livestock metabolomic studies much easier to perform, analyze and compare. The LMDB catalogues all metabolite compounds that have ever been identified and reported in the 5 livestock species (for multiple biofluids and tissues), along with concentration ranges, compound descriptions, chemical structures, reference NMR and MS spectra and other information associated with each metabolite for both healthy and a variety of abnormal physiological conditions. # Materials and methods In compiling this review and assembling the livestock metabolome database, we used a combination of web-accessible data mining tools along with manual curation to survey 2313 peer reviewed journal articles covering the period from 1930 to 2015. From this initial set of articles, we reduced the number further to cover published livestock papers reporting the measurement or characterization of ≥8 metabolites for any of the 5 major livestock species (i.e., bovine, ovine, caprine, equine, and porcine). This reduced the target number of peer-reviewed manuscripts to a total of 149. The livestock species selected for this review were based on their global population, economic impact and use in agricultural systems. Details regarding the keyword selection, search engines and databases, journals and search strategy are given below and summarized in the preferred reporting items for systematic reviews and meta- analysis (PRISMA) checklist and flow chart. ## Keyword selection As noted above, this review is focused on 5 main livestock species including cattle, sheep, goats, horses and pigs. Therefore, a combination of keywords was selected to target those specific animals and to identify the associated metabolomics studies. Keywords were divided into 3 main groups: 1) animal species, 2) sample types, and 3) metabolomic methods. Selected keywords for animal species included the name of the species and its various derivatives or synonyms, i.e., bovine, cattle, cow, calf, *Bos taurus*, etc. To target metabolomics papers in animal science, a broad range of metabolomics keywords were identified and used. These included different variations of the term “metabolomics” (such as metabolomics, metabonomics, metabolite profiling, metabolite fingerprint, chemical profile, chemical analysis, chemical composition, etc.) to target publications prior to and after 1999, as well as the names of various analytical platforms (i.e., NMR, mass spectrometry, liquid chromatography, gas chromatography-MS, etc.). Moreover, a wide variety of sample types such as different body fluids (i.e., serum, blood, plasma, urine etc.) and different organs or tissues were selected to further enrich the keyword search. ## Search engines and databases An initial comparison among many open access search engines showed that most search results are similar regardless of the search engine used. Therefore, Google Scholar (<https://scholar.google.ca/>) was selected as the primary literature search engine. In addition, a number of agriculture-specific databases such as Agricola and AGRICULTUREnetBASE were also used. Other databases included Scopus, the Web of Science, ScienceDirect (<http://www.sciencedirect.com/>) and PubMed (<http://www.ncbi.nlm.nih.gov/pubmed>). Settings for all search engines and databases were adjusted to increase search efficiency and filter irrelevant results. ## Search methods and selection criteria Different keywords were combined to target metabolomics papers in the field of animal science. For example, “cattle”, “cattle serum” or “cattle milk” was accompanied with “metabolomics”, “chemical composition” or “metabolite profiling”. Consequently, each combination of the keywords in the search engines generated a long list of results. These included various types of publications (full papers or abstracts) that contained any or all of the used keywords. A manual review was performed on all retrieved publications. Typically, the first 3–5 pages of the search results from the aforementioned search engines were manually reviewed to select for articles of interest. Among the papers identified as worth pursuing, research papers, abstracts or textbooks that showed relevance in their title or abstract were selected. In addition to papers reporting experimental results, review articles that included specific metabolite data sets were also selected. Among the selected manuscripts, only those papers that reported ≥8 metabolites were chosen for this review. The threshold of 8 or more as the minimum number of metabolites was based on a *post hoc* analysis of the retrieved papers and the need to optimize both metabolite coverage and the time devoted to manual analysis. We also determined that this selection cut-off allowed us to cover most, if not all, of metabolites reported in papers with \<8 metabolites. Based on these criteria, a total number of 149 manuscripts covering all 5 animal categories were selected for this review. Selected publications were carefully read to extract and annotate a set of 10 pieces of information including: 1) metabolite names; 2) tissue or biofluid origin; 3) quantified values (concentration) if any; 4) experimental conditions; 5) animal breed; 6) sample size; 7) analytical platform; 8) field of research, 9) physiological condition (disease or state of health), and 10) Pubmed/DOI references. ## Compilation of the livestock metabolome database In compiling the data for this livestock metabolome database or LMDB (<http://www.lmdb.ca>), all reported concentrations were transformed into a standardized concentration unit (micromolar; μM) and each entry was associated with an abbreviated description of the experimental context, the sample type, and the methodologies used for the metabolomic analyses. In identifying a metabolite for inclusion in this study the compound had to: 1) have a molecular weight \<1500 daltons; 2) it could not be a peptide, protein or oligonucleotide; 3) it had to correspond to a reasonably unique chemical entity (triglycerides and amino acids are not unique chemical entities, but LysoPC-16:2 is sufficiently unique) and 4) it had to be identified with a structurally interpretable name. This literature-based effort generated 1070 metabolites from 149 peer-reviewed papers, abstracts or textbooks. Metabolites extracted from these manuscripts were systematically categorized into the LMDB. Nearly all metabolites extracted were linked to a standard Human Metabolome Database (HMDB) identifier which provides a freely-accessible comprehensive description of each metabolite. A brief description of experimental data for each metabolite was also extracted from the articles and included in the database including information on the analytical platform, experimental conditions and field of research. A PubMed and/or DOI id was also associated with each metabolite to provide a link to the article reporting that metabolite. Additional data on each metabolite, including structure, synonyms, chemical classifications, physicochemical data, reference NMR, GC-MS or LC-MS spectra and links to other databases were obtained through an in-house annotation tool called DataWrangler. All of this information was used to construct the on-line version of the LMDB (<http://www.lmbd.ca>). The LMDB was prepared using a Ruby-on-Rails framework, modeled after other on-line species-specific metabolomic databases prepared in our laboratory. Details regarding their construction, required operating systems, browser compatibility and hardware requirements can be found elsewhere. # Results and discussion ## Growth and trends in livestock metabolomics research Based on the data collected from our literature survey, it is clear that the majority of metabolomics studies among all livestock categories have been conducted in cattle with a total of 76 articles (50% of the selected articles) focusing on various fields of bovine research and assessment. Metabolomics studies on pigs and sheep came second and third with 28% and 12% of the selected articles, respectively. The least studied group were horses with only 5 (3%) reported equine metabolomic studies. As might be expected, most livestock metabolomic studies focused on issues related to animal health, nutrition and production (65%). These studies are obviously useful for characterizing bioproduct quality, identifying biomarkers or understanding animal responses to different stressors. However, we were surprised to see relatively few efforts focused on metabolomic characterization of healthy animals with the aim of identifying baseline values for different metabolites in different biofluids or tissues. In fact, only 16 studies (10%) of this kind were reported. These “referential surveys” are foundational and are often needed before biomarker studies could/should be undertaken or fully understood. As noted earlier, most metabolomic studies of cows, sheep, goats and pigs appear to be directed towards disease detection, production and bioproduct assessment, feed efficiency determination and reproduction. In contrast, the primary focus for equine metabolomics has been on drug discovery and doping detection, specifically for thoroughbred horses. Given the large sums of money directed to horse racing, this is not unexpected. However, compared to the widespread applications of metabolomics in other livestock species for other purposes, it is clear that equine metabolomics is being under-utilized. Certainly, equine metabolomics could be used to select more desirable traits and higher value or higher performing animals, similar to what is being done for bovine metabolomics. Likewise, metabolomics could serve as a diagnostic or prognostic tool for improving equine health and disease resilience (as it has for essentially all other livestock species). Temporal categorization of all 149 published studies showed that the majority of livestock metabolomics papers were published after 1999. Less than 9% (13 articles) of the selected papers were published prior to 1999 while, \~91% (136 articles) of the papers were published thereafter. The earliest paper in our collection dates from 1930. It is noteworthy that the term “metabolomics” was not coined until 1998 therefore, metabolomics studies prior to this date had to be identified using other keywords such as “chemical composition”, “biochemical profiling”, etc. Based on our observations, it is clear that interest in livestock metabolomics is growing rapidly, especially over the last couple of years. Our data indicates that from 2000–2010 just 29 articles (19%) were published in this field, while from 2011–2015 a total of 107 (72%) articles were published. In terms of percentage growth, the most rapidly expanding subfield appears to be caprine and equine metabolomics with a growth rate of 100% over the past 5 years. In terms of overall growth, the most significant changes were in bovine metabolomics with the number of papers growing from just 10 prior to 1999 to 49 in 2011–2015. The most recent additions to the field of livestock metabolomics are studies focused on goats (starting in 2014) and horses (starting in 2007). ## Trends and gaps in livestock metabolomics applications We found that livestock metabolomics studies can be categorized in 7 main areas. These include animal health, animal nutrition, animal production, animal reproduction, animal physiology (mainly analysis of different biofluids), animal products (products originating from livestock such as milk, meat, yogurt, etc.), and human health (livestock models used for human health studies). This general categorization was based on a *post hoc* analysis of the types of articles where we manually assessed article keywords, subject headings, journal titles and the general focus of each article. Most of these categorizations (such as animal reproduction, human health and animal health) were relatively simple to make. For instance, the category “animal reproduction” obviously refers to articles using metabolomics to study reproduction in livestock. Likewise, the category “animal health” refers to articles using metabolomics to study livestock health or disease while “human health” refers to application of metabolomics to study human disease using various livestock models. Other categories proved to be somewhat more ambiguous. For instance, the field of “animal products” typically contains metabolomics investigations related to food, nutrition and human consumption of animal products, such as meat and cheese. On the other hand, “animal production” is focused on investigating the associated biochemical profile with each animal product. In some cases, we had to be fairly strict with our definitions. For instance, we limited “animal physiology” to include only those articles focused on analyzing various biofluids or characterizing the metabolite composition of specific biofluids, organs and tissues. Selected livestock metabolomics articles of 5 major livestock species were categorized based on the area of research, i.e., animal health, animal nutrition, animal production, animal reproduction, human health, animal physiology and animal products. It is noteworthy that articles in the area of human health mainly reflected animal models being used to study human related health issues. Among the seven different categories, animal health (52) and animal production (40) had the most metabolomics articles published for the largest number of animal groups. However, this varied depending on the livestock species being studied. In human health research, porcine metabolomic studies covered the majority of articles (14 articles) compared to all other livestock categories. This is not unexpected, given the comparable physiology of pigs to that of humans. In the category of animal products, bovine-based studies had the most articles published (16 articles) relative to all other groups. Some of the more interesting applications of metabolomics found in our survey include the use of metabolomics for quality control of animal products, evaluating nutritional value and impact of various feed sources on animal health and products, investigating disease biology by using animal models of human disease, investigation of potential metabolite biomarkers of animal disease, assessment of production traits, reproductive performance, and general metabolome characterization. In terms of gaps in the existing literature, it is perhaps most useful to use bovine metabolomic studies as the “gold standard” by which to compare other livestock species. While metabolomics is routinely being used to understand the biology or diagnose a few common bovine production diseases (including acidosis, mastitis, milk fever) we found no metabolomic studies looking at common diseases in sheep or goats (such as brucellosis, campylobacteriosis, pneumonia, Q fever), in horses (equine flu, equine herpes, equine sleeping sickness, anemia, laminitis, azoturia), or in pigs (respiratory diseases, swine dysentery, parvovirus). Indeed, we found only 22 metabolomic studies focused on the health of sheep, goats, pigs and horses, compared to 30 metabolomic studies for cattle alone. Of these 22 non-bovine studies, most were focused on metabolic, growth and neurodegenerative disorders. Livestock metabolomics studies also appear to be missing a number of opportunities currently being pursued in human biomedical research. One of particular note is the use of metabolomics to predict (as opposed to diagnose) or detect subclinical forms of disease. While disease diagnosis is useful, often it is too late or too costly to perform useful veterinary interventions. Detecting diseases before they manifest or predicting them before they occur allows inexpensive prophylactic or preventative measures to be taken. In human metabolomic studies, the identification of disease prediction biomarkers is becoming increasingly common. This is because metabolic changes appear to precede significant physiological changes, possibly because metabolites play an important signalling role to activate later stage (i.e. symptomatic) physiological responses. However, we could only find 2 papers (limited to cattle) that focused on disease diagnosis/prognosis or (sub)clinical detection of diseases. A similar approach could also be used towards the prediction of later-life production traits on the basis of early-life metabolic fingerprints. This, too, is an area of interest in the field of human metabolomics, where later-life health is being predicted on the basis of early-life metabolic fingerprints. Obviously the reliable prediction of economically important traits is an important tool for livestock management and strategic planning. Metabolomics is already being used in the evaluation and/or prediction of production traits such as residual feed intake (RFI), carcass merit, reproductive performance and metabolic disorders for cattle. However, there is a surprising dearth of similar studies regarding evaluation or prediction of production traits for sheep, goats and pigs. Metabolomics potentially offers a unique opportunity for indirect, inexpensive marker-assisted measurement of these economical traits. This can be achieved through non-invasive sample collection of readily accessible biofluids such as blood, urine, milk and saliva. In most cases, the standard measurement or prediction of some traits such as RFI and carcass merit requires labour intensive, invasive, costly and time consuming measurements. Metabolomic studies regarding the prediction of RFI in beef cattle have already been very promising with a reported initial prediction accuracy of 95%. Metabolomic data, when coupled with genomic data, appear to increase the accuracy of trait prediction. This combination potentially allows one to screen for individual animals with superior traits that could be used for breeding stock. Given the positive results already seen for cattle, the application of these metabolomic concepts to other livestock species is certainly worth investigating. Overall it appears that there is still a considerable body of useful metabolomic work that could be pursued with most other livestock species by simply applying or extending what has already been done in bovine metabolomics. ## Trends and gaps in sample size Nearly 50% of the selected articles for all animal species used ≤30 animals or samples (from an even smaller number of animals) to conduct their metabolomics analysis. Other sample size categories shown in account for \~10% of the peer- reviewed livestock metabolomics literature. The maximum number of samples reported from the selected papers were: 1587 (bovine), 163 (ovine), 80 (caprine), 36 (equine), and 506 (porcine). It is noteworthy that sample size does not always reflect the total number of animals used in the study. For instance, longitudinal studies typically collect multiple samples from a relatively small number of animals over an extended period of time. Relative to many reported human metabolomic studies or rodent model studies the number of samples and the number of subjects (i.e. animals) used in most livestock metabolomics studies is generally quite small. Indeed, many human and rodent model studies routinely measure 100s to 1000s of samples. This difference in sample size likely reflects the relatively high cost of performing large animal studies as well as the somewhat limited funding available to agriculture research relative to medical research. Sample size reported in livestock metabolomics papers were divided into 5 categories with papers using ≤30 samples, 31–50, 51–100, or those that have not mentioned the number of samples used in the analysis. However, it is important to note that the smaller sample sizes in livestock metabolomics also mean that statistical significance and “power” of the published results is also somewhat less than many human-subject or model organism studies. This represents a significant gap for livestock metabolomics and requires either study sizes to be increased or more effort being directed to conducting validation studies on similar-to-largely sized cohorts for confirmation previously reported results. Indeed, we found only one bovine metabolomic study reporting either independent cross validation (using a different animal cohort) or independent follow-up validation of any newly identified biomarkers or interesting metabolite findings. On the other hand, follow-up validation studies are becoming routine in human metabolomic studies . Clearly, this is a gap in livestock metabolomics that must be filled if metabolomic findings are going to be translated to practical pen-side or on- farm applications. Another consistent problem detected in the published livestock metabolomics literature is incomplete reporting. We found that 13% of all published livestock metabolomics papers did not report the number of samples used in their research. Providing information on sample size is an essential scientific measurement and reflects on the quality and reliability of published papers. Failure to report sample sizes along with failure to provide information on the numbers of animals or animal replicates indicates a major flaw in manuscript preparation and scientific work. ## Trends and gaps in biological sample types As can be seen in and, a total of 30 different sample types have been used for livestock metabolomics analyses. The most commonly used sample types include milk, plasma, serum, urine and ruminal fluid. These biofluids account for 78% of the total sample types reported. Milk and plasma are the most commonly used samples in bovine metabolomics manuscripts. Among all other animal groups, plasma was the most widely examined sample type, reflecting perhaps the ease of collection but also its potential utility as a proxy reporter for all of the organs in the body. Some of the least frequently used samples include cerebrospinal fluid, colostrum, semen, adipose tissue, kidney and kidney perfusate, feces, amniotic fluid, bile and liver. The relatively low number of papers reporting data on tissue metabolomics likely reflects the challenges and costs of animal culling especially for larger livestock, sample collection, and the need to rapidly perform metabolic quenching via liquid nitrogen (immediately after surgery or necropsy) to obtain useful tissue samples for metabolite analysis. Different varieties of samples have been used in livestock metabolomics analyses as identified by the number of published articles per sample per livestock specie. While studies on bovine milk are quite prevalent, there are essentially very few studies on sheep or goat milk. Given the importance of goat and sheep milk in the global agrifood economy, it is surprising that only a total of 6 papers have been published on goat/sheep milk metabolites. One notable study, however, is that of Park and colleagues who used LC-MS to identify/quantify 82 metabolites in sheep and goat milk. This paper reports a number of other macronutrient milk constituents including fat, protein, minerals and vitamins. In another more recent study, the effect of a specific grazing patterns and their associated dietary effect on goat milk was evaluated. These authors used GC-MS techniques to identify and quantify 25 milk metabolites. Similar trends are also seen in other biofluid or sample types, with bovine samples or bovine-related papers dominating. For instance, there are a number of metabolomic studies on bovine ruminal fluid, plasma and urine, but very few studies on these biofluids for sheep, goat, horses or pigs (43 for all 4 species and 3 sample types). Likewise, metabolomics studies on colostrum and semen are limited to cattle only with one study each. Interestingly, some of the less- frequently used sample types such as cerebrospinal fluid, synovial fluid, amniotic fluid, bile and vitreous humor are limited to the less frequently studied livestock species (sheep, goat and pig). What is also quite striking is the dearth of fecal metabolomic studies among all livestock species. With the growing interest in the microbiome and the clear role that gut (and rumen) microflora play in animal health, we were surprised by the complete absence of metabolomic papers on bovine fecal samples. Given the importance of beef, sheep and goat meat, it is also surprising to see how little metabolomic data has been collected on meat samples. Indeed, only a total of 9 papers provided data on relatively small number (140) of meat metabolites. The most comprehensive meat metabolomics study was reported by Castejón et al.. These authors profiled meat exudate using NMR to explore the effect of storage time on metabolite composition. They reported a total of 60 different metabolites. Overall, these data suggest that the livestock metabolomic literature is characterized by a significant under-representation of some important sample types, including milk, meat, fecal/rumen, semen samples and cerebrospinal fluid. These “gaps” in our knowledge and “gaps” in the published literature represent clear opportunities for livestock researchers to pursue. With regards to the number of metabolites detected, quantified and/or reported among the different sample types, we found that the broadest level of coverage was for milk, plasma and serum. Ruminal fluid, urine, feces and meat samples had slightly lower levels of coverage while the rest of the sample types reported in typically report \<60 metabolites each. It is instructive to compare these livestock metabolite numbers to data reported for human metabolites identified in similar kinds of sample types. For instance, the most comprehensive human milk metabolomics paper reports just 129 identified metabolites, which is \>3X lower than what has been reported in the livestock milk. The total number of metabolites reported for plasma/serum in humans is 4229, which is significantly more than what is reported for livestock plasma/serum (with 759). Likewise, the total number of human urine metabolites has been reported to be 445, which is more than twice that found in the urine of livestock species. Given their genomic similarity, our expectation is that the number of metabolites measurable in livestock for each of the biofluids should be comparable to the number of metabolites measured in humans. Currently, the Human Metabolome database recognized as the most comprehensive metabolomics database contains \>40,000 metabolites derived from various human biosamples. As a result, this suggests there is still a significant gap to be filled with regard to the depth and breadth of metabolome characterization in livestock. The number of metabolites detected, quantified and/or reported among the commonly used sample types in the livestock metabolomics publications up to 2016 (counting publications that reported \>8 metabolites). ## Trends and gaps in analytical instrumentation and methodologies Metabolomics uses a wide variety of analytical instruments that vary in terms of their sensitivity and breadth of coverage. Nuclear magnetic resonance (NMR) continues to be among the most commonly used analytical platforms in metabolomics. It is often chosen for its reliability and utility in absolute quantitation however, NMR is relatively insensitive and is limited to measuring substances in micromolar to millimolar (μM-mM) concentrations. Mass spectrometry (MS) platforms (especially LC-ESI-MS) can detect metabolites at nanomolar (nM) to picomolar (pM) concentrations, allowing a much higher number of metabolites to be detected. However, MS instruments are prone to frequent breakdowns and, relative to NMR, it is often difficult to quantify chemical concentrations via MS techniques. Gas chromatography-MS (GC-MS) is less sensitive than liquid chromatography (LC)-MS, but is generally more robust and more reproducible. As a result, GC-MS can sometimes be used to identify and quantify the metabolome with higher precision and reproducibility than either NMR or LC-MS. Each of the 149 livestock metabolomics papers was carefully analyzed to identify which analytical platforms (NMR, LC-MS, GC-MS) were used more frequently to conduct metabolomic analyses. In certain studies, more than one platform was used so, we simply counted the frequency that each technique or technology was used in each study. Interestingly, the most commonly used metabolomics platform for all animal categories is NMR spectroscopy, accounting for 28% of all livestock metabolomics studies. Following closely behind NMR, in terms of frequency, is LC-MS with 25% of all studies using this analytical platform. It is noteworthy that the LC-MS category includes ultra performance liquid chromatography (UPLC)-MS, high-performance liquid chromatography (HPLC)-MS, and direct flow injection (DFI)-MS. Gas chromatography-MS is the third most prevalent (15%) analytical platform used in livestock metabolomics studies. The more limited use of GC-MS is typical of other metabolomic disciplines as well. Other, less conventional or more targeted, methodologies account for the remaining 27% of the technologies used in livestock metabolomics studies. These methods include, but are not limited to, infrared spectroscopy (FTIR), silicic acid column chromatography, immunoassays, the Kjeldahl method (for organic nitrogen measurement), ELISAs, and miscellaneous, lab-specific methods. Relative to other fields of metabolomics, livestock metabolomics appears to use NMR spectroscopy somewhat more and LC-MS somewhat less. This may simply reflect the availability of instrumentation or the preferences of major research groups in livestock metabolomics. Certainly, sample abundance and supply is not a significant issue in livestock metabolomics so, the use of tools that require higher-volumes, but offer more quantitative results (such as NMR) is not unexpected. However, NMR is not the most sensitive technique and certainly if livestock metabolomics researchers wish to extend their coverage of the livestock metabolome, they will certainly need to makes use of more LC-MS methods. Another gap that was noted in livestock metabolomics research is the near complete absence of ICP (inductively coupled plasma)-MS studies to measure metal ion levels in tissues and biofluids. Indeed, only 2 studies used ICP-MS, with the most complete characterization being conducted by Saleem et al. who reported the identification and quantification of 20 metals in bovine ruminal fluid. The importance of metal ions as micronutrients for animal health and animal productivity cannot be underestimated. Therefore, it is surprising that so little metal ion data has been collected or analyzed in livestock metabolomic studies. It was also noted that the use of fluxomics or the measurement of metabolite flux using stable isotopes is completely absent in livestock metabolomics studies. Fluxomics is particularly useful in understanding metabolic sinks and sources. It is also useful for assessing nutrition and metabolic efficiency—topics, which are obviously important in livestock research. However, to conduct metabolic flux analysis, isotopically labeled (¹³C or ²H) feed needs to be used. Given the size of most livestock animals (relative to rats and mice) and the need for significant quantities of expensive, isotopically labeled feed, fluxomic studies are likely too difficult and costly to perform. Likewise, the use of imaging mass spectrometry or IMS (which is becoming very popular in human metabolomics studies) was completely absent in livestock studies. Imaging mass spectrometry is particularly useful for analyzing tissues and for understanding the metabolic changes that take place during tissue development or tissue transformation. A good metabolomics study should use more than one analytical platform, and ideally as many different (orthogonal) platforms as possible to broaden the metabolite coverage. In our analysis we found that 69% of the published studies used just 1 platform (either NMR, HPLC-UV, LC-MS, GC-MS or ICP-MS), 15% used 2 platforms and only 3% used 3 or more analytical platforms. The remaining 13% of studies used relatively non-conventional platforms or assays (immunoassays, FT- IR, etc.). The most comprehensive metabolomic analysis was a study that used 5 different platforms (NMR, HPLC-UV, LC-MS, GC-MS and ICP-MS) to characterize the bovine ruminal fluid metabolome. Looking through the more recent studies, there is a general trend towards using more than one platform and a growing trend towards using LC-MS techniques over NMR methods. However, the surprisingly high number of livestock metabolomic studies that still use only a single platform also represents a significant issue that the field must remedy. Certainly the trend in human metabolomic studies is to use at least 2 and often 3 or more different analytical platforms. Another gap that was identified from this literature analysis was the general lack of integration of other omics techniques (proteomics, transcriptomics or SNP measurements) with reported livestock metabolomic studies. Indeed, only 5 papers (3 bovine and 2 swine metabolomics studies) used metabolomics in conjunction with genomics or proteomics. One paper of note was an investigation that used genomics and metabolomics to evaluate RFI (residual feed intake) from cross breeds of dairy and beef cattle. This group of researchers used metabolomics and phenotypic data to support their genomics investigations and identified two genes (*TP53* and *TGFB1*) that were strongly associated with cellular functions driving feed efficiency. In another study by Lu and colleagues, the effect of genetic polymorphisms on dairy milk characteristics was evaluated using a combination of metabolomics and proteomics. This paper identified alterations in triglyceride composition and reported changes in the milk metabolome and proteome of dairy cows with the K232A (lysine to alanine substitution) polymorphism in the well-studied *DGAT1* gene. Given the growing trend towards systems biology research and the more “holistic” interpretations of multi-omics data in other fields of life science, the near absence of multi- omics studies represents an important gap in livestock metabolomics (and omics) research. ## Trends and gaps in metabolite quantification The majority of livestock metabolomics publications are non-quantitative or semi-quantitative (yielding relative quantification) while 28.18% of published studies provide fully (absolute) quantitative data. The metabolites tracked in this review were categorized in two main groups: 1) quantified and 2) non- quantified metabolites. Any metabolite that was associated with an absolutely quantified value (millimolar, micromolar, nanomolar, mg/mL, ug/mL, etc.) in a given sample type was placed in the quantified category. The non-quantified group consists of either metabolites with no quantified value or ones that have only relative quantification (i.e. reported as a fraction or a percentage). Over all livestock species and all sample types, we found a total of 404 quantified metabolites and 666 non-quantified. The majority of both quantified and non- quantified metabolites are lipids and lipid-like molecules. Temporal trends in metabolite quantification show that proportionally fewer livestock metabolomics papers are providing quantitative data. For instance, 69% of papers published prior to 1999 had quantitative data, while 34% of papers from 1999–2010 and just 21% from 2011–2015 generated quantitative metabolite data. Overall, livestock metabolomics still has an impressive proportion (\~28%) of publications that report absolute concentration values. In contrast, most other fields of metabolomics quantify metabolites far less frequently. Nevertheless, the steady decline in the proportion of livestock papers providing quantitative metabolomic data is not a good sign. The importance of absolute quantification in metabolomics cannot be over-emphasized. As a branch of analytical chemistry focusing on small molecule characterization, there is more than 100 years of history and a plethora of tools, standards and protocols designed specifically for absolute metabolite quantification. Absolute quantification allows facile comparisons of readings between animals, research staff, platforms, laboratories and countries. Acquiring quantified values also allows one to determine normal and abnormal ranges for disease diagnosis, prediction as well as other relevant production measures. Obtaining quantified data and recognizing normal physiological concentrations is also a requirement in biomarker discovery. Indeed, absolute quantification and the existence of normal and abnormal ranges is the foundation to the entire field of clinical chemistry. In livestock metabolomics, having a “normal” quantified range specific for each animal species or breed is critical for defining referential “healthy” conditions. Likewise, being able to quantify specific changes in an animal’s metabolome allows one to identify “abnormal” conditions such as overt disease, malnutrition, pregnancy difficulties, and most importantly subclinical conditions for which no obvious clinical indicators are visible. ## Trends and gaps in metabolite coverage Based on our analysis of the literature and the definition of a metabolite given earlier, the majority of livestock metabolomics studies report ≤50 metabolites (79% of the total selected metabolomics publications) while the other 21% report \>50 metabolites. The largest number of metabolites (or features) reported in a single paper was 647 covering multiple biofluids for bovine samples while, the fewest reported was 8 (in a variety of papers from all different livestock species). As with metabolite quantification, there is a trend for more recent livestock metabolomics papers to report a greater number of metabolites. For instance, papers published prior to 1999 averaged 29 metabolites per study, those from 1999–2010 averaged 44 metabolites per study while, papers from 2011–2015 averaged 63. Among the later publications, the recent bovine study conducted by Sun et al. who investigated potential biomarkers of milk production and quality using GC-time-of-flight/MS analyses of rumen fluid, milk, serum and urine claimed to detect the highest number of metabolites (i.e., 647). However, careful reading of the manuscript shows that they only formally identified 123. The remaining “metabolites” were unidentified MS peaks or features. In ovine metabolomic studies, Parveen and colleagues reported 168 out of 205 detected metabolites using GC-MS to investigate sheep plasma and feces. Clark et al. reported 97 metabolites out of the 571 detected features in caprine serum using a combination of both GC-MS and LC-MS. In equine metabolomics, the highest number of metabolites identified was from a study conducted by Escalona and colleagues with 102 metabolites identified via NMR analysis of plasma, urine and fecal water. A porcine metabolomics study by Metzler-Zebeli et al. reported 104 out of 132 detected serum metabolites using LC-MS. Overall, our analysis shows a total of 1070 non-redundant or unique metabolites have been detected and/or quantified in the livestock metabolomics literature. Bovine studies covered the majority of detected metabolites (i.e., 768 different compounds) over multiple sample types. Porcine and ovine studies have the next highest number of detected metabolites with 412 and 285 different metabolites, respectively. Caprine and equine studies reported 167 and 109 different metabolites, respectively. The most frequently detected metabolites with \>100 separate entries for different animals, biofluids or conditions include: alanine (124 times), valine (112 times), isoleucine (105 times), glycine (101 times), and lactate (101 times). In addition, 26 other metabolites were reported 50–100 times. Metabolites reported more than once and \<50 times add to 560 while, 479 metabolites are reported only once. It is important to provide some context to these numbers, especially with regard to metabolome studies reported for other animal or model species. The estimated size of the mammalian metabolome is \>100,000 molecules and the total number of metabolites so far reported and/or theoretically expected to be in the human metabolome or HMDB is just over 42,000. While the number of expected or theoretical metabolites is large, the actual number of experimentally identified (and/or quantified) metabolites is actually quite small. For instance, based on the HMDB, the number of experimentally identified metabolites in the human metabolome is 3821, in the *E*. *coli* metabolome it is 891 and in the yeast metabolome it is 625. Among the different livestock species, it is clear that the coverage of the bovine metabolome is quite extensive and is approaching or even exceeding that of other model organisms. However, there is an obvious gap in terms of the coverage of other livestock species with caprine and equine metabolomes being very poorly characterized. Much more work is needed on goat and horse metabolomes to bring them up to the level seen in the bovine metabolome. ## Trends and gaps in animal breeds While we have largely focused on examining metabolomics data for different livestock species, we also noticed some interesting trends with regard to the choice of specific breeds in each livestock species. Similar to other fields of bovine research, the majority (45%) of bovine metabolomic studies use either pure- or cross-bred Holsteins. A smaller amount (11%) of other studies used cross breeds to investigate various aspects of the bovine metabolome. Other common bovine breeds used in metabolomic studies include Charolais (7%) and Jersey (3%). In ovine metabolomic studies, the main breed reported is Suffolk (19%) while other breeds (i.e., Sarda) are reported only once or twice. For caprine metabolomics studies, the preferred breeds have been Norwegian (22%) with other breeds such as Saanen and Alpine being reported only once. Likewise, among equine and porcine metabolomic studies, Standardbred horses (33%) and Landrace sows (22%) were most frequently used. Interestingly, no breed information was provided in 18%, 33%, 22%, 17%, and 9% of the bovine, ovine, caprine, equine, and porcine metabolomics manuscripts, respectively. It is surprising that this essential information is not provided in the manuscripts. This suggests the reporting standards found in livestock metabolomics manuscripts still needs improvement. Based on the above statistics, one of the more obvious gaps in current livestock metabolomics research is the limited variety of breeds being used in most metabolomic studies. The vast majority of the published research appears to be focused on just one or two main breeds i.e., Holstein in cattle, Suffolk in sheep, Standardbred in horses. Evidently, assessing breed differences and their potential impacts on the metabolome has not been a priority for most livestock researchers. However, it is important to remember that the existence of dozens of livestock breeds is a consequence of centuries of selection for very unique phenotypic qualities—some of which are likely determined by their metabolism or metabolome. Different breeds will be characterized by specific production or metabolic parameters and these may be fundamentally different between breeds. While the composition of mammalian (and livestock) metabolomes is likely to be highly similar, metabolite concentrations are expected to differ substantially between different breeds. Identifying the unique aspects affiliated with each breed’s metabolome is therefore, an important component of livestock metabolomics that should be considered in future studies. This is particularly true for purebred and breeding stock herds that are limited to very few animals/herds worldwide. Breeding stock animals provide most of the genetic background found in most commercial herds, which means they have a significant influence on the metabolome associated with their progeny. We were also surprised by the very limited research on the neonatal livestock metabolome. Indeed, we found only 16 neonate metabolomic studies, with 1 study focused on calves, 4 on lambs, 1 on kids, 10 on piglets and no studies on colts or foals. ## Trends and gaps in biomarker discovery One of the strengths of metabolomics lies in its utility for biomarker discovery. Because metabolites can be more easily, cheaply and routinely quantified than most other biological molecules, they are ideal for use in biomarker panels. Indeed, metabolite biomarkers continue to be developed and used in clinical applications at a much greater rate than genes or proteins. In surveying the papers compiled for this review, we found a total of 11 livestock metabolomics papers that proposed candidate biomarkers. This included 5 papers in animal health, 1 in animal nutrition, 2 in animal production, 1 in animal reproduction and 2 for animal models of human health. These studies were limited to cattle, sheep and pigs with no metabolomic biomarker studies being reported for goats or horses. Of these papers, we observed that most reported fewer than 30 candidate biomarkers, with the lowest number being 2. A few reports used higher number of metabolites, i.e., 64, as part of a statistical model to increase the accuracy of prediction. The majority (55%) of metabolomic biomarker papers did not provide any quantitative data, but rather reported only relative metabolite trends (up or down relative to some indeterminate standard). This means that only 5 papers, all from the bovine group, effectively provided useful or verifiable biomarker data. Furthermore, only a single paper reported follow- up validation studies where the initially discovered biomarkers were subsequently validated on a separate cohort of samples. Based on our data, most biomarker studies were conducted with relatively small sample sizes with the majority of studies being done on fewer than 100 animals. The largest biomarker study was one conducted on 321 animals (1587 samples), which investigated prognostic biomarkers of ketosis in dairy cows using NMR spectroscopy. Overall, the quality of biomarker studies done for livestock metabolomics is not particularly good, especially given the standards expected of human biomarker studies. Nevertheless, among the reported biomarker studies, we did find some very interesting and compelling results. One example is a biomarker study of RFI and other feed efficiency traits in beef steers. In this study, NMR spectroscopy was used to identify and quantify plasma metabolites associated with RFI, initially in a discovery population and subsequently in the validation cohort. Karisa et al. reported 3 candidate biomarkers of RFI that significantly (P\<0.05) account for \>30% of the phenotypic variation for this trait. Other metabolites were proposed to be associated with average body weight, average feed intake, dry matter intake and average daily gain. In another interesting study, predictive biomarkers of transition diseases in dairy cows were investigated. This study monitored only 12 dairy cows over four time points during the transition (pre- and post-calving) period. Blood samples were drawn to quantify the metabolome changes associated with various periparturient diseases post-calving. Using direct flow injection (DFI)-MS, Hailemariam and colleagues profiled 120 blood metabolites of which 3 were suggested as candidate biomarkers for transition diseases, with a sensitivity and specificity of ≥85%. Another study reported by Gray et al. looked into biomarkers associated with vaccine efficacy. Using UPLC- MS metabolomic measurement of plasma derived from Holstein male calves, Gray and colleagues found 12 metabolites that were altered post-vaccination. These biomarkers are being proposed as a newer, more efficient route to optimise vaccination and to make vaccine formulation and benchmarking much more efficient and targeted. This paper emphasizes on the importance of disease prevention and vaccination procedures in livestock, especially in using new technologies such as metabolomics to enhance evaluation of vaccine efficacy. Identification of biomarkers will not only improve disease diagnosis but also allow the opportunity for disease prediction prior to manifestation of clinical signs. For example, if a metabolic disorder can be predicted well before (sub)clinical manifestation, farmers can make informative decisions with regards to their management, feeding, housing, etc., to change the cascade of biological events leading to that disease. Predictive attempts of such can make a significant financial and sustainability difference by maintaining production quantity and quality, saving on costs associated with treatment, veterinary visits, preventing animal culling and thus, maintaining longevity. ## The livestock metabolome database In assembling the material for this review, we identified a total of 1070 metabolites that have been detected and/or quantified in livestock metabolomic studies of cattle, sheep, goats, horses and pigs. This information has been systematically categorized into LMDB with all of the metabolites being fully described including information about the degree or quality of quantification (i.e., quantified, non-quantified) and the source sample types for each livestock species. All of the metabolites with quantitative data had their concentrations converted into a standardized concentration unit (i.e., μM) to improve consistency. In addition to the chemical data and source information, an abbreviated description of the experimental context for each metabolite was extracted from the articles and included in the online database (<http://www.lmdb.ca>). This information includes data on the analytical platform(s), experimental conditions, field of research, and animal breed used in acquiring the metabolomic data. All metabolites are linked to a standard HMDB (<http://www.hmdb.ca/>) identification number, which provides a freely- accessible and detailed description of the metabolite. A PubMed and/or DOI identifier is also associated with each metabolite entry, which provides a literature reference or a direct link to the article reporting that metabolite for readers who are interested in further details. Only those metabolites that had reasonably complete descriptions (i.e., unique chemical names, sample types, source information, etc.) were included in the online database. A number of metabolites or “features” were identified during the review process but not included in the LMDB. These include those compounds that have either not been characterized at all (no chemical name, no data on sample types), or not fully characterized (unknown or undefined chemical structure). This collection of 415 “unknown” metabolites will be added to the LMDB once we can obtain sufficient structural and sample source information on them. Among the metabolites entered into the LMDB, 404 compounds were quantified and 666 were not. On a species level there were 768 bovine metabolites, 285 ovine metabolites, 167 caprine metabolites, 109 equine metabolites, and 412 porcine metabolites. Detailed descriptions of each compound are provided in the LMDB “metabocard” pages. Likewise, structural images, molecular formulas, names and synonyms, chemical classification/taxonomy information, physicochemical data (molecular weights, pI’s, pKa’s, boiling/melting points), referential spectral data (both experimental and theoretical NMR, MS/MS and EI-MS spectra), links to other online databases and full reference (authors, journals, volumes, etc.) information is also provided. The LMDB has been designed so that it can be easily browsed and it supports searches through standard text queries as well as via structure, mass, and spectral queries. Most of the information in the LMDB is hyperlinked to other resources within the LMDB, allowing for a more convenient and compact route to access the data. The LMDB is available at <http://www.lmdb.ca>. This database will be constantly updated with more metabolites and more detailed metabolite descriptions as more research in livestock metabolomics is published. By assembling the LMDB and making this information freely available through both the web and this manuscript, we hoped to create a referential resource that other livestock researchers could readily use. Our past experience in assembling and maintaining the Human Metabolome Database (HMDB) clearly showed how useful a centralized, on-line resource could be in the field of human metabolomics. Therefore, our expectation with the LMDB is that it will have a comparable impact on the field of livestock metabolomics. Indeed, we believe that establishing a comprehensive repository that stores and categorizes livestock metabolome information into a standardized format will be critical for future livestock research. It will also be important for identifying potential livestock disease biomarkers, improving animal selection (via metabolomic assays), enhancing animal nutrition and understanding novel biochemical mechanisms arising from various physiological perturbations. With more and more livestock metabolomics papers appearing each year and the continued growth in metabolite coverage, it will be challenging to maintain the LMDB. However, without even attempting to create the LMDB we suspect that livestock metabolomics would continue to lag behind the metabolomics activities seen in other areas (i.e. human, plant crops, microbes, food/beverage studies) and would face significant hurdles in the coming years trying to catch up. # Conclusion Metabolomics is less than 15 years old, yet it has already delivered some remarkable achievements. This includes significant improvements in the ability to identify many environmental contaminants and toxins, significant advances in food and nutrient characterization, the identification of many novel biomarkers for disease risk including risk markers for diabetes, heart disease and cancer as well as promising leads for a variety of drugs and therapies. Metabolomics is also well-positioned to provide some important advances in both livestock research and the livestock industry, especially as it relates to livestock health, breeding and production. A number of examples were highlighted in this review including metabolome discovery for normal metabolite composition and concentrations, identification of biomarkers of transition diseases as well as production traits in dairy and beef cattle with the goal of introducing prognostic strategies in animal health as well as increasing prediction accuracies. Our observations also showed that a wide variety of biofuids have received attention for metabolomics research such as metabolic profiling of milk, plasma, serum, and urine, minimizing animal welfare concerns. However, in order for livestock metabolomics to deliver on the promise and the excitement seen in other areas of metabolomics research, it is important to carefully assess what has been accomplished, what is known and what still needs to be done. The intent of this review was to provide a critical overview of the trends and gaps in livestock metabolomics research. Specifically, we sought answers to 4 key questions: 1) What are the most common applications of metabolomics in animal science and where are they trending?, 2) What are the preferred metabolomics technologies livestock metabolomics and how are they evolving?, 3) What are the most obvious gaps or weaknesses in livestock metabolomics relative to other fields of metabolomics research? and 4) What are the known or measured metabolites for the 5 major livestock species (i.e., bovine, ovine, caprine, equine, and porcine) in different tissues and biofluids? In addressing the first 3 questions we focused on areas relating to: 1) Animal Choices; 2) Research Applications; 3) Sample Size; 4) Sample Type; 5) Instrumentation and Methodologies, 6) Quantification; 6) Metabolite Coverage; 7) Animal Breeds, and 8) Biomarker Identification. In many cases we were able to identify some clear trends while at the same time identifying important shortcoming or areas where further improvements could be made. It was apparent that livestock metabolomics appears to be ahead with regard to metabolite quantification, the diversity of research applications and its efforts in biomarker identification. On the other hand, it was also clear that livestock metabolomics (especially with regard to sample size, instrumentation and metabolite coverage) was lagging somewhat further behind than human, microbial or plant crop metabolomics. Based on our assessment of the shortcomings with current livestock metabolomics studies, it is clear that future metabolomics research should focus on expanding or extending metabolome discovery using healthy control animals, increase sample numbers, direct more effort towards metabolite quantification, perform more integrated multi-omics experiments, use a greater variety of analytical platforms or techniques (including ICP-MS, MSI and fluxomics), increase the breadth of metabolite coverage (by using more sensitive platforms, such as ESI- MS), investigate a greater and new varieties of biosamples such as semen, amniotic fluid, saliva and urine, extend the number and types of animal breeds used in metabolomic studies and be more conscientious in the design and implementation of biomarker studies. Another important outcome of this study was the collection and consolidation of livestock metabolite information into a single, centralized resource (the LMDB). It became readily apparent in conducting this review that the livestock metabolomics literature is highly diffuse and that valuable information is being “lost” or is not readily available. By compiling the LMDB and making an on-line version of the database freely available, we hope it could serve as a hub for livestock researchers and the livestock industry to further advance the field of livestock metabolomics. # Supporting information [^1]: The authors have declared that no competing interests exist. [^2]: **Conceptualization:** SAG DSW MAS GSP. **Data curation:** SAG. **Formal analysis:** SAG ACG TS. **Funding acquisition:** SAG DSW GSP. **Investigation:** SAG. **Methodology:** SAG DSW. **Project administration:** SAG DSW. **Resources:** SAG DSW. **Software:** SAG ACG TS. **Supervision:** SAG DSW. **Validation:** SAG. **Visualization:** SAG. **Writing – original draft:** SAG. **Writing – review & editing:** SAG DSW MAS GSP.

# Introduction 2-Methoxyestradiol (2ME2), an endogenous 17β-estradiol analogue, has been shown to exert antiproliferative- and antiangiogenic effects on various tumorigenic lines including sarcoma, prostate- and breast tumorigenic cell lines. However, preclinical rodent breast cancer studies and clinical trial data revealed that 2ME2 is not effective as an anticancer therapy due to its low bioavailability and rapid metabolism. Therefore, derivatives of 2ME2 such as 2-methoxyestradiol- bisulphamate (2-MEBM) have been designed and have shown encouraging results in preclinical MDA-MB-231 rodent cancer models where tumour growth inhibition and resistance to metabolism were observed. A number of research groups including ours have designed several novel 2ME2 derivatives, most of which include a sulphamoyl moiety. The sulphamoyl moiety reversibly interacts with carbonic anhydrase II allowing the sulphamoylated compound to bypass first pass liver metabolism and in so doing increases bioavailability by reducing its metabolism. These compounds are highly effective in inducing cell death in a number of cancer cell lines. In previous reports we have demonstrated that the 2ME2 derivative (8*R*,13*S*,14*S*,17*S*)-2-ethyl-13-methyl- 7,8,9,11,12,13,14,15,16,17-decahydro-6H-cyclopenta\[a\]phenanthrane-3,17-diyl bis(sulphamate) (EMBS) possesses antiproliferative effects in a variety of tumorigenic cell lines including breast tumorigenic lines including the MCF-7, MDA-MD-435 and MDA-MD-231, ovarian adenocarcinoma cell line OVCAR-3, renal carcinoma cell line SN12-C and prostate carcinoma cell line DU-145. Furthermore, similar to 2ME2, EMBS exposure blocked cell cycle progression in the G₂/M phase. We showed that EMBS exposure led to mitochondrial membrane depolarisation, caspase 6 and caspase 7 activation, while the extrinsic apoptotic pathway was also activated through caspase 8. Furthermore, 2ME2 exposure also resulted in increased reactive oxygen species (ROS) generation including hydrogen peroxide and superoxide. 2ME2-dependent ROS induction was also responsible for endoreduplication in nasopharyngeal carcinoma cells via c-Jun N-terminal kinase (JNK) and mitogen-activated protein kinase family (MAPK) activation. Controversy still exists around the mechanism through which 2ME2 increases ROS production. 2ME2 exposure in Ewing sarcoma cells resulted in reduced mitochondrial membrane potential, increased ROS production and activated the JNK pathway. Moreover, pre-treatment with JNK inhibitors or antioxidants largely reduced the effects of 2ME2 on cell proliferation suggesting that both ROS and the JNK pathways are involved in the mechanism of action exerted by 2ME2. However, in a separate report, 2ME2 exposure in rat sarcoma cells resulted in decreased mitochondrial membrane potential and an increase in ROS. Co-treatment with various antioxidants did not reduce the effect of 2ME2 on cell survival. Moreover, inhibition of caspases did not diminish the effect of 2ME2 further suggesting that, in these cells, 2ME2 induces apoptosis independently of caspases and ROS production, but still through a mitochondrial pathway. ROS, and specifically superoxide metabolism, is regulated by the superoxide dismutase (SOD) family. Researchers have reported that ROS is induced after exposure to 2ME2 as a result of the inhibition of SOD activity. However, Kachadourian., *et al* (2001) demonstrated using different assays to determine SOD activity that 2ME2 does not inhibit SOD activity. Furthermore, they speculate that 2ME2 itself could act as a free radical after conversion to a semiquinone radical and through reaction with oxygen could form superoxide. One of the major effects of oxidative stress is the formation of DNA adducts leading to DNA damage including the formation of double stranded DNA breaks (DSB). Double stranded breaks are restored through a programme of non-homologous end joining. One of the proteins involved in this process is the histone H2A which is phosphorylated at the sites of DSB to act as a beacon for the assembly of the restorative protein complex. This feature is used in cellular assays to assess the formation of DSB by quantifying H2A phosphorylation. Various sulphamoylated 2ME2 derivatives have been synthesised in order to improve on the anticancer potential of 2ME2. However, except for the effects that have been measured after exposure to these compounds, conclusive data do not exist that provide a mechanism for their action within the cell. In this report the effect of EMBS exposure on ROS production was analysed. In this report we show that ROS increase is an early event after EMBS exposure. Moreover, by using antioxidants it is shown that ROS production is essential for the major effects of EMBS exposure including cell cycle arrest, mitochondrial membrane potential effects and apoptosis induction. This indicates that ROS is involved in upstream events required for JNK activation and cell cycle arrest. Furthermore, initial ROS generation is needed for subsequent loss of mitochondrial membrane potential and the secondary increase in ROS production. # Aim and design of the study This study investigated the mode of action utilized by EMBS as a representative of the sulphamoylated 2ME2 derivatives. Moreover, the role of ROS in the induction of ROS was demonstrated to induce cell death via apoptosis. This research project is regarded as a preclinical *in vitro* study and data cannot directly be extrapolated to an *in vivo* environment. # Materials and methods ## Cell lines The estrogen receptor-positive MCF-7 cell line and the estrogen receptor- negative metastatic MDA-MB-231 cell line were obtained from Cellonex. (Johannesburg, South Africa). Both cell lines were propagated and maintained in a humidified atmosphere at 37**°**C and 5% CO₂ in Dulbecco’s Minimum Essential Medium Eagle (DMEM) supplemented with 10% heat-inactivated fetal calf serum (FCS), 100 U/ml penicillin G, 100 μg/ml streptomycin and fungizone (250 μg/l). Non-tumorigenic spontaneously immortalized adherent human breast epithelial MCF-12A cells were a gift from Professor Parker (Department of Medical Biochemistry, University of Cape Town, Western Cape, South Africa). MCF-12A cells were cultured in growth medium consisting of a 1:1 mixture of DMEM and Ham’s-F12 medium supplemented with 20 ng/ml epidermal growth factor, 100 ng/ml cholera toxin, 10 μg/ml insulin and 500 ng/ml hydrocortisone, 10% heat- inactivated FCS, 100 U/ml penicillin G, 100 μg/ml streptomycin and fungizone (250 μg/l). ## Reagents All required reagents of cell culture analytical grade were purchased from Sigma (St. Louis, United States of America) unless otherwise specified. Heat- inactivated FCS, sterile cell culture flasks and plates were purchased from Sterilab Services (Kempton Park, Johannesburg, Gauteng, South Africa). Penicillin, streptomycin and fungizone were obtained from Highveld Biological (Pty) Ltd. (Johannesburg, Gauteng, South Africa). The lactate dehydrogenase assay kit, annexin V fluorescein isothiocyanate (FITC) kit and Mitocapture mitochondrial apoptosis detection kit were purchased from BIOCOM biotech (Pty) Ltd. (Clubview, Gauteng, South Africa). Hydroethidine, 2,7-dichlorofluorescein diacetate (DCFDA), thymidine, JNK inhibitor (SP600125), N-acetyl cysteine (NAC), triton X-100, paraformaldehyde and fluoromount aqueous mounting fluid were acquired from Sigma (St. Louis, United States of America). Monoclonal mouse anti-human phospho H2A antibody and LC3B II monoclonal antibody were obtained from Abcam (Cambridge, United Kingdom) and the secondary goat anti-mouse horseradish peroxidase antibody was supplied by KPL (Maryland, United States of America). EMBS is commercially unavailable and was synthesized by iThemba Pharmaceuticals (Pty) Ltd. (Modderfontein, Gauteng, South Africa). A stock solution of EMBS dissolved in dimethyl sulphoxide (DMSO) was prepared with a concentration of 10 mM and was stored at 4°C. Previous studies conducted in our laboratory demonstrated optimal antiproliferative activity after exposure to 0.4 μM EMBS for 24 h. For this reason, for all subsequent experiments, cell lines were exposed to 0.4 μM EMBS. ## Methods ### Reactive oxygen species (ROS) generation Hydrogen peroxide generation was assessed by measuring the fluorescence of the probe DCFDA, while superoxide generation was assessed by hydroethidine. DCFDA is a non-fluorescent probe, which, upon oxidation by hydrogen peroxide, hydroxyl radical and peroxides is converted to the highly fluorescent derivative, DCF. DFCDA does not respond to nitric oxide, superoxide or hypoclorous acid. Thus, DCFDA does not directly quantify hydrogen peroxide, but an increase in DCF fluorescence suggests an increase in hydrogen peroxide generation. Superoxide oxidizes hydroethidine to form fluorescent 2-hydroethidine cation, but not ethidium. Exponentially growing MCF-7, MDA-MB-231and MCF-12A cells were seeded at 500 000 cells per 25 cm² flask. After a 24 h attachment period, medium was discarded and cells were exposed to medium containing 0.4 μM EMBS for different times. At termination cells were trypsinized and 1x10⁶ cells were resuspended in 1 ml PBS. Cells were incubated with 20 μM DCFDA for 25 minutes or 10 μM hydroethidine for 15 minutes at 37°C. DCF and the 2-hydroethidine cation fluorescent product fluorescence was measured with a FC500 System flow cytometer (Beckman Coulter South Africa (Pty) Ltd). Information generated from at least 10 000 cells were analyzed by means of Cyflogic version 1.2.1 software (Pertu Therho, Turko, Finland). ### Cell number determination The time-dependent influence of EMBS exposure on proliferation was investigated by means of crystal violet staining as described previously. Crystal violet allows for the quantification of the cell number in monolayer cultures as a function of the absorbance of the dye taken up by the cells. MDA-MB-231 cells were seeded in 96-well tissue culture plates (5000 cells/well) and were incubated for 24 h to allow for attachment. Medium was discarded and cells were incubated in medium containing 0.4 μM EMBS for different timepoints. Cells were fixed with 1% gluteraldehyde (100 μl) for 15 minutes at room temperature. Gluteraldehyde was discarded and cells were stained using 100 μl 0.1% crystal violet at room temperature for 30 minutes. Excess crystal violet was discarded and the 96-well plate was washed by rinsing the plate under running water for 2 minutes. Triton X-100 (0.2%; 200μl) was added and incubated at room temperature for 30 minutes. Solubilised crystal violet solution (100 μl) was transferred to a new microtitre plate. The absorbance was determined at 570 nm using an EL_x800 Universal Microplate. ### Mitochondrial membrane potential Mitochondrial membrane potential was measured using Mitocapture technology. The mitochondrial membrane potential is an indicator of mitochondrial function. Reduced mitochondrial membrane potential has been implicated in intrinsic apoptosis induction that is dependent on ROS induction. MDA-MD-231 cells (500 000) were seeded in 25 cm² flasks. After 24 h, cells were exposed to 0.4 μM EMBS for different times. Subsequently, cells were detached using trypsin and centrifuged at 500 *g*. Cells were resuspended in 1 ml of diluted Mitocapture solution (1 μl mitocapture: 1 ml pre-warmed incubation buffer) and incubated in a humidified atmosphere (37°C, 5% CO₂). After 20 minutes, cells were centrifuged at 500 x *g*, the supernatant discarded and cells resuspended in 1 ml of pre-warmed incubation buffer (37°C). Cells were analyzed with a FC500 System flow cytometer. Data from at least 10 000 cells were analyzed by means of Cyflogic version 1.2.1 software. ### Apoptosis induction Apoptosis induction was analysed using flow cytometry and annexin V-FITC. MDA- MB-231 cells (500 000) were seeded in 25 cm² flasks. After exposure to 0.4 μM EMBS, cells were trypsinised and cells were resuspended in 1 ml of 1x binding buffer and centrifuged at 300 x *g* for 10 minutes. Supernatant was removed and cells were resuspended in 100 μl of 1x binding buffer. Annexin V-FITC (10 μl) was added and incubated for 15 minutes in the dark at room temperature. After 15 minutes, cells were washed by adding 1 ml of 1x binding buffer and centrifuged at 300 x *g* for 10 minutes. Supernatant was carefully removed and cells were resuspended in 500 μl of 1x binding buffer solution. Immediately prior to analysis, 12.5 μl of propidium iodide (PI) (40 μg/ml) was added and samples were mixed gently and analysed using the FC500 System flow cytometer. Data from at least 10 000 cells were analyzed with CXP software. Distributions of cells within the quadrants were calculated with Cyflogic version 1.2.1 software. ### Mitochondrial metabolic activity The influence of EMBS on metabolic activity was demonstrated over time using Alamar Blue (also known as resazurin) and fluorometry. Alamar blue is stable and non-toxic and thus enables the continuous quantification of metabolic activity over time. Alamar blue (non-fluorescent and blue) is converted by mitochondrial enzymes to the reduced form, resorufin (highly fluorescent and red). MDA-MB-231 cells were seeded at 5 000 cells/well into 96 well plates. After 24 h, cells were exposed to 5% Alamar Blue and 0.4 μM EMBS. Samples were incubated at 37°C and fluorescence was measured at an excitation wavelength of 570 nm and reference wavelength 600 nm at different timepoints. ### Cell cycle progression Flow cytometry was utilized to investigate the deoxyribonucleic acid (DNA) content to determine the effects of EMBS on cell cycle distribution and confirmation of apoptosis induction. MDA-MB-231 cells (500 000) were seeded in a 25 cm² flask. Medium was discarded and cells were incubated in medium containing 0.4 μM EMBS for different timepoints. Subsequently, cells were trypsinized and 10⁶ cells were centrifuged for 5 minutes at 300 x *g*. The pellet was resuspended twice in ice-cold phosphate buffer solution (PBS). The supernatant was discarded and cells were resuspended in 200 μl of ice-cold PBS containing 0.1% FCS. Ice-cold 70% ethanol (4 ml) was added in a drop-wise manner and cells were stored at 4°C for 24 h. Cells were pelleted by centrifugation for 5 minutes. The supernatant was removed and cells were resuspended in 1 ml of PBS containing PI (40 μg/ml), 0.1% triton X-100 and RNase A (100 μg/ml) and incubated at 37°C, 5% CO₂ for 45 minutes. Cells were analysed by means of FC500 System flow cytometer. Data from at least 10 000 cells were captured with CXP software and analyzed with Cyflogic software. ### H2A phosphorylation Oxidative stress frequently results in DNA damage associated with phosphorylated H2A. H2A is phosphorylated, activated and recruited to sites of double stranded breaks (DSB) after oxidative damage as part of the DNA repair mechanism. For confocal imaging of phosphorylated H2A, cells (50 000) were plated on sterile glass coverslips in 24 well plates. The next day, cells were exposed to 0.4 μM EMBS. Cells were exposed for 4 h and 24 h respectively and were subsequently fixed in 2% paraformaldehyde for 15 minutes. Slides were washed with PBS and permeabilized with 0.2% triton X-100. Coverslips were subsequently blocked with 2% BSA in PBS for 1 h, incubated with primary antibody against phosphorylated H2A for 1 h at room temperature, washed with PBS, and incubated with FITC- conjugated anti-mouse secondary antibodies and 4',6-diamidino-2-phenylindole, dihydrochloride (DAPI) for 1 h at room temperature. After washing with PBS, coverslips were mounted in mounting fluid. Slides were examined using a Zeiss LSM510 Meta confocal microscope furnished with a 63x magnification oil objective. Representative images were made of each condition and the experiment was repeated three times. Images were analysed using Image J software where after global adjustment H2A phosphorylation was quantified using rawIntDev values divided by number of cells per image. At least 100 cells were measured per condition in each experiment. ### Autophagy induction During autophagy organelles and proteins are sequestered into autophagic vesicles that are subsequently degraded through fusion with lysosomes and recycled for sustainable cell maintenance. In order to measure the induction of autophagy microtubule-associated protein B-light chain 3 II (LC3BII) levels were determined by western blot analysis. MDA-MB-231 cells (200 000) were plated in 6 well plates, grown overnight and exposed to EMBS for 4 h, 10 h, 14 h and 24 h. As a positive control for LC3B II accumulation cells were treated with 100 nM bafilomycin A overnight. At termination, medium was removed and cells were carefully washed once in PBS. Lithium dodecyl sulphate (LDS) loading buffer with 2.5% β-mercaptoethanol was heated to 80°C and 150 μl of sample was plated into each well. Cells were scraped together and passed through a syringe to help lyse cells. Cell lysates were heated to 80°C for 10 minutes and then centrifuged at 11000 rpm for 15 minutes. Supernatants were loaded onto 4–12% SDS-PAGE gels. After resolving proteins on gel, they were blotted onto a 0.2 μm polyvinylidene fluoride (PVDF) membrane overnight at 60 V. Membranes were blocked in 2% blocking reagent before being incubated with 1:1000 dilution of monoclonal LC3B antibody or 1:5000 dilution of the monoclonal actin antibody for 1 h and secondary goat anti-mouse horseradish peroxidase antibody for 1 h with washes in between with PBS-Tween. Proteins were visualised using enhanced chemiluminescence substrate and the Biorad Chemidoc MP system (Bio-Rad Laboratories, Inc. California, United States of America). Bands were quantified using Imagelab software and ratios of LC3B II to actin were calculated using Image Lab Software developed by Bio-Rad Laboratories, Inc. ## Statistics For all experiments data were obtained from 3 independent experiments. Graphs represent the mean of three experiments with error bars depicting the standard error of the mean. Each independent experiment for crystal violet staining and extracellular lactate dehydrogenase quantification had a technical repeat of 6. Each independent Alamar Blue experiment had 3 technical repeats. Data were statistically analyzed for significance using a two-tailed Student’s *t*-test. *P*-values smaller than 0.05 were regarded as statistically significant and were indicated by an asterisk (\*). Fluorescent measurement was expressed as a ratio of the value measured for the EMBS-treated cells compared to vehicle-treated exposed cells (mean relative fluorescence). Flow cytometry analysis involved data from at least 10 000 events that were repeated thrice. Bands produced in western blot were quantified using Imagelab developed by Bio-Rad Laboratories, Inc. (California, United States of America) and ratios of LC3B II to actin were calculated. # Results ## EMBS induces oxidative stress in breast cancer cell lines The influence of EMBS (0.4 μM) on ROS was initially investigated by determining the intracellular levels of hydrogen peroxide and superoxide 24 h after exposure. At this time both, hydrogen peroxide and superoxide generation increased in MCF-7 cells (40% and 27%), MDA-MB-231 cells (46% and 33%) and in the non-tumorigenic MCF12A cell line (56% and 25%). To identify the sequence of events resulting in EMBS-induced apoptosis, ROS production was measured over time. MDA-MD-231 cells were exposed to 0.4 μM EMBS and hydrogen peroxide and superoxide production were measured at several timepoints up to 24 h after exposure. While the absolute levels of hydrogen peroxide and superoxide varied, a similar trend was observed over time for both ROS species. Both increased significantly over the first 4 h with hydrogen peroxide levels increasing rapidly up to 1.7 times above the control and then plateauing, while superoxide levels increased more gradually over the same time up to 1.3 times above initial levels. Hydrogen peroxide levels decreased suddenly and dramatically at 10 h (down to 0.5 compared to initial levels) only to recover from 12 h onwards. These levels subsequently increased consistently over the remaining time up to nearly 3-fold over initial measurements. Superoxide levels also decreased after 10 h, but more gradually and less pronounced with a recovery followed by a plateau at 1.7 fold over initial measurements after 16 h. ## EMBS treated cells undergo apoptosis correlated with decreased mitochondrial membrane potential To determine whether oxidative stress is a precursor for EMBS-induced apoptosis via mitochondrial membrane potential loss, both mitochondrial membrane potential and apoptosis levels were measured over time. Firstly, cell survival was measured using crystal violet in cells exposed to 0.4 μM EMBS. Data showed that there was a gradual decrease in proliferation starting at 2 h after exposure that continued to 6 h and eventually to 24 h with a loss in cell number compared to cells propagated in growth medium to approximately 50%. Mitochondrial membrane potential depolarisation statistically significantly increased by about 10% over the first 6 h after exposure to EMBS. A gradual incline (20%) in reduced mitochondrial membrane depolarisation was subsequently measured until 24 h. Apoptosis was measured using PI/Annexin V. Early and late apoptosis sharply increased after 4 h after which early apoptosis increased further, while late apoptosis remained stable. The population of cells present in early apoptosis dropped 18 h after exposure and late apoptosis increased indicating cells were committed to the apoptotic pathway. Data revealed minimal induction of necrosis. Alamar blue was used to determine if EMBS influences the metabolic activity of the cell. Alamar Blue reduction decreased steadily from 2 h onwards after EMBS exposure ending at a 30% decrease in fluorescence after 24 h. Therefore, EMBS exposure resulted in increased ROS production that is tightly correlated with decreased mitochondrial membrane potential and electron transport efficiency along with increased apoptosis. ## Oxidative stress is essential for EMBS induced cell death As discussed above there are opposing data regarding the role of ROS in 2ME2-induced apoptosis. Therefore, in this study ROS production induced by EMBS was inhibited by co-treating cells with the antioxidant, NAC. Cleavage of NAC at the acetyl group liberates reduced cysteine resulting in sustained production of glutathione, an essential antioxidant. Glutathione neutralizes ROS by providing a sulfyhydryl group which serves as a reducing equivalent. Glutathione also functions as a substrate in order to promote the antioxidant activity of glutathione peroxidases. To assess whether the antioxidant NAC inhibits EMBS- induced ROS, cells were co-treated with NAC and EMBS and hydrogen peroxide levels were measured. Hydrogen peroxide levels were increased within 4 h after EMBS treatment, but co-treatment with NAC abrogated this increase. To determine whether there is a causal link between ROS production by EMBS and cell death, cells were treated with a concentration range of NAC in the presence of EMBS and the effects on proliferation were measured using crystal violet. Data revealed an increase in cell viability that was correlated with increased concentrations of NAC. Furthermore, NAC abolished mitochondrial membrane potential depolarisation induced by EMBS after 4 h and 24 h. Previous studies have shown that EMBS-treated cells accumulate at the G₂/M border in the cell cycle. To determine if the block at the G₂/M border is essential for EMBS induced cell death, cells were prevented from accumulating at the G₂/M border by pretreatment with thymidine to block cells at the G₁/S border instead. Subsequently, cells were treated with NAC in the presence of EMBS in order to inhibit ROS accumulation. Crystal violet data showed that EMBS reduced cell viability by 50% in non-blocked cells. In addition, cell viability of thymidine-blocked cells was reduced by 40% after EMBS treatment. In both cycling cells and thymidine-blocked cells, 20 mM NAC treatment completely abrogated the effect of EMBS suggesting that ROS production is essential for the induction of cell death after EMBS, while the blockage of cells at the G₂/M border is not an essential prerequisite for the induction of cell death. Moreover, data indicate that thymidine had no effect on hydrogen peroxide generation after EMBS exposure suggesting that ROS generation by EMBS treatment is upstream from the cell cycle arrest. ## Oxidative stress induced by EMBS is responsible for the G₂/M cell cycle block leading to endoreduplication To determine if oxidative stress induced by EMBS is correlated to the EMBS- induced cell cycle block at the G₂/M border, the cell cycle status of cells were determined over time. Within 6 h after exposure a rapid accumulation of cells at G₂/M was observed along with a reduction in cells in the G₁ phase. A larger reduction of cells in the G₁ phase was observed from 18 h onwards that correlated with an increase in cells present in the sub-G₁ population suggesting that apoptosis was induced at this time. This data suggest that there is a close correlation between the initial increase in oxidative stress and the increase in cells at the G₂/M border. Furthermore, cell cycle data revealed that exposure to EMBS resulted not only in the blockage of cells at the G₂/M border, but also an increase of cells with a DNA content of more than 4N. The \>4N content increased even more prominently after exposure to EMBS for 48 h, while the number of cells blocked at G₂/M were reduced when compared to cells exposed for 24 h. There was also an increase in the sub- G₁ population. Importantly, when cells were co-treated with the antioxidant NAC, both the G₂/M block and the presence of \>4N cells were abolished. When cells were pretreated with a JNK inhibitor, the induction of a G₂/M block was largely abolished after 24 h and 48 h although it was less effective than the addition of NAC. This data suggests that EMBS induces a G₂/M block in cells via the induction of ROS which ultimately leads to cells death or endoreduplication. To determine whether JNK activation is involved in EMBS- dependent ROS production, hydrogen peroxide levels were quantified in EMBS exposed cells co-treated with the JNK inhibitor. Data demonstrated that the JNK inhibitor had no effect on hydrogen peroxide levels induced by EMBS, suggesting that ROS production occurs upstream from JNK activation after EMBS exposure. ## EMBS exposure leads to H2A phosphorylation and nuclear deformities The blockage in cell cycle progression and the induction of endoreduplication by EMBS can be a result of genomic instability brought about by ROS induced DNA double stranded breaks. DSB attract the protein H2A which upon binding is phosphorylated. Therefore, DSB can be quantified by measuring H2A phosphorylation. Cells exposed to EMBS for 4 h or 24 h were analysed for H2A phosphorylation by means of confocal microscopy. Expression was normalised to cell number using DAPI and results show that after 4 h of EMBS exposure there was a small, statistically significant increase in H2A phosphorylation which was blocked by co-treatment with NAC. Interestingly, after 24 h, DAPI staining of the nuclei revealed an accumulation of cells with deformed nuclei resembling multiple nuclei or micronuclei similar to that seen after radiation exposure. These nuclear deformities are also present when cells undergo endoreduplication. Again this effect was abrogated by the co-treatment of cells with 20 mM NAC. Therefore, EMBS exposure leads to a ROS-dependent increase in DNA DSB and endoreduplication. ## EMBS does not induce autophagy While autophagy is linked to starvation-induced survival there is data suggesting that starvation may induce 5' adenosine monophosphate-activated protein kinase (AMPK)-dependent autophagy induction via ROS production. To determine if EMBS-induced ROS production leads to increased autophagy, LC3B II protein levels were analysed. Cells were exposed to 0.4 μM EMBS for the time points indicated and the ratio of LC3B II to actin expression was determined. Western blot data indicated that EMBS exposure did not result in increased levels of LC3B II protein. As expected, bafilomycin A exposure did lead to an increase in LC3B II levels, since it inhibits the fusion of autophagosomes with lysosomes and the flux of LC3B. Therefore, EMBS does not induce LC3B II levels suggesting that autophagy is not involved in EMBS-induced cell death. # Discussion Previous studies have shown that EMBS inhibits proliferation and cell cycle progression by blocking cells at the G₂/M border. It has been proposed that these sulphamoylated compounds function by targeting the microtubules based on *in vitro* microtubule polymerisation assays. However, studies on the mechanism of action pertaining to 2ME2 have shown an essential role for ROS production in the induction of cell death. Therefore, this study investigated the early events after EMBS exposure to determine the role of ROS production on the cellular effects exerted by the compounds. Moreover, we elucidated a pathway leading from EMBS exposure to increased ROS production resulting in several effects including mitochondrial membrane depolarisation, metabolic activity reduction, G₂/M arrest, JNK-dependent apoptosis and endoreduplication. To determine if EMBS induces ROS production both hydrogen peroxide- and superoxide levels were quantified. Data indicate that EMBS increases ROS production significantly after 24 h in several breast cell lines. A more detailed analysis revealed a biphasic temporal induction of ROS after EMBS exposure. The initial ROS increase was correlated with reduced mitochondrial membrane potential, reduced metabolic activity and a small decrease in cell proliferation. Loss of mitochondrial membrane potential has been proposed to be essential for an increase in ROS observed in 2ME2-exposed cells, since treatment with an inhibitor of the mitochondrial respiratory chain, rotenone, inhibited 2ME2-induced ROS and apoptosis. Unexpectedly, we observed a substantial decrease in ROS production approximately 10 h after EMBS exposure. We speculate that the initial increase in ROS levels is due to a direct effect exerted by EMBS which initiates the signaling pathway resulting in the subsequent depolarisation of the mitochondrial membrane leading to the second increase in intracellular ROS generation. Furthermore, the decrease in ROS production after 10 h exposure to EMBS might represent the cells’ attempt to neutralize the ROS production through antioxidant defence activity. The later and more substantial ROS increase is correlated with a cell cycle arrest at the G₂/M border and increased apoptosis induction. This suggests that ROS production is tightly correlated with the G₂/M arrest and apoptosis induction. The possibility that ROS is not only correlated to, but is causally linked to the G₂/M arrest and apoptosis induction, was subsequently investigated. This was done by exposing cells to EMBS and NAC concurrently and determining the effects on the mitochondrial membrane potential, cell proliferation and cell cycle progression. The effects of EMBS on proliferation, mitochondrial membrane potential, cell cycle arrest and apoptosis induction were abrogated suggesting that ROS production by EMBS is essential for the induction of cell death via apoptosis and cell cycle arrest. While this study has demonstrated that ROS production is essential for cell cycle arrest and apoptosis induction by EMBS, the mechanism through which ROS is increased remains unclear. Reports have suggested that the increase in ROS after 2ME2 exposure is due to the inhibition of the superoxide dismutase SOD2. However, others have shown that *in vitro*, SOD inhibition cannot be responsible for increased ROS generation. Moreover, it has been suggested that 2ME2 could also act as a radical itself if it is converted in cells to a semiquinone radical through a reaction with cytochrome P450. In addition, it has been shown that certain estrogens can form DNA adducts which can lead to impaired respiratory chain function. Therefore, a mechanism where EMBS itself becomes a radical cannot be excluded and further investigation will have to be conducted to determine if this does indeed occur. SOD activity also needs to be analysed to determine if EMBS induces ROS production by inhibition of SOD. Data from this study show that ROS production is an early event which occurs simultaneously to a decrease in mitochondrial membrane potential, while significant increases in apoptosis induction were only observed after the second more substantial rise in ROS production 14 h after EMBS exposure. Interestingly, the other early event occurring after EMBS exposure is an increase in cells blocked within the G₂/M fraction (4N) of the cell cycle. After 4 h of exposure a significant increase in blocked cells was observed increasing further up to 24 h. However, after 48 h the fraction of blocked cells decreased with a concomitant increase in cells that underwent endoreduplication (\>4N). NAC treatment completely abrogated these effects suggesting that even the very early block in cell cycle is dependent on the production of ROS. Furthermore, ROS production does not depend on a G₂/M arrest since cells blocked G1/S by thymidine still presented with increased hydrogen peroxide levels after EMBS exposure. Therefore, EMBS induces increased ROS production resulting in subsequent G₂/M arrest. A number of papers suggest that 2ME2 and EMBS bind to microtubules which results in a G₂/M block. These reports are mostly based on *in vitro* tubulin polymerisation assays. *In vitro* assays show that 2ME2 and EMBS prevent tubulin polymerisation although tubulin preparations including microtubule associated proteins are resistant to the effect of 2ME2. In the same study cells incubated with 2ME2 for 6 h showed decreased growth rate and growth duration for microtubules while they spend more time paused resulting in an overall decrease in tubule dynamicity of 43%. However, this study suggests that EMBS primarily targets ROS production leading to G₂/M arrest and subsequent cell death induction via apoptosis. Endoreduplication is a well-documented phenomenon where DNA replication occurs without cell division. This occurs when cells are arrested by the mitotic spindle assembly checkpoint or if cells fail to pass through cytokinesis. 2ME2 was shown to induce endoreduplication which was dependent on mitochondrial oxidative stress. Our data show that EMBS also induces endoreduplication in a ROS-dependent manner. As mentioned above this mostly appears in cells exposed to EMBS for 48 h where cells blocked in the G₂/M phase were dramatically reduced. Together this data suggest that EMBS induces a blockage in the G₂/M phase of the cell cycle which results in cells either initiating apoptosis or escape cell death through endoreduplication. Several studies have shown that 2ME2 induces hypoxia inducible factor 1α (HIF1α) accumulation leading to the activation of the JNK pathway. JNK integrates extracellular stress signals and induces apoptosis in response to these signals unlike extracellular regulated kinase (ERK) which integrates extracellular signals to induce proliferation and differentiation. When JNK is activated it can phosphorylate and inactivate the anti-apoptotic protein Bcl-2. Previous data showed that apoptosis induced by 2ME2 is largely dependent on JNK activation. Our data revealed that JNK inhibition largely abrogates the effects of EMBS since cells were no longer blocked in the G₂/M phase. However, there was still a significant fraction of cells that underwent endoreduplication after 48 h indicating that the JNK pathway may not be the only pathway involved in this process. In addition, the JNK inhibitor had no effect on ROS generation induced by EMBS indicating that the ROS production is an upstream event from the JNK-dependent apoptosis induced by EMBS. # Conclusion In summary, we have shown that EMBS increased ROS production in breast cell lines. Furthermore, temporal studies showed that ROS production changed over time after EMBS exposure with initial increases followed by a short, but statistically significant drop in ROS production after which production rates consistently and dramatically increased. This second increase was correlated with mitochondrial membrane depolarisation, G₂/M block, endoreduplication and increased apoptosis induction. Co-treatment with the free radical scavenger, NAC, abolished these effects induced by EMBS suggesting that the increased ROS production by EMBS is essential for its effects on the cell survival. This study contributes to elucidating the novel oxidative stress- dependent signaling mechanism utilised by sulphamoylated 2ME2 derivatives resulting in mitochondrial damage, cycle arrest, endoreduplication and ultimately apoptosis induction. # Supporting information [^1]: The authors declare there are not any competing interests. [^2]: **Conceptualization:** MHV IVDB AMJ. **Data curation:** MHV IVDB. **Formal analysis:** MHV IVDB. **Funding acquisition:** MHV AMJ. **Investigation:** MHV IVDB. **Methodology:** MHV IVDB. **Project administration:** MHV IVDB AMJ. **Resources:** MHV IVDB AMJ. **Software:** MHV IVDB. **Supervision:** MHV AMJ. **Validation:** MHV IVDB. **Visualization:** MHV IVDB. **Writing – original draft:** MHV. **Writing – review & editing:** MHV IVDB AMJ.

# Introduction Salience refers to the state or quality of an item that makes it stand out relative to its neighbors. Accurate appraisal of salience in the environment is central to organisms' survivals in that salience detection facilitates learning and survival by enabling organisms to focus their limited perceptual and cognitive resources on the most pertinent subset of the available sensory data. Salience is also of great importance in social cognition. If we cannot determine the salient message among the large amount of information that social interactions provide, we may misunderstand other's intentions or feelings and behave in an inappropriate manner. It has been suggested that patients with schizophrenia experience a state of aberrant salience attribution, and that patients' attribution of incentive salience to irrelevant stimuli contributes to the formation of delusions or hallucinations. Previous studies of aberrant salience attribution have shown that patients with schizophrenia tend to imbue emotionality to neutral stimuli. For example, compared to normal controls, patients with schizophrenia gave higher pleasantness and unpleasantness rating scores to neutral stimuli, and showed more positive responses to neutral stimuli. In contrast, in face recognition tests, patients with schizophrenia felt more negative emotion towards neutral faces. In studying the neural bases of aberrant salience attribution in patients with schizophrenia, several structures are of particular interest. First, the dorsal anterior cingulate cortex (ACC) and insula have been identified as a salience network that functions to recognize the most relevant of several stimuli in order to guide behavior. Activities in the two regions were shown to be altered when patients with schizophrenia experienced hallucinations or delusions, which might be a consequence of emotional salience to mundane events. Second, the mesolimbic system for dopaminergic signaling and reward anticipation contributes to salience attribution. Salience attribution in schizophrenia has been associated with altered activities in the mesolimbic system including the striatum, amygdala, hippocampus, and parahippocampal gyrus. Third, the ventrolateral prefrontal cortex (VLPFC) functions to integrate cognitive and motivational information to compute behavioral significance, which can be used for goal-directed behaviors. In a previous study, salience coding in schizophrenia induced decreased activity in the VLPFC, which was linked to anhedonia. Real life situations are complex and can involve conflicting emotions, such as stimuli incorporating both positive and negative emotions together. When a person evaluates these complex stimuli, one emotion is disproportionately more influential in the holistic appraisal than the other emotion, producing affective asymmetry. The first example is “negativity bias,” which refers to a tendency for the negative system to respond more intensely than the positive system when evaluative input increases. This may aid in anticipating threatening situations and protecting from danger as soon as possible. The second example is “positivity offset,” which refers to a tendency for the positive system to respond more than the negative system when evaluative input is weak or absent. This may facilitate more active exploration of the environment. A previous neuroimaging study performed by our research group revealed that the dorsolateral prefrontal cortex (DLPFC) was involved in the processing of negativity bias and positivity offset, suggesting that affective asymmetry may be caused by integrative functions at the neocortical level. A behavioral study demonstrated that patients with schizophrenia processed affective asymmetry in an impaired manner. Based on these findings, patients with schizophrenia could potentially show a deficit in dorsolateral prefrontal function during the processing of affective asymmetry, but this hypothesis has not been examined yet. The aim of the current study was to investigate the neural correlates associated with emotional salience attribution in patients with schizophrenia when the positive and negative systems were co-activated with asymmetric manifestations. To explore these functional correlates, we used an emotion judgment task involving two pictures with similar or different emotions during event-related functional magnetic resonance imaging (fMRI). We made the following predictions based on previous findings. First, when two opposite emotions are induced simultaneously, both patients and controls will show a discrepancy in the degree of attention paid to each emotion (positive or negative), resulting in emotional salience attribution. Second, emotional salience attribution that affects the initial perception of stimuli with two opposite emotional traits will be associated with impaired processing of affective asymmetry in patients with schizophrenia. Third, when affective asymmetry is processed, compared with controls, patients will show altered activations in brain regions related to salience attribution, as well as affective asymmetry, such as the DLPFC, VLPFC, ACC, insula, striatum, amygdala, hippocampus, and parahippocampal gyrus. # Materials and Methods ## Participants Fifteen patients with schizophrenia (eight men) and 14 normal controls (six men) participated in this study. Exclusive diagnosis of schizophrenia in patients and exclusion of any psychiatric disorders in controls were made using the Structural Clinical Interview for DSM-IV. Exclusion criteria included the presence of a neurological or significant medical illness, current or past substance abuse or dependence, and left-handedness. Paranoid tendency was assessed using the paranoia scale. Ambivalence disposition was measured using the schizotypal ambivalence scale (SAS). Psychopathological symptoms were assessed using the positive and negative syndrome scale (PANSS). Demographic and clinical data are provided in. Our study was carried out under the protocols approved by the institutional review board of Yonsei University Severance Hospital, and written informed consent was obtained from all participants. ## Stimulus Materials and Experimental Task During fMRI, participants performed an emotion judgment task, in which nine positive, nine negative, and nine neutral images from the International Affective Picture System were used after modification; their mean valence was 7.53±0.18, 2.20±0.69, and 4.84±0.11, respectively. The two pictures juxtaposed represented four different conditions: ambivalent, positive, negative, and neutral, comprising pairs of positive-negative or negative-positive, positive- positive, negative-negative, and neutral-neutral images, respectively. A total of 160 pairs, with 40 pairs per condition, were presented in fully randomized order in an event-related design. All stimuli were presented for 3.5 seconds (ISI = 500 ms). The null events of crosshair fixation varied from 1.25 seconds to 10 seconds. Participants were encouraged to respond by clicking one of three buttons as quickly as possible, depending upon the subjective feeling produced by pairs of pictures as a unit. They could click the left, right, or middle mouse buttons to a positive, negative, or neither-positive-nor-negative (nPnN) response to the stimuli, respectively. Response types and reaction times were automatically transferred to the computer files. Stimuli presentation and response recordings were performed using the software IFIS SA (MRI Devices Corporation, Waukesha, WI) and E-Prime system (Psychology Software Tools, Inc., Pittsburgh, PA). ## MRI Procedure and Image Preprocessing MRI data were acquired on a 3T MR scanner (Intra Achieva; Philips Medical System, Best, Netherlands). Twenty-eight contiguous 4.5-mm-thick axial slices covering the entire brain were collected using a single-shot, T2\*-weighted echo planar imaging sequence depicting the blood-oxygenation-level-dependent (BOLD) signal (echo time = 50 ms; repetition time = 2,000 ms; flip angle = 90°; field of view = 220 mm; and image matrix = 64×64). Axial 1.5-mm-thick T1-weighted images (echo time = 4.6 ms; repetition time = 25 ms; flip angle = 30°; field of view = 240 mm; and image matrix 256×256) were also collected. Spatial preprocessing and statistical analyses were performed using Statistical Parametric Mapping 8 (SPM8). Corrections for differences in slice acquisition time were performed using user-specified sequences. Head motion was corrected by realignment, and corrected images were co-registered to the T1-weighted image for each participant. T1-weighted images were normalized to the standard T1 template, and the resulting transformation matrices were applied to the co- registered functional images. Normalized images were smoothed with an 8-mm full- width-at-half-maximum Gaussian filter. ## Statistical Analyses ### Behavioral data analysis Demographic and clinical data were compared between patients and controls by independent samples *t*-tests and chi-square tests. Response rates were analyzed using the Generalized Linear Mixed Model, in which each subject was considered to be a random effect and the factors were the group and emotional condition. The effect of the group and emotional condition on the reaction time was analyzed using analysis of covariance (ANCOVA), in which only total and the most frequent responses in each condition were included as a variable because of large differences among response rates. In all analyses, years of education were used as a covariate because they showed a significant group difference. The association between affective asymmetry including negativity bias and positivity offset and clinical measurement including SAS and paranoia scale was examined using regression analysis of General Linear Model (GLM), which was applied because there were individuals that showed response rates close to the extreme. In order to quantify affective asymmetry, negativity bias score was defined as the percentage of “negative responses” for the ambivalent condition in which evaluative input was increased, whereas the positivity offset score was calculated as the percent of “positive responses” for the neutral condition in which evaluative input was weak. ### Neuroimaging data analysis Experimental trials for all emotional conditions and null trials were modeled separately for each condition minus the null events using a canonical hemodynamic response function and its first-order temporal derivative. The group and condition effects, as well as their interactions, were analyzed using ANOVA, and a two-sample *t*-test on the whole brain volume was performed for screening group comparisons. In these two analyses, education level was used as a covariate, and the threshold was set at an uncorrected *p*\<0.001 with more than 30 contiguous voxels. Then, a two-sample *t*-test after small volume correction (SVC) with a threshold at a family-wise error-corrected *p*\<0.05 was performed in the *a priori* regions, which were related to affective asymmetry and salience attribution. The investigated regions and their coordinates (x, y, z) included the DLPFC (−36, 36, 40; −36, 54, 6), VLPFC (42, 24, −15), dorsal ACC (6,22,30; −6,18,30), insula (37, 25, −4; −32, 24, −6), putamen (16, 12, 0; −16, 12, 0), amygdala (−21, −6, −12), hippocampus (44, −24, −12), and parahippocampal gyrus (18, −22, −18). The volume comprised a sphere with a 15-mm diameter for the DLPFC and VLPFC or 10-mm diameter for the other regions. For further analysis, percentage signal changes in significant clusters from this SVC analysis were obtained in each subject using MarsBaR version 0.41 (<http://marsbar.sourceforge.net/>). Correlations of regional percent signal changes with negativity bias and positivity offset scores were examined using regression analysis of GLM, in which *p*-values were adjusted for multiple correlations using a sequential Holm-Bonferroni procedure. # Results ## Behavioral Data As shown in, for the ambivalent condition, “negative” responses were the most frequent in both groups, and there were no significant differences in response types between the two groups. For the neutral condition, “nPnN” responses were the most frequent in both groups, but patients showed significantly higher “positive” and “negative” response rates and a significantly lower “nPnN” response rate than controls (p\<0.001 in all comparisons). For the positive and negative conditions, the most frequent responses were “positive” and “negative” in both groups, respectively; however, “positive” and “negative” response rates were significantly lower in patients than in controls (p\<0.001 in both comparisons). For all conditions, missing rates were significantly higher in patients than in controls (p\<0.001 in all comparisons). The reaction times for total and the most frequent responses in each condition did not show a significant fixed effect for group and condition. In addition, there was no significant correlation between affective asymmetry scores and clinical measures. ## Imaging Data Brain regions showing a significant main effect for group and condition are summarized in. Significant group×condition interaction was found in the left medial prefrontal cortex and left inferior frontal gyrus. Post-hoc tests using regional percent signal changes revealed neither significant group difference in all conditions, nor significant condition difference in patients. In controls, however, there were significant condition differences: the ambivalent condition showed significantly higher percent signal changes than the negative (p\<0.001), positive (p = 0.004), and neutral (p = 0.003) conditions in the left medial prefrontal cortex (p\<0.001), as well as the negative (p = 0.002) and positive (p = 0.002) conditions in the left inferior frontal gyrus. Imaging results of screening two-sample *t*-test in each condition are also described in and. In the group comparison after SVC, compared with controls, patients showed significantly decreased activities in the left DLPFC, right dorsal ACC, left insula, and bilateral putamen for the ambivalent condition, as well as in the right insula for the positive condition, but demonstrated no significantly increased activities. Percent signal changes in these regions showed no significant correlation with negativity bias and positivity offset scores. In addition, no significant group difference was found in the other a priori regions, such as the VLPFC, amygdala, hippocampus, and parahippocampal gyrus. # Discussion To explore neural representations related to salience attribution during affective asymmetry in schizophrenia, fMRI was performed during the emotion judgment task in patients with schizophrenia and controls. For the ambivalent condition, both groups interpreted the emotional salience of the stimuli as a negative trait, which could be interpreted as ‘negativity bias.’ This result differed from our assumption that patients with schizophrenia would show a reduction in negativity bias, which we demonstrated in our behavioral study. This discrepancy might be due to a problem with the sample size of this study, as a relatively small sample was included in this study; however, this is more likely to stem from differences in experimental design between the two studies. In the previous study, ambivalent stimuli were presented repeatedly within a block, whereas in the current study, they were presented randomly with other types of emotional stimuli. As predictability is known to affect stimulus salience, a block design, in which participants can expect a next emotional stimulus, may not evoke significant salience attribution. Therefore, considering that salience attribution was the main focus in this experiment, we presented the stimuli in an event-related design rather than a block design. The current method may increase the mental burden because no order can be predicted. This possibility is supported by the fact that reaction times for almost all conditions were longer in the current study than in the previous study. Consistent with this, another study using the event-related design similar to the one used in the current study found no significant differences in the processing of negativity bias between patients with schizophrenia and controls. This behavioral response may be based on brain neural responses in that brain activity during the processing of emotional content is dependent not only on the type of stimuli, but also the manner in which stimuli are processed. Despite the absence of significant behavioral differences in the ambivalent condition, remarkable group differences in the salience processing triggered by ambivalent stimuli were revealed in the medial prefrontal cortex and inferior frontal gyrus, which exhibited an interaction effect between group and condition. Significantly higher activity in the ambivalent condition than the other conditions was found in controls, but not in patients. The medial prefrontal cortex and inferior frontal gyrus have been proposed to be involved in emotional regulation, and response selection process. Because regulation occurs when stimuli induce conflicting appraisals and hence incompatible response tendencies or when goal-directed activity requires suppression of task irrelevant stimulus sources, ambivalent stimuli may induce conflicting emotional appraisals and responses that require regulation. In the ambivalent condition, these two regions may need to work hard to regulate ongoing emotional reactions to visual stimuli, which contain conflicting emotions, and to select an appropriate response, whereas this need may be unnecessary in the other conditions. However, patients with schizophrenia did not show these characteristic responses, and these deficits may be related to emotional blunting, which was demonstrated by less positive and negative responses for the positive and negative conditions in this study, respectively. Emotional blunting in schizophrenia may be associated with deficits in emotional regulation. This view is supported by a previous study that showed that dysfunction in the medial prefrontal cortex may be a core of trait anhedonia in schizophrenia. The inferior frontal gyrus receives projections from the orbitofrontal cortex and subcortical areas such as the midbrain and amygdala, which are involved in processing motivational and emotional information. The inferior frontal gyrus functions to integrate cognitive and motivational information to compute behavioral significance, which can be used for elaborate decision-making or to design goal-directed behaviors. In addition, affective asymmetry-related regions such as the DLPFC were found to be decreased in patients with schizophrenia when processing ambivalent stimuli. Given that the DLPFC is involved in integrative processing during co-activation of positivity and negativity , this result suggests that patients have deficits in this integrative processing, which could result in inappropriate affective responses. Taken together, despite the intact behavioral response of negativity bias, patients with schizophrenia may not be able to integrate emotional and motivational information during processing of ambivalent stimuli. The dorsal ACC, insula, and putamen were hypoactive in patients, relative to controls, for the ambivalent condition. These three regions are part of the salience network, which functions to identify the most relevant of several internal and extrapersonal stimuli to guide appropriate behavior. Deficient activities in these regions suggest a dysfunctional salience network in the ambivalent condition. A primary role of the salience network is the integration of sensations, internally generated thoughts, and information concerning goals and plans to allow actions to be initiated or modified. A dysfunctional salience network might also lead to a defect in integration of goals and plans, resulting in the difficulty to initiate activity, which might account for psychomotor poverty syndrome in schizophrenia. Although both groups showed negativity bias in the current design, we were able to confirm our hypothesis of abnormally decreased activity in the affective asymmetry and salience related regions in patients. Compared with controls, patients showed similar negative responses in the ambivalent condition, but decreased negative or positive responses in the univalent conditions, suggesting that an underlying mechanism of negativity bias may be different between the two groups. There is evidence that patients with schizophrenia use a different prefrontal network and strategies in the executive control process for successfully performing a working memory task. Likewise, patients might take a distinct path to obtain negativity bias. One explanation may rely on an accomplishment of negativity bias by a compensatory process for deficient emotional regulations. Functional changes in the DLPFC in patients appeared as decreased activity in BA 46 for the ambivalent condition. In contrast, patients also showed increased activities in BA 8 for the ambivalent, positive, and neutral conditions, which were presented only in the supplementary table, because it was not included as the a priori regions for SVC analysis. Increased activity in an unsuspected region may indicate that this region is working harder to compensate for decreased activity in other regions. Thus, hyperfunction of BA 8 in schizophrenia may occur in order to compensate for hypofunction of the affective asymmetry or salience network. Meanwhile, patients had significantly lower nPnN response rates than controls for the neutral condition, suggesting that they have a tendency to impart emotional salience to neutral stimuli. While other emotional stimuli have characteristics of primary inducers that automatically and obligatorily elicit emotional responses, neutral stimuli have characteristics of secondary inducers, which are related to “thought,” “memories”, or “imagination”. Therefore, attributing emotional salience to neutral stimuli may be related to the abnormal thought processes of patients with schizophrenia. This behavioral characteristic may be related to previous neuroimaging findings in which patients exhibited inappropriately stronger activations in the amygdala and striatum in response to neutral stimuli, suggesting an aberrant salience attribution in schizophrenia. In this experiment, however, we could not find any significant results in the mesolimbic regions such as the amygdala, hippocampus, and parahippocampal gyrus for the neutral condition in patients. Given a report that patients showed elevated activities in the amygdala even for the baseline condition, the absence of significant group differences in the neutral condition might be due to baseline hyperactivity of the mesolimbic regions in schizophrenia. Several limitations of this study should be noted. First, the sample size was relatively small, which might explain why the expected correlations between regional activities and clinical measures were not found. Second, patients were all taking antipsychotic medication. Although the effects of antipsychotics on emotional responses are known to be negligible, psychomotor speed could be influenced by the medication. Third, there was a group difference in the level of education, which could have influenced task performance. To address this, we used years of education as a covariate in behavioral and imaging analyses. Finally, although negativity bias or positivity offset scores were calculated as any portion of responses in the ambivalent and neutral condition, respectively, imaging results were compared across all trials in the corresponding condition. This was inevitable because response rates were highly variable across subjects, and the resultant difference might reflect characteristics of the process for making the response selection, regardless of the response. In summary, emotional salience attribution during the processing of negativity bias in patients with schizophrenia was associated with multiple prefrontal hypoactivities in the medial prefrontal cortex, inferior frontal gyrus, and DLPFC, which might be related to a deficit in integration of positivity and negativity. The salience network regions, such as the dorsal ACC, insula, and putamen also showed altered activities during the processing of negativity bias in patients, suggesting abnormal emotional salience attribution in schizophrenia. The neural basis of emotional salience attribution during the processing of positivity offset in schizophrenia was not clarified in this experiment in spite of definite behavioral evidence. These regional abnormalities may underlie defective emotional salience attribution in patients with schizophrenia, which likely influences both symptom formation and social dysfunction. # Supporting Information We would like to thank Mr. Kyung Mook Choi for his valuable technical support. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: HJP YCJ JHS JJK. Performed the experiments: JWC. Analyzed the data: SKL JSL. Contributed reagents/materials/analysis tools: HJP. Wrote the paper: SKL JJK.

# Introduction Injuries to the peripheral nervous system affect millions of people around the world and reduce their motor skills. Although the peripheral nervous system has some regenerative potential, complete functional recovery is seldom achieved, especially after severe lesions that lead to interruption of nerve continuity and tissue loss. In these cases, the most common therapeutic approach is the use of an autologous nerve graft; however, this technique has a number of disadvantages, such as donor site morbidity and limited functional recovery. Different potential strategies to stimulate peripheral axonal growth have been proposed, among them is the use of cell therapy, which has been deemed efficient mainly due to the release of trophic factors. Exogenous Schwann cells have the potential to aid regeneration of nerve fibers, not only due to its key role in myelination, but also for their ability to secrete trophic factors that promote survival and axonal growth. Additionally, the use of a biodegradable conduit to guide axonal growth, insulate and protect the injury site from negative influences is also a promising alternative to nerve grafts. It has been well documented that this type of conduit represents a suitable substrate for survival and differentiation of Schwann cells. In addition, its association with different cell types such as fibroblasts, bone marrow-derived cells or Schwann cells can maintain trophic factors at the lesion site and can help to improve outcomes in peripheral nervous system repair. Another potent therapeutic strategy is treadmill training, which has been shown to improve motor function after spinal cord injury both in animals and human subjects. The effects of treadmill training on functional recovery following peripheral nerve injuries were described recently. Using different training paradigms, some authors have found that treadmill training promotes functional recovery, due to an increase in release of trophic factors and the stimulation of axonal growth. The faster regeneration occurs, the more likely it is that the target organ will be reinnervated, as the axonal growth rate tends to decrease over time. To date, few studies have given attention to combining different approaches in order to accelerate the regenerative process. The present study has tested the effectiveness of combining Schwann-cell transplantation into a biodegradable conduit, with treadmill training as a therapeutic strategy to improve the outcome of repair after mouse sciatic nerve transection. Our findings suggest that this combination of therapeutic strategies can significantly improve functional and morphological recovery. Moreover, we show that transplanted Schwann cells can be functionally incorporated into the regenerated tissue and have the ability to secrete neurotrophic factors. These factors, either induced by physical exercise or directly released by transplanted Schwann cells, are increased by the combination of therapies, most likely underlying the benefit of the association. # Methods ## Animals Thirty-two male C57Bl/6 mice weighing 20–25 g were randomly divided into four groups: those with acellular grafts (n = 8, culture medium Dulbeco's Modified Eagle Medium, DMEM group), those with Schwann cell grafts (n = 8, SC group), those with treadmill training (n = 8, TMT group), and those with treadmill training and Schwann cell grafts (n = 8, SC + TMT group). Animals were subjected to sciatic nerve transection and tubulization with either a polycaprolactone (PCL) or collagen conduit. All animal use and care protocols were approved by the Ethics Committee for the Use of Experimental Animals of the Center of Health Sciences of the Federal University of Rio de Janeiro, Brazil (Protocol DHE003). ## Schwann cells: Isolation and purification Schwann cells were obtained from C57Bl/6 mice expressing green fluorescent protein (GFP + mice), allowing the subsequent identification of these cells, as previously described. Briefly, a 10-mm segment of sciatic nerves and brachial plexus of GFP+ mice were removed and dissected, the epineurium was stripped off, and nerve segments were allowed to degenerate *in vitro*. After 3 days, degenerated segments were enzymatically and mechanically dissociated, washed, and re-suspended in culture medium containing 20 µM pituitary extract (Invitrogen, USA) and 1 µM forskolin (Sigma); dissociated cells were seeded on plates pre-coated with laminin (10 µg/mL for 1 h) (Sigma). After 24 h, the medium was changed to remove cellular debris, and 100 µM AraC (Sigma) was added for another 24 h to prevent fibroblast proliferation. The medium was changed twice a week until optimal Schwann cell confluence was reached. Three expansion cycles were performed before cells were harvested, counted and injected into the conduit, at a density of 3×10⁵ cells in 2 µL of DMEM-F12. ## Surgical procedure Surgeries were performed under deep anesthesia using i.p. injections of ketamine (100 mg/kg) and xylazine (15 mg/kg). The left sciatic nerve was exposed and transected at mid-thigh level. First, 1 mm of the proximal stump was sutured into a polycaprolactone or collagen conduit (5 mm long), through the perineurium, using a 10.0 monofilament nylon suture. Subsequently, the graft was performed in the conduit for each group as follows: groups SC + TMT and SC the conduit was grafted with Schwann cells in DMEM (3×10⁵/2 µL), and the TMT and DMEM groups received only DMEM (2 µL). Grafts were injected into the conduit using a 10-µL Hamilton microsyringe. Finally, the distal stump was sutured through the perineurium to the other end of the conduit, leaving a 3-mm gap between the stumps. Polycaprolactone conduits were provided by Dr. Chiara Tonda-Turo, from Politecnico di Torino, Turin, Italy. Collagen conduits were previously used and described by our group. Eight weeks later, the animals were anesthetized with ketamine and xylazine (as above) and euthanized by intracardiac perfusion with a fixative solution (4% paraformaldehyde in 0.1 M phosphate buffer, pH 7.4). After perfusion, the regenerated left sciatic nerve within the conduit, the L4 spinal cord, de L4 dorsal root ganglion (DRG) and de gastrocnemius muscles were harvested and processed for immunohistochemistry, light or electron microscopy. ## Treadmill training Animals were subjected to a one-week familiarization to the treadmill apparatus before experimental protocols began. Treadmill training started on the third day after transection and surgical repair of the sciatic nerve. Mice were placed on a motor-driven treadmill (Insight, Ribeirão Preto, Brazil) at a belt speed of 10 m/min and were trained in two 30-min exercise periods with a 10-min rest period between, 3 days per week, as described previously, with few modifications. Mice ran on the treadmill at this speed with little or no persuasion, even on the third post operative day. Training protocol was conducted during the 8-week survival period. The animals in the DMEM and CS groups were handled and placed on a treadmill with no motion at the same times as the exercise groups. ## Functional assessment Motor function was evaluated with the sciatic functional index (SFI) as previously described – each week after surgery, the animals' pawprints were recorded, and two measurements were taken: (i) the print length (PL), corresponding to the distance from the heel to the third toe; and (ii) the toe spread (TS), corresponding to the distance from the first to fifth toe. Both measurements were taken from injured (E, for experimental) as well as noninjured (N, for normal) sides, and the SFI was calculated according to the following equation: The sciatic function index oscillates around zero for normal nerve function, and when it is equal to −100 represents total loss of function of the nerve. We also performed the global mobility test (GMT). The animals were acclimated to the open field prior to injury, and at 2, 4, 6 and 8 weeks after surgery they were submitted to global mobility test. Global mobility was assessed by videotape with a WebCam (5 frames/s, for 1 min; KMEX, USA) using the free K3CCD software, and quantified using ImageJ free software. ## Scanning electron microscopy The ultrastructural features of the polycaprolactone conduit were investigated before and after 3 and 8 weeks of implant, its proximal portion was fixed by immersion, then washed in 0.1 M phosphate buffer (pH 7.4) and 0.1 M cacodylate buffer (pH 7.4), and postfixed for 2 h in 1% osmium tetroxide containing 0.8% potassium ferrocyanide and 5 nM calcium chloride in 0.1 M cacodylate buffer (pH 7.4). Then, it was washed in 0.1 M cacodylate buffer (pH 7.4), dehydrated in an ethanol series, gold-sputtered (FL-9496 Balzers, Union Coater) and observed in a Jeol JSM-5310 scanning electron microscope. ## Immunoelectron microscopy The regenerated sciatic nerve was divided into proximal, middle and distal portions. Middle segments of nerve were immersed overnight in fresh fixative solution (4% paraformaldehyde in 0.1 M phosphate buffer pH 7.4). Tissue was washed in phosphate buffer (pH 7.4), dehydrated through a graded ethanol series, embedded in LRWhite resin (London Resin Company) and polymerized at 50°C, for 24 h. Ultrathin sections were obtained on an RMC ultramicrotome and collected on nickel grids. All sections were blocked with a blocking solution (PBS, 1% bovine serum albumin and 0.5% milk) for 40 min and incubated in primary antibodies (anti-GFP, Millipore, 1∶50) for 3 h at room temperature. After rinsing for 20 min in blocking solution, sections were incubated in the secondary antibody (BBI, goat anti-mouse IgG conjugated to 10-ηm gold particles, 1∶100) for 2 h, washed in PBS for 30 min, washed 30 min in distilled water, and stained with uranyl acetate and lead citrate. Primary antibodies were omitted for the negative staining controls. The sections were observed and photographed in a Jeol (JEM 101) transmission electron microscope. ## Light microscopy and transmission electron microscopy After nerve middle segments were fixed by immersion in 2.5% glutaraldehyde, they were washed in 0.1 M cacodylate buffer (pH 7.4), and postfixed for 90 min in 1% osmium tetroxide containing 0.8% potassium ferrocyanide and 5 mM calcium chloride in 0.1 M cacodylate buffer (pH 7.4). The segments were then washed in 0.1 M cacodylate buffer (pH 7.4) and stained in 1% uranyl acetate overnight. Nerves were dehydrated in graded acetone, infiltrated with Embed-812 resin (Electron Microscopy Sciences) and polymerized at 60°C for 48 h. Semi-thin (500 ηm) and ultra-thin (70 ηm) cross-sections were obtained on an RMC MT-6000 ultramicrotome. The semi-thin sections were stained with 1% Toluidine Blue and examined under a light microscope (Zeiss Axioskop 2 Plus); ultrathin sections were collected on copper grids, contrasted in 5% uranyl acetate and 1% lead citrate, and analyzed in a transmission electron microscope operated at 80 kV (Jeol 900). ## Morphometric assessment To quantify the total number of myelinated fibers in each regenerated nerve, photographs of the semi-thin cross sections were taken under the light microscope at 40x magnification, using the program Axiovision Rel. 4.5. The entire nerve was photographed, and myelinated fibers were counted with the use of ImageJ Software (1.42q, USA). Samples of five areas from the same semi-thin cross sections were obtained at a magnification of 100x. For each sample the following parameters were calculated: number of blood vessels, fiber area, axon area, myelin area and G ratio. Myelin area was measured by subtracting the axon area from the fiber area. The G ratio was calculated by dividing the axon diameter by the fiber diameter, and the results were stratified in ranges of 0.0–0.4, 0.4–0.5, 0.5–0.6, 0.6–0.7, 0.7–0.8 and 0.8–0.9. For each range, the lowest value was included and the highest one excluded (e.g., the 0.0–0.4 range includes 0.0 through 0.399, excluding 0.4). ## Immunohistochemistry and imaging To investigate the integrity of neuromuscular junctions (NMJs), the gastrocnemius muscle was immunolabeled for neurofilaments and counter-stained with anti-α-bungarotoxin. After the tissue was fixed, it was washed in 0.1 M PBS (pH 7.4), cryoprotected in 10%, 20%, 30% PBS-sucrose, embedded in OCT (Tissue- Tek, Tokyo, Japan) and frozen. Sixty-µm longitudinal frozen sections were made on a cryostat (Leica CM 1850, Germany), collected and washed in PBS, treated with ammonium chloride solution (NH4Cl) for 1 h, washed in PBS for 15 min, incubated in blocking solution (10% normal goat serum, 3% bovine serum albumin, in PBS 0.1 M plus 0.3% Triton) for 1 h, and washed in PBS-Triton. Slides were incubated in the primary antibodies rabbit anti-NF200 (Sigma, 1∶200) and anti-α- Bungarotoxin (conjugated with Alexa Fluor 594, Molecular Probes, 1∶500), overnight at 4°C in a humid chamber. Slides were washed in PBS for 30 min, incubated in the secondary antibody (Alexa Fluor 488 goat anti-rabbit, Sigma, 1∶600) for 2 h at room temperature, and washed in PBS for 30 min. Finally, the slides were mounted with Fluoromount (Sigma). The sciatic nerve distal segments, the L4 dorsal root ganglion and L4 spinal cord were processed following the same protocol for cryopreservation described above. Frozen longitudinal serial sections of dorsal root ganglions (12-µm) and transverse serial sections of sciatic nerve (10-µm) and the spinal cord (20-µm) were made and collected on gelatin-coated glass slides. To investigate the presence of Schwann cells in the host tissue, slides were washed in PBS, incubated in blocking solution, and then in the rabbit anti-S-100 primary antibody (Sigma, 1∶200). Sections were washed and incubated in the secondary antibody (Alexa 488 goat anti-rabbit, Sigma, 1∶600) for 2 h. Nuclei were stained with DAPI (Molecular Probes, USA, 1∶10,000) and mounted with Fluoromount (Sigma). To analyze the levels of Brain-Derived Neurotrophic Factor (BDNF), Nerve Growth Factor (NGF), and Neurotrophins 3 and 4 (NT-3, NT-4) present both *in vitro* and *in vivo*, we used the following primary antibodies: rabbit anti-human BDNF (Preprotech, 1∶100), rabbit anti-human NGF (Preprotech, 1∶100), goat anti-mouse NT-3 (Preprotech, 1∶100) and goat anti-mouse NT-4 (Preprotech, 1∶100). After incubation in primary antibodies, sections were washed and incubated in secondary antibodies: Alexa 488 goat anti-rabbit (Sigma, 1∶600) or Alexa 488 rabbit anti-goat. Slides were washed and coverslipped with Fluoromount. Primary antibodies were omitted for all negative staining controls. Staining was analyzed in a Zeiss Axioskop 2 Plus fluorescence microscope, equipped with a Zeiss Axiocam MRC camera, and Axiovision version 4.5 was used for image acquisition. For trophic factors immunostaining analysis, a spinning disk confocal microscope (Leika TCS SP5 AOBS) was used. ## Trophic factor quantification To quantify tissue levels of BDNF, NGF, NT-3 and NT-4, images were collected from all immunostained sections using a 20x objective, and analyzed using the Image-Pro Plus software (Media Cybernetics, version 6.0). After proper threshold setting, quantification was done by measuring the ratio between the stained area and the total field area. A total of eight sections per nerve were analyzed, from five animals from each group. ## Neuronal quantification For quantification of sensory and motor neurons, slides with dorsal root ganglion and spinal cord (6–8 sections per slide) were stained with cresyl violet. Sections were photographed on a Zeiss Axioskop 2 plus microscope at a 20x magnification. All neurons nucleoli in the DRGs sections and in the ventral horn of the spinal cord were counted using ImajeJ Software (1.42q). To avoid double counting of neurons, the distance between the sections used for dorsal root ganglion and L4 spinal cord segment quantification was 96 and 200-µm, respectively. To ensure that only motor neurons were counted, a perpendicular line was drawn in the spine central axis, immediately in front of the spinal cord central channel. ## Statistical analysis All analyses were carried out using GraphPad Prism 5.01 (GraphPad Software, Inc.). Statistical analyses used One-Way analysis of variance for comparisons between four groups, followed by the Tukey post-hoc test when necessary. When the values did not assume Gaussian distribution or had large coefficients of variation, the chosen test was Kruskal-Wallis nonparametric test, followed by Dunn's post-hoc test. Two-Way analysis of variance was used for comparisons between four groups during the survival time. Data are presented as mean ± standard error of the mean (SEM). A *P*-value \<0.05 was considered significant. # Results ## Combined therapies improve nerve regeneration both functionally and morphologically Semi-thin sections of the regenerating sciatic nerves revealed the best overall organization in the group that received the combined therapies, compared with other groups. The ultrastructural analysis confirmed the light microscopy observation. Quantification of morphological parameters showed that the number of myelinated nerve fibers and myelin area differed significantly in the TMT + SC group (2060±92.10 and 8195±1090, respectively; *P*\<0.05) compared to the DMEM group (1324±279.9 and 4628±781, respectively; *P*\<0.05). However, the analyses of the axon and fiber areas (respectively) failed to show a significant difference between the groups. The number of blood vessels was significantly increased in the TMT, SC and TMT + SC groups (; 406.7±13.53, 348.4±32.51 and 392.4±28.9, respectively, *P*\<0.05; *P*\<0.01) compared to DMEM group (243±14.33). G-ratio analysis showed that DMEM and SC groups displayed many fibers in the range of 0.4 to 0.6, whereas TMT and TMT + SC groups achieved the best results, with most fibers in the range 0.5 to 0.8, the most suitable G-ratio index for the sciatic nerve. To assess the efficacy of therapies on motor function, we used the sciatic function index and the global mobility test. Although all animals showed broad loss of function in the first week after injury, the TMT + SC group showed significant improvement compared to the DMEM and TMT groups. Thereafter, all groups showed progressive improvement in motor function, but TMT, SC and TMT + SC groups showed higher values of SFI during the entire evaluation period. In the last week, all treatments resulted in a statistically significant functional improvement compared to DMEM group. Data from global mobility test progression revealed that two and eight weeks after surgery, the TMT and TMT + SC-treated animals were able to walk for longer distances than mice that received DMEM only. Quantification showed that TMT + SC animals achieved the highest velocity compared to other groups. ## Regenerating axons formed neuromuscular junctions The functional and morphological results strongly suggested that regenerated nerves were able to reinnervate target muscles of treated animals. To confirm this hypothesis the neuromuscular junctions were qualitatively analyzed in each group. In normal animals, the staining of junctions by anti-neurofilament and α-Bungarotoxin has a “pretzel-like” shape, with precise apposition of motor- nerve terminals to the motor endplates. We were able to visualize the neuromuscular junctions in all surgically transected groups, suggesting that the regenerated axons were indeed able to reinnervate target muscles. However, in the DMEM , TMT and SC groups, most of the junctions were enlarged and disorganized, with weak labeling for neurofilaments. In contrast, the TMT + SC group showed a much better apposition of motor-nerve terminals to the endplate, with neuromuscular junctions closely resembling those of the normal animal. ## Polycaprolactone tubes are a suitable substrate for the regenerating nerve and progressively degrade *in vivo* We observed the ultrastructural features of the polycaprolactone conduit prior to implantation (′), 3 and 8 weeks after being implanted (′; 6C and 6C′, respectively), by scanning electron microscopy. Cross sections of tube showed the growing nerve inside and in close contact with the tube wall. In the course of time, the tube wall became thinner, showing its biodegradable property. ## Transplanted Schwann cells functionally reintegrate into the tissue The survival and presence of green fluorescent protein in GFP⁺ Schwann cells inside the tube were evaluated 8 weeks after transplantation. In the distal segment of the regenerated sciatic nerve, GFP⁺ cells were readily visible (′, 7B and 7B′) and also expressed S-100 (′, 7D and 7D′). Immunogold images revealed that GFP⁺ cells were in direct contact with myelinated fibers. Colloidal gold particles were easily observed in both SC cytoplasm and nucleus, as well as in the myelin sheath of myelinated fibers. We also detected labeled Schwann cells exhibiting a prominent rough endoplasmic reticulum (RER) in the cytoplasm. ## Combined therapies increase levels of trophic factors *in vivo* Analysis of *in vitro* expression showed that Schwann cells were positively stained for BDNF, NGF, NT-3 and NT-4. Eight weeks after cell transplantation, sciatic-nerve, DRG and spinal cord sections from transected animals showed immunoreactivity for all trophic factors. Quantification of trophic factors immunoreactive area, relative to total nerve area, showed a significant difference between TMT + SC and DMEM groups for BDNF (0.003059±0.0005070 and 0.0009073±0.0003455, respectively, *P*\<0.05), for NGF (0.005780±0.001606 and 0.001699±0.0004659, respectively, *P*\<0.05) and for NT-4 (; 0.0008682±0.0002859 and 0.0001162±0.00001860, respectively, *P*\<0.01). Immunoreactivity for NT-3 showed no significant difference between groups. The quantification of all analyzed trophic factors in the dorsal root ganglia of the TMT + SC group revealed significantly higher levels in comparison to the DMEM group. TMT + SC group (0.002918±0.0003860) had a larger immunopositive area for BDNF compared to DMEM (0.000782±0.0002193) and SC (0.001031±0.0002140, *P*\<0.01) groups. Quantification of NGF labeling presented significant difference between TMT + SC group (0.008512±0.001935) compared to DMEM and TMT groups (0.001512±0.0001766 and 0.001989±0.0001795, *P*\<0.01, respectively;). Analysis of NT3 in the TMT + SC group (0.004458±0.0004005) showed significant difference only compared to DMEM group (0.002504±0.0003557, *P*\<0.01;). Finally, levels of NT4 staining in the TMT + SC group (0.003511±0.0005275) were significantly higher than in DMEM and SC groups (0.0006326±0.0001749 and 0.001287±0.0001189, *P*\<0.001 and *P*\<0.01, respectively). NT4 levels were also different between TMT and DMEM groups (0.002513±0.0002927 and 0.0006326±0.0001749, *P*\<0.01, respectively;). Staining of trophic factors in the spinal cord showed a significant difference in the intensity of fluorescence between TMT + SC and DMEM groups only for NT4 (0.001067±0.0002359 and 0.0005077±0.0001098, *P*\<0.05, respectively;). Immunoreactivity for the other trophic factors was similar among groups. ## Therapies influence neuronal survival To evaluate the efficiency of combined therapies on the survival of dorsal root ganglion and spinal cord neurons, the L4 lumbar segment and L4 dorsal root ganglions were analyzed eight weeks after nerve injury. The total number of motor neurons in L4 segments of the DMEM group was significantly lower (3.2±0.91) than the ones found on the SC and TMT + SC groups (15.80±2.69 and 14.20±1.80, respectively;), indicating that SC and TMT + SC treatments protected motor neurons from cell death. Counts of DRG sensory neurons revealed that the TMT group (102.6±4.73) had significantly more neurons than the DMEM group (61.0±4.14). The SC and TMT + SC (97.80±10.96 and 95.40±8.424, respectively) groups had a slight, but not significantly larger number of motor neurons compared to DMEM group. These results suggest that TMT was more effective in protecting sensory neurons than SC treatment or the combination of both. summarizes the results. # Discussion The peripheral nervous system has a well-documented ability to regenerate, and yet functional recovery in severe traumas such as complete nerve transection is still clinically unsatisfactory and often results in disability and decreased quality of life for affected individuals. Thus, in order to optimize the regenerative process and reach full functional recovery, development of new therapeutic approaches and the combination of different existing strategies are necessary. In the present study, we combined multiple repair strategies, namely: surgical repair with a tubular conduit; grafts of cultured Schwann cells; and treadmill training. This combination accelerated the regenerative process of the sciatic nerve, increasing the number of myelinated fibers and myelin area, as well as improving recovery of motor function. We also observed a better structural organization of motor-nerve terminals, which suggested a more efficient reinnervation of the gastrocnemius muscle. There is a general agreement that functional recovery is improved when the time for reinnervation is short, mainly because the regenerative potential of neuron cell bodies and the plasticity of Schwann cells decline over time. Therefore when the distance between the injury site and the target organ is too long, a reduced number of motor axons can regenerate and make functional connections with denervated muscle fibers. Hence, the growth rate and the number of axons may have direct implications for functional recovery. The g-ratio is an index that reflects the axonal myelination that leads to an appropriate conduction of action potentials. The optimal g-ratio rate for myelinated fibers from sciatic nerve was defined as 0.55–0.68. In our experiment, the animals treated with combined therapies achieved a larger number of myelinated fibers, a larger myelin area and increased g-ratio values. Together, these results may suggest an improved reinnervation of the target organ. Following sciatic nerve injury and repair, the analysis of walking tracks, quantified by the sciatic function index has proven to be a reliable method for evaluating functional recovery. It provides an easily obtained, serial and direct assessment of function. Previous studies by our group have already shown a clear correlation between morphological and functional recovery. Our work showed that TMT + SC-treated animals exhibited an improved functional recovery compared to other groups possibly due to the better morphological results. The ultimate indication of muscle reinnervation is whether the nerve is connected to the myofiber. We therefore analyzed qualitatively the morphology of neuromuscular junctions. In the TMT + SC-treated group we saw an improved structural organization of junctions which was similar to a normal animal. We also observed a better apposition of motor nerves terminals in relation to the endplate. Most junctions found in DMEM, TMT and SC-treated groups were expanded and disorganized, with deficient staining against NF-200. This deficiency can disrupt the addressing of presynaptic vesicles toward the endplate. Despite the fact that we have not performed quantification of the data, this analysis supports the functional and morphological improvement found in the TMT + SC- treated group. Tubular guides serve as a bridge between the distal and proximal segments of the nerve, and their use has been considered a promising alternative to autologous nerve graft. Two types of biodegradable tubular prostheses, collagen and polycaprolactone tubes, were used here. Both were previously tested by our and other groups, and are a suitable substrate for adhesion and elongation of the growing nerve fibers,. Ultrastructural analysis of the tubes eight weeks post- surgery, revealed clear signs of an initial degradation process. This is consistent with previous reports, and indicates an adequate degradation rate. In addition to the tubular conduit, cell therapy with Schwann cells was used to improve regeneration. These cells are the principal support cells in peripheral nervous system, and when used in cell therapy, possess the remarkable ability to promote nerve regeneration in both central and peripheral nervous systems. This therapeutic value is mainly attributed to their potential to produce trophic factors, which are important for the regenerative process in the peripheral nerves. An important issue when using cell therapy is the viability of the transplanted cells in the host tissue. Our results demonstrated the presence of Schwann cells in the middle segment of the regenerated nerves up to 56 days after injury. These findings are consistent with previous reports, which identified Schwann cells in the host tissue for up to 16 weeks after they were injected into polycaprolactone or collagen tubes. Furthermore, to the best of our knowledge, transplanted Schwann cells have never been shown to actually functionally incorporate into the tissue. Through an ultrastructural approach, we have demonstrated for the first time that transplanted Schwann cells can form a myelin sheath around regenerating axons; also, these cells exhibited a prominent rough endoplasmic reticulum, which indicates that the transplanted cells were active and probably secreting proteins. The capability of transplanted cells to fully reintegrate with the nerve represents an advantage in terms of cell-based therapies, as it suggests that exogenous Schwann cells can be recognized as native to the tissue. Treadmill training has been advocated by researchers and physiotherapists for its benefits to treat spinal cord injuries. Recently, attention has also turned to the potential of physical activity in the treatment of peripheral nerve injuries. Some experimental studies have shown that treadmill exercise, even in very small doses, can enhance axon regeneration and speed functional recovery after sciatic nerve compression in mice and in rat models. However, this is one of the first studies to provide direct evidence of the positive effects of the combination of cellular therapy seeded in a tubular conduit, with treadmill training, in peripheral-nerve regeneration after transection injury in mice. Several elements are essential for proper nerve regeneration, and neurotrophic factors are particularly important, since they can regulate neuronal survival, promote axonal and neurite regeneration, and reinnervation of target organs,. Schwann cells are described to exert neurotropic and guiding effects for axonal projection by secreting neurotrophins and adhesion molecules. The effects of exercise found in this study may also be attributable to the upregulation of neurotrophins, since exercise is known to enhance the levels of BNDF in regenerating neurons. We observed that Schwann cell and treadmill training treatments increased the expression of neurotrophins in the sciatic nerve (BDNF, NGF and NT-4), DRG (BDNF, NGF, NT3 and NT4) and spinal cord (NT4). This effect might be attributed to direct secretion of trophic factors by the injected cells and/or to release promoted by the treadmill training. Taken together, these evidences may explain the greater degree of fiber myelination observed in the TMT + SC group. Given that one of the roles of neurotrophins is to regulate neuronal survival, we also analyzed the number of dorsal root ganglion and spinal cord neurons. Our untreated group (DMEM) showed a substantial reduction in both DRG and spinal cord neuronal survival. Meanwhile, all the employed treatments, albeit not always significant, exerted some degree of protection to sensory and motor neurons. The translation of pre-clinical studies to an alternative therapeutic approach combining therapies to treat PNS injuries in humans is necessary, but requires further investigation. Developing a protocol to expand human Schwann cells from a small biopsy sample to a large number of cells is still a challenge, and the ability of cultured human Schwann cells to produce myelin in the tissue requires further demonstration. Moreover, it is also important to establish an individual treadmill-training protocol appropriate for each type of injury. In summary, this study provides evidence that combined therapies improve sciatic-nerve regeneration after a traumatic nerve injury in mice. These strategies are promising in aiding regeneration after nerve damage, and might benefit patients with neurodegenerative diseases in the future. We are grateful to Daniella de Freitas Pereira Angelo Durço for help with image processing and analysis, and to the Biology Institute of the Fluminense Federal University by JEM 1011 Transmission Electron Microscope used in this work. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: COG AMBM. Performed the experiments: COG SJ AS JTO SM FA SdL. Analyzed the data: COG AMBM. Contributed reagents/materials/analysis tools: CTT. Wrote the paper: COG SJ AMBM. [^3]: Current address: Dominick P. Purpura Department of Neuroscience, Albert Einstein College of Medicine, Bronx, NY, United States of America.

# Introduction Enterovirus 71 (EV71) is a member of human enterovirus group A (HEV-A) species, belongs to *Enterovirus* genus in *Picornaviridae* family. Since it was first identified from a child with neurologic symptoms in California in 1969, outbreaks associated with EV71 have been reported worldwide. During late 20^th century, several large outbreaks of hand-foot-and-mouth disease (HFMD) associated with EV71 infections have been reported in Eastern and Southeastern Asian countries and regions, e.g., Malaysia in 1997, Singapore in 1998, and Taiwan in 1998. The large scale and a number of fatal cases made these outbreaks distinguishing and attract the global attentions. Since then, outbreaks of EV71 frequently occurred in Asian countries, with high incidence of neurologic infection and fatality,. In China, large scale outbreak of HFMD associated with EV71 emerged in 2007. which occurred in Linyi City, Shandong Province, with 1,149 mild and 3 fatal cases reported. The nationwide epidemics of EV71 started in 2008, beginning as an outbreak of unknown viral infection in Anhui province, and spreading into other provinces quickly. Totally 488,955 children were suffered from this epidemic and 126 lives were claimed due to neurogenic cardiopulmonary failure and brain stem encephalitis. To respond to the epidemic, a national HFMD surveillance system was established in May 2008 so an effective prevention strategy can be implemented. In 2009, the epidemics of EV71 continued and became much more rampant. VP1 gene is of major significance for virological surveillance of enterovirus because it contains a number of importance neutralization epitopes. In this study, we performed genetic analyses on VP1 sequences using isolates detected in mainland China from 1987 to 2009 to investigate the evolution and molecular epidemiology of EV71 in this country. Although there are already a few previous reports about evolutionary analysis on global EV71, the data on subgenotype C4 especially from China after 2007 were little included. Through the analyses of sequence data from mainland China, we aimed to find some clues to explain the sudden outbreak of EV71 infections in China, and provide useful advice from virology for the disease control in the future. # Results ## 1. Surveillance of HFMD in China To collect epidemiologic information and understand the recent epidemics, HFMD was listed as a Class C notifiable disease in the national surveillance system on May 2, 2008. It requires the information of every case be submitted to the National Notifiable Disease Reporting System (NNDRS) within 24-hour upon diagnosed. The National Laboratory Network (excluding Hong Kong, Macau, and Taiwan) for HFMD was organized in 2008 to incorporate 331 prefectural laboratories at prefectural Centers for Disease Control and Prevention (CDCs), 31 provincial laboratories at provincial CDCs, and 1 national laboratory at China CDC. Collection of patient information and specimen followed the guidelines in the *Manual for Hand, Foot, and Mouth Disease control and prevention* released by the Ministry of Health ([www.moh.gov.cn/publicfiles/business/htmlfiles/mohbgt/s3 582/200906/41047.htm](http://www.moh.gov.cn/publicfiles/business/htmlfiles/mohbg t/s3582/200906/41047.htm)). Prefectural laboratory conducts tests using RT-PCR or real time RT-PCR on clinical samples to identify the infections of EV71, CA16, and other enterovirus. Provincial laboratory performs virus isolations and serotyping on positive samples from prefectural laboratory. The identified EV71 and CA16 isolates are forwarded to National laboratory for sequencing and molecular analysis. A total of 488,955 HFMD cases including 1,165 severe and 126 fatal cases were reported to NNDRS in 2008. In 2009, the outbreak continued. There was an apparently increase of the reported national incidence rate of HFMD, 87 hundred- thousandth in 2009 to 37 hundred-thousandth in 2008. Totally 1,166,131 cases with 13,810 severe cases and 353 fatalities were reported throughout 2009. Children younger than 5 years old were the majority of the victims in the outbreaks, which accounted for 91.3% (446,263 cases) of the reported cases in 2008, and 93.2% (1,086,793 cases) in 2009, respectively. Laboratory results revealed that EV71 and CA16 were the major pathogens for the recent HFMD outbreaks. In 2008, 54.4% and 14.6% of the cases confirmed were associated with EV71 and CA16 infections, respectively. In 2009, EV71 and CA16 were still the major pathogens, responsible for 48.1% and 31.5% of the confirmed cases, respectively. However, EV71 infections were predominant for severe and fatal cases which caused 81.6% of the severe cases and 95.3% of the fatalities in 2008, and 80.6% and 92.8% accordingly in 2009. ## 2. Molecular epidemiology of EV71 in China The complete VP1 gene sequence data of the available EV71 strains isolated in China were used for this study, including released sequences from GenBank before February 23^rd, 2010 and sequences from our laboratory. In China, the first EV71 strain was isolated from an adult HFMD case in 1987. Successive molecular virologic data on EV71 was available since 1998. After the establishment of laboratory network for HFMD in 2008, a large amount of data on etiology and molecular epidemiology become available, Strains isolated from 18 provinces during 1987-2009 were included, 87% (283/326) of the sequences data collected were between 2007 and 2009. In our study, a total of 326 complete VP1 genome sequences of EV71 isolates from mainland China were collected. Phylogenetic dendrograms were constructed based on these complete VP1 gene sequences. It was shown that all the EV71 circulating in China since 1998 were clustered with strains of subgenotype C4, except for 0667-CHN-87 (isolated in 1987, belonged to genotype C but not any subgenotype) and 97-56-CHN-97 (isolated in 1997, belonged to subgenotype C3). The first strain of subgenotype C4 (SHZH98) documented in mainland China was detected in 1998 in Guangzhou. Since 1998, the predominant subgenotype C4 has been circulating in China for at least 11 years, the epidemic pattern of which was very different from other countries and regions. In other neighboring countries and regions, EV71 of several subgenotypes circulated simultaneously during the outbreaks, and in some regions subgenotype transitions happened for several times throughout the period ,. For example, in Taiwan, C2 was the major subgenotype co-circulated with B4 during the epidemics in 1998; in the subsequent outbreaks in 2000, B4 became a dominant subgenotype; then C4 in 2004, and B5 and C5 in 2006. Although mainland China is neighboring to Taiwan, subgenotype C4 was the unique type detected in China since 1998, during which there was no transition or co-circulation of any other genotypes. This characteristic made the successive data of EV71 of subgenotype C4 in China invaluable for understanding the evolution of EV71. The sequences of C4 were analyzed further. All sequences were grouped by the year of sampling and showed with different colors in the phylogenetic tree. It was revealed that all the EV71 of subgenotype C4 from China could be split into two clades, which were designated as C4a and C4b, respectively. The strains of C4b were detected only before 2004, and C4a included all the isolates since 2005 and some isolates from 2002 to 2004. Briefly, subgenotype C4 had displayed 3 major evolutionary phases in China: the first phase, from 1998 to 2001, in which all isolates belonged to clade C4b; in second phase, during clade transition, from 2002 to 2004, isolates of both clade C4b and C4a co-circulated; in third phase, after 2005, only C4a were detected. The clade transition suggested that evolution of EV71 of subgenotype C4 occurred during persistent circulation in mainland China within 11 years. The genetic progression showed that the genetic distance to SHZH98 (AF302996), the first isolate of subgenotype C4 in China, became larger over time, which also suggested the evolution of subgenotype C4 in China. The strains detected during the recent outbreaks, from 2007 to 2009, interspersed in clade C4a with the isolates before 2007, but most of them clustered with each other and located at the terminal of the phylogenetic tree, which meant that the most recent descendant from C4a was responsible mainly for the current large outbreaks. It suggested that evolution might play some roles in recent epidemics of EV71 infections. ## 3. The evolution of subgenotype C4 in China ### Origin and evolutionary rate To study the evolutionary characteristics of subgenotype C4 in China, 82 VP1 sequences of subgenotype C4, sampled from 1998 to 2009, were selected for divergence time and substitution rate estimation with Bayesian Markov chain Monte Carlo (MCMC) method. The 82 sequences selected for evolutionary analysis (detailed) were all typical representatives in phylogenetic tree, which was labeled with blue-block in. Totally 8 different models were used for data analyses. It was found that the GTR and uncorrelated lognormal distribution clock model (GTR+UCLD) fitted to our data best. The coefficient of variation of the evolutionary rates among branches in the UCLD model was 0.358 (95% highest posterior density \[HPD\]: 0.176–0.550), significantly different from 0, which indicated that rates heterogeneity existed among different branches and the evolution of EV71 was not clock-like in China also, consistent with results from global data. According to our analyses, the most recent common ancestor of subgenotype C4 strains in China can be traced to 15.1–16.1 (95% HPD: 13.0–18.4) years before 2009. The divergence time was deduced to be from Oct. 1992 to Oct. 1993 (95% HPD: Jul. 1990–Jan. 1996), approximately equal to the global subgenotype C4 divergence time estimated in previous study. This evidence suggested that subgenotype C4 strain might begin to circulate in mainland China immediately as it diverged as a novel subgenotype from other genotypes. And EV71 of subgenotype C4 had been circulating for about 5–6 years in China before it was first detected in 1998, consistent to a previous report which estimated that genogroup B and C had each circulated cryptically in the population for up to 5 years before causing large HFMD outbreaks. Subgenotype C4 evolved in China at inconstant rate, and the estimated mean evolutionary rate among different branches was 4.6×10⁻³ and 4.8×10⁻³ substitutions per site per year inferred by the models of GTR+CS+UCLD and GTR+EG+UCLD, respectively, which was approximate to estimates of global genotype B and C estimated also by MCMC method previously. ### Epidemic history referred by molecular sequence To exclude that the recent outbreak of HFMD is an artifact of increased reporting attributed to the establishment of NNDRS in 2003, the demographic history of subgenotype C4 in China was reconstructed with Bayesian skyline plot method based on the VP1 sequences data. The plot showed a sharp increase, indicating an exponential growth of the population size (reflected the effective infections) occurred from 2007 to 2008, which was consistent with the epidemiological data about the large-scale HFMD outbreak in 2008. This finding revealed that the outbreak of EV71 in 2008 was actually ascribed to the explosion of viral population but not the improved surveillance sensitivity. ### Natural selection analysis We conducted natural selection analyses with SLAC and FEL method respectively on the Datamonkey website service. The results showed a low mean *dN/dS* ratios (0.055) of subgenotype C4 in VP1 (297 codons). Totally 127 codons and 202 codons (details was not shown) were identified as negatively selected sites by SLAC and FEL respectively, and only the codon 98 of VP1 protein was under positive selection pressure (P-value \<0.1). The results suggested that VP1 protein was under strongly negative selection pressure. Because of the importance of VP1 protein on neutralizing epitopes, negative selection pressure on VP1 meant that variations of VP1 protein weren't motivated by selection pressure from host immune system but limited by its important functions. In other words, the variation of VP1 sequence was mainly ascribed to the lack of correcting function during the genome replication of RNA virus, and the deleterious mutations were cleared. The evolution of VP1 protein was mainly due to the accumulation of random mutations during genome replication; a small portion of non-synonymous mutations induced amino acid changes but not directed variations. # Discussion HFMD epidemic has become a serious public health concern in China since 2007. Laboratory and epidemiology surveillance system of HFMD had been established in mainland China in 2008. Virologic data from laboratory network revealed that EV71 infections were the main reason for severe and fatal cases, although both EV71 and Coxsackievirus A16 were the major pathogens during the widespread outbreaks in China from 2008 to 2009. Before 2008, only several small outbreaks with mild cases were documented and no compulsory surveillance on HFMD in China. In our study, both the epidemiology and epidemic history referred by molecular sequences support the recently sudden outbreak of EV71 infections as an emerging one in China but not the artifact of more sensitive reporting. The emergence and reemergence of infectious disease could be attributed to either accumulated susceptible population or appearance of pathogens with new biological characteristics. For EV71 infections, children less than 5 years old are the majority of susceptible group. Our retrospective seroepidemiology study reported in previous study revealed that the circulation of EV71 had been established widely in China before 2005. Our unpublished study (discussed in reference 39) on EV71-seroprevalence rates among young children aged 1-5 years in Fuyang during and after the outbreak in 2008 showed that the post-epidemic seroprevalence rate had a slight increase. The seroepidemiologic results in China, which are consistent with previous studies from Taiwan, suggested that susceptible populations may play some role in the nationwide epidemics, but might not be the major reason. This study performed molecular epidemiologic and evolutionary analyses on EV71 detected in China through 2009, trying to find some explanations for the recent outbreaks from the virology point of view. The results indicated that EV71 genotype C4 was the unique subgenotype detected in mainland China since 1998, which originated from 1992 to 1993 and has been circulating persistently for more than 12 years. But it was associated with nationwide HFMD outbreak in last three years, and causing quite a few severe and fatal cases. Evidence above suggested that the recent outbreak in China was an emerging one caused by the *native* virus but not the importation of new genotype. Differently to mainland China, emergence and reemergence of EV71 outbreaks in other countries and regions were associated with establishment of the new genotype. According to our phylogenetic analyses on VP1 sequences, in China evolution and clade transition of subgenotype C4 occurred during the circulation of more than 12 years, and the most recent descendants of subgenotype C4 were the majority EV71 strains detected during the outbreaks since 2007. It suggested that the evolution of EV71 strains might be responsible for the recent HFMD outbreak which associated with subgenotype C4 virus. After the long and persistent circulation, the recent EV71 strains in China might have gained new characteristics making it more aggressive, such as stronger transmissibility, infectivity and (or) virulence. But the mechanism is still unclear because of the lack of suitable animal model, advanced researches are still needed. Fortunately, the recent discovery of cellular receptors of EV71 provides us better understanding on the binding and infection of EV71, also presents promising opportunity to develop transgenic animal model of EV71. For enterovirus control, e.g., polio eradication, the trivalent oral poliovirus vaccine (tOPV) is the only one using in China currently. For prevention and control of HFMD, vaccine is probably the most effective method to fight against viral infections. Inactivated EV71 vaccine would be the most hopeful so far, attributing to the simpler technology and higher safety. However, whether the incipient vaccine strains could maintain effective for long term like polio vaccine or the strains should be optimized periodically according to surveillance as in the case of influenza virus? So far, Limited information on antigenic evolution of EV71 is available. Our estimated mean evolutionary rate of EV71 was ∼ 4.6×10⁻³ substitutions/site/year, which is approximate to that of the influenza and much less frequently than polio virus (1.03×10⁻²), the typical species of enterovirus with antigenic stable for long. However, results on natural selection revealed that VP1 protein, the most important region of antigenic epitopes, of subgenotype C4 in China was under strongly negative selection pressure, similar with that of polio but influenza (which is also discussed in reference). Moreover, various subgenotypes could be neutralized by the antibodies induced after EV71 infection although several subgenotypes of EV71 have emerged, since first identified in 1963 attributed to the genomic evolution, indicating that the antigenicity of EV71 still keep stable during evolution. Accordingly, it was speculated that the antigenicity of EV71 in China would keep stable for a long period like polio. This characteristic would make the development of EV71 vaccine much simpler, which means that it is not necessary to optimize the vaccine strains periodically due to the antigenic variation. However, still a few codons (only codon 98 in this study) of VP1 were demonstrated to be under positive selection by both this and previous studies. Besides the selective pressure on VP1 codons, potential recombination could also induce antigenic variation in EV71. Additionally, there are still 3 other capsid proteins (VP2-VP4) presence. The antigenic evolutionary characteristics of EV71 was not fully comprehended as described by a previous research. Therefore, a long term antigenic surveillance for EV71 is necessary. The persistent circulation of subgenotype C4 in China made our data invaluable for exploring the evolutionary characteristics of EV71, which should remain as the priority in our future researches. Because of its wide territory and large population in mainland China, the epidemics of EV71 might still need another several years to sweep out the susceptible population. Additionally, the new born population which refresh the susceptible pool annually and the nonsynchronous epidemics of EV71 among different provinces also determine that the current epidemic is still going to continue in the near future. The epidemics of EV71 infections would still harass China for some time as prediction, and a harsh combat in control and prevention is yet to come. # Materials and Methods ## 1. Virus isolation and identification After processing, specimens from clinically diagnosed HFMD patients were inoculated into appropriate cell cultures including HEp-2, RD, and Vero cells. EV71 strains were identified by RT-PCR and sequencing of VP1 genome (891 nucleotides). ## 2. Sequencing and sequence collection of EV71 RT-PCR amplification and sequencing of VP1 encoding gene were performed as previously described. Thirty-one sequences have been deposited in GenBank (accession number HM212436- HM212466). Other EV71 strains with complete VP1 sequences isolated from mainland China were downloaded from GenBank which were released before February 23^rd, 2010. The sequence information was checked manually to exclude the laboratory adaptive strains, clones, and strains with high passage numbers. Eventually 295 complete VP1 gene sequences with known collection dates between 1987 and 2009 were obtained. Totally, 326 EV71 isolates were included in this study as shown in. ## 3. Phylogenetic analyses For phylogenetic analysis, reference strains of A, B, and C1-C5 genotype/subgenotype, were obtained from GenBank as listed in. Sequence alignment of Chinese and reference strains was conducted with ClustalW program in MEGA 4 (Molecular Evolutionary Genetics Analysis software; Tamura, Dudley, Nei, and Kumar 2007). A phylogenetic tree based on complete VP1 coding sequence was constructed with Kimura 2-Parameter model and Neighbor-Joining method, using MEGA. The bootstrap test was performed with 1,000 replications. ## 4. Evolutionary analyses based on Bayesian MCMC method The evolution rate, time of most recently common ancestor (t_MRCA), molecular clock phylogeny, and demographic history of the EV71 strains circulating in mainland China were inferred using Bayesian Markov chain Monte Carlo (MCMC) method in BEAST version 1.4.8 ([www.beast.bio.ed.ac.uk](http://www.beast.bio.ed.ac.uk)). Because of the large epidemic of EV71 in China since 2007, a great amount of the sequence data were collected from 2007 to 2009, to reducing the computation load, sequences with high homogeneity (e.g., from the same outbreak) were deleted. Thus, 82 representative sequences were selected for the evolutionary and natural selection analyses. We analyzed the data using both the Hasegawa-Kishino-Yano (HKY) and the General Time Reversible (GTR) Nt substitution models with the gamma distribution of among-site rate variation with four categories estimated from the empirical data. The codon positions were partitioned into 1^st +2^nd and 3^rd, in which 3^rd codon position had a separate relative substitution rate parameter with others. Also, two different models of rates variation among branches: the strict clock and the uncorrelated lognormal distributed (UCLD) relaxed molecular clock were implemented in our analyses. Both constant and exponential growth population size coalescent were used as tree prior to avoid impact from different demographic models. For each model the MCMC chain was run for 30,000,000 steps and sampled every 1,000 steps. The first 3,000,000 steps of each run were discarded as burn-in. This ensured that the effective sample sizes for all the parameters were more than 500 for all the models. To reconstruct the demographic history of EV71 in China, Baysian skyline plot was estimated , inferring the tendency of effective infections since t_MRCA in China. ## 5. Natural selection analyses The nonsynonymous and synonymous substitution rates (dN and dS, respectively), and the *d*N/*d*S ratio were computed to estimate the natural selection pressure on each VP1 codon of EV71. Both Single-likelihood ancestor counting (SLAC) and fixed-effects likelihood (FEL) methods at the Datamonkey website ([www.datamonkey.org](http://www.datamonkey.org)) were performed. In SLAC method, if dN/dS\>1 (or \<1), a codon was called positively (or negatively) selected, p-value (0.1) derived from a two-tailed extended binomial distribution was used to assess significance. P-value (0.1) from one degree of freedom likelihood ratio test was used in FEL method to assess whether the site was under positive or negative selection. ## 6. Ethics Statement HFMD is a notifiable disease in China, and the pathogenic surveillance of HFMD without private information referred is required by the Law of the PRC on the Prevention and Treatment of Infectious Diseases. The data used in this study was obtained from samples as part of this monitoring program and according to this law. No identifying data were used in this study. Thus, the requirement for written informed consent was waived. This study was approved by the second session of Ethics Review Committee in Chinese Centre for Disease Control and Prevention. # Supporting Information We thank all the staffs throughout Chinese surveillance network for HFMD, especially those in provincial and prefectural Centers for Disease Control and Prevention, for their enormous contributions to surveillance. We thank D. Yang in CDC, U.S.A., for assistance in revising the manuscript. We thank Y. Feng, in National Center for AIDS/STD Control and Prevention, China CDC, China, for helpful discussions. We thank C. Kang, in Beijing University, China, for regression analysis. [^1]: Conceived and designed the experiments: WX XT YZ. Performed the experiments: XT X. Huang SZ HC QY HW X. Huo JZ YW DY DW AC HA. Analyzed the data: XT. Wrote the paper: XT WX. [^2]: The authors have declared that no competing interests exist.

# Introduction Neurodegeneration with brain iron accumulation (NBIA) disorders are a set of clinically analogous neurological diseases characterised by neuropathology of the basal ganglia coinciding with iron deposition. Patients display pyramidal and extrapyramidal movement disruption as well as cognitive decline. Pathological examination highlights either axonal swellings with ubiquitinated aggregates, tau tangles or Lewy bodies depending on the NBIA subtype. Mutations in 12 genes have been shown to cause NBIA and each protein has a seemingly disparate cellular function. These functions include iron metabolism, mitochondrial metabolism, lipid homeostasis and autophagy. The most common NBIA subtype is pantothenate kinase-associated neurodegeneration (PKAN), caused by recessive mutations in the *PANK2* gene. This accounts for 35–50% of all NBIA cases. Pantothenate kinase (PANK) catalyses the first step of coenzyme A (CoA) biosynthesis from dietary vitamin B5. CoA has critical roles in multiple mitochondrial metabolic pathways, including the TCA cycle, β-oxidation and fatty acid synthesis. There are four human PANK isoforms; PANK1 and PANK3 are cytosolic, whereas PANK2 is localised to the mitochondria. There is still some contention over the localisation of mouse Pank2 between the mitochondrial membranes and the cytosol. PANK4 is an isoform presumed to lack catalytic activity. CoA is present in the mitochondrial matrix at 1000-fold higher levels than in the cytosol and PANK2 is the major active PANK isoform in the human brain. Despite rodent brain tissue being less enriched for Pank2 than human brain, its central role is demonstrated as *Pank2* knockout mice have 60% reduced total PANK activity in neural tissue. These data support a primary role for PANK2 and CoA in neuronal mitochondria. However, mitochondrial CoA is yet to be measured in patient-derived or mouse model brain tissue. The mechanism by which *PANK2* mutations lead to neurodegeneration is not known but several animal models have been generated to facilitate investigation of disease mechanisms. *Drosophila* have one Pank orthologue and deletion partially recapitulates some of the movement phenotypes and reduced lifespan observed in PKAN. Addition of human mitochondrial PANK2 is able to rescue this phenotype. Interestingly, while *Drosophila* cytosolic Pank isoforms are not able to rescue knockout fly phenotypes, addition of the human cytosolic isoforms provide a partial rescue. *Pank2* knockout mice also show a similar phenotype, but only when metabolically stressed with a ketogenic diet. These animal models fail to display iron accumulation. Patient fibroblasts have been shown to display defective iron handling, increased reactive oxygen species damage (ROS) and mitochondrial physiological deficits–findings that were replicated in human neurons for the first time after direct reprogramming from patient fibroblasts and subsequently reinforced in iPSC-derived neurons. PKAN patients display iron accumulation in the globus pallidus and have cellular pathology, namely axonal swellings and gliosis, affecting the cortex as well as the neurons of the globus pallidus. Despite being an essential element of cell survival, it is unclear whether iron accumulation is causative or consequential to neurodegeneration. Many cellular enzymes make use of heme iron for normal folding and function as well as iron-sulphur clusters for enzymatic function; notable examples are the complexes of the electron transport chain of oxidative phosphorylation. Deregulated iron can be potentially harmful to the cell as, depending on its oxidative state, it can lead to free radical formation via the Fenton reaction. Therefore, tight cellular mechanisms for import, storage and export of iron from the cell exist. Neuronal iron is predominantly imported via endocytosis of the Transferrin Receptor (TfR) and either stored intracellularly through complexes such as Ferritin, consisting of both heavy (FTH) and light (FTL) chains, or utilized immediately by the cell for normal function. For its use in aerobic respiration, iron is imported into the mitochondria via mitoferrin transporters (MFRN1/2) and mitochondrial specific ferritin (MTFT) stores mitochondrial iron until it is required or when in excess. Iron is exported from neurons via Ferroportin (FPN), which requires cell surface stabilization through binding to β-amyloid precursor protein. Iron within a healthy cell is predominantly contained within mitochondria and lysosomes and upon entering the mitochondria it is thought to accumulate as no mitochondrial iron exporter has been described thus far. This has led to the suggestion that mitophagy may be a mechanism for liberating iron from mitochondrial stores. The present study sets out to investigate the consequences of *PANK2* mutations on iPSC-derived cortical neuronal cells in culture. Fibroblasts from three atypical PKAN patients were reprogrammed to iPSCs and, along with three control pluripotent stem cell (PSC) lines, differentiated into cortical neuronal cells using a highly efficient differentiation paradigm. Mitochondrial dysfunction was observed, namely altered NADH and FADH supply to oxidative phosphorylation as well as increased reactive oxygen species (ROS) production and oxidative damage. Changes to iron and CoA metabolism were not witnessed. Additionally, it was shown that iron chelation led to increased oxidative damage in patient and control neuronal cells. These findings enable analysis of early pathological events in PKAN without the context of aging and complex late-stage disease. # Materials and methods ## Cell culture and reprogramming of fibroblasts All culture reagents were purchased from Thermo Fisher unless otherwise stated. Patient biopsies were taken using a skin punch under informed consent (ethical approval from the NHNN and IoN joint research ethics committee, study number 10/H0721/87). Fibroblasts were cultured as previously described. Briefly, 5mm biopsies were taken with a skin punch and then allowed to expand in DMEM supplemented with 10% FBS. Fibroblasts were passaged using 0.05% trypsin/EDTA. Fibroblasts were reprogrammed using the episomal plasmids as described by Okita et. al.. The episomal plasmids were obtained from Addgene (plasmids \#27077, \#27078 and \#27080) and fibroblasts were nucleofected using the Lonza P2 nucleofection kit (Amaxa). Nucleofected cells were changed to iPSC culture media after 7 days and colonies were manually picked after they appeared around 30 days post nucleofection. iPSCs and hESCs were maintained in Essential 8 media on Geltrex coated plates. Cells were routinely passaged using 0.5 mM EDTA. Karyotype counts and G banding analysis was performed by Cell Guidance Systems (Cambridge, UK). At least two clonal iPSC lines from each patient were taken forward for experimentation and compared to two control iPSC lines and one hESC line; termed Control 1, Control 2 and hESC Control respectively. The hESC line Shef6 was obtained from the UK Stem Cell Bank, Control iPSC line 1 was generated from a neurologically normal individual in the lab of Dr Tilo Kunath and Control iPSC line 2 was obtained from the Coriell repository. Stem cells were differentiated to cortical neuronal cells using the protocol described by Shi et. al.. Briefly, cells were subjected to 10 days dual SMAD inhibition using 1 μM dorsomorphin (Tocris) and 10 μM SB431542 (Tocris), followed by extended neurogenesis in N2B27 media containing retinoids. The final time point for all experiments was taken as 100 days post neural induction. ## Immunocytochemistry Cells were fixed in 4% PFA, washed three times in PBS with 0.3% Triton X-100 (PBST) to permeabilize the cells and blocked in 3% BSA. Primary antibodies were incubated overnight at 4°C in blocking solution. Cells were then washed three times in PBST and secondary antibodies (Alexa Fluor, Thermo Fisher) were added for 1 hour at room temperature in the dark in blocking solution. Finally, cells were washed once with PBST containing 1 μM DAPI and then twice in PBST before being fixed and mounted. Images were taken on a Zeiss LSM microscope or a Zen confocal microscope. Counting data was taken from 5 images per replicate; areas were randomly selected in the DAPI channel and automated counting was performed using the ITCN nuclear counting plugin for Image J, using the same threshold setting throughout. ## qPCR RNA was isolated from samples using Trizol reagent and purification was performed following the manufacturers’ instructions (Thermo Fisher). Reverse transcription was performed on 2 μg of RNA using Superscript reverse transcriptase III and random hexamer primers. Power Sybr Green mastermix (Thermo) was used for the qPCR reaction on the Agilent MX3000P qPCR system with annealing temperatures of 60°C for all primers used. All results are relative to three housekeeping genes *GAPDH*, *Cyclophilin* and β*-actin*. ## Western blot analysis Samples were treated with 50 μM FAC for 18 hours in normal cell culture media. Samples were lysed in RIPA buffer containing 10 mM Tris pH 8, 140 mM NaCl, 1 mM EDTA, 0.5 mM EGTA, 1% Triton X-100 0.1% sodium deoxycholate, 0.1% SDS plus protease and phosphatase inhibitors (Roche), for one hour on an orbital shaker at 4°C, followed by centrifugation at 10,000 g for 15 minutes at 4°C. Protein concentrations were measured using the BSA assay (Biorad) and samples were separated on a NuPage 10% SDS polyacrylamide gel (Novex) before being transferred onto a nitrocellulose membrane for western blotting. Membranes were blocked in 3% milk in PBS containing 0.1% Tween 20. Primary antibodies were added to the membranes in blocking solution overnight at 4°C. Blots were then washed three times before secondary antibodies were added in blocking solution for one hour. After final washes, images were captured and densitometry analysis performed on the Li-Cor Odyssey imaging system (Li-Cor). ## HPLC CoA species were extracted from cells with ice-cold perchloric acid (PCA, 3.5%). After centrifugation at 21,000 g for 5 min at 4°C, the supernatant, containing CoASH and short-chain CoA esters, was collected and 1 M triethanolamine (TEA) was added to a final concentration of 100 mM. The pH was adjusted to pH 6 with 5 M K₂CO₃ and potassium perchlorate pellet was removed by centrifugation at 21,000 g for 3 minutes at 4°C. For the quantification of the total level of acid-soluble CoA esters (combined level of unesterified CoA and short chain CoA), 5 M KOH and 100 mM tris (2-carboxyethyl) phosphine were added to neutralized PCA extracts to final concentrations of 0.5 M and 10 mM, respectively. KOH hydrolyses all PCA-soluble esters into unesterified CoA which were then measured by HPLC. After incubation at 25° C for 5 minutes, the pH was adjusted to pH 6 with 5% PCA. CoASH and short-chain CoA esters were measured by HPLC as previously described except EDTA was omitted from the injection mixture. For the quantification of total long-chain acyl CoAs, the PCA pellets were solubilised in 89 mM TEA, 0.44 mM KOH and 11.1 mM DTT by gentle sonication to hydrolyse long-chain CoA esters to unesterified CoA. After incubation for 5 min at 25° C, proteins were precipitated by PCA and pelleted by centrifugation at 21,000 g for 10 min at 4°C. The supernatant was collected and the pH was adjusted to 6–7 with 0.5 M K₂CO₃ and centrifuged again at 21,000 g for 3 min at 4°C. The supernatant was collected and CoA was measurement by the CoA recycling assay adapted to a plate reader format. ## Mass spectrometry The UPLC-MS/MS instrument consisted of a Waters ACQUITY UPLC system coupled to a Xevo TQ-S triple quadrupole mass spectrometer with an electrospray ionization source. The mass spectrometer was operated in negative ion mode and data were acquired using MassLynx V4.1 software. Chromatographic separations were achieved using a Waters CORTECS C₁₈ column (1.6 μm, 2.1 x 50 mm), with a CORTECS C₁₈ VanGuard pre-column (1.6 μm), which was maintained at 40°C. Binary gradient profiles were developed using water with 0.01% formic acid (A) and methanol (B) (HPLC grade, Merck) at a flow rate of 700 μL/min. Separations were conducted under the following chromatographic conditions: 100% solvent A for 1 min, decreased to 10% over 1 min, maintained for 1 min at 10% before being increased to 100% over 0.1 min. Column equilibration time was 0.9 min, with a total run time of 4 min. The injection volume was 10 μL. Mass spectrometric conditions were as follows: capillary voltage 2.5 kV, cone voltage 60 V, source temperature 150°C, desolvation temperature 600°C, cone gas flow 150 L/h, desolvation gas flow 800 L/h, collision gas flow 0.25 L/h and nebulizer gas flow 7 bar. Dwell time was set at 8 msec for each analyte. The quantitation of vitamin B5 and isotopically-labelled vitamin B5 (Sigma) was then performed using the multiple reaction monitoring (MRM) method described in. It is important to note for isotopically labeled pantothenate, treatment was performed in media deficient of pantothenate: HBSS media supplemented with N2 and B27 supplements, NEAA and L-glutamine as the above media recipes. Sample preparation. Cell pellets were reconstituted in 20 μL of buffer (pH 7.8), consisting of 100 mM Tris base, 6 M urea, 2 M thiourea and 2% ASB-14 (adjusted to pH 7.8 with HCl). Samples were shaken for 30 minutes (1000 rpm; 37⁰C) before being diluted 1:100 with water and analyzed via UPLC-MS/MS. ## Inductively-coupled-plasma mass spectroscopy (ICP-MS) Cellular iron content was analyzed by ICP-MS using the protocol previously reported. Briefly, 150 μg of total protein as measured by Bradford protein assay, was lyophilized before resuspension in 100 μl nitric acid (69% v/v; ultraclean grade, Aristar) overnight at room temperature (RT). Samples were then heated for 1 hour at 90°C, before the addition of an equivalent volume of hydrogen peroxide (30%, Merck). Sample was incubated for 15 min at RT before a further 30 min at 70°C. To evaluate metal content against calibration standards (#IMS-102; Ultra Scientific) samples were diluted in double-distilled water until within quantifiable parameters using a NexION 350X inductively coupled plasma mass spectrometer (PerkinElmer, Waltham, MA, USA). Each sample was measured in triplicate and normalized to total protein concentration. ## Live cell imaging For live cell imaging iPSC cells were incubated with 25 nM TMRM (tetramethylrhodamine, methyl ester; a cell permeant, cationic, red-orange fluorescent dye sensitive probe for mitochondrial membrane potential and used in the redistribution mode) for 40 minutes in a HEPES-buffered salt solution (HBSS) composed of (mM): 156 NaCl, 3 KCl, 2 MgSO₄, 1.25 KH₂PO₄, 2 CaCl₂, 10 glucose and 10 HEPES; pH 7.35. Images were obtained using a Zeiss 710 Laser Scanning Microscope (CLSM) with an integrated Meta detection system and a 40x oil-immersion objective. Illumination intensity was kept at the minimum of laser output and the pinhole was set to give an optical slice of ∼1 μm. TMRM was excited using the 560 nm laser line and fluorescence measured above 580 nm. Calcein-AM based cell area measurements were used for normalization. The NADH autofluorescence was measured with excitation at 405 nm and emission at 440–480 nm. FAD autofluorescence was determined using 458 nm Argon laser line and fluorescence was measured from 505 to 550 nm. FAD and NADH redox indexes and mitochondrial pools were estimated by sequentially applying 1 μM of the mitochondrial uncoupler FCCP (carbonyl cyanide p-trifluoromethoxyphenylhydrazone), followed by 1 mM of the complex IV inhibitor sodium cyanide. Lipid peroxidation experiments were performed using, C11-BODIPY (581/591, 2 μM, Molecular Probes) was excited at the 488 and 565 nm laser lines and fluorescence measured from 505 to 550 nm and above 580 nm (40x objective). GSH levels were assessed using monochlorbimane, MCB (Molecular Probes, Invitrogen) fluorescence with excitation at 351 nm and emission at 435–485 nm. Data were acquired and analyzed using ZEN2009 software. Superoxide generation was measured with Dihydroethidium (HEt; 2 μM, Invitrogen). Fluorescent images were acquired with a frame interval of 10 s. Data were analyzed using software from Andor IQ (Belfast, UK). Statistical analysis and data analysis were performed using Origin 9 (Microcal Software Inc., Northampton, MA, USA) software. Results are expressed as means ± standard error of the mean (SEM). ## Sanger sequencing gDNA was extracted from iPS cells and differentiated neuronal cells from all utilised patients and control clones using a standardized phenol extraction. Primers and touch-down PCR programmes as available in were utilised to Sanger sequence the specified mutations within the PANK2 gene as given in for the respective disease and control clones to confirm their presence/absence. ## Statistical analysis Data represent three control PSCs and at least two clones from three unrelated PKAN patients. Number of independent inductions is depicted in histograms. Where appropriate, statistical significance was calculated using the Student’s *t* test or ANOVA, as stated. Histograms represent mean values and error bars represent standard error of the mean. \*p\<0.05, \*\*p\<0.01, \*\*\*p\<0.001. # Results ## Generation of patient-derived iPSCs and cortical neuronal cells iPSC lines from three NBIA patients with confirmed *PANK2* mutations were generated to study the mutation effect in human neurons *in vitro*. These patients each have compound mutations leading to atypical PKAN with later disease onset and less severe progression than classical PKAN. Clinically, these patients display iron accumulation evident by T2\* MRI and a generalised dystonia phenotype. Fibroblasts from the 3 patients were reprogrammed to iPSCs using the non- integrating episomal reprogramming method. At least two iPSC clones from each patient were picked and expanded for further characterisation, alongside two control iPSC lines and one control hESC line. To confirm pluripotency in the newly derived iPSC lines, expression of the pluripotency markers OCT4 and SSEA4 were immunocytochemically identified. *PANK2* iPSCs displayed similar expression of these pluripotency markers and characteristic colony morphology to control lines. qPCR analysis also showed that the gene expression of the core pluripotency markers *OCT4*, *NANOG* and *SOX2* was comparable to hESCs in contrast to nascent fibroblasts and that the newly formed iPSCs faithfully silenced fibroblast markers *S100A4* and *VIMENTIN*. The genomic integrity of the newly formed lines was confirmed by karyotype stability and G-band analysis and all disease lines were confirmed to carry the compound heterozygous mutations by Sanger sequencing, as specified in. Absence of these mutations was confirmed in control PSC lines. To generate neurons, we subjected the iPSCs to a cortical differentiation protocol due to the very high efficiency of differentiation (\>95%). All lines tested from the patient-derived iPSCs were able to faithfully generate forebrain patterned neural precursor cells, expressing the telencephalic marker OTX2. After 100 days of neurogenesis, control and patient-derived PSCs generated deep layer (TBR1-positive), upper layer (SATB2-positive) and middle layer (CTIP2-positive) cortical neuronal cells. The ability of lines to undergo cortical neurogenesis showed some variability but was comparable between all patient lines, consistent with a non-developmental disease. Sanger sequencing was performed on genomic DNA from terminally differentiated cells and all patient-derived heterozygous mutations were confirmed. Western blot analysis demonstrated a reduction of mature PANK2 protein in patient-derived neuronal cultures in comparison to control cell cultures, (patient 1 32.8±4.1%; patient 2 36.5±4.5%; patient 3 30.2±2.1%), despite only one patient displaying a premature stop codon mutation. ## CoA homeostasis in patient-derived neuronal cells is comparable to control Only a subset of mutations in *PANK2* that lead to PKAN reduce enzyme activity and affect protein folding. Indeed, changes in CoA levels in human cell lines or brain tissue have yet to be demonstrated. Therefore, absolute levels of CoA and acetyl-CoA were investigated in the cultures via HPLC. Primary analysis of the free CoA to acetyl-CoA ratio in iPSC cultures and differentiated cells indicated an undefined switch in the metabolic states between the two cell types, seen via a relative decrease in free CoA and an increase in acetyl-CoA levels in neuronal cells. This ratio change between iPSCs and neurons was independent of *PANK2* mutation. No change in short chain CoA species (free CoA (CoASH), acetyl-CoA and short- chain CoA esters) was observed by HPLC. As it has been previously reported that *PANK2* mutations could lead to the disruption of β-oxidation of fatty acids, long chain CoA derivatives were also evaluated. While some variability arose between cell lines, no overall significant difference was observed between control and mutant lines. Together, these data signify a comparable steady state of total CoA levels in the patient-derived neuronal cells to control cultures. To infer changes of synthesis and breakdown of CoA in patient-derived neuronal cultures, stable isotope labelled vitamin B5 was used to measure neuronal uptake and handling. Cells treated with labelled vitamin B5 (pantothenate) for up to 24 hours displayed a similar rate of uptake between control and patient-derived neuronal cultures. No change also inferred that the cellular lifetime of vitamin B5 was unaffected by the presence of the *PANK2* mutation, shown via similar homeostasis of the labelled pantothenate. ## Patient-derived neuronal cells exhibit dysfunctional mitochondrial respiration In order to test whether the mutations in *PANK2* affect the mitochondrial health in our iPSC-derived neuronal cells, mitochondrial membrane integrity and live cell imaging experiments were performed. Mitochondrial membrane potential (Δψ_m), is an indicator of mitochondrial function, and was assessed using TMRM fluorescence. The presence of the *PANK2* mutation increased Δψ_m in both iPSC (pooled controls: 470.3±72.2, n = 48 versus pooled patients: 1207.0±95.3, n = 72. \*\*\*p\<0.0001) and neuronal cells (pooled controls: 1254.3±67.7, n = 48; versus pooled patients: 1685.6±135.1, n = 106; p = 0.005, ⁱⁱ). Changes in mitochondrial membrane potential have been reported in other PKAN models, albeit with opposing results. Here, altered maintenance of Δψ_m is likely to represent a compensatory mechanism to mitochondrial deficits. Some variability between cell lines was apparent, as demonstrated by patient 2 and 3 derived cells displaying significantly higher Δψ_m compared with controls, whereas patient 1 derived cells consistently demonstrated Δψ_m similar to control lines. Due to this variability, data for individual cell lines is presented, and significant comparisons are depicted by the pooled data (ⁱⁱ). The known association of PKAN pathology with brain iron accumulation led us to examine whether iron was involved in changes to the mitochondrial membrane potential observed in patient cells. Pre-treatment with the iron chelator deferoxamine (DFO) at 50 μM for 30 minutes increased neuronal Δψ_m (control 1: 1317.7±65.8 to 2023.8±104.3, n = 48; control 2: 1025.7±62.8 to 3741.2±149.3, n = 49; patient 1: 1987.3±96.2 to 3999.4±270.7, n = 33; patient 2: 1864.0±256.1 to 2272.3±238.3, n = 49; patient 3: 2272.3±238.3 to 3989.1±51.6, n = 49. \*\*\*p\<0.001) representing a further increase in membrane potential above non-treated PKAN-associated levels as a result of iron chelation. The basis for differences in Δψ_m maintenance was further investigated by exposing the cells to known specific inhibitors of the electron transport chain complexes. A comparable decrease (\~70% of basal levels) in control and PKAN neuronal cells of TMRM signal after oligomycin treatment (complex V inhibitor), suggested complex V in these cells was working in reverse as an ATPase to maintain mitochondrial membrane potential. Rotenone (a complex I inhibitor) also induced a decrease of TMRM fluorescence, however PKAN cells responded slower and to a reduced extent than control cells. This suggested that either complex I in PKAN neuronal cells was ineffective at generating a membrane potential or NADH substrate availability from the TCA cycle was decreased. To elucidate whether complex I was dysfunctional or starved of substrates, mitochondrial NADH redox levels were measured and found to be significantly reduced in PKAN neuronal cells under basal conditions (pooled controls: 60.1±9.5%, n = 24; versus pooled patients: 24.3±10.6%, n = 36. \*p = 0.0139). The application of the mitochondrial uncoupler FCCP, to fully deplete NADH levels by stimulating mitochondrial respiration, and NaCN, to block mitochondrial respiration and thus NADH consumption, enabled the rate of NADH production and maximal pool of NADH in mitochondria to be estimated. A trend towards a reduced total pool of NADH was evidenced in PKAN neuronal cells. Together these observations were thus able to explain the reduction in complex I activity due to lack of NADH substrate availability. Similar analysis of the complex II substrate FAD revealed that PKAN neuronal cells displayed a significantly lower FAD redox index (pooled controls: 89.5±8.2%, n = 24; versus pooled patients; 42.4±8.0%, n = 36. \*\*\*p = 0.0001), indicating an inhibited rate of complex II dependent respiration. Altogether, these investigations provide evidence for a defective electron transport chain in neurons carrying the *PANK2* mutation. ## Increased ROS production and oxidative stress in PKAN patient-derived neuronal cells Using the fluorescent dye monochlorobimane (MCB), levels of the reduced form of the antioxidant glutathione (GSH) were measured in control and patient-derived neuronal cells. Compared with controls cells, MCB fluorescence intensity was significantly decreased in both undifferentiated patient-derived iPSCs (pooled controls: 512.3±73.5, n = 90; versus pooled patients: 431.0±17.6, n = 127. p = 0.3022) and PKAN neuronal cells (pooled controls: 1151.0±36.6, n = 105; versus pooled patients: 675.6±21.7, n = 120. \*\*\*p\<0.0001), representing a decreased intracellular antioxidant pool. The lower reduced glutathione level in iPSCs compared to neuronal cells is consistent with the low oxidative phosphorylation status and ROS production of iPSCs that metabolically correspond to an early embryo. ROS generation rates in the neuronal cells were measured by dihydroethidium oxidation as a direct readout of superoxide (O₂^-) production. Under basal conditions, O₂^- generation was consistently higher in PKAN neuronal cells compared with control cells (pooled controls: 41.7±2.6, n = 46; versus pooled patients: 83.2±3.0, n = 54. \*\*\*p\<0.0001). Lipid peroxidation, assessed via the ratiometric dye BODIPY-C11, was also found to be significantly higher in PKAN neuronal cells from two of the patients whereas cells derived from patient 2 showed a similar trend but did not reach significance (0.59683±0.06744, n = 196 for pooled patients, compared to 0.26±0.022, n = 86 for pooled controls; \*\*\*p\<0.0001). Following the findings in, the relationship with iron and PKAN neuron associated oxidative damage was explored. Surprisingly, 30 minutes pre-treatment of neuronal cells with DFO increased ROS generation in both control and patient- derived cells (pooled controls: 112.5±4.6; n = 46; versus pooled patients: 122.0±2.4; n = 53. p = 0.0782). Taken together, these data provide evidence that the mitochondrial dysfunction observed in patient-derived neuronal cells leads to increased ROS generation and downstream oxidative damage. In our conditions, iron chelation exacerbated ROS generation and worsened the PKAN neuronal phenotype. ## Cellular response to iron in PKAN patient-derived neuronal cells Iron accumulation is a progressive pathological characteristic of PKAN seen via MRI during life, however, it is unclear whether this is a cause of neurodegeneration or a consequence. Here, the iron response pathway was assessed in control and patient-derived neuronal cells. Response of the cellular iron-handling pathway was analyzed in response to 18 hours iron treatment (50 μM ferric ammonium citrate (FAC)). Across all cell lines and consistent with a reduction in cellular iron import, TfR expression was reduced at the transcriptional and expression levels. The cytosolic iron storage molecule Ferritin (FTH and FTL) exhibited increased protein levels in all lines, consistent with its translational control by iron response genes. Interestingly, *Ferroportin* expression was significantly increased in patient- derived neuronal cells versus controls, in basal conditions and in response to iron treatment, suggesting a cellular response for increased iron export. No significant transcriptional change in response to iron was observed for *MFRN1/2 or PANK2*, however, a trend to increased *FTMT* expression was witnessed in patient-derived cells, hinting to increased mitochondrial iron storage. It is noteworthy that *PANK2* expression is similar between control and patient- derived neuronal cells, whereas protein levels are reduced in patient-derived cells. ## Total iron content is unchanged in patient-derived neuronal cultures To investigate whether the small differences in expression of *FTMT* and *FPN* in response to extracellular iron related to altered cellular iron content, absolute metal ion content was measured by inductively coupled plasma mass spectrometry (ICP-MS). In all neuronal lines, elevated intracellular iron content upon incubation with FAC (50 μM; 18 hours) was confirmed and no change was observed between control and PKAN neuronal cells. Data indicates a largely appropriate iron response from PKAN neuronal cells that preserves intracellular levels of iron even in an elevated extracellular iron environment. An exception is the consistent increase in *FPN* expression in patient-derived cells versus controls in both basal conditions and under iron stress, which may indicate a compensatory measure to maintain intracellular iron levels by enhancing iron export. # Discussion In this study, PKAN patient-derived neuronal cells have been generated in an attempt to identify early mechanisms of neurodegeneration in NBIA. These cells represent a human model with appropriate gene dosage of clinically proven pathogenic mutations in which to study the earliest underlying consequences of *PANK2* mutations in developmentally immature neuronal cells. Cortical neurons represent a relevant model of PKAN, as one of the cell types affected by axonal swellings and gliosis, additional to the main site of pathology in the globus pallidus. There currently exists no stem cell differentiation protocol towards pallidal neurons and the cortical differentiation protocol employed here is highly efficient and generates very homogenous cortical cultures due to the default nature of this developmental paradigm. It should be noted that other cell types exist in the culture, for example glial cells and some progenitor cells that persist, together representing less than 5% of cells. These cells represent a minority, however we cannot discount contributions of these cells to our data, for example the provision of metabolites from astrocytes to neurons for oxidative phosphorylation. We observed a significant degree of variability in cellular responses to many experimental paradigms; including the fact that cells derived from patient 1 often responded similar to controls. These variabilities may represent tissue culture artefact and a consequence of the cells being cultured outside of their natural context and one example of this is the substantial change in transcription witnessed in primary microglia just six hours after being placed in culture. Our observation that the relative proportion of free CoA to acetyl-CoA changes through differentiation suggests a metabolic alteration between the stem cells and neurons. This finding calls for future investigations to describe the metabolic states of the two cell types and changes through neuronal differentiation. We find that point mutations associated with atypical PKAN lead to unaltered *PANK2* expression in neuronal cultures but a reduction in PANK2 protein levels. Hypothetically, this could be explained via alterations in folding and degradation of the putative unfolded protein; further investigation is required. These findings are in line with iPSC-derived neurons harbouring premature stop codons; showing unaltered *PANK2* transcription but a total lack of protein. It is noteworthy that biochemically, a number of peptides harbouring PANK2 mutations display normal enzyme function. Thus protein dosage may be central to the disease mechanism, with juvenile onset displaying no PANK2 protein and atypical PKAN a reduced level. The data presented here demonstrates that pathogenic point mutations in *PANK2* do not alter neuronal CoA levels or the metabolic flux of pantothenate. This suggests that there is either a distinct cellular function for mitochondrial PANK2 or that cytosolic PANK enzymes may be compensating for a PANK2 deficiency. The ability of mitochondrial human PANK2 and only mitochondrial isoforms of drosophila Pank to rescue phenotypes in Drosophila Pank knockouts, supports the former concept, whereas the ability of human PANK3 and PANK4 to partially rescue knockout flies supports the latter. Alternatively, CoA flux might respond in a faulty manner in relation to stress in the setting of the ageing or diseased brain. Dysfunctional mitochondrial oxidative phosphorylation may be a key component in the PKAN brain and altered mitochondrial membrane potential has been seen in other PKAN models. Reduced mitochondrial membrane potential and defective ATP production has been described in PANK2 knockout mouse models, patient-derived fibroblasts as wells as induced neuronal models. The increased membrane potential seen in this study may represent a compensatory response of the cells to mitochondrial deficits, in the context of atypical disease mutations. Mitochondrial and metabolic deficiencies could explain the adult onset of neurodegeneration through altered environmental stresses and diet in this highly energy demanding cell type. This is reinforced by mutations in *PANK1* being linked with hyperglycaemia. We report a metabolically immature neuronal phenotype in PANK2 mutants as well as in control lines, as seen by depolarization of mitochondria in response to oligomycin application after 100 days of differentiation. This result highlights that all cultures analysed in this study are partially glycolytic, contrary to mature primary neurons in culture. Due to this observation, some metabolic consequences of *PANK2* mutations may be masked by the fact that our cultures are not entirely dependent on mitochondrial oxidative phosphorylation. Closer analysis of the mitochondrial physiology demonstrated that the cells were defective in complex I of the electron transport chain. Our data indicate that a lack of NADH substrate provided by the TCA cycle may be central to this deficiency. In addition to NADH, *PANK2* mutations reduce the amount of FADH production in complex II. This observation is reinforced by changes in NADH in tissue homogenate in *Pank1*/*Pank2* double knockout mice and validates the hypotheses gleaned in studies in iPSC-derived neurons. Importantly, irrespective of *PANK2* mutations, our cultures rely on glycolysis to compensate for inefficient oxidative phosphorylation and for increased ATP consumption by the ATPase to maintain the mitochondrial membrane potential. It is interesting that undifferentiated iPSCs display some mitochondrial abnormalities in addition to neuronal cells, namely increased TMRM fluorescence and reduced antioxidant levels. Levels of antioxidants have been shown to be altered in PKAN patient- derived fibroblasts and transdifferentiated neurons, including lower levels of reduced glutathione. However, in the brain an increase in levels of glutathione- cysteine have been described, suggesting increased oxidised glutathione and potentially oxidative stress. These consistencies again suggest a selective vulnerability of certain neurons to a consistent mutation-associated phenotype. The reported ultrastructural disruption of cristae organization and mitochondrial swelling in a PKAN transgenic model and in iPSC-derived neurons could occur as a cause or result of the metabolic dysfunction described here. This physical disruption of the respiratory chain could lead to electron leakage and could be another reason for the reversal of the ATP synthase to maintain mitochondrial membrane potential described here. ROS is a normal cellular signal for multiple physiological processes; however prolonged exposure to high levels combined with environmental stresses will inevitably lead to damaged cellular components. Increased ROS generation and subsequent lipid peroxidation as a result of altered oxidative phosphorylation in PKAN neurons could potentially explain the post-developmental onset of the NBIA. Of note, is the finding that iron chelation increases mitochondrial membrane potential and ROS generation in both control and mutant cells. This can be explained as DFO is a specific Fe(III) chelator, altering the equilibrium between the two redox states. In turn, DFO thus favours the oxidation of Fe(II) to Fe(III), leading to the release of electrons and the formation of ROS. This is an important finding with respect to the relevance of iron to mitochondrial homeostasis, however, it is important to consider that we do not see iron accumulation in our model. For this reason, we cannot comment on the effectiveness of iron chelation with respect to ongoing clinical trials in PKAN using the cell permeable iron chelator deferiprone. Current findings calls for further investigations in other model systems that display iron accumulation and elucidating the role of iron in healthy mitochondria. Iron deposition is a characteristic feature of NBIAs such as PKAN and is increasingly apparent as an early pathological feature in other neurodegenerative diseases such as AD and PD. It is still unclear whether defective iron homeostasis is a cause or consequence of the neuropathological events in these diseases, but brain imaging has identified that its accumulation is clearly progressive during life and may occur prior to symptom onset. The homeostatic response of iron regulatory proteins and total intracellular iron levels appear largely normal in patient-derived neuronal cells under these conditions. However, altered *FPN* expression reported here and in *PANK2* knockout cell lines suggest an increase in the iron export pathway may exist. A trend to increased *MTFT* expression in *PANK2* mutant neuronal cultures not only reinforces a mitochondrial defect, but also may indicate a further attempt by the cell to sequester excess iron safely. It is tempting to speculate that a mitochondrial defect could lead to altered iron storage in the mitochondrial matrix and increased iron export via FPN, leading to iron dyshomeostasis and a potential accumulation over time. Thus, mitochondrial phenotypes may theoretically precede iron accumulation and underlie PKAN disease progression. Orellana et. al. performed investigations into PKAN in cortical neurons derived from iPSCs. This study focused on early onset-associated mutations that lead to a lack of mature PANK2 protein. The authors also see mitochondrial abnormalities and increased ROS production in iPSC-derived cortical neuronal cultures, albeit it with reduced mitochondrial membrane potential. This may hint to different compensatory mechanisms between early-onset and late-onset-associated mutations in PANK2. The authors also hypothesise altered NADH supply to the mitochondria; here we have shown this link to be valid via reduced NADH redox ratios. Orellana et al elegantly show that the wild-type PANK2 transgene can reverse the disease phenotypes, but also put forward extracellular CoA supplementation as a novel therapeutic avenue. This strategy has also been shown to reverse disease phenotypes in model organisms of CoA imbalance, as developmental, vasculature and metabolic deficiencies from Coasy and Pank2 knockdown zebrafish are also reversed via CoA supplementation. In conclusion, the generation and characterisation of PKAN patient-derived iPSC neuronal cells has provided new insights into the underlying mechanisms of NBIA with relevance to other diseases exhibiting iron accumulation. Reduced cofactor supply for oxidative phosphorylation can explain the mitochondrial defects in patient-derived neuronal cultures, which in turn precedes iron accumulation. Additionally, the effects of iron chelation described here call for careful consideration in the future therapeutic strategies. # Supporting information This work was supported by the Medical Research Council (MRC UK); a pathfinder grant from the Wellcome Trust and Eli Lilly; the Brain Research Trust (BRT) and also supported by the NIHR Queen Square Dementia Biomedical Research Unit and the NIHR UCLH Biomedical Research Centre (BRC). The authors would like to thank all the participants of the study for their essential help with this work. [^1]: This work was, in part, funded by a Pathfinder Grant from Eli Lilly and the Wellcome Trust. This does not alter our adherence to PLOS ONE policies on sharing data and materials.

# Introduction Colorectal cancer (CRC) was the fourth most common cancer following lung, breast, and prostate cancer, and was the second leading cause of cancer related deaths around the world in 2018. According to the annual report of cancer statistics in Korea in 2017, CRC is the second most common cancer type, with 28,111 new cases and the third leading cause of cancer-related deaths behind lung and liver cancer. Mutiple surgical approaches, including hemicolectomy (right or left), anterior resection, and low anterior resection, are commonly indicated for patients with CRC. Complications of colorectal resection can be divided into intraoperative and postoperative ones. Intraoperative complications include bleeding, bowel injuries, ureteral lesions, and bladder injuries. The most frequent postoperative complications include surgical site infection, anastomotic leakage (AL), intraabdominal abscess, ileus, and bleeding. Among these, AL is considered to be the most detrimental in terms of the impact on morbidity, mortality and quality of life. AL leads to anastomotic stricture and impaired colorectal function including reduced neorectal capacity, evacuation problems, fecal urgency, and incontinence. More seriously, AL is associated with an increase in local recurrence and lower long-term survival. Alongside these complications and the potential need for reoperation, AL management consumes extensive healthcare resources and expenses. The rate of AL has been reported to range from 5% to 19% globally in colorectal/coloanal surgery depending upon perioperative factors such as anatomical site, surgical method, operators’ experience, and patient’s clinical characteristics. Although a multicenter study reported an AL rate of 6.3% after laparoscopic rectal cancer surgery in Korea, the AL rate at each hospital showed wide variation ranging from 2.0% to 10.3%. To identify the generalizable economic burden caused by AL, a nationwide population study is necessary but has not been carried out yet in Korea. This retrospective, nationwide claims analysis was conducted to identify clinical and economic outcomes among Korean patients with cancer who had undergone anterior resection (AR), low anterior resection (LAR), or ultra-low anterior resection (uLAR) with a manual circular stapler as the current standard of care. This study sought to assess an association between such factors as patients, procedures, and providers and the incidence of AL, and its economic burden in terms of length of stay (LOS), readmissions, and total costs. # Methods ## Data sources Under the public and single-payer system, health claims data have been generated in the archives of the Health Insurance Review and Assessment Service (HIRA) in the process of reimbursing healthcare providers, covering all Korean citizens. As these reimbursements are predominantly based on the fee-for-service system, claims data contain comprehensive information about treatments, pharmaceuticals, procedures, and diagnoses for almost 50 million beneficiaries. This study was approved by the Public Institutional Bioethics Committee designated by the Ministry of Health & Welfare (MOHW) (No. P01-202007-21-024) including the waiver of informed consent as it was a retrospective study based on a de-identified claims database. ## Study population The HIRA database identified 156,545 patients who underwent AR, LAR, or uLAR (procedure codes: Q2921, QA921, Q2922, QA922, Q2928, QA928) between January 1, 2007 and January 31, 2020, of whom only adult patients (≥19 years) who had undergone surgery as the principal treatment for CRC (ICD-10 code: C18, C19, C20) using manual circular staplers (device code: B1022XXX) were included. After excluding patients who had prior colorectal surgery within a 1-year lookback period, the analysis dataset included a total of 120,245 patients. The “index surgery” was defined as the first colorectal resection for each patient, and the “index hospitalization” as admission for the index surgery. ## Clinical outcomes The following variables were identified as the indicators of surgical complications within 30 days after the index surgery: AL, blood transfusion, urinary tract injury, pulmonary embolism, acute renal failure. Since there was no specific diagnosis code for AL, the following procedures were required for AL occurrence during in-hospital care to meet operational AL definitions: (1) Imaging study including computed tomography scans (2) Administration of antibacterial drugs (more than 7 consecutive days after the surgery) (3) Abdominal reoperation and (4) image-guided percutaneous drainage. AL occurrence is defined by the combination of the above four items. Combinations to meet the definition of AL are any including both (1) and (2), any including (3), or any including (4). Among these combinations to define AL, those that include (3) or (4) were designated as additional intervention cases (AIC), because reoperation and percutaneous drainage are potentially more definitive treatments for AL compared to imaging and antibiotics. Blood transfusion was detected using the procedure code. Urinary tract injury, pulmonary embolism, and acute renal failure were all also identified through the ICD-10 codes. ## Economic outcomes The economic outcomes identified were as follows: LOS for the index hospitalization, LOS for all-cause readmissions within 30 days, total costs for the index hospitalization, total costs for during the period from the index surgery to 30 days post-operation, and total costs for readmissions. To comprehensively understand the economic burden on society, total costs included both out-of-pocket expenditures for patients and costs covered by payer. ## Covariates The following covariates of interest were included: demographics (age, gender), patient characteristics (primary diagnosis for the index surgery, Charlson Comorbidity Index (CCI), other comorbidities (diabetes, chronic obstructive pulmonary disease (COPD), metastatic disease, ischemic heart disease, ischemic stroke), pre-operation radiation therapy, procedure characteristics (surgical procedure, surgical approach, protective ostomy, multiple circular stapler use, multiple linear stapler use), provider characteristics (number of beds) and others (blood transfusion, urinary tract injury). ## Statistical analysis All study variables were analyzed descriptively. Continuous variables were presented as mean, median, and standard deviation, while categorical variables were expressed as frequencies and proportions. The overall incidence of each complication (AL, additional intervention cases, blood transfusion, urinary tract injury, pulmonary embolism, acute renal failure) was identified, and the incidence of AL was further summarized by surgical procedure (AR, LAR, or uLAR), primary diagnosis (C18, C19, C20) and surgical approach (open or laparoscopic). To figure out the factors associated with AL, all covariates considered in this study were compared according to the incidence of AL. Student’s t-test or Wilcoxon’s rank sum test was applied for comparison of continuous variables while a chi-squared test or Fisher’s exact test was conducted for categorical variables as appropriate after testing assumptions such as normality. The univariable and multivariable logistic regression models were fitted to assess the risk factors associated with AL so that the odds ratio of each risk factor could be calculated. The multivariable model included factors that showed the statistical significance based on univariable analysis. LOS and costs were estimated during a 30-day period after the index surgery. Costs were summarized by surgical procedure and approach and compared with the Kruskal-Wallis (K-W) test. When the costs difference was significant by the K-W test, then the Dwass, Steel, Critchlow-Fligner method was applied for post hoc comparison. All analyses were conducted using SAS (version 9.2, SAS Institutes, Cary, NC, USA) and R (version 4.0.3, The R Foundation for statistical computing, Vienna, Austria). A p-value of 0.05 or less was considered to indicate statistically significant difference. ## Ethical considerations The study was reviewed by the Public Institutional Review Board designated by the Ministry of Health and Welfare and determined to be exempt from IRB approval (Review number: P01-202007-21-024). As the study involves no more than minimal risk to patients, IRB approved a request to waive informed consent under the Bioethics and Safety Act of Korea. # Results ## Incidence of anastomotic leakage and additional interventional cases based on the operational definition Among 156,545 patients who underwent AR, LAR, or uLAR from January 2007 to January 2020, a total of 120,245 patients met the eligibility criteria for this study. Among these patients, 33.97% had imaging studies, 7.54% received antibacterial drugs for more than 7 consecutive days, 0.36% had reoperations, and 6.99% had image-guided percutaneous drainage. 5.98% of patients satisfied the operational definition of AL having events corresponding with the defined combination within 30 days after surgery. When applying a stricter definition that requires reoperation or image-guided percutaneous drainage, 4.56% of patients fulfilled the definition of additional interventional cases (AIC). ## Demographics and perioperative clinical characteristics Patient demographics and perioperative characteristics are displayed in. Their mean age was 64.09 (standard deviation \[SD\] ± 11.45) years, and there was no statistically significant difference between patients with and without AL. The proportion of males was significantly higher in patients with AL (66.68% vs. 62.41%, *p*\<0.0001). More patients with AL had LAR or uLAR than those without AL (LAR: 61.44% vs. 53.57% and uLAR: 2.93% and 1.61%, *p*\<0.0001). Patients with AL had more open surgery (32.50% vs. 25.47%, *p*\<0.0001), protective ostomy (27.15% vs. 15.83%, *p*\<0.0001), multiple linear stapler use (57.83% vs. 48.94%, *p*\<0.0001), multiple circular stapler use (1.08% vs. 0.76%, *p* = 0.0027), and blood transfusion (32.44% vs. 17.00%, *p*\<0.0001) than those without AL. The following comorbidities were more frequently observed in patients with AL than those without AL; diabetes (24.48% vs. 20.62%, *p*\<0.0001), COPD (7.17% vs. 6.33%, *p* = 0.0046), metastatic disease (12.65% vs. 7.41%, *p*\<0.0001), ischemic heart disease (10.19% vs. 8.22%, *p*\<0.0001), and ischemic stroke (5.35% vs. 3.97%, *p*\<0.0001). The CCI score in the AL group was significantly higher than that in patients without AL (mean ± SD, 3.75 ± 2.90 vs. 3.18 ± 2.43, *p*\<0.0001). ## Risk factors for anastomotic leakage The results from logistic regression predicting the odds of AL are shown in. Most of the variables had an association with AL in univariable analysis, except for age and the number of hospital beds. In multivariable analysis, male gender had the increased odds of AL (odds ratio \[OR\]: 1.176, 95% confidence interval \[CI\]: 1.117 to 1.239), and older age was associated with the reduced odds of AL (OR 0.995, 95% CI, 0.993 to 0.997). A primary diagnosis of C19 (rectosigmoid junction cancer) decreased the odds of AL (OR 0.888, 95% CI, 0.821 to 0.96, reference: C20 (rectal cancer)), but C18 (colon cancer) did not have a statistically significant association with AL. Diabetes (OR 1.124, 95% CI, 1.059 to 1.192), metastatic disease (OR 1.44, 95% CI, 1.335 to 1.553), ischemic heart disease (OR 1.156, 95% CI, 1.064 to 1.257), and ischemic stroke (OR 1.237, 95% CI 1.107 to 1.382) each increased the odds of AL. Among operative characteristics, protective ostomy (OR 1.546, 95% CI 1.448 to 1.65) and multiple linear stapler use (OR 1.192, 95% CI 1.133 to 1.255) were associated with the higher odds of AL. Meanwhile, AR and LAR were associated with the lower odds of AL versus uLAR (OR 0.619, 95% CI 0.524 to 0.731 and OR 0.691, 95% CI 0.594 to 0.805), and the laparoscopic approach was also associated with the lower odds of AL (OR 0.849, 95% CI 0.804 to 0.897). ## Economic outcomes We compared the total healthcare costs (the index hospitalization alone, the period from the index date to last follow-up date, and readmissions) by procedure and approach. All cost categories showed the highest costs in patients who received uLAR compared to AR and LAR. For the index hospitalization, uLAR with an open approach had the highest mean cost (11,212 ± 6,198 USD), and AR with a laparoscopic approach had the lowest mean cost (6,689 ± 3,256 USD). While open approaches were more costly than laparoscopic approaches for the index hospitalization for both AR and uLAR, an open approach for LAR was associated with lower index hospitalization costs than the laparoscopic approach. ## Economic burden of anastomotic leakage Economic outcomes for patients with AL and without AL are summarized in. The mean costs for the index hospitalization were significantly higher for patients with AL compared to those without AL (8,991 vs. 7,153 USD; *p*\<0.0001). Including the 30-day follow-up period, the mean costs were 10,971 USD for patients with AL and 7,531 USD for those without AL (*p*\<0.0001), reflective of a marked increase in post-discharge resource use among patients in the AL cohort. Patients with AL also required prolonged LOS for the index hospitalization versus those without AL (16.78 days vs. 14.22 days; *p*\<0.0001). Readmission costs for patients with AL were higher than those for patients without AL (3,160 vs. 1,316 USD; *p*\<0.0001). AL patients also required longer mean LOS for readmissions compared to those without AL (20.83 days vs. 13.93 days; *p*\<0.0001). The shows economic outcomes depending on the presence of AIC. # Discussion This study demonstrates that AL after surgery for colorectal cancer is associated with increased costs for the index hospitalization and readmissions. this study is one of the largest studies to measure the economic burden of AL after colorectal surgery, and using a nationwide population dataset, it could overcome selection bias inherent to smaller, single institution studies. Various clinical parameters were significant predictors of AL in our analysis. Male gender, diabetes mellitus, ischemic heart disease, ischemic stroke, uLAR, multiple stapler usage, blood transfusion, and urinary tract injury are reported to be well-known risk factors for a higher rate of AL. However, there were several associations that were a matter of controversy. Previous prospective randomized clinical trials for rectal cancer showed no significant difference in the incidence of AL between open and laparoscopic surgeries. In contrast, one cohort study revealed an increased rate of AL in laparoscopic surgery versus open surgery (10.8% vs. 3.4%, p = 0.012) in patients with mid and low rectal cancer. This result, however, needs to be interpreted with caution, and the learning curve inexperience in the introduction period of laparoscopic surgery for rectal cancer may affect the higher AL rate. Our study showed that the odds of AL were statistically lower for laparoscopic surgery than for open surgery. Basically, there are inevitably significant differences in patient selection, surgical procedure, and complication risk between the patient group who underwent laparoscopic surgery and the group who underwent open surgery. A decision on whether to select laparoscopic or open surgery necessarily involves various preoperative evaluations and surgeon’s concerns about anastomotic leakage risks. In addition, the selection of a surgical method can play a role in preventing the occurrence of anastomotic leakage, but it was very difficult to verify this point in a retrospective study using claim data. Another important confounding factor may be a conversion from laparoscopic to open surgery, which could be an indicator of difficulties encountered during surgery. During such conversions anastomosis may be performed manually, which may help to prevent AL. Unfortunately, conversion surgeries could not be ascertained using claims data, so our findings with respect to surgical approach and incidence of AL should be interpreted with caution. The influence of preoperative chemoradiotherapy (preop-CRT), especially for rectal cancer, on the incidence of AL is also up for discussion. Jang et al. reported that preop-CRT did not have any incremental impact on the occurrence of AL, as the incidence of AL in patients with and without preop-CRT was 7.5% and 8.1%, respectively, after propensity matching (*p* = 0.781). Hu et al. reported that neoadjuvant therapy did not statistically increase the incidence of AL (OR 1.16, 95% CI 0.99–1.36; *p* = 0.07, random effects model) in a meta-analysis. However, other reports revealed that preop-CRT was a significant independent factor for AL. Our large cohort study showed preoperative receipt of radiation therapy to have no statistical association with the odds of AL. Further studies are required to elucidate the true impact of preop-CRT with AL. With respect to the influence of age on the occurrence of AL, Zaimi et al. reported that increased age every 5 years was protective for AL after colorectal resections in the multivariable analysis (OR 0.965, 95% CI 0.941–0.985, *p*\<0.001) of 45,488 Dutch colorectal cancer patients. Similarly, other population-based studies revealed that old age was associated with a lower risk of AL. In contrast, several studies did not support this or demonstrated that old age was a risk factor for a higher AL rate. In this study, we found that age was associated with a lower incidence of AL. Surgical intervention performed by surgeons with caution for elderly patients or healthy survivor effect, indicating that unhealthy patients die before reaching older age might be suggested as possible reasons for less AL in elderly patients. Our study is one of the largest studies that demonstrated the protective effect of age on AL in Asian groups. Although our study did not focus on the impact of age on AL, this observation might be helpful in planning adequate surgical procedures for elderly patients. A previous study found short-term outcomes to be dependent on hospital caseload and a degree of specialization. Zheng et al. demonstrated this hospital-center effect with respect to hospital LOS and in-hospital mortality for patients with stage I-III colon cancer who were treated with laparoscopic colectomies. Among the 120,245 patients who were enrolled in our study, only 17,325 (14.41%) underwent surgeries in hospitals with a small number of beds (less than 500 beds). This phenomenon might be attributed to the concentration of hospitals selected by patients in Korea, as more patients from rural areas tend to travel to large metropolitan hospitals for surgery. We did not detect differences in the odds of AL as a function of the hospital size (over 1000 beds, 500–1000 beds, and less than 500 beds). Therefore, surgical processes and/or population risks may not vary across hospitals within these size categories. Curative resection for CRC is regarded as a complex procedure which requires highly trained and experienced surgeons. It was not possible to conduct the analysis of this factor, since the source data lack information on hospital and surgeon caseload. Hospital LOS was longer among AL patients than those without AL in our study, though the gap between the two groups was relatively small (mean 16.78 days in the AL group vs. 14.22 days in the no AL group, *p*\<0.001). In contrast, Lee et al. observed a much larger incremental difference in the USA (12 days in the AL group vs. 5 days in the no AL group, *p*\<0.0001). In Korea, many patients may not seek early discharge, given that hospital room charges are reimbursed by the national insurance system. The impact of AL on hospital LOS may therefore depend on historical practices and incentives within local healthcare systems. Despite modest differences in LOS, the total costs for the index hospitalization were significantly higher in patients with AL, so were the total costs for readmissions; these observations corroborate findings from a previous study. One of the critical limitations of this study is that the presence of AL was not identified using individual patients’ medical records. The characteristics of claim data archived by the Korean Health Insurance Review and Assessment Service (HIRA) are not suitable to find a definitive case of AL. There is a precedent, however, of using surrogate indicators of AL from similarly structured data sources. Ashraf et al. reported the economic impact of AL using 23,388 patients registered at the Hospital Episode Statistics (HES) dataset from various National Health Service (NHS) hospitals in England. In that study, researchers defined the AL event as “relaparotomy within 28 days of surgery” because no valid diagnostic code for AL was available. In a recent study using a large claims dataset in the United States, Lee et al. inferred the occurrence of AL from the presence of an abscess, septicemia, peritonitis, or infection among 239,350 patients undergoing colorectal surgery. As with these prior studies, we selected possible clinical patterns representing ALs according an “operational definition”. Nonetheless, we could not make definitive conclusions about the presence of AL, leading us to define additional interventional cases (AIC) in order to overcome the shortcomings. Repeated analyses using the definition of AIC confirmed that the relevant indicators and economic aspects were quite similar. However, as a prior study noted, not all of the relaparotomy within 28 days was done due to AL only; in the case of our study, surgical relaparotomy and radiologic intervention were not always performed on account of AL only. Another major limitation of this study relates to a lack of cancer staging data. Surgical extent, combined resection, and receipt of chemotherapy are largely dependent on cancer stage. Thus, a stage-specific comparison would be required to eliminate this confounding factor. Since claims data lack staging information, further research with a data source inclusive of surgical morbidity, leakage status, and cancer stage is required. In conclusion, this research demonstrated AL to be associated with a significantly increased economic burden in the nationwide dataset. Despite limitations inherent to the reliance on surrogate endpoints to find AL cases, this work underscores an opportunity to reduce the economic burden associated with treatments required for AL. # Supporting information 10.1371/journal.pone.0267950.r001 Decision Letter 0 Meyer Alberto Academic Editor 2022 Alberto Meyer 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 Nov 2021 PONE-D-21-34976Risk factors and economic burden of postoperative anastomotic leakage related events in patients who underwent surgeries for colorectal cancerPLOS ONE Dear Dr. Lee, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. As you can see, the article needs improvement, especially, in the light of methodology of how the findings compare colon to rectal cancer and surgery with neoadjuvant treatment. Please submit your revised manuscript by 01/02/2022. 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, Alberto Meyer, MD, 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\. Thank you for stating the following financial disclosure: (This work was supported by Johnson & Johnson Medical Korea(<https://www.jnjmedicaldevices.com/ko-KR>). HK and HJP, employees of the sponsor, participated in study design and preparation of the manuscript. BL who was paid for the statistical analysis from the sponsor.). 3\. We note that one or more of the authors is affiliated with the funding organization, indicating the funder may have had some role in the design, data collection, analysis or preparation of your manuscript for publication; in other words, the funder played an indirect role through the participation of the co- authors. If the funding organization did not play a role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript and only provided financial support in the form of authors' salaries and/or research materials, please do the following: A. Review your statements relating to the author contributions, and ensure you have specifically and accurately indicated the role(s) that these authors had in your study. These amendments should be made in the online form. B. Confirm in your cover letter that you agree with the following statement, and we will change the online submission form on your behalf: “The funder provided support in the form of salaries for authors \[insert relevant initials\], but did not have any additional role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript. The specific roles of these authors are articulated in the ‘author contributions’ section. C. Please note that in order to use the direct billing option the corresponding author must be affiliated with the chosen institute. Please either amend your manuscript to change the affiliation or corresponding author, or email us at <plosone@plos.org> with a request to remove this option. 4.Thank you for stating the following in the Competing Interests section:  (I have read the journal's policy and the authors of this manuscript have the following competing interests: HK and HJP are employees of Johnson & Johnson Medical Korea. BL is an employee of RexSoft, which was paid for the statistical analysis from Johnson & Johnson Medical Korea.).  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>).  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. 5\. In your Data Availability statement, you have not specified where the minimal data set underlying the results described in your manuscript can be found. PLOS defines a study's minimal data set as the underlying data used to reach the conclusions drawn in the manuscript and any additional data required to replicate the reported study findings in their entirety. All PLOS journals require that the minimal data set be made fully available. For more information about our data policy, please see <http://journals.plos.org/plosone/s/data- availability>. ""Upon re-submitting your revised manuscript, please upload your study’s minimal underlying data set as either Supporting Information files or to a stable, public repository and include the relevant URLs, DOIs, or accession numbers within your revised cover letter. For a list of acceptable repositories, please see <http://journals.plos.org/plosone/s/data-availability#loc-recommended- repositories>. Any potentially identifying patient information must be fully anonymized. Important: If there are ethical or legal restrictions to sharing your data publicly, please explain these restrictions in detail. Please see our guidelines for more information on what we consider unacceptable restrictions to publicly sharing data: <http://journals.plos.org/plosone/s/data-availability#loc- unacceptable-data-access-restrictions>. Note that it is not acceptable for the authors to be the sole named individuals responsible for ensuring data access. We will update your Data Availability statement to reflect the information you provide in your cover letter. 6\. 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> Additional Editor Comments: Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. As you can see, the article needs improvement, especially, in the light of methodology of how the findings compare colon to rectal cancer and surgery with neoadjuvant treatment. We look forward to receiving your revised manuscript. King regards, Alberto Meyer 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\. 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 \*\*\*\*\*\*\*\*\*\* 2\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: I Don't Know Reviewer \#2: Yes Reviewer \#3: 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 Reviewer \#3: 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 Reviewer \#3: 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 English language needs to be polished for correcting some writing and alphabetical errors thorough text. It is suggested that statistical analyses are reviewed/re-checked by a statistician. The statistical values and numbers are suggested to be reviewed again to confirm by authors. Reviewer \#2: The authors retrospectively analyzed the risk factors and economic burden of anastomatic leakage (AL) after colorectal cancer. The study included large samples from Korean patient database, the analysis method was proper, and the manuscript was well organized. However, the limitations of the study were also obvious. There are some important issues to be further addressed before acceptance: 1\. The study include rectal cancer and sigmoid colon cancer, which were highly heterogeneous in surgery and the AL risk. Why did the author mixed such two kinds of disease together? 2\. In the method part, the author clarified that "all patients underwent surgery as the primary treatment", that means all the included patients did not receive neoadjuvant chemoradiotherapy. So the conclusion of this study is not representative to all the CRC patients who got surgery, but only to those who got surgery first (without neoadjuvant chemoradiotherapy). 3\. The AL has classification criteria of AL (grade), which was helpful to evaluate the severity of AL, and also helpful to explain the cost and LOS data, but the authors did not show the grade data. 4\. Due to the big difference in therapy, the result has its own bias and limitation. For example, the open surgery and laparoscopy surgery has huge diffence in patient selection, surgical procedure and complication risk; so the authors need fully discuss the limitation of the results. Reviewer \#3: Authors conducted a retrospective, nationwide research about the clinical and economic burden caused by anastomotic leakage (AL) in Korea. Of 156,545 patients undergoing anterior resection (AR), low anterior resection (LAR), or ultra-low anterior resection (uLAR) for colorectal cancer (CRC) between January 1, 2007 and January 31, 2020 were included. Among 120,245 patients who met the eligibility criteria, 7,194 (5.98%) patients had AL within 30 days after surgery. Male gender, comorbidities, protective ostomy, and multiple linear stapler use were associated with a higher odds of AL. Older age, rectosigmoid junction cancer, AR, LAR, and laparoscopic approach were associated with reduced odds of AL. Patients with AL incurred higher costs for index hospitalization compared to those without AL (8,991 vs. 7,153 USD; p \<0.0001). Patients with AL also required longer LOS (16.78 vs. 14.22 days; p \<0.0001) and readmissions (20.83 vs. 13.93 days; p \<0.0001). In summary, they concluded that patients requiring resection for CRC, the occurrence of AL was associated with significantly increased costs and LOS. Preventing AL could not only provide for superior clinical outcomes, but also reduce the economic burden for patients and payers. The results seems interesting and appealing; however, there are a lot of criticisms and have several issues that the authors need to address before the manuscript is suitable for publication. Major Compulsory Revisions: 1\. Clinical outcomes paragraph. The following variables were identified as indicators of surgical complications within 30 days after index surgery: AL, infection, blood transfusion, urinary tract injury, ileus, pneumonia, pulmonary embolism, acute renal failure. Blood transfusion is defined as surgical complications? The transfusion units should be considered. Acute myocardial infarction, wound infection and stroke should be also included as surgical complications. 2\. Since there is no specific diagnosis code for AL, presence of the following procedures was required when AL occurrence during in-hospital care comprised operational AL definitions: (1) Imaging study including computed tomography scans (2) Administration of antibacterial drugs (more than 7 consecutive days after the surgery), the above two procedures were hard to be defined as AL. In addition, the nonsynchronous creation of colostomy after AR, LAR, and laparoscopic approach should be considered as AL. 3\. In Table 2: Demographics and Perioperative Clinical characteristics. How did authors could identify some variables were surgical complications or risk factors of AL? For example, ischemic heart disease, ischemic stroke, etc. 4\. Table 3: Risk factors of anastomotic leak from logistic regression. Age, (years) was an independent variable by multivariate analysis but not by univariate analysis? The subgroup analysis of CCI Score should be mentioned here, and no multivariate analysis for CCI score? Protective ostomy is often performed in ultra-low anterior resection (uLAR), of which might be considered as a compounding factor but not as a risk factor. Furthermore, robotic-assisted surgery vs laparoscopy vs open surgery is suggested to be analyzed if this procedure is related to AL? 5\. Table 4. Healthcare costs by procedures & approaches. Only open vs laparoscopic approaches? How about in the comparison with robotic-assisted surgery? The relevant information regarding robotic-assisted surgery is important in recent years. 6\. Table 5. Economic outcomes by the presence of anastomotic leakage. Mean LOS and Median LOS for index hospitalization, (duration) was 14.22 ± 8.36 and 12 days, respectively. In fact, it was relatively longer compared to Western countries and even longer than some Asian countries. 7\. If authors could use Health Insurance Review and Assessment Service (HIRA) archives claim data to analyze the difference of overall survival between AL vs non-AL patients? 8\. In Abstract section: Male gender, comorbidities, protective ostomy, and multiple linear stapler use were associated with a higher odds of AL. The above statement should be amended according to complete results in Table 3. Minor Essential Revisions: 1\. Please correct the typos and grammatical error by English-editing with the certificate enclosed. 2\. Abbreviations in the Tables must be shown their corresponding full name in the footnotes. \*\*\*\*\*\*\*\*\*\* 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.0267950.r002 Author response to Decision Letter 0 31 Mar 2022 Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. As you can see, the article needs improvement, especially, in the light of methodology of how the findings compare colon to rectal cancer and surgery with neoadjuvant treatment. Please submit your revised manuscript by 01/02/2022. 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>. Journal Requirements: When submitting your revision, we need you to address these additional requirements. 1.Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at <https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main _body.pdf> and <https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_titl e_authors_affiliations.pdf> 2\. Thank you for stating the following financial disclosure: (This work was supported by Johnson & Johnson Medical Korea(<https://www.jnjmedicaldevices.com/ko-KR>). HK and HJP, employees of the sponsor, participated in study design and preparation of the manuscript. BL who was paid for the statistical analysis from the sponsor.). 3\. We note that one or more of the authors is affiliated with the funding organization, indicating the funder may have had some role in the design, data collection, analysis or preparation of your manuscript for publication; in other words, the funder played an indirect role through the participation of the co- authors. If the funding organization did not play a role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript and only provided financial support in the form of authors' salaries and/or research materials, please do the following: A. Review your statements relating to the author contributions, and ensure you have specifically and accurately indicated the role(s) that these authors had in your study. These amendments should be made in the online form. Response: Thank you for your comments. We reviewed the author contribution and modified it. This statement was applied in the revised manuscript and online form. B. Confirm in your cover letter that you agree with the following statement, and we will change the online submission form on your behalf: “The funder provided support in the form of salaries for authors \[insert relevant initials\], but did not have any additional role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript. The specific roles of these authors are articulated in the ‘author contributions’ section. Response: Thank you for your comments. We agree with your suggestion to revise the statement because the funders did not have any additional role in the study and manuscript. C. Please note that in order to use the direct billing option the corresponding author must be affiliated with the chosen institute. Please either amend your manuscript to change the affiliation or corresponding author, or email us at <plosone@plos.org> with a request to remove this option. 4.Thank you for stating the following in the Competing Interests section: (I have read the journal's policy and the authors of this manuscript have the following competing interests: HK and HJP are employees of Johnson & Johnson Medical Korea. BL is an employee of RexSoft, which was paid for the statistical analysis from Johnson & Johnson Medical Korea.). 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>). 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. Response: We used the claim data generated in the Health Insurance Review and Assessment Service (HIRA). HIRA has strict regulations to provide the data for research. We described the legal restrictions to sharing the minimal data set in the cover letter. 5\. In your Data Availability statement, you have not specified where the minimal data set underlying the results described in your manuscript can be found. PLOS defines a study's minimal data set as the underlying data used to reach the conclusions drawn in the manuscript and any additional data required to replicate the reported study findings in their entirety. All PLOS journals require that the minimal data set be made fully available. For more information about our data policy, please see <http://journals.plos.org/plosone/s/data- availability>. ""Upon re-submitting your revised manuscript, please upload your study’s minimal underlying data set as either Supporting Information files or to a stable, public repository and include the relevant URLs, DOIs, or accession numbers within your revised cover letter. For a list of acceptable repositories, please see <http://journals.plos.org/plosone/s/data-availability#loc-recommended- repositories>. Any potentially identifying patient information must be fully anonymized. Important: If there are ethical or legal restrictions to sharing your data publicly, please explain these restrictions in detail. Please see our guidelines for more information on what we consider unacceptable restrictions to publicly sharing data: <http://journals.plos.org/plosone/s/data-availability#loc- unacceptable-data-access-restrictions>. Note that it is not acceptable for the authors to be the sole named individuals responsible for ensuring data access. We will update your Data Availability statement to reflect the information you provide in your cover letter. Response: We used the claim data generated in the Health Insurance Review and Assessment Service (HIRA). HIRA has strict regulations to provide the data for research. We described the legal restrictions to sharing the minimal data set in the cover letter. \<Legal restrictions to sharing the minimal data set\> Under the public, single-payer system, the claim data were generated in the Health Insurance Review and Assessment Service (HIRA) archives in the process of reimbursing healthcare providers, covering all Korean citizens. In order to use the big data of HIRA, a researcher should submit the application with a study proposal. HIRA review the application for approval of the access to data. Once the approval is given, HIRA retrieves the data from the data warehouse system, which is then uploaded to the system of HIRA. The system is accessible only to the researcher for the study through the designated computer of the HIRA datacenter or authorized remote access only for a limited period. As taking the raw data out is forbidden, the researcher can take out only the analysis results. Therefore, we cannot share the minimal data set of this study. The application process and documents are specified in the Healthcare Bigdata Hub homepage of HIRA. HIRA homepage address: <https://opendata.hira.or.kr/or/orb/useGdInfo.do> 6\. 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> Response: Thank you for your detailed explanation. 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 Reviewer \#3: Yes \_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_ 2\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: I Don't Know Reviewer \#2: Yes Reviewer \#3: Yes \_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_ 3\. Have the authors made all data underlying the findings in their manuscript fully available? The PLOS Data policy 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 \_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_ 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: 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 English language needs to be polished for correcting some writing and alphabetical errors thorough text. Response: Thank you for your comments. We made corrections to the manuscript accordingly. It is suggested that statistical analyses are reviewed/re-checked by a statistician. Response: As you commented, the statistician reviewed/re-checked the overall statistical analysis parts. The statistical values and numbers are suggested to be reviewed again to confirm by authors. Response: We reviewed all the statistical values and numbers again. Reviewer \#2: The authors retrospectively analyzed the risk factors and economic burden of anastomatic leakage (AL) after colorectal cancer. The study included large samples from Korean patient database, the analysis method was proper, and the manuscript was well organized. However, the limitations of the study were also obvious. There are some important issues to be further addressed before acceptance: 1\. The study include rectal cancer and sigmoid colon cancer, which were highly heterogeneous in surgery and the AL risk. Why did the author mixed such two kinds of disease together? Response: Basically, we agree that colon and rectal cancer have different anastomotic leakage rates. Therefore, there have been many previous studies where the analyses were performed separately. Our study was conducted with the intention of identifying the trend of anastomotic leakage for all colorectal cancer patients nationwide. As you pointed out, we analyzed the differences in anastomotic leakage rates by location such as the colon and the rectum as shown in Table 2. We also presented the healthcare costs by tumor location in Table 4. Thank you for your comments. 2\. In the method part, the author clarified that "all patients underwent surgery as the primary treatment", that means all the included patients did not receive neoadjuvant chemoradiotherapy. So the conclusion of this study is not representative to all the CRC patients who got surgery, but only to those who got surgery first (without neoadjuvant chemoradiotherapy). Response: Thank you for your comments. The sentence "all patients underwent surgery as the primary treatment” was meant to include all the patients who underwent surgeries. As you pointed out, however, this may cause misunderstandings about patients who received neoadjuvant treatment. Therefore, it would be better to modify the above expression to "all patients underwent surgery as the principal treatment" as follows. Thank you for your comments. 3\. The AL has classification criteria of AL (grade), which was helpful to evaluate the severity of AL, and also helpful to explain the cost and LOS data, but the authors did not show the grade data. Response: As you commented, hospital costs and length of stay vary depending on the grade of AL, so if we can know this accurately, we can perform more detailed analysis. However, it was difficult to obtain data on the grades of AL due to the nature of the claim data, so we could not include it in our analysis. Thank you for your comments. 4\. Due to the big difference in therapy, the result has its own bias and limitation. For example, the open surgery and laparoscopy surgery has huge diffence in patient selection, surgical procedure and complication risk; so the authors need fully discuss the limitation of the results. Response: Thank you for your comments. We agree with your comments. We added the paragraph below in the Discussion section Included in the Discussion section Basically, there are inevitably significant differences in patient selection, surgical procedure, and complication risk between the patient group who underwent laparoscopic surgery and the group who underwent open surgery. A decision on whether to select laparoscopic or open surgery necessarily involves various preoperative evaluations and surgeon's concerns about anastomotic leakage risks. In addition, the selection of a surgical method can play a role in preventing the occurrence of anastomotic leakage, but it was very difficult to verify this point in a retrospective study using claim data. Reviewer \#3: Authors conducted a retrospective, nationwide research about the clinical and economic burden caused by anastomotic leakage (AL) in Korea. Of 156,545 patients undergoing anterior resection (AR), low anterior resection (LAR), or ultra-low anterior resection (uLAR) for colorectal cancer (CRC) between January 1, 2007 and January 31, 2020 were included. Among 120,245 patients who met the eligibility criteria, 7,194 (5.98%) patients had AL within 30 days after surgery. Male gender, comorbidities, protective ostomy, and multiple linear stapler use were associated with a higher odds of AL. Older age, rectosigmoid junction cancer, AR, LAR, and laparoscopic approach were associated with reduced odds of AL. Patients with AL incurred higher costs for index hospitalization compared to those without AL (8,991 vs. 7,153 USD; p \<0.0001). Patients with AL also required longer LOS (16.78 vs. 14.22 days; p \<0.0001) and readmissions (20.83 vs. 13.93 days; p \<0.0001). In summary, they concluded that patients requiring resection for CRC, the occurrence of AL was associated with significantly increased costs and LOS. Preventing AL could not only provide for superior clinical outcomes, but also reduce the economic burden for patients and payers. The results seems interesting and appealing; however, there are a lot of criticisms and have several issues that the authors need to address before the manuscript is suitable for publication. Major Compulsory Revisions: 1\. Clinical outcomes paragraph. The following variables were identified as indicators of surgical complications within 30 days after index surgery: AL, infection, blood transfusion, urinary tract injury, ileus, pneumonia, pulmonary embolism, acute renal failure. Blood transfusion is defined as surgical complications? The transfusion units should be considered. Acute myocardial infarction, wound infection and stroke should be also included as surgical complications. Response: Thank you for your comments. Blood transfusion was included as one of the variables in Table 3. Since the situation requiring blood transfusion may reflect the difficulty of surgery, we checked the relationship between blood transfusion and anastomotic leakage. However, like your opinion, there are parts that it is not easy to refer to as risk factors directly, so we have changed the title of Table 3 into "Risk factors and clinical parameters associated with the anastomotic leak from logistic regression." However, it was very difficult to include the transfusion volume, acute myocardial infarction, wound infection and stroke as covariates due to the characteristics of claim data. This would be one of the main limitations of our study. 2\. Since there is no specific diagnosis code for AL, presence of the following procedures was required when AL occurrence during in-hospital care comprised operational AL definitions: (1) Imaging study including computed tomography scans (2) Administration of antibacterial drugs (more than 7 consecutive days after the surgery), the above two procedures were hard to be defined as AL. In addition, the nonsynchronous creation of colostomy after AR, LAR, and laparoscopic approach should be considered as AL. Response: We agree with your comments. We believe that abdominal CT is performed again during the period of discharge after surgery when it is really necessary to identify a certain serious situation in the abdomen. Therefore, we determined that such examination after colorectal cancer surgery would be required mostly when complications such as leakage are suspected. In addition, if antibiotics are used for more than a week, in Korea, the use of antibiotics is usually limited to 1-2 days after elective colorectal cancer surgery, according to the evaluation criteria of the National Review and Assessment Service. This standard was published as an index that most hospitals satisfy. Therefore, it can be judged that the clinical situation in which antibiotics are used continuously for one week or more after surgery is an emergency in which the use of antibiotics is essential in light of these indicators. For this reason, two items were included in the operational definition. Nevertheless, as we described the limitations in defining the accurate anastomotic leakage using nationwide claim data, it was impossible to collect the correct cases only. Because nonsynchronous creation of colostomy could not be identified in the claim database, it was impossible to include it as an independent variable. However, we believe that our operational definitions could cover most of the colostomy or ileostomy formation cases because those patients usually underwent APCT again or antibiotics were prescribed longer than usual. Thank you for your comments. 3\. In Table 2: Demographics and Perioperative Clinical characteristics. How did authors could identify some variables were surgical complications or risk factors of AL? For example, ischemic heart disease, ischemic stroke, etc. Response: We agree with your opinion. In Table 3, ischemic heart disease and ischemic stroke were not surgical complications. These variables were identified in each patient within 1 year before surgery. Thank you for your comments. 4\. Table 3: Risk factors of anastomotic leak from logistic regression. Age, (years) was an independent variable by multivariate analysis but not by univariate analysis? The subgroup analysis of CCI Score should be mentioned here, and no multivariate analysis for CCI score? Protective ostomy is often performed in ultra-low anterior resection (uLAR), of which might be considered as a compounding factor but not as a risk factor. Response: Thank you for your comments. We constructed an initial multiple model using variables that were significant based on a significance level of 0.15 in the univariable model. Because of this, in multiple models, the number of beds was excluded. At this time, CCI and comorbidities (Diabetes, COPD, Metastatic disease, Ischemic heart disease, Ischemic stroke) were variables with strong multicollinearity, and only one of them had to be included in the model. Therefore, the model with higher explanatory power was determined to be the final model by comparing multiple models with comorbidities excluded and multiple models with CCI excluded and comorbidities included. We described this procedure in the statistical analysis part in the manuscript as follows: The multivariable model included factors that showed the statistical significance based on univariable analysis at the significance level of 0.15, and it was evaluated by generalized variance inflation factor(VIF) for multicollinearity and the final model was chosen based on Akaike Information Criteria(AIC). Furthermore, robotic-assisted surgery vs laparoscopy vs open surgery is suggested to be analyzed if this procedure is related to AL? Response: Thank you for your comments. Robotic-assisted surgery for colorectal resection is not covered by insurance in Korea. Since we used the nationwide claim data and robot surgery is not covered by insurance, it was not included in this analysis by default. 5\. Table 4. Healthcare costs by procedures & approaches. Only open vs laparoscopic approaches? How about in the comparison with robotic-assisted surgery? The relevant information regarding robotic-assisted surgery is important in recent years. Response: Thank you for your comments. Robotic-assisted surgery for colorectal resection is not covered by insurance in Korea. Since we used the nationwide claim data and robot surgery is not covered by insurance, it was not included in this analysis by default. 6\. Table 5. Economic outcomes by the presence of anastomotic leakage. Mean LOS and Median LOS for index hospitalization, (duration) was 14.22 ± 8.36 and 12 days, respectively. In fact, it was relatively longer compared to Western countries and even longer than some Asian countries. Response: Thank you for your comments. We agree with your opinion. It is not possible to accurately analyze the cause of hospital stay being considerably longer than in other cases. Although there is a difference in the room rate in Korea, when it is covered by insurance, the cheapest room rate is about 10 dollars a day. Therefore, patients want to stay in the hospital as long as possible, rather than being discharged early. This is thought to be one of the main causes. But, over the last 5-6 years, each hospital has been applying ERAS, etc. to discharge patients as early as possible to improve the management of the hospital, so the length of stay has been shortened. Therefore, it seems that the long-term hospitalization was probably reflected in the past, and a relatively large number of patients who underwent open surgery are also considered one of the reasons. 7\. If authors could use Health Insurance Review and Assessment Service (HIRA) archives claim data to analyze the difference of overall survival between AL vs non-AL patients? Response: As the HIRA database includes only the information on reimbursement, there is no information on death, so we could not identify the mortality. Thank you for your comments. 8\. In Abstract section: Male gender, comorbidities, protective ostomy, and multiple linear stapler use were associated with a higher odds of AL. The above statement should be amended according to complete results in Table 3. Response: Thank you for your comments. As mentioned in the previous section, CCI and the rest of the comorbidities had high multicollinearity, and the multiple model with comorbidities showed better explanatory power than that with CCI. Thus, we just showed the effect size of CCI in the univariable model and excluded that in the final multivariable model. Included in the revised manuscript Male gender, comorbidities (diabetes, metastatic disease, ischemic heart disease, ischemic stroke), protective ostomy, and multiple linear stapler use, blood transfusion, and urinary tract injury were associated with the higher odds of AL. Older age, rectosigmoid junction cancer, AR, LAR, and laparoscopic approach were related with the reduced odds of AL. Minor Essential Revisions: 1\. Please correct the typos and grammatical error by English-editing with the certificate enclosed. Response: Thank you for your comments. We made corrections to the manuscript accordingly. 2\. Abbreviations in the Tables must be shown their corresponding full name in the footnotes. Response: Thank you for your comments. We made corrections to the manuscript accordingly. \_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_ 6\. PLOS authors have the option to publish the peer review history of their article (what does this mean?). 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. 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.0267950.r003 Decision Letter 1 Meyer Alberto Academic Editor 2022 Alberto Meyer This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 20 Apr 2022 Risk factors and economic burden of postoperative anastomotic leakage related events in patients who underwent surgeries for colorectal cancer PONE-D-21-34976R1 Dear Dr. LEE, 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, Alberto Meyer, MD, PhD Academic Editor PLOS ONE Additional Editor Comments (optional): Dear Authors, Thank you for accepting our recommendations for revision and incorporating the relevant changes. I am satisfied with your response and thus happy to recommend in favour of publication of your study. Kind regards Reviewers' comments: 10.1371/journal.pone.0267950.r004 Acceptance letter Meyer Alberto Academic Editor 2022 Alberto Meyer 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 May 2022 PONE-D-21-34976R1 Risk factors and economic burden of postoperative anastomotic leakage related events in patients who underwent surgeries for colorectal cancer Dear Dr. Lee: I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department. If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact <onepress@plos.org>. If we can help with anything else, please email us at <plosone@plos.org>. Thank you for submitting your work to PLOS ONE and supporting open access. Kind regards, PLOS ONE Editorial Office Staff on behalf of Professor Alberto Meyer Academic Editor PLOS ONE [^1]: The authors have read the journal’s policy and have the following competing interests to declare. HK and HJP are employees of Johnson & Johnson Medical Korea. BL is an employee of RexSoft, which was paid for the statistical analysis from Johnson & Johnson Medical Korea. There are no patents, products in development or marketed products associated with this research to declare. This does not alter our adherence to PLOS ONE policies on sharing data and materials.

# Introduction In striated muscle, electrical excitation activates ryanodine receptors (RyR) located in the sarcoplasmic reticulum (SR) membrane which in turn mediate the massive release of intracellular Ca²⁺ required for activating the contractile system. Electron microscopy studies indicate that cardiac (RyR2) channels could interact among themselves as they are physically connected in organized arrays at the terminal cisternae of SR. Indeed, it has been shown that multiple RyRs synchronously activate and deactivate during excitation- contraction (EC)-coupling. Moreover, under resting conditions, brief elementary events of Ca²⁺ release (“Ca²⁺ sparks”) arise as a result of the concerted activation and deactivation of six to twenty RyR2 in brief bursts lasting ∼5–20 ms. These functional channel-channel interactions seem to survive isolation and reconstitution in bilayers, where multiple RyRs often display synchronicity named “coupled gating”. It is also apparent that, in the cytosolic environment, RyR1 and RyR2 may be modulated via physical interactions with other associated proteins, such as the L-type Ca²⁺-channels. The nature of the interactions between neighboring RyRs and/or with associated proteins has not been fully defined, but it is likely that electrostatic interactions may play a role as the vestibule of RyRs contain negatively charged regions that could be a target for cationic ligands. Indeed, it is well known that RyR channel function can be modulated by positively charged moieties, including polycationic peptides such as protamine, histone and polylysine, which seem to display a variety of actions including activation and block of RyR- mediated Ca²⁺ release. Furthermore, in failing heart as well as in skeletal muscle pathologies, RyR-mediated SR Ca²⁺ release was found to have increased sensitivity to activation by polylysine. Protamine is a mix of highly cationic (arginine rich) peptides with molecular weight of ∼5.1 kDa (major component) which has been previously used as a tool to study how RyRs are modulated through electrostatic interactions. In this early study, large doses of protamine (\>20 µg/ml) were found to fully inhibit skeletal (RyR1) channels regardless of the cytosolic Ca²⁺ levels. We extended these studies to cardiac RyR2 reconstituted into planar lipid bilayers and tested a wider range of protamine levels (0.02 to 20 µg/ml). Our results indicate that the action of protamine added to the cytosolic surface of RyR2 is complex. It includes voltage-dependent activation and block as well as transitions to subconductance states (substates). Some of the results have been presented in preliminary form. # Methods ## Drugs and Chemicals CaCl₂ standard for calibration was from Word Precision Instruments Inc. (Sarasota, FL). Phospholipids were obtained from Avanti (Alabaster, AL). Ryanodine was from Calbiochem (San Diego, CA). Imperatoxin A (IpTx_A) was from Alomone Labs (Jerusalem, Israel). Ryanodol was obtained from hydrolyzed ryanodine as previously described. Protamine and all other drugs and chemicals were either from Sigma-Aldrich or were reagent grade. ## Sarcoplasmic Reticulum Microsomes All procedures with animals were designed to minimize pain and suffering and conformed to the guidelines of the National Institutes of Health. The committee on the Use and Care of Laboratory Animals of Southern Illinois University School of Medicine reviewed and approved the protocols for animal use. Sarcoplasmic reticulum (SR) microsomes were obtained from pig heart ventricle using heart homogenization and ultracentrifugation steps that follow the procedures published by Chamberlain *et al.*. SR pellets obtained after high speed centrifugation were resuspended in 290 mM sucrose - 5 mM Imidazole buffer (pH = 7) and were aliquoted in cryovials (300 µl each) and kept in liquid nitrogen (better and safer long-term storage). Every month, a few cryovials are used to generate smaller aliquots of membranes (15 µl each) which were stored at –80°C for easy access. For experiments, aliquots were quickly thawed in water, kept on ice and used within 3–5 hours. ## Bilayer Technique Reconstitution of RyR2 in planar lipid bilayers, was performed as previously described. Briefly, planar lipid bilayers were formed on 80 to 100 µm-diameter circular holes in teflon septa, separating two 1.3 ml compartments. The *trans* compartment was filled with HEPES-Ca²⁺ solution containing HEPES 250 mM and Ca(OH)₂ 53 mM, pH 7.4. The *trans* compartment was clamped at 0 mV using an Axopatch 200B patch-clamp amplifier (Axon Instruments, Foster City, CA). The *cis* compartment (ground) was filled with HEPES-Tris solution containing HEPES 250 mM and TrisOH 118 mM, pH 7.4. Bilayers of a 5∶4∶1 mixture of bovine brain phosphatidylethanolamine, phosphatidylserine and phosphatidylcholine (45–50 mg/ml in decane) were painted onto the holes of teflon septa from the *cis* side. Sarcoplasmic reticulum microsomes (5–15 µg) were then added to the *cis* solution followed by 500–1000 mM CsCl and 1 mM CaCl₂ to promote vesicle fusion. After RyR currents (or Cl⁻ currents \>100 pA at 0 mV) were observed, the *cis* chamber was perfused with HEPES-TRIS solution for 5 min at 4 ml/min. A mixture of BAPTA and dibromo-BAPTA was used to buffer free \[Ca²⁺\] on the cytosolic surface of the channel (\[Ca²⁺\]_cyt). As previously done, RyR channels were identified by current amplitudes (∼3.5 pA at 0 mV), slope conductance (∼100 pS), reversal potential (∼−45 mV, *trans* - *cis*) and response to diagnostic ligands (e.g., ryanodine, Ca²⁺, ATP, caffeine and Ruthenium Red). RyR channel currents are depicted as positive (upward deflections of the current) in figures and reflect cation flux from the *trans* (lumenal) to the *cis* (cytosolic) compartment. Membrane voltages always represent the difference between *trans - cis* compartments (in mV). ## Single Channel Analysis Channel currents were first filtered through the Axopatch 200B low-pass Bessel filter at 2 kHz, digitized at 20 kHz with an analog to digital converter (Digidata 1320, Axon Instruments) and stored on DVD. Recordings were analyzed using pClamp9 software (Axon Instruments). Analysis with this program included open times, closed times and open probabilities (P_o), which were determined by half-amplitude threshold analysis of single-channel recordings as done before. Recordings were digitally filtered at 500 Hz in order to estimate the probabilities of substates using two different methods: ### 1) Manual analysis of the traces Frame by frame analysis (100 ms/frame) was performed using pClamp9 on each 4-min recording. In most of our recordings, we were able to distinguish different levels of current that included the baseline, full openings and substates induced by protamine, ryanodol or imperatoxin A. Events, at different levels of current lasting more than 3 ms, were manually selected. The parameters (mean amplitude and duration) of each event were collected and averaged. The probability of substate occurrence, P_substate, was estimated from the ratio: time spent in substate/total recording time. P_{full open} was estimated as the fraction of time spent in the maximal current level (full opening). ### 2) Current-amplitude distributions All-points current-amplitude histograms (band width = 0.01 pA) were obtained from each 4-min recording. Histograms were normalized so that total histogram area = 1. In most of our experiments, we were able to detect peaks (components) corresponding to the baseline, full opening and substate levels. Each component was fitted with a Gaussian function using the Levenberg-Marquardt method. When the fitting is good, the fitted area of each component can be used as an estimation of its probability. In some cases, the small signal-to-noise ratio precluded the unequivocal resolution of all individual components from the amplitude distribution histograms. In these cases, the P_substate were only estimated using frame by frame analysis. It is important to point out that in previous studies we have established that frame by frame analysis of higher signal-to-noise recordings yielded the same results as histograms (i.e. we can be confident when comparing P_o values calculated using different methods). ## Statistical Analysis Data are shown as means±S.E.M. of *n* measurements. Statistical comparisons between groups were performed with Student's t-test of paired differences. Differences were considered statistically significant at P\<0.05. # Results In this work, we studied the action of protamine added to the cytosolic surface of cardiac RyR2 reconstituted into planar lipid bilayers. High concentrations of protamine (\>20 µg/ml) have been tested on partially active skeletal RyR1 channels. Here, we extended these studies testing the effect of lower levels of protamine (0.02–20 µg/ml). As previous studies reported an inhibitory effect of protamine, we conducted most of our experiments using fully activated RyR2 (locked open by the combined effect of high cytosolic Ca²⁺ and caffeine) in order to minimize the interference of RyR2 intrinsic gating events (closures) with the expected effect of protamine (full or partial block). However, other studies have suggested that polycationic peptides (histone, polylysine) could induce SR Ca²⁺ release under cellular resting conditions. Consequently, we also tested the effect of protamine on RyR2 with low P_o at low (resting) cytosolic Ca²⁺ levels. ## Protamine Induces Block and Substates As shown in, at high levels of Ca²⁺ and caffeine, increasing concentrations of protamine induced sub-conductance states (substates) of several current levels (indicated by dashed lines). With 0.02 µg/ml protamine, infrequent transitions to a substate of ∼50% of the full open conductance were observed. With 0.2 µg/ml protamine, the probability of this substate dramatically increased. Addition of protamine to reach a concentration of 1 µg/ml resulted in the formation of an additional substate of ∼20% the full open conductance. Under these conditions, the probability of full openings decreased to virtually zero and the channel fluctuated between the two substates. With 2 µg/ml protamine, the substate with the smaller conductance became predominant. It is possible that even smaller substates arise at high concentrations of protamine, but they cannot be unequivocally resolved from the baseline current noise. As previously reported for RyR1, we observed full block of RyR2 in the presence of 20 µg/ml protamine. Addition of heparin (250 µg/ml), known to bind protamine with very high affinity, reversed the effect of protamine on RyR2. This reversibility suggests that the action of protamine was not related to the dissociation of any cofactor or modulatory subunit bound to the RyR2 channel. At 0 mV, showed that 1 µg/ml protamine induced transitions between two subconductance levels. show that at positive voltages (SR lumen - cytosol), 1 µg/ml protamine induced the formation of substates of only one current level. The probability of this substate decreased with increasing positive holding voltages. This resulted in an increase of the probability of the full open state at higher voltages, as the channels were fully activated by the combined action of high cytosolic Ca²⁺ and caffeine prior to protamine addition. As previously done, we analyzed the voltage dependence of protamine block using a modified version of the equation derived by Woodhull for block of a one-site, two-barrier channel model:where F, R, and T have their usual meanings, P_relative is the ratio of open probabilities (presence of protamine/absence of protamine) measured at each voltage and K_Pro is the protamine concentration at which block is half-maximal at 0 mV. The “equivalent valence” of the blocker (dz) is influenced by z = 22 (average valence of protamine molecules) and d (fraction of the membrane potential acting at the site). This term includes binding site location within the membrane field, Ca²⁺ flow – protamine interactions and the possibility of multi-sites. Fitting of Woodhull equation to our data render a K_Pro = 50±23 ng/ml and d = 0.10±0.03. According to the Woodhull model, a d ∼0.1 would indicate that protamine interacts with a RyR2 region that “weakly senses the field”. This resembles previously reported observations for Imperatoxin A (IpTx_A) and neomycin. Comparatively, peptide blockers and even ryanodol, a neutral ryanoid, have sharper voltage-dependence. As shown in and, the difference in current amplitude (full open – protamine substate) remains the same at all positive voltages. This is not the case for other known conductance-modifiers, where the substate amplitude changes proportionally with the amplitude of the full open channel. As an example, we show how a change in the holding voltage (from 0 to +40 mV) affects the amplitude of the substates induced by ryanodol and imperatoxin A (IpTx_A) (respectively). Notice that at 0 mV ryanodol and IpTx_A induced the formation of substates which current levels of ∼45% and ∼33% the current through the full open channel. The same proportions were observed when the holding voltage was +40 mV. These results are reflected in the I-V curves were the slopes for the full open state, the ryanodol-induced substate and the IpTx_A-induced substate are, respectively 129±1, 56±2 and 43±3 pS (open circles, triangles and squares). In contrast, the current levels of the protamine-induced substate at 0 mV and +40 mV were, respectively, ∼50% and ∼75% of the full open current. Since the drop in current induced by protamine is the same regardless of the holding voltage, the I-V curves for the open state and the protamine-induced substate display very similar slopes (129±1, 123±3 pS, respectively) (open circles vs. filled circles). To determine the reversal potential for the full open state and the protamine- induced substate, experiments were conducted in the presence of Cs⁺ and Ca²⁺ in the *cis* and *trans* chambers, respectively. shows that the I-V curves for the full open state and for the protamine-induced substate meet at the same reversal potential. This indicates that the selectivity Ca²⁺/Cs⁺ during the full openings is the same as during protamine-modified events. ## Protamine Affects the Probability of Imperatoxin A and Ryanodol Substates The peptide imperatoxin A (IpTx_A) is an agent that has similarities to the II-III loop of DHPR, and binds to a specific site at the open RyR to induce the formation of long-lasting, voltage dependent and reversible substates. Here, we tested possible interactions between the sites responsible for the formation of protamine- and IpTx_A-induced substates. We defined *“s”* as small substate that, in the absence of protamine, has ∼30% of the full opening current amplitude. In the presence of protamine, *s* is defined as the “sub-substate” and its amplitude is ∼30% of the protamine-induced highest conductance substate. As shown in, no transitions to any substate were observed under control conditions (absence of IpTx_A and/or protamine). shows that at a holding voltage of +20 mV, IpTx_A alone induced frequent long-lived substates. The effect of 1 µg/ml protamine alone is shown in the channel spent most of the time in the protamine-induced substate with infrequent transitions to short lived sub-substates. The combined action of protamine and IpTx_A was tested in two sets of experiments, switching the order in which these agents were applied. The result was always the same regardless of which agent was added in the first place: the channel spent most of the time in the protamine-induced substate with infrequent transitions to sub-substates. In this case the sub-substates were kinetically heterogeneous: there were short- lived events, similar to those observed in absence of IpTx_A and a few events of much longer duration that were only observed upon addition of IpTx_A. These events could represent either IpTx_A-induced substates in the protamine-modified channel or IpTx_A stabilization of the protamine-induced small-conductance substates. Still, the global probability of sub-substates (which includes both populations of events) was not significantly different in absence versus presence of IpTx_A (filled circles *versus* filled triangles). Thus, IpTx_A has little or no effect on RyR2 behavior when protamine is present. Ryanodol binds to open RyRs at a specific “ryanoid site” inducing the formation of reversible substates of ∼45% of the full open channel conductance. Previous studies indicated that the ryanoid site senses the electric field. Here, we tested the possibility that protamine induces substates by reversibly interacting with the “ryanoid site”. If this is the case, then we should not observe protamine-induced substates in ryanodol-modified channels (the “ryanoid site” would be unavailable to protamine when it is occupied by ryanodol). As shown in, in the presence of 10 µM ryanodol, RyR2 reversibly fluctuated between the full opening and the ryanodol-modified state. After addition of protamine the channel displayed virtually no full openings and it alternated between the protamine-induced substate and a long lasting (11±3 seconds at Vm = 20 mV) smaller substate which was never observed in the absence of ryanodol. Thus, protamine and ryanodol are not likely to act at the same binding site given that their effects do not exclude each other. However, in the presence of protamine, the ryanodol effects appear to be potentiated as events are longer and more frequent than those observed in the absence of protamine (legend). Therefore, addition of protamine increased the probability of ryanodol-induced substates (P_Ryanodol). As expected, subsequent addition of heparin reversed this effect, the channel turned back to fluctuate between the full open state and the ryanodol substate, and P_Ryanodol decreased to control values. ## Protamine Increases the Activity of RyR2 at Low Cytosolic Ca²⁺ We tested the effect of protamine on RyR2 with low P_o in the presence of 100 nM cytosolic Ca²⁺ as this condition better mimics the channel environment in a resting cell. In the control, at V_m = 20 mV, a few short and infrequent openings were observed and the open probability (P_o) was 0.042±0.029. Upon addition of 1 µg/ml protamine, most of the openings were to the protamine-induced substate reaching a P_o = 0.946±0.013. As expected, upon addition of heparin, the protamine-induced substate was no longer observed and channel activity returned to the levels observed under control conditions, with a P_o = 0.009±0.006. Protamine activating effect on RyR2 was voltage- dependent and P_o only reached 0.583±0.050 at V_m = 40 mV. The activating effect of protamine was not affected by further decreasing cytosolic Ca²⁺ levels to 25 nM (P_o = 0.950±0.012 and 0.901±0.079, before and after decreasing Ca²⁺ levels, respectively; V_m = 20 mV; n = 5). Addition of 5 mM cytosolic Mg²⁺ decreased the amplitude of the full open state as well as the substate by ∼15% at 20 mV, which is expected from the decrease in driving force for divalent flux. However, Mg²⁺ did not affect protamine induced activation (P_o = 0.961±0.006 and 0.951±0.018, before and after addition of Mg²⁺, respectively; V_m = 20 mV; n = 4). Notice that in the absence of protamine, at low cytosolic Ca²⁺ and high Mg²⁺, RyR2 channels had P_o∼0, even when exposed to 10 mM caffeine. The activating effect of protamine did not require lumen-to-cytosol Ca²⁺ flux as it was also observed after replacing lumenal Ca²⁺ with Ba²⁺ (not shown), a divalent cation that does not activate RyR2. shows a continuous recording of a RyR2 in the presence of protamine at low cytosolic Ca²⁺. Notice that when the RyR2 activates, it preferentially switches from the closed state to the protamine-induced substate rather than to the full open state (grey arrows). In contrast, when the channel closes, it is more likely to do it from the full open state than from the protamine-induced substate (black arrows). Estimations (from 4 minutes of recordings at +40 mV) of the probability of full openings, closures and protamine-induced substates are shown in. Analysis of these results suggests that there is a preferential sequence of events during protamine-induced RyR2 activation. First, protamine binds to the closed channel and induces activation. Transitions from closed to the substate (1423) are 8.3 times more probable than transitions from closed to full openings (171). Once activated, the channel fluctuates between the substate and the full open state, probably as a result of rapid binding-unbinding of protamine to its site. Relative rate of transition (number of transitions A→B/probability A) for Full open→Closed (9165) is nearly ten-fold that for Protamine substate→Closed (919). If we assume that protamine is unbound during full openings, then the RyR2 would be no longer under the activating influence of protamine and it would be more likely to close. # Discussion Protamine did not simply block RyR2 in an all-or-none fashion. Rather, it affected the RyR2 channel conduction pathway by producing substates and full block in a voltage-dependent and dose-dependent manner. At low concentrations, protamine induced a drop in the current amplitude to a large conductance substate. The ratio between this amplitude drop and the amplitude of the full opening decreases as the positive holding voltage (V_m) increases. This is reminiscent of the rectification of some K⁺ channels observed in presence of polyamines. Furthermore, protamine activated channels at low resting cytosolic Ca²⁺ levels (25–100 nM) even in the absence of lumenal Ca²⁺ or in the presence of high cytosolic Mg²⁺ levels, suggesting that protamine-induced RyR2 activation mainly results from electrostatic interactions and does not seem to require cytosolic or luminal Ca²⁺ binding to the respective activating sites. ## Comparison with Previous Studies In this work, protamine added to the cytosolic surface of RyR2 had a complex action as it affected RyR2 channel conduction (multiple substates/block) and gating (activation). Our observation of full block of RyR2 induced by high levels of protamine (20 µg/ml; V_m = 0 mV) are consistent with previous findings in single RyR1 channels. However, lower protamine concentrations that would have allowed to observe protamine-induced RyR1 activation have not been tested in this previous report. Moreover, the substates and full block of fully activated RyR2 observed here at V_m = 0 mV would lead to a decrease in current flow through the channels, which is in agreement with previous reports were protamine was found to inhibit caffeine- induced RyR1-mediated Ca²⁺ release from skeletal muscle SR microsomes (notice that the expected SR lumen voltage in microsomes is ∼0 mV). Although this previous study did not test the effect of protamine in conditions homologous to those used here to detect activation, it provided evidence that RyRs can be activated by another polyarginine (Histone IIS) and by a synthetic polylysine. These agents were found to activate Ca²⁺ release from resting RyR1 and to induce block of drug-activated channels. Similarly, polylysine has been reported to activate RyR2-mediated Ca²⁺ release in cardiac microsomes regardless of the presence/absence of Mg²⁺. The literature suggests that the activation of RyR by polycations is relatively restricted to peptides like polylysines and polyarginines. Protamine/histone/polylysine-induced activation of SR Ca²⁺ release would result from Ca²⁺-independent and Mg²⁺-insensitive transitions from closed to the high subconductance state in resting RyRs. Protamine and polylysine have been reported to produce diverse effects that do not include activation on other ionic channels. Thus, protamine- and polylysine- induced activation seems to be specific for RyRs. Inhibition of drug-stimulated RyR-mediated SR Ca²⁺ release would result from voltage-dependent transitions from the full open state to multiple subconductance and block-like states. A large variety of polycationic molecules were found to produce these effects on RyRs, as well as on other channels. ## Effect of Protamine on RyR2 Channels Conductance Protamine is essentially a poly-arginine molecule with a charge of ∼+22. Thus, it is reasonable to expect voltage-dependent effects on the RyR channels as they contain rings of negatively charged amino acids in the cytosolic vestibular regions that sense the electrical field. Indeed, at V_m≤0 mV protamine added to the RyR cytosolic surface produced several subconductance states as well as full block. However, at V_m ranging from +10 to +80 mV, block events and smaller substates disappeared, suggesting sharp voltage-dependence, while the substate with the highest level of conductance remained with a significant probability. This difference in the rate of decline of the probability for the different substates could indicate that multiple protamine molecules, despite their large charge and size, would bind to multiple sites in the RyR vestibule that would have differential sensitivity to the electric field. This would be in agreement with the apparent multiple binding components that have been suggested for polyamine-induced rectification of channels. The nature and functional implication of substates in RyRs is not known (discussed). Protamine substates are different from those induced by known RyR conductance modifiers such as ryanoids and Imperatoxin A (IpTx_A), which change channel conductance by inducing a drop in the current amplitude that is proportional to the amplitude of the full opening (i.e., these agents produce a “rigid” conformational change in the RyR's conduction pathway). The nearly constant drop in current amplitude observed at positive voltages for the protamine-induced transition from full opening to substate is unusual and it implies that the proportion by which the conductance is reduced (relative to the size of the full open channel) is less as the holding voltage increases. As the driving force of Ca²⁺ fluxes increases, it could better counteract an electrostatic screening phenomenom produced by bound protamine and/or change the orientation of protamine putative coiled-coil structure in a rigid vestibule. It is also possible that protamine binds to a flexible loop in the RyR that modulates the ability of protamine to act as a plug within the vestibule in a voltage-dependent manner. The cationic peptide IpTx_A, induces long-lived substates by binding to sites putatively located at RyR regions that sense the electrical field. More recently, a 3D mapping study has located the IpTx_A binding to a RyR region between the clamp domain and the central handle domain. Here, protamine greatly decreased the probability of IpTx_A-induced substates suggesting that protamine may bind to the same RyR region as IpTx_A. Protamine was effective at inducing substates in ryanodol-modified channels, suggesting that the binding sites for protamine and for ryanoids do not overlap. The ryanodol-protamine interactions seem to be cooperative since protamine increased the probability of ryanodol-induced substates. This result resembles previously reported findings, where IpTx_A, a segment of the DHPR II- III loop (Peptide A) and other positively charged “ball peptides” affected the action of ryanodol in a cooperative fashion. As ryanoids and IpTx_A were shown to bind to different regions in the RyR molecule, our finding that protamine affects the action of both agents would suggest multiple sites of action or allosteric effects. Increase in ryanodine binding induced by polylysine has also been reported for skeletal and cardiac RyRs. In summary, protamine affects RyR2 channel conduction (substates and block) and the binding affinity of agents that interact with vestibular regions of RyR2. This could be the result of allosterically-induced flexible conformational changes, although steric interference cannot be ruled out. This effect of protamine on conduction does not appear to be highly specific as similar effects were found for other polycationic molecules. ## Protamine Activates RyR2 Protamine (1 µg/ml) activated RyR2 at cytosolic Ca²⁺ levels below those found in resting cells (when the RyR2 are normally closed). Regarding the steps involved in protamine-induced activation, our results show that there is a preferential sequence of events. The channels first open to reach the protamine substate level and then they fluctuate between the full open sate and the substate. We also observed that the probability to close is higher when the channel is fully open than during substates. Like protamine, IpTx_A- and ryanoids-induced substates rarely transition to the closed state. Unlike protamine, IpTx_A and ryanoids do not activate RyR2 under resting Ca²⁺ conditions. The efficacy of protamine to activate RyR2 was not affected by addition of up to 5 mM Mg²⁺, known to compete with Ca²⁺ at the RyR2 cytosolic binding activating sites. Moreover, protamine remained effective after we replaced lumenal Ca²⁺ with Ba²⁺, which would prevent “feed through” activation, as Ba²⁺ can permeate through the channel but does not activate it. The results support the idea that protamine-induced activation of RyR2 is mostly mediated by interactions that are Ca²⁺- independent within levels found in the cytosol and SR lumen of myocytes. Although RyRs are highly sensitive to the activating action of protamine and polylysine, they are not gated open by a variety of other cationic molecules, including peptides, polyamines and macrolide antibiotics, which only produce substates and/or block. The reason for this differential sensitivity is still unclear but it is possible that the high efficacy of cationic polypeptides is related to the large magnitude of their charge density (e.g., ∼+22 for protamine), which would allow them to strongly interact with a negatively charged gating domain. Previous studies have found that Ca²⁺ dependence of RyR gating is insensitive to carbodiimide titration of negative surface charges located either on the luminal or the cytosolic channel surface. Likewise, \[³H\]ryanodine binding remained Ca²⁺ dependent even in presence of very high salt levels (1 M NaCl or KCl) which should greatly affect electrostatic interactions. Thus, it is possible that the action of protamine may be more than just a surface-charge effect. Accordingly, previous reports with histones and polylysines of different molecular weights and charges did not clearly indicate whether the most charged molecules are better RyR activating agents. In principle, poly L-arginines would be better activating agents than poly L-lysines. Indeed, we reach near maximal activation of RyR2 with ∼250 nM protamine, which is about one tenth of the concentration of a larger molecular weight polysysine required to elicit the same effect. It is possible that the arginines present in protamine are more suitable to form strong π-cation interactions with aromatic aminoacids present at the RyR gating domain, resulting in a higher binding affinity. ## Is There a Role for Modulation of RyRs through Electrostatic Interactions Protamine is clinically utilized to reverse heparin overdose and has been shown to display significant cardiotoxic effects, including decreased cardiac output and electrocardiographic disturbances. Studies performed using isolated myocytes revealed multiple effects of protamine, including changes in cardiomyocyte contraction (depressed contraction at 3 Hz but enhanced at 0.5 Hz or at lower rates), partial membrane depolarization, rise in resting tension and appearance of rested state rapid cooling contractures. The effects of protamine have been associated to calcium overload and impairment of sarcoplasmic reticulum functions. Our results open the possibility that direct protamine binding to RyR2 could mediate, at least in part, the observed abnormalities in SR Ca²⁺ homeostasis. It is plausible that the strong effect of protamine on channel conduction and gating results from its interaction with the rings of charges in the vestibular region of RyRs. However, the physiological role for the existence of a RyR channel domain that confers susceptibility to modulation by agents like protamine is still unknown. A physiological implication of our findings is the possibility that protamine mimics the action of positively charged peptide motifs, present in the vicinity of the RyR2 channels in cells which might modulate channel activity (activation and deactivation). In favor of this possibility, previous observations suggested that RyRs hypersensitivity to polylysine is associated with cardiac diseases or with modifications of skeletal RyR1 interdomains that trigger malignant hyperthermia. Studies of multichannel behavior indicated that physical RyR-RyR interactions between domains of neighboring channels could shape their coupled gating. It is possible that protamine-sensitive domains are involved in this process. RyR also interact with DHPRs. Indeed, during the action potential, physical DHPR-RyR1 interactions activate RyR1-mediated Ca²⁺ release in skeletal fibers. The very limited effects of IpTx_A and DHPR peptides on sparks suggest they have restricted access to the RyR2. An interesting observation is that protamine also has very limited effect on RyR1 in skeletal fibers, suggesting that the protamine-sensitive domains of RyR1 are not accessible in the cellular environment. An interpretation is that there might be an endogenous polycationic peptide, which could move when propelled by a change in the electric field at the T-tubule during excitation to induce activation of RyR1. The same moiety could subsequently block the RyR1 when the T-tubule voltage returns to resting values. In the heart, Ca²⁺ entry is known as an absolute requirement for cardiac EC coupling and the presence of a voltage-dependent component in the activation is still an open question. Still, the observation of Ca²⁺-entry-independent negative modulation of local events of Ca²⁺ sparks by specific DHPR agonist and blockers suggests the existence of physical RyR2-DHPR interactions of a still unknown nature that regulate RyR2 function. The protamine-binding domain in RyR2 might be involved in this modulation. In summary, our studies indicate that protamine, can induce Ca²⁺-independent RyR2 activation and deactivation. The possibility of similar moieties shaping RyR-RyR interactions for functional coupling or DHPR- RyR interactions for triggering and/or terminating SR Ca²⁺ release opens the field for further investigation. We thank Prof. D.M. Caspary, Ph.D, SIU-SOM, for his critical input. We also thank Mr. R. Herndon, Research Administrator and Mrs. J. Bryan, Office System Specialist, SIU-SOM, for proofreading the manuscript. [^1]: Conceived and designed the experiments: PLDS JAC. Performed the experiments: PLDS JAC. Analyzed the data: PLDS JAC. Contributed reagents/materials/analysis tools: PLDS JAC. Wrote the paper: PLDS JAC. [^2]: The authors have declared that no competing interests exist.

# Introduction Vestimentiferan tubeworms (Polychaeta; Siboglinidae) are among the foundation species in deep-sea chemosynthetic communities worldwide. The 10 currently recognized genera in this well-defined siboglinid clade are commonly partitioned into subgroups based on habitat types and evolutionary relationships: the hydrothermal vent genera (*Riftia*, *Ridgeia*, *Tevnia*, *Oasisia*, *Alaysia*, *Arcovestia*) and the hydrocarbon seep genera (*Lamellibrachia*, *Escarpia*, *Paraescarpia*, *Seepiophila*). Vestimentiferans living at vents experience frequent ecological disturbances and unpredictable changes in physicochemical conditions, which has led to the evolution of 'weedy' species with short lifespans and fast growth rates. In contrast, tubeworms living in seeps experience relatively stable habitat conditions with more constant and less toxic fluid flows, which has led to long-lived and slow-growing species. Nonetheless, some of the putative 'vent' species have been found at cold seeps and some of the 'seep' species have been found at vents, whale falls and even shipwrecks. Thus, a strict ecological separation of 'vent' and 'seep' tubeworm genera does not appear to exist. All adult vestimentiferans lack a functional digestive system and must therefore acquire their nutrition entirely from chemosynthetic bacterial endosymbionts that inhabit specialized cells within a complex, anatomically adapted organ of the tubeworm trunk, the trophosome. These bacteria oxidize reduced sulfur compounds to gain energy for the production of organic matter, part of which is shared with the tubeworm host. Studies on *Riftia pachyptila* indicate that tubeworms produce aposymbiotic larvae that acquire their symbionts horizontally via infection by free-living bacteria from the local environment in which the larvae settle. Penetration of the larval epidermis by infecting bacteria triggers metamorphosis to a gutless juvenile stage and initiates a profound renewal of the skin, thereby preventing further symbiont infections. After the death of the tubeworm the symbionts are released to the free-living population and regain the potential to infect new hosts, which ensures the persistence of this partnership in subsequent generations. The potential for multiple infections by different locally adapted bacterial strains creates opportunities for selective enrichment of potentially beneficial strains, but it decouples dispersal of the host larvae from dispersal of the symbionts, thereby increasing the risk of failing to acquire a suitable symbiont. Infectious environmental acquisition, as seen in *Riftia*, is expected to result in heterogeneous symbiont populations within host individuals. Nevertheless, most *16S* rRNA studies to date suggest that individual tubeworms harbor symbiont populations composed of a single gammaproteobacterial phylotype, but see. Tubeworm symbionts form a monophyletic clade that is subdivided into two major habitat-specific phylotypes. The seep tubeworms (e.g., *Lamellibrachia* and *Escarpia*) host *16S* rRNA phylotype I, which comprises three subgroups that show different depth distributions. By contrast, vent tubeworms (e.g., *Riftia* and *Tevnia*) host the closely related phylotype II, *a*.*k*.*a*. *Candidatus Endoriftia persephone*. A third, distantly related phylotype group was recently described from low-diffuse vents in the Caribbean Sea. A number of studies questioned the lack of intra-host variation of the tubeworm symbionts and provided evidence that distinct phylotypes and subtypes can exist in a single tubeworm individual. The diversity of symbiont populations within individual hosts and among geographically disjunct vent and seep localities is likely to be underestimated. Although genomic references for seep and vent endosymbionts are now available, population genetic studies of tubeworm symbionts have relied on direct sequencing or cloning of PCR products, methods that are biased towards the most abundant bacterial types in mixed symbiont infections. Given these limitations Zimmermann et al. argued that advanced molecular techniques would be needed to uncover symbiont variation at the *16S* level. Such methods would probably also be needed to reveal specificity between host species and symbiont phylotypes. A few studies indicated that some degree of host-symbiont specificity might exist in vestimentiferan symbioses. For example, co-distributed tubeworm species do not always take up identical symbiont subtypes, but small-scale patchiness in local environments might account for these results. In the present study we used high-throughput *16S-V4* amplicon sequencing and complementary CARD-FISH analyses in co-occurring eastern Pacific populations of the seep-associated tubeworm species *L*. *barhami* and *E*. *spicata* to shed more light on symbiont diversity and specificity in vestimentiferan tubeworms. We chose this gene region for our analyses because (1) it is the most commonly used fragment for bacterial diversity analyses and (2) publicly available *16S* sequences from tubeworm symbionts contain a few single nucleotide polymorphisms in the *V4* region. Our assumption was that this variation is actually higher and has so far been obscured by low-throughput methods. In contrast to these expectations, our sequencing data indicated that the *V4* hypervariable region is monomorphic between and within *L*. *barhami* and *E*. *spicata* hosts, while additional CARD-FISH in the *V6* region provided evidence for phylotypic diversity. # Methods ## Sample collection and DNA extraction Tubeworm specimens were collected with remotely operated vehicles (ROVs *Tiburon* and *Hercules*) from two widespread locations in the eastern Pacific Ocean. Permissions for animal collections during the R/V *Western Flyer* 2002 cruise in US territorial waters were not required. Sampling permits for seeps in the Gulf of California during the R/V *Western Flyer* 2003 and E/V *Nautilus* 2017 (NA090) cruises were obtained by the Monterey Bay Aquarium Research Institute and the Ocean Exploration Trust from Mexico's Secretariat of Foreign Affairs (SRE: DAN-00254, EG0072017), the Secretariat of Environment and Natural Resources (SEMARNAT: SGPA/DGVS/5152) and the National Aquaculture and Fishing Commission (CONAPESCA: 13103.613-03/0057, PPFE/DGOPA-010/17) where necessary. As soon as possible after recovery of the vehicles, tubeworms were removed from their tubes, dissected on individual dishes and frozen at –80°C. Only individuals with intact trophosomes were considered for analysis. For *16S* amplicon sequencing DNA from 21 *L*. *barhami* and 12 *E*. *spicata* individuals was extracted at the Monterey Bay Aquarium Research Institute (Moss Landing, CA, USA) with the DNeasy Blood & Tissue kit (Qiagen, Hilden, Germany). We followed the manufacturer's instructions, except for adding a second elution step to increase DNA yield. Because symbiont distributions can vary across the length of the trophosome, we sampled serial tissue sections and used the homogenates for DNA extractions. The PowerClean Pro DNA clean-up kit (Mo Bio, Carlsbad, CA, USA) was used to remove contaminants that might inhibit PCR. For CARD-FISH analyses trophosome pieces of three *L*. *barhami* and three *E*. *spicata* individuals from the E/V *Nautilus* 2017 cruise to the Guaymas Transform Fault were fixed in PFA overnight, washed three times in PBS and then stored in PBS:ethanol. The PFA fixed specimen were not used for DNA analyses. ## Host species identification Mitochondrial *COI* sequences were used to verify the tubeworm species identifications that were initially based on morphology while on shipboard. A \~700 bp fragment of the *COI* gene was amplified on a Veriti thermal cycler (Applied Biosystems, CA, USA) using 10 *p*mol of primers jgLCO1490 and jgHCO2198, 12.5 μl AmpliTaq Gold Fast Master Mix (Thermo Fisher Scientific, Foster City, CA, USA) and \>20 *n*g of DNA adjusted to a total volume of 25 μl with PCR grade water. Negative controls without template were included on each PCR plate to check for sample contamination. Cycling and sequencing protocols followed. Sequence analysis was performed with GENEIOUS v9.1.8 (<http://www.geneious.com/>) as described in. All sequences were compared to the NCBI Nucleotide Collection with MEGABLAST to determine species identities. POPART v1.7 ([http://popart.otago.ac.nz](http://popart.otago.ac.nz/)) was used to draw haplotype networks based on the median joining algorithm. ## Symbiont *16S* amplicon sequencing Barcoded amplicon libraries targeting a \~250 bp fragment of the hypervariable *V4* region of the *16S* rRNA gene were prepared using the primer pair 515f/806r in a single PCR step. PCR amplifications were done in triplicate with the AmpliTaq Gold Fast Master Mix (Thermo Fisher Scientific, Foster City, CA, USA) in a volume of 25 μl. We used the same PCR protocol as, but decreased the number of cycles from 35 to 26 to reduce PCR induced mutations and chimera generation. Three negative controls were included to check for cross-contamination. After cleanup with AMPure XP beads (Beckmann Coulter, Brea, CA, USA), final libraries were quantified with the Quant-iT PicoGreen dsDNA assay (Thermo Fisher Scientific, Eugene, OR, USA) and then mass normalized to 240 ng per sample. Pooled libraries were sent to SeqMatic (Fremont, CA, USA) for 2x250 bp paired- end sequencing on one lane of an Illumina HiSeq2500 platform. ## Sequence analysis and identification of operational taxonomic units (OTUs) Demultiplexed paired-end reads were quality checked in FASTQC v0.11.5 and then adapter clipped with TRIMMOMATIC v0.38. Reads were merged, filtered, dereplicated and clustered with USEARCH v11. In addition, we tried the QIIME1 and QIIME2/DADA2 OTU clustering pipelines. The minimum %id of alignment and the maximum number of mismatches during merging were set to 80 and 10, respectively, while the merging length was constrained to 230–270 bp. Merged reads that had an error rate \>0.001 and were shorter than 230 bp after truncation at a base quality threshold of 20 were discarded. Dereplication, denoising and clustering of reads into ZOTUs was performed with the *fastx_uniques* and *unoise3* commands. All merged reads were then mapped to the clustered sequences with the *otutab* command to generate an OTU table. Taxonomic classification was performed in QIIME2 ([https://qiime2.org](https://qiime2.org/)) with the silva-132-99-515-806-nb-classifier. Samples with less than 1000 reads, OTUs with less than 100 reads as well as singletons were filtered from the OTU table. This approach recovered one main symbiont OTU in all samples, which we hereafter call Seep symbiont 1. To identify whether this OTU could be separated into different genotypes we applied the OLIGOTYPING v2.0 pipeline on the merged and padded symbiont reads. ## CARD-FISH To verify the occurrence of Seep symbiont 1 in the trophosomes of *L*. *barhami* and *E*. *spicata* we designed specific horseradish peroxidase (HRP) labeled oligonucleotide probes for the *V4* region of the *16S* rRNA. Unlabeled helper probes flanking the target regions were designed to open up the *16S* rRNA secondary structure and to make the binding sites accessible, while unlabeled competitor probes were used to avoid unspecific hybridizations of the HRP- probes. A 0–60% formamide series with 10% increments was performed to identify the formamide working concentrations for the different oligonucleotide probes. Clear signals were only observed at 20% formamide. The general eubacterial probe EUB338I-III and the nonsense probe NON338 were used as positive and negative controls, respectively. To further investigate symbiont variability in the trophosome, we used the probe L_mars1, which targets the hypervariable *V6* region. An alignment showing the binding specificity of the Seep symbiont 1 and L_mars1 probes is given in. All probes were used on the same sample preparations. Although we processed each of the six PFA-fixed samples for CARD- FISH, the *E*. *spicata* trophosome pieces were mostly lost during processing of the microscope slides, so that we focused our analyses on *L*. *barhami*. Optimal tissue sections and preservations were obtained for *L*. *barhami* \#45 and we therefore performed essentially all double hybridizations on this individual. The PFA-fixed tubeworm samples were dehydrated through serial incubations in 70% ethanol/PBS, 80% ethanol/PBS and 96% ethanol for 30 min. Embedding was done in Steedman's wax by incubating the tissues in 96% ethanol/wax and three times in pure wax for 60 min at 38°C. The wax blocks were cut into 4 μm sections, mounted on Polysine-coated glass slides and incubated at 30°C overnight to improve adherence of the tissue sections. Slides were washed three times in 96% ethanol (5 min) and then rehydrated by incubation in 80% and 70% ethanol for 10 min. Endogenous peroxidases were inactivated by washing the slides in 0.2M HCl for 12 min. Permeabilization was performed by incubation in 20 mM Tris/HCl (pH 8) for 10 min, lysozyme solution (0.01 g/ml lysozyme, 0.05 M EDTA, 0.1 M Tris/HCl pH 8) for 30 min at 37°C and 20 mM Tris/HCl (pH 8) for 10 min. Sections were circled with a Pap-pen (Kisker Biotech, Steinfurt, Germany) to avoid leakage of the hybridization mixture across sections. CARD-FISH was performed according to with modifications. 1 μl of each probe (50 ng/μl) was diluted in 150 μl hybridization buffer and then incubated for 3–24 hours at 46°C in dark humidity chambers. Slides were washed in pre-warmed washing buffer (48°C) and 1x PBS for 15 min before amplification. Amplification was done for 60 min at 46°C with Alexa-488 labeled tyramides (Molecular Probes, Leiden, the Netherlands). Subsequently, slides were washed in 1x PBS for 15 min and Milli-Q water for 10 min. For double CARD-FISH four probe combinations were tested: Seep_symbiont_1 + EUB338I-III, EUB338I-III + Seep_symbiont_1, Seep_symbiont_1 + L_mars1 and L_mars1 + Seep_symbiont_1. The HRP enzyme of the first probe was inactivated by incubation in 0.5% H₂O₂ in methanol for 30 minutes prior to hybridization of the second probe. Amplification for the second probe was done with Alexa-594 labeled tyramides (Molecular Probes, Leiden, the Netherlands). After air-drying the tissue sections were counterstained with 4,6-diamidino-2-phenylindole (DAPI) for 8 min, washed in Milli-Q water for 2 min and then embedded for microscopy in Vectashield (Vector laboratories, Burlingame, CA, USA). Overview images of whole tissue sections were taken on an Olympus BX53 compound microscope (Olympus, Tokyo, Japan) equipped with an ORCA Flash 4.0 (Hamamatsu Photonics K.K, Hamamatsu, Japan) camera using a 20× Plan- Apochromat objective and the software cellSens (Olympus, Tokyo, Japan). Close-up images were taken on a Zeiss LSM 780 confocal laser-scanning microscope (Carl Zeiss, Jena, Germany) equipped with an Airyscan detector (Carl Zeiss, Jena, Germany) using a 63× Plan-Apochromat oil-immersion objective. For image acquisition the Zen-Black software (Carl Zeiss, Jena, Germany) was used. Details of the image acquisition settings can be found in. Raw pictures were further processed in Fiji. # Results ## Host cytochrome-c-oxidase subunit I (*COI*) sequencing Mitochondrial *COI* sequences revealed two highly divergent haplogroups that identified the two tubeworm species *E*. *spicata* and *L*. *barhami*. Each haplogroup consisted of one major haplotype and two to four minority haplotypes that differed only by a few mutations. For the two *L*. *barhami* populations no significant geographic structure could be detected as–with the exception of three site-specific sequences–haplotypes were shared among localities. ## *16S* amplicon sequencing and OTU clustering *16S* amplicon sequencing resulted in an average of 73,133 raw paired-end reads per sample. After merging an average of 21,872 reads/sample remained. The USEARCH denoising pipeline identified three ZOTUs in the dataset, two of which were excluded due to low abundance and/or occurrence in only a single sample, which is indicative of sequence artifacts or contaminants. Although one of these ZOTUs seemed to be related to an *E*. *spicata* endosymbiont, attempts to differentiate this and other sequences from the remaining dominant symbiont OTU with the OLIGOTYPING method were unsuccessful as entropy values for each nucleotide position were much smaller than 0.2, implying that sequence polymorphisms were related to sequencing or PCR errors rather than biological variation. After all filtering steps and OLIGOTYPING analyses we could confirm only one tubeworm symbiont OTU–Seep symbiont 1 –that occurred in all samples independent of species or locality. ## CARD-FISH We investigated the presence of two symbiont phylotypes (targeted by the Seep symbiont 1 and L_mars1 probes) in a total of 151 throphosome sections (4 μm section thickness) on 28 slides of three *L*. *barhami* and two *E*. *spicata* specimens (LB45: 113 sections; LB33: 7 sections; LB39: 7 sections; ES9: 11 sections; ES19: 13 sections). Our analyses generally confirmed the presence of Seep symbiont 1 in the trophosomes of all investigated *L*. *barhami* individuals as well as in one *E*. *spicata* individual (;, and Figs;). The Seep symbiont 1 cells are coccoid-shaped with a diameter of approximately 5 μm. *L*. *barhami* sections examined using the Seep symbiont 1 probe produced a positive signal in 56 of 94 cases, and four out of 16 *E*. *spicata* sections produced a signal. All symbionts that hybridized with the Seep symbiont 1 specific probe were also stained by the general EUB338I-III probe in the double CARD-FISH analyses (and Figs;). No signals were detected for the negative control probe NON338. Technical issues or poor quality of the tissue sections produced hybridization failures for 45 sections. To further assess symbiont diversity in the trophosome we used the previously validated symbiont probe L_mars1 that was designed for the *16S-V6* region on the same samples and tissue sections in both mono and double CARD-FISH. In six sections out of 40 this probe stained localized aggregations of symbiont cells in the *L*. *barhami* trophosome, indicating the presence of at least one other seep symbiont phylotype that was not observed in the *16S* OTU set. Similar to Seep symbiont 1 and in agreement with the results by Zimmermann et al. for peripheral symbiont cells, this phylotype had a coccoid morphology with a cell diameter of \~5 μm. As positive CARD-FISH signals for this second phylotype were only observed in consecutive sections or sections within close proximity, we suppose that this symbiont was compartmentalized in a specific region of the trophosome in concordance with previous descriptions. The repeated observation of these colonization aggregates and the differences in target sequences between symbiont phylotypes further indicate that these patterns are a true signal and do not result from unspecific binding of the L_mars1 probe (see also). No other hybridizations with this probe exhibited fluorescence. In 22 of the 40 sections a lack of hybridization signal was most likely observed because the symbiont type was not present in the investigated tissue sections. In these cases, a fluorescence signal was observed for the Seep symbiont 1 or EUB338I-III probes either in the same (for double CARD-FISH) or in adjacent sections (for mono CARD-FISH), indicating that the overall CARD-FISH experiment was successful. In the remaining 12 sections the hybridizations failed. These analyses were all double CARD-FISH analyses, which can fail at various steps due to e.g., insufficient binding or competition between probes, ineffective amplification reactions of the HRP, removal of probes during consecutive washing steps etc. We also tried the L_mars1 probe on three *E*. *spicata* slides (three sections), but did not obtain any fluorescence. Since in these cases positive signals were obtained for the EUB338I-III probe, we assume that this symbiont type was absent in the investigated sections. In summary these outcomes support the presence of a second, highly localized phylotype in the investigated *L*. *barhami* individual, which is consistent with previous findings in *L*. *anaximandri*. # Discussion Horizontal transmission presumably occurs during a narrow window of susceptibility, when locally occurring stages of symbiotic bacteria infect the settling tubeworm larvae. Compared to vertical transmission, which leads to bottleneck effects and co-speciation in the symbiont population, the environmental infection mode is expected to result in symbiont heterogeneity within host individuals and the absence of specificity between a particular host and a particular symbiont. In vestimentiferan tubeworms most traditional *16S* sequence analyses have shown that every tubeworm individual contains only one gammaproteobacterial symbiont phylotype, whereas evidence for host specificity has been inconclusive. A limited number of studies challenged the finding of genetic homogeneity in the symbiont population. For instance, *16S* clone libraries showed that *Lamellibrachia anaximandri* from the Mediterranean Sea as well as *Escarpia laminata* and *Lamellibrachia* sp. 2 from the Gulf of Mexico can harbor two distinct symbiont phylotypes. Given that conventional PCR- based approaches are biased towards the most abundant DNA template, Zimmermann and colleagues argued that these data likely underestimate the true degree of symbiont heterogeneity and that improved molecular analyses would be needed to identify the level of hidden diversity. To address these issues, we used high- throughput *16S* amplicon sequencing in co-occurring populations of the tubeworm species *L*. *barhami* and *E*. *spicata* from the eastern Pacific Ocean. In contrast to Zimmermann et al.’s expectations our OTU clustering results revealed the presence of only one symbiont phylotype–Seep symbiont 1 –in *L*. *barhami* and *E*. *spicata*, which confirms previous notions that the tubeworm symbiosis is highly selective. Although uptake of multiple strains would provide the advantage of selecting symbionts that are optimally adapted to the local environment, increased discrimination for a symbiotic partner might reduce the risk of infections by bacteria that do not provide any benefit to the host, given that horizontal transmission favors the evolution of cheaters. The molecular mechanisms that underlie symbiont selectivity in vestimentiferan tubeworms are poorly understood. Perez and Juniper hypothesized that the type-VI secretion system, found in the *Ridgeia* symbionts in their study, might be involved in partner choice as it is in the rhizobium-legume symbiosis. Other types of secretion systems that might have similar roles in host selectivity have recently been discovered in the draft symbiont genomes of the seep tubeworms *Lamellibrachia*, *Escarpia* and *Seepiophila*. Evidence from gene expression analyses further suggests that the host immune system interacts directly with the symbionts to govern cell growth, as transcripts for peptidoglycan recognition proteins and toll-like receptors were significantly upregulated in the trophosome compared to symbiont-free tissues. Despite the high selectivity observed in the tubeworm symbiosis relative to general microbial communities, our results imply a lack of specificity between host species and symbiont phylotype, given that Seep symbiont 1 associated with both *L*. *barhami* and *E*. *spicata*. These data agree with a previous study by Vrijenhoek et al., which showed that co-occurring *E*. *spicata* and *L*. *barhami* from the Gulf of California contain the same *16S* phylotype–a remarkable finding because these host species belong to phylogenetically very distantly related vestimentiferan taxa. A limited number of studies provided evidence for host-symbiont specificity in other co-occurring tubeworm species, although these investigations usually included comparisons of a typical seep species with a typical vent species. Vent tubeworms such as *Riftia pachyptila* acquire their symbionts from the surrounding water during a short period between larval settlement and metamorphosis to the juvenile stage. The mechanisms of symbiont acquisition and durations of infection susceptibility have not been similarly investigated in seep tubeworm taxa, which might influence their symbiont compositions relative to those of their vent counterparts. In contrast to vent tubeworms seep tubeworms possess long body extensions called 'roots', with which they obtain sulfide from the seafloor sediments for their symbionts. Cordes et al. suggested that the root system plays other important roles in the tubeworm symbiosis. They hypothesized that seep tubeworms use their roots to pump sulfate byproducts into the surrounding sediments, where it is consumed during sulfate-driven anaerobic methane oxidation. This, in turn, would produce sulfide that the tubeworm can deliver to its endosymbiont. Perhaps the root system has more far reaching functions and continuously acquires new symbiotic bacteria, as it appears to do in bone-eating *Osedax*, a related siboglinid tubeworm and in the legume-rhizobia symbiosis. Such potential differences in symbiont uptake mechanisms could be alternative explanations for the observed differences in symbiont compositions between vent and seep tubeworms without the presence of genetic specificities. Disentangling the roles of phylogenetic, genetic, physiological, behavioral and ecological factors affecting the acquisition and enrichment of symbiont phylotypes by various tubeworm species will be difficult. One limitation of our study is that we only investigated a small fragment of the evolutionarily conserved *16S* rRNA gene, which might not have sufficient resolution to reveal patterns of symbiont specificity and variability if they exist. Metagenomic studies in the deep-sea hydrothermal vent mussel *Bathymodiolus septemdierum* have recently shown that this species harbors a single endosymbiotic phylotype with a monomorphic *16S* rRNA sequence, but that the population of this phylotype varies significantly in the composition of key metabolic genes for hydrogen oxidation and nitrate reduction—a crucial finding that helps to understand how host animals might adapt to the dynamic environmental conditions at vents and which remained undetected by previous methods. Interestingly, we observed another seep symbiont phylotype when targeting a different region of the *16S* rRNA with CARD-FISH. This result could indicate that the *V4* region might not be an appropriate target to investigate diversity in tubeworm endosymbionts, at least in this sample set. Studies on pathogenic bacteria have previously shown that the *V4* region can be more conserved than other hypervariable regions. Our findings are still unexpected given that minor variations in the *V4* regions can be found between publicly available seep symbiont *16S* sequences (e.g., GenBank). One other limitation of our study could be that we investigated only adult tubeworms but not larval stages in which the symbionts are acquired. For example, it is possible that multiple infections occur in tubeworm larvae, but that different symbiont types outcompete each other during the development of the tubeworm or that some strains are lost due to drift and only a subset of the initially acquired symbiont diversity is retained. PCR-free metagenome analyses in a sufficient number of samples and developmental stages might be best suited for further research that intends to address these concerns and uncover symbiont diversity and specificity in vestimentiferan tubeworms that might be hidden below the *16S* rRNA level. # Supporting information We thank the ship crews and ROV pilots for their able assistance in sampling tubeworm specimen for this study, Bob Vrijenhoek for providing samples from his collection as well as SeqMatic (44846 Osgood Rd, Fremont, CA 94539) for sequencing our *16S* amplicon libraries. This research used samples provided by the Ocean Exploration Trust’s Nautilus Exploration Program, Cruise NA090. The Symbiosis Group at the Max Planck Institute for Marine Microbiology Bremen is gratefully acknowledged for their excellent advice and lab support on the CARD- FISH analyses. Especially, we thank Silke Wetzel, Miriam Sadowski and Martina Meyer for helping with the lab procedures. We also want to thank Roxanne Beinart, Peter Girguis and Jessica Mitchell for providing the PFA fixed material. Finally, we want to acknowledge an anonymous reviewer for thoughtful comments that helped to improve this manuscript. [^1]: The authors have declared that no competing interests exist. [^2]: Current address: University of Rhode Island, Graduate School of Oceanography, Narragansett, RI, United States of America

# Introduction Imatinib Mesylate, commonly known as Gleevec, is a selective tyrosine kinase inhibitor which targets the Abelson tyrosine kinase, also known as Abl1 and c-Abl, platelet derived growth factor receptor (PDGFR), transmembrane receptor tyrosine kinase, and ABL-related genes. In recent years, Imatinib has been a marketed effective drug for chronic myelogenous leukemia (CML) and gastrointestinal stromal tumors (GIST), and the two listed disorders are caused by BCR-ABL and c-kit oncogenes, respectively. Interestingly, recent studies have implicated Imatinib in ameliorating symptoms of diabetes, a disorder with a completely different pathogenesis from CML and GIST. For a number of patients who have both type 2 diabetes and CML, Imatinib has improved symptoms for both disorders. This finding prompts investigations into the mechanism of action of the Imatinib drug at the cellular level, and whether it is also therapeutic for type 1 diabetes. Type 1 diabetes, also known as insulin-dependent diabetes is a chronic autoimmune disorder affecting approximately 1/300 persons in the United State. It is caused by genetic and/or environmental factors interacting to induce an autoimmune response that destroys the islet β cells within the pancreas. This process involves autoreactive CD4⁺ and CD8⁺ T cells, B lymphocytes, and activation of the innate immune system. Insulin- dependent (type 1) diabetes ensues when this autoimmune attack coordinated with a pro-inflammatory environment results in the death of β cells to the extent that the residual islet β cells are unable to produce adequate insulin to regulate blood glucose levels in a normal range. When Imatinib was orally administered to nonobese diabetic mice (NOD), a typical human type 1 diabetes mouse model, the drug dramatically prevented NOD mice from developing type 1 diabetes. The most striking finding in this report was that Imatinib rapidly reversed diabetes in NOD mice with new onset diabetes showing that around 50% of the diabetic mice became euglycemia within a week of drug administration, and almost all diabetic mice were reversed within 10 days. While the authors attributed this effect to the possible anti-inflammatory activities of Imatinib to protect the residual β cells, the pancreatic histological study results failed to support this idea because there was no difference in terms of insulitic lesions between the treated and untreated groups. Also, it is an unwarranted conclusion that anti-inflammation leads to diabetes reversal under the situation that diabetes recurs within a week in the majority of the mice upon the discontinuation of Imatinib which has already been administered for three weeks. Imatinib’s rapid diabetes-reversing effect is also difficult to be explained by its anti-apoptotic effect on β cells demonstrated by a previous study because in diabetic NOD mice the residual β cells are not able to maintain euglycemia even if they are no longer undergoing apoptosis unless the ability of the residual β cells to secrete insulin is enhanced simultaneously. We believe the mechanism underlying Immatinib’s diabetes-reversing effect is still unclear. Based on the findings that Imatinib can quickly reverse new onset diabetes, two possibilities likely exist: one is that Imatinib enhances insulin production by the limited number of residual β cells in new-onset diabetic mice; the other is that Imatinib improves the sensitivity of peripheral tissues to insulin so that insufficient insulin would still be able to control blood glucose. Although the latter is supported by the previous study showing that Imatinib improves insulin sensitivity and glucose disposal rates in rats fed a high-fat diet, a recent study shows the contradictory, demonstrating that c-Abl activation, and not inhibition, enhances insulin sensitivity in liver, muscle and fat cells , the major tissues responding to insulin in our body. Therefore, it is questionable whether Imatinib truly improves insulin sensitivity. We think the most plausible explanation would be that Imatinib enhances insulin production by the residual β cells. In this study, we hypothesize that tyrosine kinase c-Abl negatively regulates insulin expression in β cells, therefore, inhibition of c-Abl enhances insulin production by β cells. To test this hypothesis, in this study, we investigated how c-Abl inhibition using different approaches affected the insulin production in β cells. Our results indeed demonstrated that Imatinib up-regulated insulin expression not only in glucose-stimulated β cells but surprisingly in non-stimulated resting β cells. SiRNA interference of the c-Abl gene markedly enhanced insulin production. Further analysis demonstrated that c-Abl tyrosine kinase affected insulin gene expression without involvement of PDX-1, which is known to be the major transcription factor regulating insulin gene expression. However, overexpression of c-Abl in β cells led to down-regulation of insulin gene expression accompanied by the reduction of NKx2.2, a positive regulator of insulin gene expression. Imatinib treatment of β cells leads to increased levels of NKx2.2. Intriguingly, we also unraveled that c-Abl inhibition markedly promoted the expression of glucose transporter, GLUT2, on β cells. To our knowledge, this is the first report on the role of C-Abl in inhibiting β cell insulin expression possibly via negatively regulating NKx2.2 and GLUT2. # Materials and Methods ## Mouse Insulin and C-peptide ELISA Experimentation Mouse pancreatic β cell line NIT-1 (derived from NOD β cells, originally purchased from ATCC and maintained in our laboratory) were cultured in 1 ml of DMEM media with 4.5 mM glucose and supplemented with fetal bovine serum (FBS), HEPES, MEM non-essential amino acids, and penicillin/streptomycin. The numbers of cells utilized in the various experiments were different but kept consistent in the same experiment. The cells were cultured for various times as indicated in the specific experiments under different conditions: media only, high concentration of glucose, imatinib, glucose and imatinib. Supernatant in some experiments was collected from each culture condition and measured for insulin and/or C-peptide with mouse ultrasensitive ELISA kits (ALPCO Diagnostics, Salem, NH). The assay was analyzed using spectrometer (BioTek Instruments, Inc., Winooski, VT) at 450 nm and 630 nm and insulin concentration was calculated according to instructions from the manufacturer. ## RT-PCR Experimentation NIT-1 cells were cultured in 1 ml of DMEM media with low glucose (4.5 mM) supplemented with FBS, HEPES, MEM non-essential amino acids, and penicillin/streptomycin. The cells were cultured for 6 hrs under different culture conditions as indicated in each experiment. After cell culture, mRNA was extracted from conditioned NIT-1 cells using Qiagen RNeasy Mini Kit (Qiagen Inc., Valencia, CA). The mRNA samples were reverse transcribed and evaluated via real-time RT-PCR. For insulin, pdx-1 and neuro D1 real-time PCR, samples were processed with optimized SYBR green protocol (Sigma-Aldrich, St. Louis, MO), and data was analyzed using Opticon MJ software (Promega Biosciences, Inc., San Luis Obispo, CA). Alternatively, for c-Abl real-time PCR, the cDNA samples were processed and evaluated following the TaqMan protocol, using hydrolysis probes for detection (Roche Ltd., Basel, Switzerland). ## C-Abl Cytoimmunological Staining in NIT-1 Cells NIT-1 cells were spun to the slides by cytospinning. The cells in the slides were fixed with 4% paraformaldehyde for 5 minutes and then blocked for 10 minutes at room temperature with 10% normal goat serum. The cells were spun onto the glides and stained with c-abl antibody for 1 hour at room temperature (1∶200, Cell Signaling Technology) followed by secondary antibody, goat anti rabbit AF555 for 1 hour at room temperature (1∶1000, Invitrogen Technology). All the washing steps were used with 1X TBS solution. Imaging was done using Zeiss Axioskop microscope. ## Mouse c-Abl siRNA Transfection on NIT-1 Cells NIT-1 cells were cultured in low glucose DMEM media as mentioned above. The c-Abl or control siRNA was incubated with NIT-1 cells following the protocol from the siRNA transfection kit (Santa Cruz Biotech, Inc., Santa Cruz, CA). Thereafter, the siRNA transfected NIT-1 cells were incubated for 6 hours in the presence of 16 mM glucose. Insulin gene expression in NIT-1 cells transfected with control siRNA and cultured in low glucose DMEM media served as an additional control. The supernatants were collected from the cultures and assayed for insulin expression by ELISA as aforementioned. Additionally, mRNA was extracted from c-Abl or control siRNA transfected NIT-1 cells, and c-Abl and insulin gene expression was evaluated using real-time RT-PCR as described above. ## Luciferase Assay In a set of experiments, cells of the 293 cell line (provided by Dr. Lijun Yang at University of Florida) were co-transfected with mouse Pdx-1 promoter-driven firefly luciferase reporter gene along with control plasmid (pCDNA3), *c-Abl* plasmid or *pdx-1* plasmid (self-regulated gene). Twenty-four hrs later, the luciferase activities in the above conditioned cells were measured using the Dual Luciferase Reporter Kit (Promega). To evaluate the gene specific changes in expression, firefly luciferase activity was normalized to that of *Renilla* luciferase, included in kit. ## Overexpression of c-Abl in β Cells by Retrovirus-*c-Abl* Transfection and Examination of Insulin, PDX-1, NKx6.1 and NKx2.2 Gene Expression The retrovirus-c-Abl and lentivirus-GFP plasmids were prepared according the method previously reported. Then, NIT-1 cells (β cell line) were transdued with retrovirus carrying the *c-Abl* gene (2×10⁶ transducing units (TU)/ml) or control vector (lentivirus-GFP) (2×10⁶ TU/ml). Forty- eight hours later, the transduced β cells were examined by real-time PCR for the expression of insulin-I and II, PDX-1, NKx2.2, and NKx6.1. The data were normalized to β actin and fold-change was calculated relative to NIT-1 cells transfected with Lentivirus GFP, validated using Opticon MJ software. ## Western blot Assay for NKx2.2 and PDX-1 Expression in NIT-1 Cells NIT-1 cells were cultured in low glucose DMEM or in DMEM with glucose 16 mM in the presence of Imatinib for 24 hrs. Thereafter, the cells were harvested and processed using Western blotting lysis buffer (Qiagen). The cell lysates were used to run SDS-PAGE gel (10%). The proteins were transferred to a nitrocellulose blotting membrane, which was then used for blotting by anti- NKx2.2 and PDX-1 antibodies (rabbit polyclonal antibody, Abcam, Cambridge, MA), respectively at a dilution of 1∶500. Horseradish peroxidase (HRP)-conjugated goat anti-rabbit secondary antibody was used at 1∶2000 diluation. The procedure of western blot was performed following the instruction from the manufacturer (Qiagen). The densitometric analysis of the target proteins relative to the level of β-actin or GAPDH was performed using a densitometry analysis program (ImageJ 1.47, NIH). ## Measurement of GLUT2 Transcription Levels Using Real-time RT-PCR NIT-1 cells were cultured in low glucose DMEM with or without 3 µM Imatinib for 6 hrs. Then the cells in each condition were harvested and the RNA was extracted using Qiagen RNeasy Mini Kit (Qiagen Inc., Valencia, CA). Real-time RT-PCR for GLUT2 gene expression was performed following the instruction from the manufacturer (Qiagen). The expression level of GLUT2 in Imatinib-treated NIT-1 cells was calculated relative to the level of GLUT2 in NIT-1 cells cultured in the absence of Imatinib. ## Western Blotting for GLUT2 Expressed on β Cells NIT-1 cells were cultured in low glucose DMEM alone or with 8 mM glucose, 16 mM glucose, 3 µM Imatinib, 8 mM glucose and 3 µM Imatinib, or 16 mM glucose and 3 µM Imatinib for 24 hrs. Thereafter, the cells in each culture condition were harvested and processed using Western blotting lysis buffer (Qiagen). The cell lysates were used to run SDS-PAGE gel (7.5%). The proteins on the gel were transferred to a nitrocellulose blotting membrane, which was then used for blotting by anti-GLUT2 antibodies (goat polyclonal antibody, LifeSpan BioScience) at a dilution of 1∶500. Rabbit anti-goat secondary antibody was used at 1∶2000 diluation. The densitometry of each GLUT2 band relative to the level of β-actin band of the corresponding sample was performed using a densitometry analysis program (ImageJ 1.47, NIH). ## Statistical Analysis The unpaired student *t* test was used for comparison of two independent samples. For the experiments with multiple groups, we employed one-way ANOVA with post hoc test, or two-way ANOVA followed by a Bonferroni post-test. The difference with p\<0.05 was considered to be significant. # Results ## Inhibition of Tyrosine Kinase c-Abl Synergizes with Glucose Stimulation in Inducing Insulin Expression Previous evidence showed that administration of the c-Abl tyrosine kinase inhibitor, Imatinib, reverses diabetes in new onset diabetic NOD mice, and improves type 2 diabetes. Although this diabetes-reversal effect of Imatinib has been suggested to be associated with its anti-inflammatory effect as well as improving insulin sensitivity, whether Imatinib has direct effects on β cells to promote insulin production is yet to be addressed. If β cells were shown to increase insulin levels, then Imatinib can be a candidate therapy for diabetes including type 1 and type 2 diabetes. To determine the role of Imatinib in regulating insulin expression, we treated mouse β cell line, NIT-1 cells, with high concentrations of glucose and with or without Imatinib to assess insulin expression. As expected, glucose significantly promoted insulin expression at both the protein and mRNA levels. Surprisingly, we discovered that Imatinib treatment along with glucose stimulation led to further increase in insulin expression by the β cells. In line with these results, independent experiments showed that Imatinib enhanced the capacity of β cells to secrete C-peptide, a more stable form derived from pro-insulin. Because Imatinib is a specific c-Abl inhibitor, the results suggest that c-Abl is a negative regulator for glucose- induced insulin production. To further validate the role of c-Abl in β cell insulin expression, we knocked down the *c-Abl* gene by transfection of *c-Abl* siRNA in NIT-1 β cells to study the influence of c-Abl on insulin production induced by glucose. In corroboration with the above findings, the knockdown of *c-Abl* significantly enhanced glucose-induced insulin production. ## C-Abl Plays an Important Role in Controlling Insulin Production in Resting β Cells It is known that β cell produces little insulin when insulin is not needed, which is when the β cells are exposed to the low levels of glucose. However, the mechanisms regulating insulin expression in the resting β cells are not well investigated. To determine whether c-Abl also plays a role in regulating insulin expression in resting β cells, we assessed the effect of c-Abl suppression on insulin production in resting β cells. Intriguingly, we found that inhibition of c-Abl by Imatinib dramatically enhanced insulin production of NIT-1 cells cultured in media with low glucose (4.5 mM) in a dose-dependent manner. To support these findings, we consistently observed that Imatinib treatment significantly increased insulin mRNA levels in resting NIT-1 cells. Furthermore, knockdown of *c-Abl* using siRNA led to approximately 70% reduction of c-Abl mRNA by real time RT-PCR (data not depicted), and enhanced insulin production by NIT-1 β cells. Furthermore, when *c-Abl* siRNA transfected NIT-1 cells were treated with Imatinib, insulin production could be further enhanced. The above findings strongly indicate that c-Abl is a negative regulator that controls insulin expression in β cells under resting state when insulin is not needed. ## Glucose Stimulation of NIT-1 β Cells Up-regulates Both c-Abl and Insulin Expression As shown above, the inhibition of c-Abl up-regulated insulin production induced by glucose, it is of interest to investigate how glucose stimulation influences c-Abl gene expression. We postulated that glucose stimulation would suppress c-Abl, the inhibitor of insulin gene expression so that insulin production could be enhanced. To our surprise, glucose stimulation significantly enhanced both c-Abl and insulin expression. This result implicates that glucose induces up- regulation of insulin and c-Abl probably via two different pathways, and it is likely that c-Abl up-regulated by glucose stimulation in turn acts on certain elements in the insulin production pathway to regulate and maintain insulin expression at proper levels. ## C-Abl Attenuates Insulin Expression without Involvement of PDX-1 PDX-1 is a major insulin transcription factor in β cells by binding to A boxes (AT-rich elements), upstream of the insulin gene to enhance gene expression. To determine whether increased insulin production in Imatinib-treated β cells is associated with PDX-1, we studied PDX-1 mRNA levels by real-time PCR in NIT-1 cells under different conditions. Surprisingly, we failed to observe any influence of Imatinib on PDX-1 transcriptional levels. To confirm that c-Abl inhibition by Imatinib does not affect PDX-1 expression, we performed experiments for the protein levels of PDX-1 expressed by NIT-1 cells treated by different concentrations of Imatinib using Western blot. Consistent with the results shown in, Imatinib did not affect PDX-1 protein expression. To further confirm that PDX-1 is not involved in c-Abl-regulated insulin gene expression, we then overexpressed c-Abl by transfecting a *pdx-1* promoter- driven luciferase reporter gene into 293 cell line along with control plasmid or *c-Abl* plasmid or positive *pdx-1* plasmid. As shown in, *c-Abl* transfection did not affect *pdx-1*-promoter-driven luciferase activity compared to control. While these data would not be able to rule out the regulatory effect of c-Abl on PDX-1 gene expression in real β cells because 293 cell is not β cell-derived cell line, our findings at least indicate that c-Abl does not directly interact with transcriptional regulatory elements of PDX-1 gene. Consistent with the previous report showing that PDX-1 serves as a positive regulator for the gene expression of its own, we found that transfection of *pdx-1* plasmid markedly induced PDX-1-driven luciferase activity. ## C-Abl Regulates Insulin Gene Expression in β Cells Likely via Regulating NKx2.2 To further establish the role of C-Abl in regulating β cell insulin gene expression, we used NIT-1 cells to create a β cell line that overexpresses c-Abl by transfection of a *c-Abl*-retrovirus, and then studied the changes of insulin gene expression and the expression of insulin-related genes. As shown in, we demonstrated that the overexpression of c-Abl in β cells significantly suppressed β cell insulin gene expression. In line with the data presented in, c-Abl overexpression did not affect PDX-1 gene expression and its downstream gene of NKx6.1. We observed, however, that another insulin gene expression regulatory factor, NKx2.2 was dramatically reduced in β cells with c-Abl overexpression. The NKx2.2 is a strong activator of NeuroD1 gene, which controls insulin gene expression in response to glucose stimulation. Mouse insulin has two forms (insulin I and II) expressed by two different insulin genes located in chromosome 19 and 7, respectively, so we examined the expression levels of both insulin genes. Consistent with the findings described above, we found that the gene expression levels of both insulin I and II were dramatically decreased in c-Abl overexpressed NIT-1 cells in contrast to control cDNA-transduced cells. The above data indicate that c-Abl suppresses insulin gene expression likely through NKx2.2 without the involvement of PDX-1. To support NKx2.2 is indeed involved in Imatinib treatment-induced insulin up-regulation, we further demonstrated that c-Abl inhibition by Imatinib markedly enhanced NKx2.2 protein levels in NIT-1 cells. Consistent with the above findings, we also found that Imatinib treatment markedly up-regulated gene expression of NeuroD1 (supplemental), which is the down-stream of, and positively regulated by NKx2.2. ## C-Abl Inhibition by Imatinib Promotes GLUT2 Expression on β Cells It is well known that glucose is a strong stimulator for β cells to produce insulin. Thus, the increase of glucose concentration in cytosol of β cells would stimulate insulin expression. To determine whether glucose intracellular transportation participates in the Imatinib-induced insulin expression, we treated NIT-1 cells with Imatinib and then GLUT2 mRNA and protein levels were evaluated by real-time RT-PCR and western blot, respectively. We found that c-Abl inhibition by Imatinib significantly enhanced GLUT2 expression at both protein and mRNA levels, suggesting that GLUT2 may participate in up-regulation of insulin induced by Imatinib. # Discussion Imatinib was first used in the treatment of chronic myelogenous leukemia (CML) to inhibit c-Abl tyrosine kinase which is constantly activated because of the fusion protein encoded by bcr-abl fusion gene due to chromosome 22 and 9 translocation. Later, it was also used for other diseases with enhanced c-Abl activity, such as GIST. Recently, it was evaluated in both type 1 and type 2 diabetes for its potential application in diabetes management because it was noted that CML patients with diabetes who took Imatinib to control CML also showed improvement in their diabetes. It has demonstrated that CML patients with type 2 diabetes have markedly elevated levels of adiponectin, implicating a mechanism for improved insulin sensitivity in the peripheral tissues. Research in streptozotocin-induced diabetes or type 1 diabetes mouse models has demonstrated that Imatinib prevents β cell apoptosis, and islet inflammation. Recently, Mokhtari et al reported that Imatinib improved survival of insulin- producing β cells via inducing phosphatidylinositol 3-kinase signaling. However, there has been no report on whether Imatinib directly acts on β cells to promote insulin production. It could be possible that Imatinib promotes β cell insulin secretion leading to amelioration of diabetes. Supporting this idea, clinical observations demonstrate that CML patients with type 2 diabetes have enhanced levels of serum C-peptide while taking c-Abl tyrosine inhibitor. It is also likely that Imatinib treatment improves diabetes through different mechanisms working together, e.g. providing β cell survival signals and at the same time promoting insulin production. In the present study, we report a novel mechanism for β cells to regulate their insulin expression. To our knowledge, this is the first to demonstrate that c-Abl tyrosine kinase, as a negative regulator, controls insulin expression in β cells. Tremendous volume of work has been done to elucidate the mechanisms that stimulate insulin production, however, the mechanisms that negatively regulate insulin expression, especially when β cells are under resting state, have drawn much less attention. In fact, positive and negative regulators for β cell insulin production are equally important and both have to be physiologically active and working in concert to control insulin production within a normal range. Through studying the effect of c-Abl tyrosine kinase inhibitor, Imatinib on β cell insulin production, we reveal that c-Abl is an important negative regulator for β cell insulin expression. This regulatory mechanism functions not only in glucose-stimulated β cells but also in β cells at resting state. The detailed mechanisms of how c-Abl negatively regulates insulin expression are still not clear. However, we have ruled out the possibility that c-Abl down- regulates the expression of PDX-1 which is the major factor promoting insulin production by showing that inhibition of c-Abl using Imatinib does not alter the levels of PDX-1 expression. To support this, we further show that c-Abl tyrosine kinase does not directly affects *pdx-1* promoter-driven luciferase reporter gene expression in a gene transfection cell model using the 293 cell line. To further determine what factors are being involved in down-regulation of insulin by c-Abl, we investigated NKx2.2, which regulates insulin-gene expression- associated transcriptional factor, neuroD1. We found that overexpression of c-Abl in β cells reduced gene expression for both insulin and NKx2.2 to similar degrees. In line with the finding that c-Abl activity had no relationship with PDX-1, we did not observe any change in the expression of PDX-1 associated transcription factor, NKx6.1. Our western blot results provide further support of c-Abl’s negative regulatory role for NKx2.2 because inhibition of c-Abl by Imatinib markedly enhances NKx2.2 levels. It is known that glucose is the most important stimulator for insulin secretion physiologically. Our data show that inhibition of c-Abl by Imatinib works in synergy with glucose stimulation to promote insulin production. It is of interest to know how glucose stimulation affects c-Abl expression. Given the data presented above, we postulated that glucose would suppress this insulin negative regulator to promote insulin production. Surprisingly, we observed that glucose stimulation dramatically enhanced the expression of c-Abl in β cells. This finding reveals an important regulatory mechanism for insulin secretion under high-level glucose conditions, where glucose not only activates insulin positive regulators to increase insulin production, but also induces negative regulators such as c-Abl to prevent overexpression of insulin, so that an appropriate level of insulin can be maintained. Thus, c-Abl may serve as an important gatekeeper for β cell insulin expression and secretion. The second level of regulating insulin expression by c-Abl may be associated with affecting glucose transporter GLUT2 on β cells. Indeed, we observed that c-Abl inhibition by Imatinib considerably promoted GLUT2 expression on β cells, which would enhance intracellular glucose transportation and then stimulated insulin expression. As shown in, we propose the following pathway that in response to high-level glucose, the glucose transporter GLUT-2 on β cells first transports glucose molecules inside the cell, whereupon glucose then activates insulin gene transcriptional regulators, such as PDX-1, NeuroD1 and MafA – to up-regulate insulin expression. One the other hand, glucose also activates c-Abl gene expression as negative feedback to control insulin gene expression and to limit levels of glucose transporter GLUT-2. The timing of glucose initiating these two opposing processes is still unclear. Both could be initiated simultaneously or sequentially. It is likely that glucose and c-Abl affect insulin expression independently but cooperatively. Another important point of this study is that c-Abl likely remains active to control insulin gene expression in resting β cells when insulin is not needed. The molecular interactions among the molecules associated with regulating insulin gene expression need to be further investigated, such as how NKx2.2 expression is regulated by c-Abl and how c-Abl inhibition by Imatinib up- regulates GLUT2 on β cells. Importantly, animal studies will also be needed to study how c-Abl inhibition by Imatinib affects β cell function in vivo. These issues as well as the results demonstrated in the current study are of importance in guiding the development of new therapeutic approaches for type 1 and type 2 diabetes. Given the discovery of cell permeable small-molecule c-Abl tyrosine kinase activator, insulinoma can be potentially managed by using c-Abl activator to inhibit insulin production. # Supporting Information We would like to thank Anna Chernatynskaya for her assistance in preparing NIT-1 cells. We thank Chao Liu, a current master student in a training program of statistics at University of Florida for his assistance in statistical data analysis. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: CQX. Performed the experiments: PZ SWL LY TX CX. Analyzed the data: CQX SWL PZ MC. Wrote the paper: PZ CQX.

# Introduction Tsps are a superfamily of integral membrane proteins of 20–30 kDa that were first identified in mammals as cell-specific antigens and later in insects, worms, sponges, fungi (but not in yeast) and plants. To date, 33 distinct Tsps have been found in humans, 37 in *Drosophila melanogaster*, and 20 in *Caenorhabditis elegans*. Members of the Tsp family derive their name from four transmembrane domains (TMs). They have cytoplasmic tails at the N- and C-termini, a small extracellular loop (EC1), a small intracellular loop (ICL), a large extracellular loop (EC2) containing a conserved Cys-Cys-Gly (CCG)-motif and two to six additional cysteines. Mainly localized in the plasma membrane, Tsps form complexes, so-called tetraspanin-enriched microdomains (TEMs), by interacting with a variety of proteins including other Tsps, integrins, growth factor receptors, intracellular signaling molecules and receptor tyrosine kinases. Most of these protein-protein interaction sites as well as most monoclonal antibody epitopes map to the extracellular loop EC2. Several lines of evidence indicate that Tsps, respectively Tsps as a part of TEMs, play roles in physiological processes such as cell differentiation, adhesion, motility, cell signaling and sperm-egg fusion as well as in pathophysiological processes, including cancer metastasis and infections caused by pathogenic organisms. Some of these functions have been linked to post-translational modifications of Tsps. Palmitoylation of cytoplasmic, juxtamembrane cysteines is thought to be required for initiating TEM formation and *N*-glycosylation has been shown to play important roles in protein interactions, cell adhesion and motility. Due to their plasma membrane location with key interacting domains on the extracellular side, Tsps represent tumor markers and potential drug targets that could be addressed by monoclonal antibodies as well as by small molecules. Nearly all types of cells and tissue contain multiple Tsps, often expressed at 30,000–100,000 copies per cell. Despite their large number and implication in a broad spectrum of important cellular activities, only a relatively small number of Tsps have been studied in detail. The major obstacles to understanding Tsps are their subtle effects and functional redundancy. Recently, completed genome projects revealed that *Tsp* genes are also found in a few protozoan amoebae such as *D*. *discoideum*, *Trypanosoma brucei* and *Entamoeba histolytica*, albeit with only a few genes per cell type. Especially *D*. *discoideum* is a well-established model for studying the cellular role of proteins that have human orthologues. *D*. *discoideum* normally lives as a free amoeba but when starved, the cells aggregate to form a multicellular fruiting body. Therefore, this organism provides the opportunity to unravel basic Tsp functions in both, unicellular and multicellular contexts. Yet, surprisingly, physiological and functional data on Tsps in *D*. *discoideum* are completely absent. In this study, we set out to characterize the predicted Tsps from *D*. *discoideum*. RNA from all five genes was detected in the multicellular slug stage, whereas only three *Tsps* were found to be expressed in vegetative cells. We raised specific antibodies directed against the vegetative Tsps and could detect TspA and TspC by Western blotting. We show by fluorescence microscopy, that the vegetative Tsps co-localize with the V-H⁺ ATPase on contractile vacuoles (CVs). At least for one gene, namely TspC, a gene disruption sensitizes cells for osmotic stress, most likely by delaying the exocytosis of CVs. # Materials and Methods *D*. *discoideum cell culture*, *growth and development assays*. *D*. *discoideum* AX2 cells were cultured axenically at 22°C. For growth assays, amoebae in mid-log phase were diluted to an OD₆₀₀ of \~ 0.1 in HL5 medium (Formedium) with 0.5% glucose and agitated at 22°C with 150 rpm. The cell density was monitored by photometric measurements for 96 h. The doubling time was calculated from four independent cultures. To induce development, cells were washed twice in ice-cold Sörensen phosphate buffer (SPB, 15 mM KH₂PO₄, 2 mM Na₂HPO₄, pH 6.0) and 5 × 10⁶ cells were spread on 1% KK2 agar plates (20 mM KH₂PO₄/K₂HPO₄, pH 6.8, with 1% agar). Plates were incubated in a moist-chamber at 22°C. Pictures were taken in 4-h intervals for 24 h. *Preparation of cDNA and PCRs*. For cDNA preparation, 10⁷ vegetative cells in late-logarithmic phase were harvested (1000 × *g*, 4°C, 5 min) and lysed in 1 ml of TRIzol (Invitrogen). For the preparation of slug lysates, development was initiated, slugs were harvested after 16 h and the pellet resuspended in 1 ml of TRIzol. Total RNA was isolated according to the TRIzol protocol. The RNA was isopropanol-precipitated, washed with ethanol 70%, and used for cDNA synthesis (First strand cDNA synthesis kit, Fermentas) using (dT)₁₈ primers. PCRs were performed with One Taq DNA polymerase (New England Biolabs) and sequence-specific primers. For details on the PCR primers used see (primers 1–10). The PCR programm was as follows: 95°C for 5 min, 30 cyclces of 95°C for 30 s, 55°C for 40 s, and 68°C for 2 min, and 68°C for 10 min. *Cloning of TspA*, *TspC and TspD and site-directed mutagenesis*. The coding sequences of *TspA*, *TspC* and *TspD* were amplified by PCR from *D*. *discoideum* amoeba cDNA. We used PCR Primers 1–10. For generation of TspA-His and TspC-His we complemented the respective reverse primer with a 6 bp linker and a sequence coding for a 6x His-tag. The PCR products were ligated into pBluescript II SK(−) for sequencing. DNA point mutations were introduced according to the QuickChange protocol (Stratagene) using Pfu Turbo DNA Polymerase AD (Stratagene). For details on primer sequences see (primers 11–14). The PCR programm was as follows: 95°C 30 s, 16 cycles of 95°C 30 s, 55°C for 1 min, and 68°C for 8 min, and 68°C for 20 min. For the generation of GFP fusion constructs, the *Tsp* genes were ligated into pTX-GFP (N-terminal GFP,) using *Bam*H I/ *Xho* I or, after removal of the *Tsp* stop codon, into pDM323 (C-terminal GFP,) using *Bgl* II/*Spe* I. *Western blot and glycosylation analyses*. *D*. *discoideum* amoebae were harvested in mid-log phase, resuspended in 1 ml of water and lysed by four freeze-thaw cycles at − 80°C and 37°C. Proteins from *D*. *discoideum* cells (30 μg per lane) were separated by SDS-PAGE and transferred to PVDF membranes (Macherey & Nagel). The membranes were incubated with primary antibody. For protein detection we used a commercial polyclonal anti-GFP (Santa Cruz Biotechnology, 1:10,000) and a monoclonal anti-penta-His antibody (Qiagen, 1:2,000) or custom made affinity-purified polyclonal antisera directed against EC2 derived peptides (BioGenes, Berlin, Germany, 1:200). The Western blots were developed with horseradish peroxidase conjugated goat anti-rabbit (Jackson Immuno Research Laboratories, 1:5,000), respectively with conjugated goat anti- mouse (Jackson Immuno Research Laboratories, 1:2,000) antisera using ECL (Amersham Biosciences). For enzymatic deglycosylation analysis, a membrane protein fraction was collected by centrifugation of debris-free amoebae lysates (100,000 × *g*, 4°C, 45 min). *N*-glycosylation of TspA and TspC was assayed by incubation with 1,000 units of *N*-glycosidase F (New England Biolabs) for 60 min at 37°C. *Expression and localization of Tsp-GFP fusion constructs*. 7 × 10⁶ *D*. *discoideum* amoebae were harvested (2,000 × *g*, 10 min, 4°C), washed twice with H50 buffer (20 mM HEPES, 50 mM KCl, 10 mM NaCl, 1 mM MgSO₄, 5 mM NaHCO₃ and 1 mM NaH₂PO₄, pH 7.0) and incubated for 5 min in 700 μl of H50 with 4–10 μg of plasmid DNA. Electroporation was done in 0.4 mm electroporation cuvettes with two pulses (5 s delay) of 1.2 kV, 50 μF and a time constant of 1–2 ms using a Gene Pulser II (Bio-Rad). After addition of axenic HL5 medium (Formedium) with 0.5% glucose, cells were incubated at 22°C for 24 h before adding 10 μg ml^-1 G418 for antibiotic selection. Fluorescence imaging was done by confocal microscopy (Leica TCS-SP, 100x NA1.4 objective). Tsp-GFP expressing amoebae were fixed with 2% paraformaldehyde and 15% picric acid in 20 mM PIPES buffer pH 6.0 (Sigma). Cells were washed with PIPES and Phosphate-buffered saline (PBS, 137 mM NaCl, 2.7 mM KCl, 8.1 mM Na₂HPO₄ and 1.5 mM KH₂PO₄, pH 7.4) with 100 mM glycine and incubated for 10 min in 70% ethanol. CVs were stained as described in with a monoclonal antibody 221-35-2 directed against a subunit of the vacuolar-H⁺ ATPase, or protein disulfide isomerase 221-135-1 (undiluted culture supernatant) and with a second polyclonal goat-anti mouse Cy3 antibody (Dianova; 1:1,000). GFP fluorescence was excited with an argon ion laser at 488 nm and Cy3 fluorescence with a He/Ne laser at 543 nm. *Generation of tspC*^*−* *and tspD*^*−* *strains*. Gene deletion was done as described previously, using electroporation with a blasticidin S resistance cassette flanked by two homologous fragments of the *TspC* or *TspD* gene for homologous recombination. Briefly, 5´ and 3´ fragments were amplified from AX2 genomic DNA using specific forward and reverse primers (for details on the PCR primers see, primers 15–22). The PCR programm was as follows: 95°C for 5 min, 30 cycles of 95°C for 30 s, 53–56°C for 40 s, and 68°C for 2 min, and 68°C for 10 min. The PCR products were cloned into pLPBLP via the *Sal* I/*Hin*D III (5´ fragment) and *Pst* I/*Bam*H I sites (3´ fragment). The construct was excised with *Sal* I/*Bam*H I, purified and electroporated into AX2 cells. Transformants were selected for seven days in HL5 containing 10 μg ml^-1 blasticidin S. Individual colonies were picked with a pipette, transferred into 24-well plates and screened by PCR. For details on the primers used see (primers 23–34). The PCR programm was as follows: 95°C for 5 min, 30 cylces of 95°C for 30 s, 50–54°C for 40 s, and 68°C for 2 min, and 68°C for 10 min. *Adhesion and random motility assays*. Cell-substrate adhesion was performed as described by Garcia *et al*. with slight modifications. A suspension of 1 × 10⁶ cells in 1 ml of SPB was placed in a 125 ml glass culture flask and put on a gyratory shaker for 10 min at 120 rpm. The cells were allowed to adhere to the glass substrate for 2 h. The flask was then gently agitated for 3 min at 60 rpm. The supernatant was collected and the number of cells was counted with a hemocytometer. Random motility was monitored by time-lapse imaging of 7.9 × 10⁴ growth-phase cells per 24-well (Sarstedt, \#83.1836) in SPB using an ImageXpressMicro High Content Screening System (Molecular Devices). In each experiment, the x/y positions of 50–100 cells were determined in 15 s intervals for 15 min. Cell motility was analyzed using ImageJ Manual Tracking and Chemotaxis software. Experiments were performed with two independent knockout clones for each *Tsp*. *Osmoregulation assays*. Exponentially growing cells were placed in a well of a 96 well plate (Greiner bio-one, \#655090) and left to adhere for at least 30 min. Pictures were taken with an ImageXpressMicro High Content Screening System (Molecular Devices) with a 100x objective in HL5 medium with 0.5% glucose (0 min). After exchanging the media with distilled water, images were captured at 5, 10, 20, 30 and 40 min. Circularity was measured offline for each cell with ImageJ. Each data point is the mean of 75–116 cells. Error bars represent S.E.M.. Experiments were performed with two independent knockout clones for each Tsp. Chambers for *in vivo* microscopy of contractile vacuoles were made from four fused silicon rings (Flexiperm, Sarstedt) pressed on a 50 x 50 mm square custom-made glass coverslip (Hecht) and maintained in a moist chamber. Cells from shaking culture were seeded into the wells and allowed to adhere for at least 60 min in growth medium. After the medium was withdrawn, cells were rinsed once with SPB and immediately imaged for up to 30 min in fresh SPB containing 1 μg ml^-1 of the styryl dye FM 4–64 (Invitrogen) using argon ion laser excitation (488 nm) and recording emission in a window from 600 to 800 nm on a Leica TCS-SP confocal microscope. # Results ## Cloning and sequence analysis of *D*. *discoideum* Tsps BLAST searches in the genome sequence of *D*. *discoideum* yielded five novel putative *Tsp* genes that had been annotated by dictybase as *TspA-E* ([dictybase.org](http://dictybase.org)). *TspA-D* are located on chromosome 1, *TspE* on chromosome 5. *TspA-E* consist of \~700 bp with an A + T content of \~ 70%. The flanking non-coding regions have an A + T content of more than 85%, which is typical for intergenic regions in the *D*. *discoideum* genome. The genes of *TspA-E* have one to three introns. We were able to amplify the ORFs of *TspA*, *TspC* and *TspD* from cDNA of *D*. *discoideum* AX2 vegetative amoebae. We also managed to amplify the ORFs of all five *Tsps* from AX2 slugs, indicating that the genes are transcribed and do not represent pseudogenes. All PCR products had the predicted size of approximately 700 bp. Expressed sequence tag data (EST) and expression time courses from dictyExpress ([dictybase.org](http://dictybase.org)) provide further support for the gene transcription pattern we describe here. The three Tsps from vegetative amoebae, TspA, TspC and TspD, were used for further studies. The DNA sequences obtained experimentally were identical to the database entry. A topology prediction on the basis of a protein sequence alignment shows that *D*. *discoideum* TspA-E contain four TMs, a short extracellular loop EC1 (20–30 aa), a very short intracellular loop (ICL, 4 aa) and a large extracellular loop EC2 (76–139 aa), flanked by relative short N- (14–19 aa) and C-terminal (9–18 aa) tails. The EC2 is divided into a constant region (light yellow shading) containing α-helices A, B and E, that is important for homodimerization and a variable region (blue shading), that represents putative protein-protein interaction sites. TspE contains the “CCG”-motif present in all human Tsp EC2s, while the other four Tsps show a modified “CCK/Y/C” motif, which is more similar to some variants present in the plant lineage. *Dictyostelium* Tsps contain the typical four cysteines (yellow circles) that form intramolecular disulfide bonds crucial for correct folding and protein-protein interaction of metazoan EC2s. We also identified multiple consensus sites for secondary protein modifications of *Dictyostelium* Tsps: two potential palmitoylation sites adjacent to TM2 and TM4 (pink wavy lines) and potential *N*-glycosylation sites (black pins) in EC1 and EC2. Most amino acids indicated as important by structural and mutational analysis of the mammalian proteins are also present in *Dictyostelium* Tsps. Specifically, this includes conserved polar (Asn/Glu/Gln) and Gly residues in the TMs (indicated in italic letters). In TM3, *Dictyostelium* Tsps have a Leu/Gly where human Tsps have conserved Glu/Gln. However, mutation of Glu103 in TM3 of human CD9 to Ala had no effect on cell surface expression of the protein and its dimerization, hinting at a tolerable amino acid exchange. ## Expression and glycosylation of Tsps from *Dictyostelium* amoebae To test for *in vivo* expression by Western blot analysis, we raised antibodies against TspA, TspC and TspD. We used peptide sequences derived from EC2 for antibody production, as the respective N- and C-terminal tails are too short and most useful Tsp antibodies have been generated against the EC2. To test the three affinity-purified antibodies, we generated Tsp proteins fused to green fluorescent protein (GFP) in *D*. *discoideum* vegetative cells. We detected two bands in our Western blots for each Tsp-GFP fusion protein using a polyclonal α-GFP antibody. We found an 80 kDa band, representing an unspecific signal (\*) that was also detected in cells transformed with a control plasmid as well as a specific band of 70 kDa. This is a higher mass than the theoretically expected 52 kDa, hinting at potential post-translational modifications, a common feature of Tsps. We used the same cell lysates to test the three affinity-purified antibodies. The TspD antibody depicted no signals. The affinity purified TspA antibody detected the 70 kDa band of TspA-GFP and three additional bands of 22, 40 and 44 kDa that might represent endogenous TspA. For example, the molecular weight of human CD63 has been observed to be 32, 35 and 50 kDa with *N*-linked glycosylation in Western blot experiments, although the predicted molecular weight of CD63 is 25 kDa. We addressed the specifity of the TspA antibody by preincubating it with the immunization peptide and found that all bands disappear (preblock). Transformation of *D*. *discoideum* vegetative cells with TspA-His lead to an increase in the intensity of the 40 kDa band compared to the signal from untransfected cells. The same band was also detected by an α-His antibody (right). These results strongly suggest that at least the 40 kDa signal represents endogenous TspA. The α-TspC antibody detected the TspC-GFP fusion protein of 70 kDa but, surprisingly, not the endogenous TspC. A weak signal at 40 kDa was only found in TspC-GFP expressing cells but not in wildtype cells, most likely representing a degradation product. We saw no difference in antibody recognition for N- and C-terminal GFP fusion. As we found two potential *N*-glycosylation motifs in the TspC-EC2 (Asn143 and Asn164) that lie in close proximity to the antibody recognition site Cys171-Tyr185, we figured that the difference in antibody recognition may be due to a difference in glycosylation of endogenous and GFP-TspC. To test this hypothesis and also check for *in vivo* glycosylation of TspA, we used an enzymatic approach. Incubation with *N*-glycosidase F (PNGase) shifted the molecular weight of the 40 kDa TspA band and the 70 kDa GFP-TspC band, hinting at *N*-glycosylation of both Tsps. For TspC we proceeded using a mutational approach. We expressed GFP-TspC mutants that lack one (N143A) or both (N143A, N164A) asparagine residues and found a stepwise molecular weight shift of the 70 kDa band, indicating that both positions are glycosylated *in vivo*. Even lane loading is seen by the unspecific signals (\*) throughout all lanes (left). Using the α-TspC antibody we also obtained a weak signal at 40 kDa for the GFP- TspC degradation product that gained intensity and shifted to lower molecular weight in the respective glycosylation-deficient mutants (arrow). This observation may hint at the fact that glycosylation in the neighborhood of the epitope interferes with antibody recognition. ## TspA, TspC and TspD co-localizate with V-H⁺ ATPase Next, we set out to characterize the subcellular localization of the Tsps from *D*. *discoideum* vegetative amoebae. Our affinity-purified antibodies did not recognize endogenous TspC and TspD and produced several bands in case of TspA. This rendered our antisera unsuitable for immunolocalization. As an alternative, we used N- and C-terminal GFP fusion constructs for the experiments. Unlike the plasma membrane localization typically seen for mammalian Tsps, we found a mainly intracellular localization of the GFP fusion proteins in *D*. *discoideum* amoebae. Close to the surface of the cell that adheres to the coverslip, tagged Tsps accumulated on large vacuoles and reticular structures that radiate from them. This network was distinct from the endoplasmic reticulum (ER), because it did not coincide with the marker protein disulfide isomerase (PDI), but instead strongly overlapped with the vacuolar proton pump (PP, to), a well-known marker of the contractile vacuole (CV) system. Because N-terminally ( to) and C-terminally ( to) GFP-fused Tsp-proteins showed the same localization pattern, we are confident that all the endogenous proteins represent true constituents of the CVs of *D*. *discoideum*. Farther away from the adherent suface, Tsps were seen in patches within the plasma membrane, indicative of exocytosed CVs, and in clouds of vesicles in the vicinity of the cell nuclei most likely representing the Golgi apparatus, where the Tsps are thought to receive their glycosylation. ## *TspC*^*−* mutant cells have defects in osmoregulation Because the localization of GFP-tagged Tsps to the CV suggests that they might play a role in the function of this organelle, we generated *D*. *discoideum* strains by disrupting the respective genes. In case of *TspA*, we were not able to generate viable knockout cells from several independent transformations. At present, we cannot say if this is due to an unsuitable construct, or if a lack of *TspA* affects viability of *D*. *discoideum* amoebae. In contrast, *tspC*^*−*and *tspD*^*−* mutants were easily obtained. The genomic integration of the resistance cassette was validated by PCRs. We performed each of the following experiments with two independent knockout clones, which provided identical results. The *tspC*^*−* and *tspD*^*−* cells were morphologically intact and grew normally with doubling times of about 10 h in axenic medium. Both mutants maintained the ability to complete a full development cycle within 24 h and to generate viable spores (24 h). To quantify the motile properties of wildtype and *tspC*^*−* or *tspD*^*−* *D*. *discoideum* amoebae, we tracked random motility of individual cells in low osmolarity phosphate buffer using time-lapse video recording. The average speed of knockout and wildtype cells was comparable at 7.0 μm min^-1, respectively 6.8 μm min^-1 for *tspC*^*−*and 7.2 μm min^-1 for *tspD*^*−*. Chemotaxis towards cAMP seemed equally unaltered, as deduced from developmental assays where *tspC*^*−* *and tspD*^*−* cells exhibited normal streaming behavior (Figs and). To test for adhesion defects of the deletion strains, wildtype and mutant cells were allowed to adhere to a glass surface and then gently agitated. The non- adhered cells in the supernatant were counted and subtracted from the total cell number as a measure of cell-substrate adhesion. *TspC*^*−* *and tspD*^*−* cells did not differ from wildtype cells in their ability to adhere to the substratum (89%). Finally, we tested for osmoregulation defects. Wildtype *D*. *discoideum* cells can cope with severe hypo-osmotic stress by rapidly expelling water through the contraction of vacuoles. Wildtype as well as *tspC*^*−* and *tspD*^*−* cells looked healthy and were amoeboid shaped in standard HL5 media (0 min). Then, we challenged the cells by suspending them in distilled water. After initial swelling (10 min), wildtype and *tspD*^*−* cells regained their normal shape after 20 min, whereas *tspC*^*−* cells were affected more strongly at 10 min and remained round for significantly longer periods of time. Representative images of the cells for the crucial time points can be seen in. However, the cells show a tendency for full recovery after 40 min. For statistical analysis we performed one-way ANOVA with a Bonferroni post-hoc test and found particularly highly significant differences for *tspC*^*−* and wildtype cells at 20 and 30 min (\*\* = *P* = 0.001) and significant differences for 40 min (\* = *P* \< 0.05). To illustrate the *tspC*^*−* phenotype, we calculated Δcircularity between 20 and 10 min as a quantitative parameter for initial cell recovery. (right) shows that the initial recovery rate of wildtype (black bar) and *tspD*^*−* (white bar) is comparable whereas the value for *tspC*^*−* (grey bar) is 2.5x smaller. In order to analyse the functionality of the contractile vacuoles, living cells were stained with the styryl-dye FM4-64 and followed by confocal microscopy. The dye labelled the plasma membrane immediately and CVs of similar size accumulated in both, wildtype cells and *tspC*^*−* mutants, indicating that their filling efficiency is similar. For quantification, we counted the CVs per cell. Over half an hour, the number of CVs in the mutant (grey bars) increased at almost double the rate of wildtype cells (black bars). In the time-window from 21 to 30 min, where the circularity of cells diverged most clearly, *tspC*^*−* cells had significantly more CVs than the wildtype (\* = *P* \< 0.05, evaluated with two-tailed student t-test), suggesting that empying of their lumen was less efficient. To test whether TspC can rescue the *tspC*^*−* knockout phenotype, we recombinantly expressed TspC-His (*tspC*^*−*::TspC-His) in *tspC*^*−* cells (*tspC*^*−*). Both cell types looked healthy and were amoeboidal-shaped in HL5 media (0 min). Suspending the cells in distilled water lead to the anticipated initial swelling (10 min). Unlike *tspC*^*−* cells, *tspC*^*−*::TspC-His were able to recover their normal shape after 20 min (student t-test for 20 min \*\* = *P* \< 0.001). *TspC*^*−*cells did not even fully recover after 30 min (\* = *P* \< 0.05). Two-tailed student t-tests showed no significant difference in circularity for *tspC*^*−* and *tspC*^*−*::TspC-His after 40 min, reflecting the tendency we saw in the initial osmoregulation experiment. However, as we also found a not significant but slight increase in circularity for *tspD*^*−* cells after 40 min (*P* = 0.282, evaluated with one-way ANOVA with a Bonferroni post-hoc test), we can not exclude other cell effects that contribute to cell shape at this time point. Calculation of the initial recovery rate shows a 7x increased ability of *tspC*^*−*::TspC-His to recover after 20 min. This high rate might be explained by overexpression of TspC. Altogether, these data suggest that TspC plays a role mainly in the initial phases of osmoregulation (10 and 20 min) of *D*. *discoideum* vegetative cells, which is consistent with TspC localization in CVs. However, we currently do not now whether this is really a TspC isoform-specific effect or if each tetraspanin is potentially capable of supporting CV function, if only expressed at a high- enough level. # Discussion In most organisms, Tsps usually localize to the plasma membrane where they self- organize in TEMs. Besides, some Tsps are also found in the endosomal system, mainly within late endosomes, lysosmes and in lysosome-related organelles such as α-granules in platelets, melanosomes in melanocytes, cytotoxic granules in T-cells, Weibel–Palade bodies in endothelial cells and Major Histocompatibility Complex II-compartments in dendritic cells. In this study, we show that the *D*. *discoideum* vegetative amoebae *TspA*, *TspC* and *TspD* are absent from the plasma membrane and rather co-localize with vacuolar proton pump on CVs. Importantly both, N- and C-terminally GFP tagged Tsps, show the same co- localization with the V-H⁺ ATPase, excluding interference of the GFP with potential N- or C- terminal signaling sequences. In this study, we used endogenous *D*. *discoideum* Tsp sequences for overexpression, ensuring that natural partner proteins as well as post-translational modifications are available. However, it would be interesting to see whether any of the mammalian plasma membrane Tsps would follow a correct trafficking itinerary to the *Dictyostelium* plasma membrane. The phenotype we found for *tspC*^*−* is highly consistent with an intracellular localization in the CV, because the *tspC*^*−* mutant displayed a clear defect in osmoregulation. The osmoregulatory organelle in *Dictyostelium* and other free-living amoebae, such as *Trypanosoma*, is the CV network. It consists of tubes and bladders. Under hypo-osmotic conditions, as water enters into the cytoplasm, H₃O⁺ and, most likely driven by antiport HCO₃⁻ ions are pumped into the CV lumen. Following the osmotic gradient, water streams into the CV, presumably via aquaporins as we have shown previously. When the CV reaches its maximal diameter of 2–4 μm, it discharges its content through a pore at the plasma membrane in a giant kiss-and run process and quite a number of proteins have been shown to contribute to this intricate mechanism. The precise step in this process, to which TspC contributes, is unknown. However, our data suggest an involvement of TspC in the emptying-phase of CVs, which appear to be filled normally before and also reach a size similar to wildtype CVs. Recently it was shown that CVs display functions in addition to coping with hypo-osmotic stress. For instance, CVs facilitate the transport of DdCAD-1, a Ca²⁺ dependent cell-cell adhesion molecule, to the cell surface. DdCAD-1 is synthesized as a soluble protein in the cytoplasm and is then transported to the plasma membrane via CVs. From there exocytosis ensues and the protein binds back to the extracellular face of the plasma membrane. Whether one of the Tsp proteins plays a role in this pathway remains to be established. It was also shown that CVs in *Dictyostelium* represent a highly efficient acidic Ca²⁺-store that is required for cAMP induced Ca²⁺ influx. Merely judging from the expression pattern , TspB could be a candidate protein involved in this process. Clearly, more experimental work is necessary to determine the roles that Tsps play in CVs of *D*. *discoideum*. In this context, TspA should be taken into account. Our Western blots as well as expression time courses from dictyExpress and hint at high expression levels of *TspA* in *D*. *discoideum* amoebae. Until now, we were not able to delete the *TspA* gene by homologous recombination. Further experiments are necessary to address whether the apparent lethality of *tspA*^*−* cells can be explained by a requirement for TspA in CV function. Studying phenotypes generated by downregulation of TspA, either by gene knockout or by knockdown via RNAi or by using inducible promotors will be our next attempts to characterize the role of TspA in *D*. *discoideum* vegetative physiology. Taken together, the characterization and localization of Tsps in *D*. *discoideum* amoebae fills a gap in *Dictyostelium* research, and gives an excellent starting point for further investigations on the functions and protein interactions of Tsps. Although the role of TspC in the CV of *Dictyostelium* may not appear relevant for mammalian cell research, it should be investigated whether similar processes operate in the CVs of pathogenic *Trypanosomes*. In the end, a highly specialized Tsp may one day constitute a valuable drug target for treating sleeping sickness, one of the most threatening tropical diseases. # Supporting Information We thank Dina Quitzau, Manuela Kramp, Björn Henke, Heike Otto and Heide Gärtner for technical support, and the Stock center at dictybase for providing the pDM and pTX plasmids. [^1]: The authors have declared that no competing interests exist. [^2]: **Conceptualization:** JvB. **Data curation:** JvB TA MM. **Formal analysis:** JvB MM. **Investigation:** TA JvB MM. **Methodology:** JvB TA MM EB. **Project administration:** JvB. **Resources:** EB. **Supervision:** JvB. **Validation:** TA JvB MM EB. **Visualization:** TA MM JvB. **Writing – original draft:** JvB MM TA. **Writing – review & editing:** MM TA EB.

# Introduction What is the semantic structure of free-flowing thought? How do meanings come up in our thoughts, and how are they linked over time? Human experience is filled with this structure, while we stand in line for our groceries, wait in line at the bus station or even in a moment of mind wandering while conversing with a friend. In this paper, we explore recent events surrounding COVID as a domain to tap into this semantic structure. We devised a “stream-of-consciousness” task in which participants imagined the future beyond COVID, and quantified how their free-flowing thought varied as a function of how they were prompted before this writing. We find that if participants are prompted with the present COVID situation vs. the pre-COVID times, their structure of thought changes. Our exploratory analyses suggest language from the future-oriented responses reflects its temporal priming, i.e., the pre-pandemic vs. during the pandemic prompt that came before. These results offer hints at the semantic patterns that characterize these self-reflections, and how context is central to the forms they take. We end by arguing that a generalized notion of “self-communication” may organize phenomena such as these in intriguing ways. # Background When the COVID-19 global pandemic spread rapidly in late 2019 and early 2020, major life impacts reverberated globally. These effects were felt across many aspects of everyday life, from direct health impacts to more indirect effects on the economy and social life. People began referring to life before the pandemic colloquially as the “before times,” and there was a sense of a new normal. These effects were also significant in the lives of younger individuals, such as students, with virtual schooling, diminished social interaction, and limited hands-on learning. Evidence suggests an alarming impact of the pandemic on the mental health of college students, including increased stress moderated by self- regulation efficacy, increased depression and anxiety, and negative changes to student relationships. The effect of perceived threat of COVID-19 on mental well-being appears to be mediated by future anxiety as well, showing the potential to impact mental health in the long run as decisions about the future may be impacted. The pandemic thus provides a unique opportunity to study how major events influence perceptions of life and mental health, and their relationship to other dimensions. To study these perceptions, we investigate how student participants construct a narrative text about their lives. Our approach is inspired by methods used in essay writing and journaling, self-talk, and think aloud. These domains suggest that when we speak to ourselves, ruminate, or reflect on aspects of our lives, our linguistic styles and strategies may be a signature for underlying mental or emotional states and processes. Intriguingly, such intrapersonal communication has been frequently the topic of discussion, yet remains largely an understudied construct. Self-communication occurs when both sender and receiver of a communicative instance are contained within a single individual, such as in dialogical self-talk, and can include transcending across time and space. Here we use this process as a source of data about these life perceptions. While not all aspects of intrapersonal communication may be easily accessible for study, such as the seemingly endless streams-of-consciousness we engage in every day, various methods have been employed to tap into self-talk through writing or speech. Raffaelli et al. used a think aloud paradigm in which individuals are instructed to speak aloud their thoughts. Negative valence in the words used correlated with a narrowing of conceptual scope, such that thoughts became more semantically similar when they were more negative. Social context may play a role, too. Oliver et al. used a think aloud task to study mental health outcomes. Changing social context to more supportive environments led to greater use of positive emotion words, fewer negative emotion words, fewer swear words, and fewer first-person references compared to a control condition that lacked emotional recognition or meaningful rationale. Recent work has shown that self-talk also links to mental health outcomes during COVID-19. In a questionnaire study conducted on an Iranian sample, there were significant relationships found between self-talk, death anxiety, obsessive-compulsive disorder, and coping strategies in relation to the pandemic. As noted above, the onset of COVID for some may present a distinct point in time at which everyday life changed. This temporal effect of COVID may alter the way we contextualize and think about events before, during, and after this distinct transition. Changing the temporal framing of one’s thoughts or self-talk may have an influence on the language that we use. For example, construal-level theory hypothesizes that increased psychological distances are linked to increased abstractness of hypotheticals. The further away something is in time, space, or relatability (e.g., feeling similar to or different from an individual), the further the perceived psychological distance and the less concrete related thoughts become. Similarly, increasing concreteness, such as through writing about a given event, may decrease psychological distance to that event in time. High-level construals, or higher abstractions when mentally representing objects or concepts, may better facilitate self-control, such as attenuating the impact of future discounting on economic decision making. Following the September 11, 2001 attacks on the United States, entries in an online journaling platform used increasingly psychologically distant language in their daily writings, suggesting major events can impact our relationships with time. However, prior work suggests that writing about emotional experiences such as trauma may provide benefits for both mental and physical health. Considering these findings, one might expect that writing about “the before times,” pre-pandemic, would influence how one perceives their future, perhaps with increased abstraction leading to greater possibilities and allowing distance from such a troubling and disruptive event. Relatedly, a focus on life during the pandemic may have negative impacts on one’s perceptions of their future after the pandemic, with lower levels of construal leading to ruminative tendencies. To test this idea, we collect and analyze a text-based “stream-of-consciousness” dataset. Participants carried out this open-ended response task online, typing in their thoughts about COVID-19 life under different temporal conditions. The task was designed to elicit a naturalistic and uninterrupted flow of thought. Participants were first told to consider and write about life either before or during the pandemic. This prompt (before vs. during) served as a frame for a subsequent writing prompt, where participants were instructed to share their thoughts about a post-pandemic life. This future-oriented prompt is the same for all participants and is the focus of our analysis, and participants only differed in which writing prompt preceded this one (before vs. during the pandemic). This open-ended writing task generates a large and rich dataset of text. We thus took a preliminary, exploratory approach to investigate the influence of this temporal framing on their responses. In the *Analyses* section, we consider prior research that frames some factors guiding our exploratory analysis, and we introduce the ways in which these texts can be measured and analyzed. # Methods Data collection was conducted during two separate college quarters: in the first quarter (fall, 2021), classes were hybrid (both in-person and online), with students returning to campus for the first time since the pandemic began. This research was approved by the UCLA North General Institutional Review Board. 134 undergraduate students (female = 95, male = 38, other = 0) contributed data to this first phase of sampling. The second phase of sampling occurred the following quarter (winter, 2022), which had returned to online-only for the first four weeks due to the rapid spread of a particularly contagious strain of the virus, labeled “omicron.” In this phase, 91 undergraduate students (female = 70, male = 19, other = 2) contributed data. The students completed the study online for course credit in an introductory communication course. The goal of the study was to collect a rich dataset for exploratory analyses, and several aspects of the data were not included for analysis. Because of the pandemic- related constraints participants encountered at the start of the winter quarter, we first use this second phase dataset for our main analyses. We then use the fall dataset for exploratory comparison. The experiment was built using jsPsych in conjunction with [https://cognition.run](https://cognition.run/) to store the data. First, participants encountered a consent page, then click to continue only if they agree to consent to participate in the study. After the initial consent page, participants selected on a slider where in the COVID-19 timeline they considered the current moment to be. For the main portion of the study, participants wrote in a stream-of-consciousness style manner for ten minutes per prompt, responding to three total writing prompts. The first prompt asked participants to write either as if it is before the pandemic, or as if during the pandemic. Following the initial prompt, participants responded to a similar prompt asking to write as though it is after the pandemic. The final prompt included whichever temporal framing was not responded to in the first prompt. This resulted in two possible conditions: before-after-during the pandemic, or during-after-before the pandemic. Our focus here is on how the before vs. during prompt, chosen randomly as the first temporal framing, influences the way participants write about the future, after the pandemic. During all writing prompts, a countdown timer was visible on the screen during writing, and on multiple pages throughout the study, mental health resources were provided. After completion of the three prompts, participants responded to questions about demographics, COVID-19 experiences, journaling experience, as well as three individual difference measures: a rumination scale, an 18-item adaptation of the need-for-cognition scale, and a social connectedness and belonging scale. ## Measures and analyses In the analyses that follow we focus on document-wide features, taking an exploratory approach. Linguistic Inquiry and Word Count (LIWC) categorizes the words in a text based on a range of concepts, including emotions, cognitive tension words, causal words. LIWC provides one methodological tool for enabling indirect inferences about mental states. The most recent version of LIWC at the time of analysis features over 100 word categories, capturing a large variable space. This version of LIWC was tested and validated using a “Test Kitchen Corpus” of around 31 million words pooled from a wide range of corpora, including blogs, emails, movie dialogues, transcribed speech, natural conversations, social media posts, and more. Analysis of language data can be challenging due to the complexities at play. However, LIWC has been successful at predicting a variety of psychological and social measurements from language usage. For example, course performance has been generally predicted based on the written self-introductions of undergraduate students. Several of the specific LIWC word categories also map neatly onto well-studied and meaningful dimensions of language. LIWC includes several sentiment related categories, including positive and negative emotion and tone, as well as several discrete emotions, such as sadness, anger, and anxiety. Sentiment of language may provide hints at wellbeing. In one pair of studies, improvements in physical health were linked to a greater use of positive emotion words and a moderate number of negative emotion words (neither very high nor very low), as well as increased use in both causal and insight words throughout a writing task. Pronoun usage may also hint at different psychological processes. Greater use of first person singular pronouns is associated with interpersonal distress, as well as depressive symptoms and negative emotions. LIWC additionally includes categories relating to time, such as a past or future focus, and health categories, all concepts highly relevant for the topics of interest in this dataset. Content words (e.g., nouns, regular verbs, and various adjectives and adverbs) and function words (e.g., pronouns, prepositions, articles, conjunctions, and so on) are also detectable using LIWC, and may reveal information about one’s social inclination—the use of function words often requires understanding shared knowledge between interlocutors, for example. For each future-oriented text produced by participants, LIWC generates a set of semantic category measures that reflect the percentage of these categories represented in that text. This can be understood as a multivariate vector of measurements of how positive, negative, etc., a text is based on a calculation of the percentage of words that fall under these categories. To evaluate the influence of temporal framing (i.e., writing about pre-pandemic or during pandemic life in the first prompt) on writing about life after the pandemic, we evaluated the LIWC features present in the post-pandemic documents (the LIWC data for all documents and the analyses script are available at <https://github.com/conbainbridge/covid_thoughts>). Because LIWC has over 100 of these categories, we face the challenge of multivariate analysis without simply deploying an analysis pipeline on each of the 100 separate dimensions. To conduct a more global analysis of LIWC features, we used principal components analysis (PCA) to determine components that best predict the temporal framing condition. PCA permits the analysis of many intercorrelated variables, characterizing the structure of both the observations in the dataset and the variables themselves (for a detailed explanation, see). PCA has been used successfully in prior work to reduce LIWC dimensions. PCA thus extracts a conceptual space across all LIWC dimensions, but at a lower dimensionality. Another way to think of this process is that PCA reveals this lower dimensionality based on how normalized LIWC scores cluster across students’ writing. For example, instead of the three LIWC features “positive tone,” “negative tone,” and “emotion,” the PCA model may infer that these three features load onto just one principal component (PC). This example is intuitive, but finding clusters across texts and many features yields subtler and more complex patterns of correlation. Our main test is whether these lower- dimensional PCs distinguish temporal prompts at all. This analysis is done based solely on the post-pandemic-oriented texts, to see how the framing of a preceding prompt may echo into thoughts about the future. Put simply: It would show that participants primed by the past or present (pre- and during pandemic) alter the way they think (or write about) the future. # Results The PCA recovers as many PCs as there are variables, ranked by the strongest component to the weakest. In cases where the number of participants is less than the number of variables, the number of PCs is limited to match this sample size. Because the winter dataset’s sample size (*n* = 91) is less than the total LIWC variables (117), the PCA yields 91 total PCs. The LIWC features cluster across texts as we found a nonlinear rise in PC strength, expressed through cumulative proportion of variance accounted for. The first 20 PCs account for almost 70% of the cumulative variance in the dataset. As noted above, we tested whether these LIWC PCA components from the future- oriented prompt relate to the temporal frame of the prior prompt (before vs. during the pandemic). With a logistic regression predicting prior condition from these 20 components, we found seven PCs that were significant or approached significance. We chose a liberal initial threshold of *p* = 0.1 to ensure we captured a wide range of possible semantic structures in the future-oriented writing. In a secondary generalized linear model, six of these remained significant (PCs 1, 4, 5, 10, 11, and 13). We also included the seventh (PC18) in our selected components because it trended towards significance in that follow-up model alone. Note that these results reflect coefficients from a single regression model–not independent tests. The *p*-values for each PC in the model, and the top ten most influential LIWC features per PC (i.e., highest absolute values in loading scores, ordered from most to least influential) are in. LIWC categories in bold, italic font feature positive loading scores, indicating their tendency to characterize the *during*-pandemic framing, while negative (normal font) loading scores characterize the *pre*-pandemic framing. Loading scores are available in the. For details on what the different LIWC categories entail, see. PC5 is the most significant component of the selected components from the generalized linear model. By itself, it is able to predict which temporal framing preceded the post-pandemic prompt, based solely on the linguistic characteristics in that post-pandemic prompt (*p* =.011). PC5 shows that positive emotion, tone, want, and discrepancy are found more in post-pandemic contemplations when they are preceded by reflections about *during* the pandemic. Conversely, when prompted with *before* the pandemic, PC5 shows focus on the past and use of personal pronouns. Interpreting LIWC loadings may be subjectively influenced, and interpretive assessment must be done with caution. In this particular case, PC5 indicates that thoughts about experience during the pandemic prompt positivity (i.e., the presence of the “positive tone” and “positive emotion” LIWC categories) that is desired (“want”, “tone,” “emotion,” and “discrepancy”–which includes words like would, can, and want). On the other hand, the pre-pandemic priming may focus on what was lacking (“lack”) in the past (“past focus”) and may be expressing themselves more spontaneously (“authenticity”). An interpretation of PC1 could indicate people think more negatively (“negative tone”) and analytically (“analytic”) about health (“health”) as a result of thinking about the experience of the pandemic (e.g., “illness” and “article”, perhaps the result of noting “*the* pandemic”). Other PCs may hint at pre- pandemic priming leading to thinking enthusiastically about social life (PC4 –“fulfill,” “prosocial,” “exclamation,” “social behavior”), expressing sadness over remembering one’s lifestyle from the past (PC 10 –“memory,” sad emotion,” “leisure,” “lifestyle,”), or perhaps more episodically inspired thoughts guiding future projections (PC13 –“mental,” “visual,” “perception,” “past focus”). The during pandemic priming may lead to frustration (PC10 –“risk” and “swear”) and one’s needs and their justifications (PC13 –“Cause,” inclusive of words like how, because, and why, and “Need”). Importantly, participants were *not* prompted to contrast the future and the present/past; the temporal prompt simply alters the semantic patterns in their writing, revealed by the PCs shown in. One way to quantify these overall linguistic trends is to assess them using network analysis. This method takes the PCs and visualizes the relationships among the LIWC categories. These more visual, geometric relationships among the dimensions may help to interpret the overall shift taking place in participant writing after the prompts. We built a network model using the “igraph” R package to explore which LIWC features are shared across the selected principal components. The nodes represent the top 50 most influential LIWC features across the selected components (i.e., PCs 1, 4, 5, 10, 11, 13, 18). Edges are formed as the result of shared presence of the linked LIWC features across components, suggesting recurring themes in distinguishing the conditions. The color represents the level of influence that feature has in distinguishing conditions, such that the lighter the purple, the greater the influence across these components. This influence is calculated as the sum of the loading score absolute values for a given LIWC feature across components, and rather than being specific to the condition captures that feature’s overall distinguishing influence. Therefore, the lighter purple nodes represent features that are more influential across the semantic landscape. The manner in which features cluster may then represent the distinctive set of semantic factors that are combined during our particular task. For example, positive emotion, positive tone, want, tone, emotion, and discrepancy, all features found to characterize the during-pandemic condition in PC5, cluster together even across these components and appear to hint at longing for better times. Other clusters suggest livelihood elements (work, tech, lifestyle, culture), health (health, illness, physical), and sociality (she/he, friend, affiliation). When a feature within a cluster also exhibits a lighter tone, it may be the case that feature is particularly unifying of the cluster’s concepts (e.g., “perception” being linked to “visual” as well as “motion”), although the feature itself may merely have more influence independently. Because semantic graphs of this kind can represent how one “moves” through meaning space (cf.), a potential future application of this network-based technique is to visualize and characterize the set of potential semantic paths induced by a given prompt or frame of mind. In our case, thoughts about COVID induce particular sorts of ideas, such as health and social connection. When we prompt participants with a prior temporal frame (i.e., pre- pandemic or during), they appear to take different paths on this network. Such methods may facilitate characterization of the streams-of-consciousness and internal thought processes that open this paper. Importantly though, any such graph structure should be compared to a baseline to ensure that we are not interpreting a chance outcome. To test whether this network was structured meaningfully relative to a baseline, we extracted some measures and performed a permutation. We analyzed the mean degree (number of adjacent edges), mean betweenness (number of shortest paths going through a vertex or edge), and clustering coefficient (probability that adjacent vertices of a vertex are also connected) of the network. The mean degree is 1.92, mean betweenness 29.02, and transitivity 0.49. We then ran the same network analyses on 10,000 random permutations of the normalized LIWC data to see where the original data falls on this random distribution. This is to confirm whether such semantic structure arises specifically as a result of condition, as opposed to random clustering. The probability of the mean degree in the permutations distribution is.038.185 for mean betweenness, and.004 for the clustering coefficient. Given both the mean degree and clustering coefficient are outside a 95% confidence interval, this suggests structure meaningfully departs from what would be expected by chance. We next conducted a PCA on the fall (first phase) dataset. Our initial focus on the winter dataset was because we expect a more intensive response from participants–students who just had another disruption to their class activities during more lock down. To match the winter dataset, we ran a generalized linear model on the first 20 components, which account for 66% of the cumulative variance. Of these 20 components, only PC4 is significant (p =.008). The top ten loading scores for LIWC features characterizing PC4 are word count (-0.205), positive tone (0.182), impersonal pronouns (-0.181), death (-0.179), conversation (-0.17), conflict (-0.169), social references (-0.168), anger (-0.161), social (-0.157), and technology (-0.156), with each associated with the past-primed condition with the exception of positive tone. Because there is only one significant PC, we did not conduct network analysis on this dataset. In general, there are some small effects in the fall dataset but far less pronounced than the structure we find in the winter. We return to this below. # Conclusions and discussion These exploratory analyses showcase the influence of temporal framing on college students as they envision their post-pandemic lives. Based only on what participants wrote when imagining their post-pandemic lives, LIWC features reduced into principal components can predict which temporal framing participants received. A few significant components hint at different categories of words that aid in making these distinctions. A tentative interpretation suggests that there is an extra focus on health (PC1) and a longing for better times (PC5) when primed by pandemic life, and more socially-oriented thinking (PC4) when primed by pre-pandemic life. In a network analysis, a semantic structure appears to arise, particularly in comparison to a distribution of random permutations of the original LIWC data. Interpretive assessment of the semantic network confirms the results on individual components. This visualization also reinforces the idea that the temporal framing leading into a stream-of-consciousness might shape the conceptual structures that participants work with. These explorations offer an initial foundation for understanding the influence of temporal thinking, and in this particular study on construing imagined futures after a major global crisis. While interpretations of the components are speculative, the LIWC categories may inform deeper studies into the specific ways COVID-19 has shaped the content of students’ imagined futures. Regardless of the meaning behind these semantic spaces, this work highlights that shifts in life triggered by COVID-19 can have an impact on immediate thoughts about the future. With this in mind, interventions may be developed to explore how re-framing thoughts, such as temporally, can encourage shifts where future thoughts may be more hopeful and positive, and less dire or pessimistic. When conducting PCA on the comparison fall dataset, we find only one component is significant. This could suggest that the winter dataset induced more complex semantics due to the emotional experiences associated with the constricted context, when students returned to remote learning due to the pandemic. Indeed this was our expectation, and motivated our initial focus on that winter data set. However, there are several factors that limit any strong conclusions. First, such environmental contextual influences need to be studied in more depth in future work. These represent just two time points, and a wider sample of data from multiple timepoints may suggest these differences were due to noise. Second, the datasets do have slightly different properties. While the winter dataset had a limited sample size, constraining the total components in the PCA, the fall dataset had a larger sample size, enabling the number of components to match the number of LIWC variables. Because of limited prior work using such methods, we did not have strong priors for an optimal sample size, which may additionally limit power in these analyses. Given our analyses were exploratory in nature, future work will benefit from taking insights gained here to formulate a priori hypotheses and planned analyses. Priming participants to write in a stream-of-consciousness style seems less common in the literature in favor of journal paradigms, where more editing and refining of language may limit inferences about inner psychological and emotional processes, perhaps especially their dynamics. Nevertheless there are some important limitations to our own design that should be acknowledged. One limitation of the data collection using this paradigm was the online context of the study, which may include extraneous variables that would be valuable to measure and control for in future work. For example, this could include aspects of their state in the moment (e.g., exhaustion, mood), ease of technology use, and the environment when completing the task (such as presence of others in the room). Participants may also have still performed some editing, or struggled to understand or adhere to a free-flowing style of writing. To overcome these issues, it may be useful to integrate content analysis like this with typing dynamics. Indeed, we collected individual keypresses and timings, including the use of the delete key. It may thus be possible to reconstruct some deleted content, and give a full portrait of the stream-of-consciousness exercise. These typing data may also be used to validate and refine this paradigm to study finer-grained psychological events. Dynamic typing data may also reveal memory search, rumination through recurrent themes or word sequences. These data may also reveal document-wide typing rates that signal cognitive signatures that relate to global features such as overall sentiment and mood. In addition to word categories such as the LIWC dictionaries, other natural language processing techniques and analyses may reveal further insights. LIWC-22 also includes a measure called “narrative arc.” Narrative arc includes proposed stages of composition such as staging, plot progression, and cognitive tension, and appears to follow different patterns depending on text or transcription formats, such as fictional-style writing versus *New York Times* science articles. Whether journalistic or stream-of- consciousness writing follows certain narrative arc patterns, or varies depending on the topic, sentiment, or other features, remains an open question. Topic modeling or recurrence analyses can explore how possibilities become constrained (or not) by temporal framing. Further explorations into associations across LIWC categories could also contribute to understanding meaningful differences caused by temporal framings. Individual differences in the experience of COVID-19 life would seem to be a critical ingredient here that we do not yet explore. Future analyses may consider such differences in more detail, such as comparing the framing texts to the post-pandemic texts. If an individual writes particularly optimistically about their life *during* the pandemic, they may be more likely to then write positively about the future, while greater negativity may similarly bleed into perceptions of the future. A rumination scale, a need-for-cognition scale, and a social connectedness and belonging scale were included in data collection, though these measures were not factored into the present exploratory analyses. First-person singular pronoun use is increased in the self-focused attention typical of rumination, and thus may have potential for predicting rumination levels based on streams-of-consciousness. The interplay between language and rumination may result in, for example, pervasive use of such pronouns in future- oriented texts regardless of temporal framing. The need-for-cognition scale may also predict how much semantic space one covered in their streams-of- consciousness to begin with, and may inform language-oriented interventions if one temporal framing or the other encouraged greater cognitive exploration. Aside from the social connectedness and belonging measure, we asked questions about actual social experience during COVID-19. Taken together, these measures may explain some of the semantic space that the PCA revealed (e.g., PC4, which included the LIWC categories of “prosocial” and “social behavior” characterizing the pre-pandemic condition). Given that this dataset was a college sample, factors such as age or other demographics remain open for study as they relate to global crises. For example, experiences of age-related change appear to influence perceptions of the future, and in turn mental health. Age also appears to be a factor in influencing in- the-moment perceptions of COVID-19, although it may not have had as much influence on perceptions of the future. While the pandemic marked a sudden major lifestyle shift globally, it will be valuable to evaluate similarities and differences to other health concerns experienced personally, such as chronic health issues, injury, or a terminal illness diagnosis. Whether the global, collective experience of COVID-19, or concern about one’s own experiences drive differences in future projections remains an open question. Understanding the relationships between personal health, global health, and how perceptions of their impact on the future change across the lifespan may clarify how different kinds of interventions may perform better or worse for different health profiles. This study examines the effects of temporal framing on perception of the future, all within individuals; however, undoubtedly many external factors will also shape how individuals have experienced the COVID-19 pandemic. The media landscape and how the pandemic has been framed to different audiences will certainly have some influence. Indeed, psychological distance has been found to influence the evolution of misinformation regarding COVID-19 when the threats appear more distant. Future work may examine the effects of the media on self- talk relating to global crises. Additionally, COVID-19 panned out to be a highly politically polarizing event. Political affiliation and intensity of an individual’s antagonistic views will shape how this global event influences thoughts of a post-pandemic life. Social media usage and the makeup of one’s social network both on- and off-line likely moderate perceptions of pandemic life. The language of social media posts across different styles of platforms and through different media (e.g., written such as in a tweet versus video, such as on TikTok) may show differences in socially oriented communication, which may then turn inwards when engaging in self-talk. How inter- and intra-personal communication are linked appears to be a ripe area for research. Our findings suggest that streams-of-consciousness could have rich dynamic properties. In broader terms, the contexts of a person’s present thoughts offer a kind of momentum, propelling them into the next stream-of-consciousness. In the language of dynamical systems, there is *hysteresis*, when “the subject remains longer in the initially perceived interpretation” (p. 373). This hysteresis property characterizes psychological dwell time in many domains, from motor control to categorization. Even high levels of cognitive complexity, like streams-of-consciousness, may be given to these properties of complex, dynamic systems. The results here suggest this hysteresis occurs in temporal framings. When participants engage in thought about the past vs. the present, it may set their mind on a given trajectory, giving it some momentum and remaining longer in the initial perceived interpretation. Importantly though, the effects of such framings are confounded with the psychological boundary of a major global event. How context and psychological distance in the temporal domain interact or differently influence hysteresis could be the subject of future work. The global nature of COVID-19 may provide a valuable comparison point to other world events or disasters, such as the findings from Cohn, Mehl, and Pennebaker of language shifts after the September 11, 2001 attacks on the United States. Future stream- of-consciousness prompts may also explore open-ended thoughts rather than anchoring on a specific event, to further clarify the temporal element alone, instead manipulating how far back or forward in time one is projecting their thoughts. In the particular study presented here, we took an exploratory step into how streams-of-consciousness may color our views as we look towards the future. Our conscious thoughts, once expressed, are not divorced from measurable effects of thoughts that came before. The words we write and the things we think of before considering the future have a non-trivial influence on the way we frame that future to ourselves, and this can be seen even when considering a global crisis that has shaken our worlds as we knew them. # Supporting information We thank Greg Bryant and Anne Warlaumont for feedback on the project and writing, as well as the participants for contributing their streams-of- consciousness. 10.1371/journal.pone.0285200.r001 Decision Letter 0 Zimmerman Steve Staff Editor 2023 Steve Zimmerman 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 Jan 2023 PONE-D-22-30477 Thinking about life in COVID-19 The influence of temporal framing on streams of consciousness PLOS ONE Dear Dr. Bainbridge, 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 10 2023 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <plosone@plos.org>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript: A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols>. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at <https://plos.org/protocols?utm_medium=editorial- email&utm_source=authorletters&utm_campaign=protocols>. We look forward to receiving your revised manuscript. Kind regards, Steve Zimmerman, PhD Associate Editor, PLOS ONE Journal Requirements: When submitting your revision, we need you to address these additional requirements. 1\. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at  <https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main _body.pdf> and  <https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_titl e_authors_affiliations.pdf> 2\. Please provide additional details regarding ethical approval in the body of your manuscript. In the Methods section, please ensure that you have specified the name of the IRB/ethics committee that approved your study. 3\. Please amend your current ethics statement to address the following concerns: a\) Did participants provide their written or verbal informed consent to participate in this study? b\) If consent was verbal, please explain i) why written consent was not obtained, ii) how you documented participant consent, and iii) whether the ethics committees/IRB approved this consent procedure. 4\. Please ensure that you include a title page within your main document. You should list all authors and all affiliations as per our author instructions and clearly indicate the corresponding author. 5\. Please include your full ethics statement in the ‘Methods’ section of your manuscript file. In your statement, please include the full name of the IRB or ethics committee who approved or waived your study, as well as whether or not you obtained informed written or verbal consent. If consent was waived for your study, please include this information in your statement as well.  6\. 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: The manuscript has been evaluated by three reviewers, and their comments are available below. The reviewers have raised a number of concerns that need attention. They request additional information on methodological aspects of the study and they request revisions to the discussion section. Could you please revise the manuscript to carefully address the concerns raised? \[Note: HTML markup is below. Please do not edit.\] 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 Reviewer \#3: Yes \*\*\*\*\*\*\*\*\*\* 2\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes Reviewer \#2: Yes Reviewer \#3: 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 Reviewer \#3: 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 Reviewer \#3: 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 appears in a qualitative form but in a quantitative form, but there are many methodological problems related to the subject of the study, which is that the subject was not studied purely empirically to ensure that the changes are due to the Corona epidemic. Including, for example, that the study was conducted online, so how can all the extraneous variables be controlled such as the ease of students’ use of technology, as well as students’ specialization, and the degree of their fatigue or exhaustion. Therefore, despite the importance of the topic, the results will be difficult to verify since the experimental steps were not followed So I suggest putting the word exploratory study in the title Reviewer \#2: Introduction: well-written. It explains the rationale very well. Data collection: well described Results section: Table 1: I would suggest to group the features by condition (i.e., colour) to make it easier to discern patterns line 235: fraught is an odd choice of words here, what do you mean by this? Do you mean that interpreting the LIWC loadings may be subjective and thus risky? Please clarify a little. line 235–236: what do you mean by qualitative assessment? Do you simply refer to the interpretation? If yes, the why suddenly using this term for it? It makes it seem like you mean something else than interpretation. Please make it clear what you mean (interpretation or something else) line 237–240: I keep reading the paragraph, but keep on having trouble following it. Could you formulate your thoughts in a more accessible way? Maybe shorten the sentences? You offer a complex interpretation of the components, and it could be clearer. Figure 3: I’m not an expert in network science, and it is not quite clear to me what the plot represents. I feel that it could be improved to be better understandable. The nodes that were most influential are coloured in a brighter red. But this information is not too useful if there is no indication of the PC the word primarily belongs to. It you added this information, it would be visible which PCSs have somewhat overlapping/related linguistic features line 313: minor typo: while should be uppercased Discussion section: The discussion is very broad and touches on a very different research topics and ideas. While I do appreciate the scope, I would also be interested in how you would use the insights from this exploratory work to perform a follow-up on the ideas you present in the main body of the paper. How could the self-report measures be incorporated to supplement your findings? What could they explain in the variance of written responses/resulting LIWC dimensions? In lines 375–389, the discussion is almost too broad and strays very far from what the content of the paper. Of course, “more work could be done in understanding other communication tools such as movement, music and other creative presentations”, however the research you just presented leaves already so many open questions that it seems a little inconsequential or far-fetched to dream of research into movement expression. Instead, I would recommend more focus in the discussion on the implications of the exploration you just presented, the paragraph starting at line 390 seems fruitful in that regard, as it presents more comprehensible implications of your work here. The preceding paragraph (375–389) could in my opinion be removed. Please view my suggestions as well-meaning. I think the paper is well-done and a valuable, thought-provoking contribution to the field and an interesting application for LIWC. With my comments I wish to offer improvements to something that, in my eyes, is already very good and I would be happy to see it in its best possible form so that other researchers can get the most from it. Reviewer \#3: In this study, the authors examined essays written by students about life after the Covid-19 pandemic in a stream of consciousness fashion. Critically, before being prompted to write about post-pandemic life, participants were prompted to think about either 1) life before the pandemic or 2) life during the pandemic. The authors then used LIWC in combination with Principal Components Analysis to examine whether the earlier before/during prompt changed how participants wrote about the future. Overall, I think this is a really interesting study. I liked the “stream of consciousness” paradigm and the application of LIWC, and I also think understanding how people think about the future as related to the Covid pandemic is important. I think this manuscript is already in good shape, but I did find that the interpretation of the results was a little shallow in the results section. For instance, the authors do a good job of explaining PC5, but don’t really go into any detail about the other 6 PCs that were found to be significant in the regression model. I’m wondering what these PCs say? Based on the LIWC features shown in Table 1, the PCs seem a little hard to interpret. Maybe PC13 is related to Time in some way, but the other PCs seem a little opaque. I think more discussion of these PCs and why they might be important for distinguishing between during vs. pre-pandemic framing. Similarly, I think more interpretation of the network analysis would be useful as well. There is a lot going on with Figure 3 as it shows the clusters of LIWC categories, but then also has colors indicating the categories that distinguish between the prompts. I wasn’t really sure what to make of this in terms of the main question of the study, which is how does thinking about pandemic vs. pre- pandemic life affect thoughts about the future. The authors again mention the words related to PC5, but I’m wondering about how some of the other categories relate? For instance, “polite”, “ppron”, “pronoun” seem to be at least somewhat clustered, and are quite red. Why would during- vs. pre-pandemic framing result in differences in these categories? Also, without having a lot of knowledge about the LIWC, it is difficult to know what some of these categories even are (e.g., I’m not really sure what “conj”, “Dic”, or “Clout” mean). Overall, I think this is an interesting and well-written paper, and I believe that some deeper interpretations of the results would improve the manuscript and make it more interesting to readers. \*\*\*\*\*\*\*\*\*\* 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: **Hanaa Shuwiekh Reviewer \#2: **Yes: **Bettina Manuela Johanna Kern 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.0285200.r002 Author response to Decision Letter 0 10 Mar 2023 Dear Dr. Zimmerman, Thank you for inviting us to revise and resubmit our manuscript, formerly entitled “Thinking about life in COVID-19 The influence of temporal framing on streams of consciousness.” The three reviewers shared some valuable criticism and we’ve addressed all their points directly in changes to the manuscript. We carefully detail our updates below with page numbers for convenience. In summary, our changes mainly included the following: 1\. Highlighted the exploratory nature of our work in the title and various passages, especially in response to comments from Reviewers \#1 and \#2. 2\. Various clarifications to the table, figures and main text in response to helpful suggestions from Reviewers \#2 and \#3. 3\. Expanded theoretical discussion, especially in the conclusion, guided by comments from Reviewers \#2 and \#3. We feel that our manuscript is much improved thanks to your input. Thank you very much to you and reviewers, and we look forward to further feedback on our resubmission. Sincerely, Constance Bainbridge and Rick Dale Reviewer \#1: This study appears in a qualitative form but in a quantitative form, but there are many methodological problems related to the subject of the study, which is that the subject was not studied purely empirically to ensure that the changes are due to the Corona epidemic. Including, for example, that the study was conducted online, so how can all the extraneous variables be controlled such as the ease of students’ use of technology, as well as students’ specialization, and the degree of their fatigue or exhaustion. Therefore, despite the importance of the topic, the results will be difficult to verify since the experimental steps were not followed So I suggest putting the word exploratory study in the title. Response: We have followed the reviewer’s advice and added new remarks regarding these issues (see pg. 18, lines 433-436), and we’ve also added “exploratory study” to the title. Reviewer \#2: Introduction: well-written. It explains the rationale very well. Data collection: well described Response: Thank you for these encouraging remarks! Results section: Table 1: I would suggest to group the features by condition (i.e., colour) to make it easier to discern patterns Response: We’ve updated the table as suggested (see pg. 12). A value discrepancy in Table 1 was also updated (the p-value for PC4, fixed from 0.037 to the correct value of 0.027). line 235: fraught is an odd choice of words here, what do you mean by this? Do you mean that interpreting the LIWC loadings may be subjective and thus risky? Please clarify a little. Response: We now clarify this. We agree that “fraught” may be too dire a descriptor. So we’ve reframed this passage to simply convey that any interpretations should be considered as tentative proposals for what these loadings may mean (pg. 13, line 275). line 235–236: what do you mean by qualitative assessment? Do you simply refer to the interpretation? If yes, the why suddenly using this term for it? It makes it seem like you mean something else than interpretation. Please make it clear what you mean (interpretation or something else) Response: This is a very helpful suggestion. We’ve updated the wording in this passage and elsewhere from “qualitative” to “interpretative,” and we agree that “qualitative” could be misinterpreted as referring to a particular family of research methods. (pg. 12, line 275; pg. 16, lines 389, 393). line 237–240: I keep reading the paragraph, but keep on having trouble following it. Could you formulate your thoughts in a more accessible way? Maybe shorten the sentences? You offer a complex interpretation of the components, and it could be clearer. Response: We’ve now revised this paragraph substantially to clarify our interpretation of PC5 (pg. 13, lines 276-281). Figure 3: I’m not an expert in network science, and it is not quite clear to me what the plot represents. I feel that it could be improved to be better understandable. The nodes that were most influential are coloured in a brighter red. But this information is not too useful if there is no indication of the PC the word primarily belongs to. It you added this information, it would be visible which PCSs have somewhat overlapping/related linguistic features Response: We’ve made a few changes to clarify this. First, we’ve expanded the accompanying description in the main text (pg. 14-15, lines 309-336), including the addition of a couple references (pg. 25, lines 695-696; 703-704). We’ve also updated the network model to provide maximally relevant information, with updated colors to make clear that it represents a different aspect of the loading scores than the condition-specific characterization. line 313: minor typo: while should be uppercased Response: Fixed. Discussion section: The discussion is very broad and touches on a very different research topics and ideas. While I do appreciate the scope, I would also be interested in how you would use the insights from this exploratory work to perform a follow-up on the ideas you present in the main body of the paper. How could the self-report measures be incorporated to supplement your findings? What could they explain in the variance of written responses/resulting LIWC dimensions? Response: We thank the reviewers for this great suggestion. We’ve now expanded this section of our discussion. These revisit some of the key ideas of the study and how we might integrate new measures and conduct future work on related themes across COVID, health, and other domains (pg. 17, lines 404-411; pg. 19, lines 469-480 with a new reference, pg. 26, lines 734-737; pg. 20, lines 491-498; pg. 20, lines 555-560). In lines 375–389, the discussion is almost too broad and strays very far from what the content of the paper. Of course, “more work could be done in understanding other communication tools such as movement, music and other creative presentations”, however the research you just presented leaves already so many open questions that it seems a little inconsequential or far-fetched to dream of research into movement expression. Instead, I would recommend more focus in the discussion on the implications of the exploration you just presented, the paragraph starting at line 390 seems fruitful in that regard, as it presents more comprehensible implications of your work here. The preceding paragraph (375–389) could in my opinion be removed. Response: We’ve taken the reviewer’s helpful advice and limited the scope of the discussion to ideas more topically relevant, removing the noted paragraph (pg. 20). We are excited about the potential for this general experimental paradigm and related theoretical framework, but we agree that readers may feel jolted by the jump from our focus on pandemic-related matters into these broader topics. Because we also expanded the other discussion (see prior remark), we hope this strikes a better balance. We thank the reviewers for this suggestion. Please view my suggestions as well-meaning. I think the paper is well-done and a valuable, thought-provoking contribution to the field and an interesting application for LIWC. With my comments I wish to offer improvements to something that, in my eyes, is already very good and I would be happy to see it in its best possible form so that other researchers can get the most from it. Response: We greatly appreciate this constructive and helpful feedback! We hope our revisions improved the paper, and that you feel we’ve heeded your thoughtful challenges in these updates, especially in matters of methodological clarification, future directions and theoretical discussion. Reviewer \#3: In this study, the authors examined essays written by students about life after the Covid-19 pandemic in a stream of consciousness fashion. Critically, before being prompted to write about post-pandemic life, participants were prompted to think about either 1) life before the pandemic or 2) life during the pandemic. The authors then used LIWC in combination with Principal Components Analysis to examine whether the earlier before/during prompt changed how participants wrote about the future. Response: Thank you for this summary, it is helpful to see how reviewers translate the main findings into a core summary. We also hope we’ve addressed your suggestions faithfully, as we summarize below. Overall, I think this is a really interesting study. I liked the “stream of consciousness” paradigm and the application of LIWC, and I also think understanding how people think about the future as related to the Covid pandemic is important. Response: Thank you for these encouraging remarks, we are also enthusiastic about this framing for open-ended tasks of this sort. So we’ve kept some of these broader points in both the introduction and conclusion, and also highlighted specific issues in response to Reviewer \#2 and your comments below. I think this manuscript is already in good shape, but I did find that the interpretation of the results was a little shallow in the results section. For instance, the authors do a good job of explaining PC5, but don’t really go into any detail about the other 6 PCs that were found to be significant in the regression model. I’m wondering what these PCs say? Based on the LIWC features shown in Table 1, the PCs seem a little hard to interpret. Maybe PC13 is related to Time in some way, but the other PCs seem a little opaque. I think more discussion of these PCs and why they might be important for distinguishing between during vs. pre-pandemic framing. Response: This is helpful, and like Reviewer \#2 we think Reviewer \#3 highlights some needed expansion of our discussion. This expansion is framed around the particular study and its results, and we now elaborate a bit more about the PCs (pg. 13, lines 282-291) and also, as suggested by Reviewer \#2 as well, add clarification on the implications of the specific study in this domain in the general discussion (pg. 17, lines 404-411; pg. 19, lines 469-480). Similarly, I think more interpretation of the network analysis would be useful as well. There is a lot going on with Figure 3 as it shows the clusters of LIWC categories, but then also has colors indicating the categories that distinguish between the prompts. I wasn’t really sure what to make of this in terms of the main question of the study, which is how does thinking about pandemic vs. pre- pandemic life affect thoughts about the future. The authors again mention the words related to PC5, but I’m wondering about how some of the other categories relate? For instance, “polite”, “ppron”, “pronoun” seem to be at least somewhat clustered, and are quite red. Why would during- vs. pre-pandemic framing result in differences in these categories? Also, without having a lot of knowledge about the LIWC, it is difficult to know what some of these categories even are (e.g., I’m not really sure what “conj”, “Dic”, or “Clout” mean). Response: It is very helpful to get these remarks, and indeed Reviewer \#2 shares similar suggestions (see points above). We mentioned above and we note again here for convenience: We’ve now expanded our discussion of the network diagram and updated the diagram to provide more meaningful information via color, shifting to a different color to avoid confusion with condition-related color choices in the Table and Fig. 2. We’ve also elaborated more on the interpretation of the network (pg. 14-15, lines 309-336). Additionally, we note a reference to the LIWC-22 manual (Boyd et al., 2022), which includes further details on what the LIWC categories encapsulate (pg. 11, lines 231-232). Overall, I think this is an interesting and well-written paper, and I believe that some deeper interpretations of the results would improve the manuscript and make it more interesting to readers. Response: Thank you! We hope our revisions have achieved this important direction, and we greatly appreciate your feedback. 10.1371/journal.pone.0285200.r003 Decision Letter 1 Ptaszynski Michal Academic Editor 2023 Michal Ptaszynski 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 Apr 2023 PONE-D-22-30477R1Thinking about life in COVID-19: An exploratory study on the influence of temporal framing on streams-of-consciousnessPLOS ONE Dear Dr. Bainbridge, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. Please submit your revised manuscript by May 18 2023 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <plosone@plos.org>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript: A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'. A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'. An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols>. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at <https://plos.org/protocols?utm_medium=editorial- email&utm_source=authorletters&utm_campaign=protocols>. We look forward to receiving your revised manuscript. Kind regards, Michal Ptaszynski, PhD Academic Editor PLOS ONE Journal Requirements: Please review your reference list to ensure that it is complete and correct. If you have cited papers that have been retracted, please include the rationale for doing so in the manuscript text, or remove these references and replace them with relevant current references. Any changes to the reference list should be mentioned in the rebuttal letter that accompanies your revised manuscript. If you need to cite a retracted article, indicate the article’s retracted status in the References list and also include a citation and full reference for the retraction notice. \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation. Reviewer \#1: All comments have been addressed Reviewer \#2: All comments have been addressed 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 \#1: Partly 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 study seems quantitative, but it was dealt with qualitatively, and therefore the results are more like suggestions and not facts, which confirms that if the application is repeated again on these students, will the same results be obtained? I expect not. Therefore, I suggest adding the word exploratory study in the title of the study Reviewer \#2: Dear authors, I'm very impressed by your work, congratulations! Thank you for considering my suggestions and for your kind remarks on them. I hope to see your paper published soon. All the best! Reviewer \#3: The authors have addressed all of my comments in the previous round, and I believe the manuscript, which was already in good shape, is even stronger now. In particular, I found the additional paragraphs on interpreting the PCs and network analysis were very helpful. Overall, I think this will make a very nice contribution to PLOS ONE, and I appreciate the opportunity to review this work! \*\*\*\*\*\*\*\*\*\* 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: **Hanaa Shuwiekh Reviewer \#2: **Yes: **Bettina MJ Kern 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.0285200.r004 Author response to Decision Letter 1 7 Apr 2023 Reviewer \#1: The study seems quantitative, but it was dealt with qualitatively, and therefore the results are more like suggestions and not facts, which confirms that if the application is repeated again on these students, will the same results be obtained? I expect not. Therefore, I suggest adding the word exploratory study in the title of the study Response: We hope to clarify this point: we had updated the title of the manuscript to include “exploratory study” in our previous revision, honoring this very helpful comment from Reviewer \#1. We noticed the “short title” we submitted into the system still used the former title, so we have additionally updated the short title to now read “Thinking about life in COVID-19: An exploratory study on the influence of temporal framing.” If there is an alternate spot in the manuscript this comment intended to address, we are happy to make any additional updates to make sure this aspect of the research is clear. We have resubmitted again and double checked that “exploratory” is indeed in the title of the manuscript. If this is not visible to reviewers or editor, we welcome any advice about correcting it. Thank you again for your suggestion! Reviewer \#2: Dear authors, I'm very impressed by your work, congratulations! Thank you for considering my suggestions and for your kind remarks on them. I hope to see your paper published soon. All the best! Reviewer \#3: The authors have addressed all of my comments in the previous round, and I believe the manuscript, which was already in good shape, is even stronger now. In particular, I found the additional paragraphs on interpreting the PCs and network analysis were very helpful. Overall, I think this will make a very nice contribution to PLOS ONE, and I appreciate the opportunity to review this work! Response to all reviews: Thank you once again for your valuable feedback in strengthening the manuscript! 10.1371/journal.pone.0285200.r005 Decision Letter 2 Ptaszynski Michal Academic Editor 2023 Michal Ptaszynski 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 Apr 2023 Thinking about life in COVID-19: An exploratory study on the influence of temporal framing on streams-of-consciousness PONE-D-22-30477R2 Dear Dr. Bainbridge, 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, Michal Ptaszynski, 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: (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 Response) \*\*\*\*\*\*\*\*\*\* 3\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: (No Response) \*\*\*\*\*\*\*\*\*\* 4\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: (No Response) \*\*\*\*\*\*\*\*\*\* 5\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: (No Response) \*\*\*\*\*\*\*\*\*\* 6\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: Thank you for accept my suggestion and adding "“exploratory study”" " in the title, Thank you again \*\*\*\*\*\*\*\*\*\* 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: **hanaa Shuwiekh \*\*\*\*\*\*\*\*\*\* 10.1371/journal.pone.0285200.r006 Acceptance letter Ptaszynski Michal Academic Editor 2023 Michal Ptaszynski This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 20 Apr 2023 PONE-D-22-30477R2 Thinking about life in COVID-19: An exploratory study on the influence of temporal framing on streams-of-consciousness Dear Dr. Bainbridge: 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. Michal Ptaszynski Academic Editor PLOS ONE [^1]: The authors have declared that no competing interests exist.

# Introduction CD97 is a member of a new subgroup of seven-span transmembrane (TM7) molecules which belong to the secretin receptor superfamily. CD97 is produced as alternatively spliced forms that contain three (EGF1,2,5), four (EGF1,2,3,5), or five (EGF1-5) repeated EGF-like domains which mediate binding to its cellular ligand, decay accelerating factor (DAF, CD55), a regulatory protein of the complement cascade. CD97 was originally found to be expressed by hematopoietic cells, then abundantly detected in various normal tissues and advanced stages of thyroid, colorectal, gastric, pancreatic, esophageal and oral squamous cell carcinomas. CD97 protein is over-expressed in majority of gastric adenocarcinomas (60%–88%) and mostly located in the tumor cells at the invasion front, which has higher motility as compared to the cells in the solid formation. Various studies reported that elevated expression of CD97 in gastric cancer is associated with the dedifferentiation and aggressiveness of tumor cells and directly correlates with clinicopathological parameters such as TNM classification. Recently, the interaction between the small isoform of CD97 (CD97/EGF1,2,5) and its ligand CD55 led to increased motility and elevated proteolytic activity of matrix metalloproteinases or chemokine secretion was revealed in colorectal cancer cells. Our previous studies revealed that out of three CD97 isoforms, only the small one was associated with increased invasive behavior of gastric cancer cells in vitro. However, knowledge of the role of CD97 isoforms, especially CD97/EGF1,2,5, in tumor metastasis in vivo is still lacking. In this study, by employing the cells with stable CD97 small isoform knock-down and orthotopic gastric cancer mouse model, we further investigated the role of CD97 small isoform in gastric cancer progression in vivo, focusing on tumor development and metastatic potential. # Results ## Generation of Transfectants with Stable Knockdown of CD97 Small Isoform For this purpose, four human gastric cancer cell lines, SGC-7901, AGS, BGC-823 and MGC-801 were investigated for the CD97 gene expression. RT-PCR revealed that all cell lines investigated expressed CD97, however with different isoform distribution (EGF1,2,5; EGF1,2,3,5 and EGF1-5). As demonstrated before, BGC-823 cells had the lowest intensity of CD97/EGF1,2,5 but the strongest intensity of CD97/EGF1-5, while AGS showed the strongest expression of CD97 small isoform but the weakest of CD97 big isoform. SGC-7901 which expressed moderate levels of both CD97 small and big isoforms was selected to investigate the specific effects of CD97 small isoform. After we transfected SGCwt cells with shRNAs targeting four different sites of CD97/EGF1,2,5 and selected stable clones by G418, the expression of CD97 small isoform was evaluated by RT-PCR and western blot. As compared with wild-type cells or the clones bearing non-silencing shRNAs, the clones expressing the forth shRNA (CD97iso3-Si4) displayed a 40% loss of CD97/EGF1,2,5 mRNA expression and a nearly total loss of CD97 protein (∼80 kDa). The silencing of CD97/EGF1,2,5 by employing other sh-RNA site induced similar reduction and effects as compared to CD97iso3-Si4. It is worth to note that due to the knock-down of CD97/EGF1,2,5, the other two isoforms (CD97/EGF1,2,3,5 and CD97/EGF1-5) were not significantly affected. ## Effect of CD97 Small Isoform on Proliferation, Migration and Invasion of Human Gastric Cancer Cells Increased proliferation and migration are both important parameters defining cancer cells and their reductions may serve as potential anti-cancer therapy. In this study, the proliferative ability of CD97/EGF1,2,5 kd clones was significantly higher as compared to the wild-type and control groups, which coincided with our previous research. Using the scratch-wound assay, a continuous movement was observed in all the three groups, but a more significant movement of the CD97/EGF1,2,5 kd clones migration front was evident at 24 h, in which the cell-free “scratch” region was almost full confluent. Upon comparison with wild-type and empty control groups, the migration of CD97/EGF1,2,5 kd clones was significantly enhanced after 24 h of incubation (P\<0.01). These observations revealed that the down regulation of CD97 small isoform did not result in an inhibition of proliferative or migrative ability of gastric cancer cells as previously expected, but caused a statistical significant enhancement of both the proliferative and migrative ability of SGC-7901 cells. Soft agar colony formation assay revealed the different invasive ability of CD97/EGF1,2,5 kd clones. CD97/EGF1,2,5 kd clones generated 5 times less colonies as compared with wild-type and empty control cells. Furthermore, the colonies formed by CD97/EGF1,2,5 kd clones were bigger in size and showed the preference of aggregated growth rather than the pattern of detachment from the main tumor bulk and disseminated growth. To confirm whether CD97 small isoform may affect the extracellular matrix by alterations of penetrating ability, transfectants investigated were seeded on filters coated with gelatin. After 24-h incubation in the upper chamber, the cells that had migrated to the other side of the filter were stained and counted. The number of penetrated CD97/EGF1,2,5 kd clones was significantly decreased when compared with empty control, which indicated its partial loss of invasive ability after the knock down of CD97 small isoform. ## CD97 Small Isoform Supports Gastric Cancer Local Growth Aiming at verifying the invasion promoting role of CD97 small isoform in vivo, a mouse orthotopic gastric cancer model with a high frequency of lymph node metastasis was employed. Six-eight weeks after the hypodermic inoculation of SGC wt, SGC-NS and CD97/EGF1,2,5 kd clones, 100% of gastric cancer cells (10/10) grew as subcutaneous implants in nude mice. The size of subcutaneous formed tumor masses of the SGC wt (1.05±0.3 cm) and SGC-NS (1.01±0.2 cm) groups were significantly bigger than the CD97/EGF1,2,5 kd group (0.4±0.05 cm) as indicated in and no regional or distant metastasis was observed. Hypodermic tumor masses were harvested and orthotopically transplanted as described. Primary tumor masses and metastases which included local and distant lymph nodes, the liver, the lung and the pancreas were weighted and harvested on post-operative days 7–42 and histologically confirmed. Similar with the hypodermic yielded tumor masses, the weight of primary tumor masses of CD97/EGF1,2,5 kd group (0.6±0.14 g) were significantly lower as compared to SGC wt (2.8±0.26 g) and SGC-NS (2.6±0.28 g) groups on post operative days 42, which again suggested the tumor supporting effect of CD97 small isoform. ## CD97 Small Isoform Supports Metastatic Spread of Orthotopically Implanted Gastric Cancer To further evaluate the role of CD97 small isoform in lymph node metastasis, metastatic tumor cells within lymph nodes along the lesser gastric curvature were counted by C4.4A staining and flow cytometry. Though only small amount of tumor cells were detected on post operative days 7 (2–3±1×10⁴/LN), the frequency of lymph node metastasis was 100% (60/60). The tumor cell number within CD97/EGF1,2,5 kd group (69±10×10⁴/LN) was distinctly decreased on post operative days 42 as compared to SGC wt (250±31×10⁴/LN) and SGC-NS (268±25×10⁴/LN) groups. Further investigations were performed on EPCAM, VEGFR and CD44, tetraspanins and integrins previously reported to participate in tumor progression by preparation of (pre)metastatic niche. In this study, CD44, VEGFR, CD31 together with CD97 antibodies were employed to examine the metastasis-supporting effect of CD97 small isoform and the alterations of transmembrane receptors protein expression by down regulation of CD97/EGF1,2,5. Strongly down regulated CD44, VEGFR, CD31, as well as CD97 expression in early metastatic regional lymph nodes (lymph nodes along the lesser gastric curvature of post-operative day 7) of CD97/EGF1,2,5 kd group were demonstrated by immunohistochemistry, which indicated the potential involvement of CD97 in the preparation of (pre)metastatic niche. # Discussion Previous studies demonstrated that CD97 isoforms play dual roles in gastric cancer cell migration and invasion, and only the CD97 small isoform is associated with increased invasive behavior of gastric cancer. To verify this theory, we generated stable knockdown clones of CD97 small isoform and established orthotopic gastric cancer mouse model with metastasis. We found in this study CD97 small isoform not only supported gastric cancer local growth, but also promoted metastatic spread in orthotopically implanted mouse model. Involvement of the CD97 small isoform in the preparation of (pre)metastatic niche could be one of the reasons. It had been shown that CD97 protein was over-expressed in tumor tissues, especially in scattered tumor cells at the tumor invasion front, possessing much higher motile abilities than the tumor cells in the solid formation. Our previous studies revealed that strong co-expressions of CD97 and its ligand CD55 were exclusively localized at the tumor invasion front of gastric cancer, and additionaly correlated with its TNM status. These findings further support the participation of CD97 in gastric cancer cell migration and invasion. However, no relations with lymph node metastasis or lymphoid/blood vessel infiltration were found by clinicopathological investigations. We demonstrated previously that over-expression of CD97/EGF1,2,5 or CD97/EGF1-5 in low CD97 cell line BGC-823 triggered the mechanisms resulting in quite different cell behaviours. The cells bearing CD97/EGF1,2,5 insert revealed increased invasive behavior, while the cells over-expressing CD97/EGF1-5 long isoform demonstrated tumor suppressive properties. Other investigations concering CD97 variants revealed that over-expression of CD97/EGF1,2,5 insert propagated invasion of colon cancer cells, while C-terminal insertion reduced it. Besides, the motility of the cells remained almost unaffected when the full length isoform was over-expressed, what is with agreement to our findings. Thus, the CD97-induced augmentation of cancer invasiveness is closely related to its small isoform. Based on these observations and our previous findings, we focused further investigations on the specific role of CD97 small isoform in gastric carcinoma progression and metastasis. SGC-7901 cells which expressed moderate levels of both CD97 small and big isoforms were selected as a suitable model to investigate the specific effects of CD97 small isoform. After knock-down of CD97/EGF1,2,5, the proliferative ability of CD97/EGF1,2,5 kd clones was significantly higher as compared to the wild-type or control groups, while the invasive ability was dramatically decreased. It conformed to our previous research that CD97 small isoform is associated with the invasive behavior, while the CD97 full length isoform is correlated with the proliferative property; and there existed an intrinsic balance between the two isoforms, when the expression of one isoform elevated, the other decreased. The elevation of the gelatinolytic abilities of CD97/EGF1,2,5 kd clones was considered due to the alternation of activity of MMPs. Galle J reported HT1080 cells which over-expressing CD97 small isoform possessed the highest activity of matrix metalloproteinases (MMPs) MMP-2 and MMP-9, but HT1080 cells over-expressing CD97 full length isoform responded with almost unaltered or even decreased levels of MMPs as compared with wild-type or empty plasmid cells. Although multivariate analysis reported no significant relations between CD97 expression and lymph node metastasis, results in this study showed noticeably retarded primary tumor growth as well as fewer metastatic tumor cells in regional lymph nodes in CD97/EGF1,2,5 kd group of the orthotopically transplanted metastatic mouse model of gastric carcinoma. This phenomenon conformed to the tumor invasive promoting role of CD97 small isoform. But how can CD97 small isoform facilitate lymph node metastasis? In recent years, developed evidence based on Stephen Paget’s “seed and soil” hypothesis has emerged. Growth factors secreted by the primary tumor prime certain tissues for tumor cell engraftment. In response to these soluble factors, tumor associated cells such as macrophages and haematopoietic progenitor cells cluster at some functional microenvironment, which is also known as “metastatic niches” that supports metastatic tumor cell maintenance and actively regulates cell proliferation and invasion. This microenvironment is considered to comprise supportive stromal cells, soluble factors, vascular networks, nutrients and metabolic components, and the structural extracellular matrix (ECM) architecture. VEGF, critical regulator of tumor angiogenesis, is thought to mobilize the bone marrow derived cells (BMDCs), which may subsequently be recruited and facilitate tumor growth and metastasis 888. It was reported that BMDCs express VEGFR localized to pre-metastatic sites before the arrival of tumor cells, and inhibition of VEGFR1 could prevent the BMDCs infiltration and “metastatic niche” formation in lungs. CD44 is also known to be essential for the homing and engraftment of the cancer stem cells. It was reported after knocking down of CD44v6, it failed to assemble a soluble matrix in pre-metastatic organ, which allows a highly metastatic pancreatic cell line ASML embedding and growth. It strongly indicated both VEGFR and CD44 were involved in the preparation of (pre)metastatic niche. In this study, VEGFR, CD44 along with platelet endothelial cell adhesion molecule (CD31) were together employed in an early lymph node metastatic model (post operative day 7) of gastric caner to investigate the possible mechanism of the metastasis-facilitating role of CD97 small isoform. In regional lymph nodes along the lesser gastric curvature prior to metastasis, the control group showed already existed, aggregated, high intensity of CD97 protein expression, as well as the elevated intensity of CD44, VEGF and CD31 expression; while the CD97/EGF1,2,5 kd clones showed comparatively scattered and down regulated transmembrane receptors protein expression, which indicated CD97 small isoform may also contribute to the metastasizing tumor cells settlement and involved in the preparation of the metastatic niche formation. However, information of how CD97 bearing tumor cells interacting with the metastatic organ surroundings and how the long-distance communication established is still lacking. Furthermore, it was reported the expression of CD97 was consistently suppressed in glioblastoma cell lines along with the silencing of WT1, which suggested the possible upregulation of CD97 and its invasiveness promotion role were mediated by the regulation or expressional changes of other genes. But there was very little data describing the regulatory relationship of CD97 with other genes hitherto, which also merit further investigation. In summary, taking advantage of the establishment of orthotopically transplanted metastatic mouse model of gastric carcinoma and the knocking down clones of CD97/EGF1,2,5 which showed poor metastatic ability, we demonstrate CD97 small isoform not only supports gastric cancer local growth, but also facilitates metastasis in a mouse model. Involvement in the preparation of the metastatic niche formation could be the reason for the contribution of CD97 small isoform to the metastasizing tumor cells settlement. Although exemplified in an animal model, the findings in this study are required to be controlled for their relevance in human cancer progression. # Materials and Methods ## Cell Lines and Animals All cell lines employed in this study were purchased from ATCC ([www.atcc.org](http://www.atcc.org)). Stomach adenocarcinoma cell line SGC-7901, MGC-801 and AGS cells were propagated in RPMI-1640 medium (Genom Biologic, Hangzhou); BGC-823 cells were propagated in Dulbecco’s minimal essential medium (DMEM)/Ham’s F12 (Genom Biologic, Hangzhou). All media were supplemented with 10% fetal bovine serum (FBS, Gibco) and 1% penicillin/streptomycin (Genom Biologic). The cells were grown in standard humidified incubator in 5% CO2 at 37°C and passaged every 4–7 days using trypsin-EDTA. All the animal studies were 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 and the protocol was approved by the Animal Research Committee of Zhejiang University, Hangzhou, China. All surgery was performed under sodium pentobarbital anesthesia, and all efforts were made to minimize suffering. Mouse protocols were conducted in accordance with stringent regulations laid out by Zhejiang University Laboratory Animal Center. Seventy- five 6–8-week old male BALB/c nu/nu mice weighing 18–22 g used for subcutaneous (15/75) or orthotopic tumor implantation (60/75) were randomly divided into control and CD97/EGF1,2,5 knockdown groups. Animals were housed in a sterile environment, cages and water were autoclaved, bedding and food was γ-ray- sterilized. All animals were maintained on daily 12-h light/12-h dark cycle, which was controlled by qualified staff in the Zhejiang University Laboratory Animal Center. ## Knockdown of CD97 Isoforms by RNA Interference (RNAi) Four RNAi candidate target sequences of human CD97 isoform 3 (CD97/EGF1,2,5) were designed following the procedure of Dharmacon siDESIGN™ center, and were cloned into pGCsilencer™ U6/Neo/GFP vector (Shanghai GeneChem) using Lipofectamine 2000 reagent according to manufacturer’s protocol (Life Technologies). Selection of the clones was initiated 2 days after transfection by employing 700 µg/ml of G418. Two weeks after transfection, positive clones were selected and maintained in fresh medium containing G418 at final concentration of 500 µg/ml. Medium was changed every 2–3 days and transfectants were passaged every 5–6 days. Expression of CD97 isoforms was verified by RT-PCR and Western blot analysis. Nonsilencing (NS)-siRNA was used as a control. ## Orthotopic Transplantation of Human Gastric Cancer Subconfluent SGC-NS (control) and SGC-CD97/EGF1,2,5 RNAi clone cells were harvested with trypsin-EDTA and resuspended to a final concentration of 1×10⁸/ml PBS. Hypodermic inoculation: 0.1 ml of cell suspension was subcutaneously injected into the right flank of respective mouse. Six-eight weeks later, when the size of tumor was around 1 cm³, tumor mass from each group was taken out and minced into pieces of approximately 1 mm³ for use in transplantation. Nude mice were anesthetized with pentobarbital sodium solution (Sinopharm Chemical Reagent, Beijing) via intraperitoneal injection (45 mg/kg). A left lateral laparotomy was performed. The stomach wall was carefully exposed, and a mechanical serosal injury was made in the middle of the greater curvature. One tumor piece was then fixed on the injured serosal surface with a 5–0 Dexon transmural suture. ## Evaluation of Growth and Metastasis of Orthotopically Implanted Tumors To evaluate the growth and metastasis of orthotopically implanted tumors, mice were sacrificed according to the institutional guidelines on postoperative days 7, 14, 28 and 42 before they developed signs of distress. Autopsies were performed immediately, and primary tumors growing on the stomach wall were excised, weighed and examined histologically. The lungs, liver and enlarged regional or metastatic lymph nodes in peritoneal cavity were collected and processed for careful immunohistochemical examination. ## Total RNA Extraction and Reverse Transcriptase-polymerase Chain Reaction (RT-PCR) Total RNA from SGC-7901 wild-type cells (SGCwt), stable SGC-NS and SGC- CD97/EGF1,2,5 RNAi clones was extracted using TRIzol reagent according to manufacturer’s instructions (Life Technologies). To exclude genomic amplification of prepro-CD97, specific intron-spanning oligonucleotide primers, which are suitable for amplification of all 3 CD97 isoforms (sense- actctgccgggagctgaaac; antisense-tggatggtgacctcggctga), were employed. RT-PCR reactions were performed in a 25-µl solution containing 4 µl of cDNA, 2.5 µl of 10× Advantage2 polymerase mix buffer, 10 nmol/l of dNTP, 20 pmol of the primer, and 2 U TaqDNA-polymerase (Life Technologies). PCR cycles consisted of an initial denaturation step for 5 min at 95°C, followed by 35 cycles of denaturation at 94°C for 30 sec, annealing for 45 sec at 61°C, elongation for 45 sec at 72°C, and a final extension cycle for 5 min at 72°C. PCR products were visualized on a 1% agarose gel containing 0.05% ethidium bromide. Semiquantitative RT-PCR analysis was performed with the Bio 1D software (LTF, Wasserburg, Germany) and β-actin served as normalizing marker. ## Cell Proliferation Assay (MTT Assay) SGCwt, stable SGC-NS and SGC-CD97/EGF1,2,5 RNAi stable transfectants were plated in 200 µl of RPMI-1640 with 10% FBS at 2.5×10⁴cells/well in 96-well plates. After overnight incubation, the growth medium was replaced with serum- free medium. At each time-point (24, 48 and 72 h), 20 µl of 5 mg/ml MTT (Sigma) was added to each well and incubated at 37°C for 4 h to allow the reduction of MTT to formazan. Formazan crystals were dissolved in 100 µl DMSO and measured at 540 nm using ELISA reader (TECAN, Austria GmbH). ## Scratch-wound Assay Cells were seeded at 2×10⁵ cells/well in 6-well plates and cultured as 70% confluent monolayers and deprived of serum for 24 h. A scratch was performed across the well with a standard 200 µl pipette tip and subsequently washed extensively with RPMI-1640 (supplemented with 1% FBS). The rate of scratch closure was determined by inverted phase contrast microscope (Leica, Germany) immediately and 24 h later. The migration rate was expressed as a percentage of the control (SGC-NS), and was calculated as the proportion of the mean distance between both borderlines caused by scratching to the distance that remained cell-free after re-growing. Two independent series of experiments were performed in quadruplicates. ## Soft Agar Colony Formation Two-layered soft agar assays were performed in 6-well plates. The bottom layer of agar (1.5 ml per well) consisted of 2.5 ml of 3% agar (Roth) in sterile water, 1.5 ml of FBS, 150 µl of G418 (20 µg/ul), 150 µl of a 1∶1,000 dilution of amphotericin B (0.25 µg/ml; Sigma), and 450 µl of a 1∶1,000 dilution of gentamicin (10 µg/ml; Sigma) added to 15 ml with RPMI-1640 medium. Once solidified at room temperature for 10 min, 20,000 of SGCwt, SGC-NS and CD97/EGF1,2,5 RNAi clones were separately mixed into 1 ml of upper agar layer prepared from a stock consisting of 0.8 ml of 3% agar, 10% FBS, 75 µl of G418 (20 µg/ul), 75 µl (1∶1,000 dilution) of amphotericin B, 225 µl (1∶1,000 dilution) of gentamicin in 15 ml of culture medium. This cell suspension was carefully layered on the top of the bottom layer. Once the top agar layer had solidified, 1 ml of the culture medium was carefully added and changed once a week. After 4 weeks, cell colonies in the agar were stained overnight at 37°C in a 5% CO2 atmosphere with 200 µl of iodo-tetrazolium chloride (5 mg/ml; Sigma). Stained cell colonies were examined by light microscopy (Zeiss). ## Invasion and Migration Assays The invasion assays were evaluated in 24-well Transwell™ chambers (Costar). The upper and lower culture compartments were separated by polycarbonate filters with 8 µm pore size. For invasion assay, the upper site of the filters (upper chamber) was coated with 1 mg/ml of gelatin (Sigma) before seeding the cells. SGCwt, SGC-NS and CD97/EGF1,2,5 RNAi clones were seeded at 2×10⁴cells/well in RPMI-1640 medium without FBS in the upper chamber and the lower chamber was filled with medium with 10% FBS. After 24-h incubation in a 5% CO2 atmosphere at 37°C, cells remaining on top of the filter were wiped off with cotton swabs and those transfectants that had traversed the membrane pores to the lower surface of the membrane were washed with chilled PBS, incubated for 5 min in 1∶1 PBS/methanol and 15 min in methanol before staining with 0.1% toluidine blue (Sigma) in 2.5% sodium carbonate. Migrated cells were counted by light microscopy (Zeiss) in four separate high-power fields per filter. For the migration assay, subconfluent monolayer was scratched with a pipette tip. Wound healing was evaluated after 24–72 h by light microscopy. All experiments were performed at least in triplicates and were expressed as mean±SEM. ## Western Blot Analysis CD97 protein expression was analyzed by western blot analysis. SGCwt, SGC-NS and CD97/EGF1,2,5 RNAi clones were seeded at 1.5×10⁵/well in 25 cm² flasks and cultured in medium with 10% FBS until 70% of confluency. Total proteins were isolated with 2× extraction buffer (125 mM Tris- HCl pH 6.8; 4% sodium dodecylsulfate (SDS); 20% glycerol; 10% mercaptoethanol (ME); 2% bromophenol blue) plus protease inhibitors cocktail (all reagents from Sigma). Proteins were resolved on a 12% SDS gel, blotted at 1 mA/gel cm2 for 2 h onto Hybond-ECL nitrocellulose membranes (Amersham). Membranes with blotted proteins were blocked for 2.5 h with 5% milk in PBS plus 0.02% Tween-20 (PBST; Sigma) and incubated overnight at 4°C with the mouse polyclonal CD97 antibody (1∶5,000; Abnova). Secondary HRP-conjugated rabbit anti-mouse IgGs was used at 1∶20,000 for 1 h at RT. Immunoreactive protein bands were visualized with the ECL kit (Amersham). The same membranes were reprobed with mAbs specific to human β-actin prior to incubation in stripping solution (2% SDS; 125 mM Tris-HCl, pH 8.0; 0.7% ME) and the block buffer (5% milk in PBS plus 0.02% Tween-20). β-actin was visualized with secondary goat anti-mouse antibodies (Sigma, 1∶20,000 in blocking buffer for 1 h) and ECL-kit. ## Flow Cytometry Lymph nodes (LN) aseptically removed were cut into small pieces and meshed through fine gauze. Suspended cells (2–5×10⁵) were stained with fluorochromeconjugated mAbs against C4.4A (Santa Cruz Biotechnology) according to routine procedures. For intracellular staining, cells were fixed and permeabilized using Cytofix/Cytoperm (BD PharMingen) according to the manufacturer’s instructions. The number of tumor cells (C4.4A +) in LN was evaluated by LSR II flow cytometer (Becton Dickinson, San Jose, CA), and data were analyzed using Flowjo software (Tree Star, Inc., Ashland, OR). ## Immunohistochemistry Immunohistochemistry was performed on frozen sections using the AEC Chromogen Kit (Sigma) according to manufacturer’s protocol. Frozen tissues were sectioned (5–7 µm thickness), mounted and air-dried. After having been fixed in cold acetone and washed with PBS, the sections were incubated in 0.03% hydrogen peroxide for 15 min to inactivate endogenous peroxidase. Slides were then incubated with antibodies against mouse CD97 (1∶200; Abnova), CD44 (1∶200, Abcam), CD31(1∶200, Abcam) and VEGF(1∶400, Abcam). Positive reactions (rose-red insoluble precipitates) were developed by incubating the slides with AEC substrate reagent after treatment with the corresponding peroxidase-conjugated secondary antibodies. The sections were then counter-stained with Mayer’s hematoxylin (invitrogen). ## Statistical Analysis Statistical analysis was performed with the SPSS 10.0 software. Student’s t-test and one-way analysis of variance were used. All experiments were performed at least in triplicates and were expressed as mean±SEM with P-values of \<0.05 considered as statistically significant. We thank Dr. Ting Fu (Institute of molecular biology/Magdeburg University, Germany) for insightful discussions and critical reading of the manuscript; Dr. Li Liu (Experimental Surgery/MOC, University of Heidelberg, Germany) for excellent administrative assistance and technical assistance. [^1]: Conceived and designed the experiments: LC. Performed the experiments: DL LY CL LZ XL GL. Analyzed the data: DL BT YZ. Wrote the paper: DL BT. [^2]: The authors have declared that no competing interests exist.

# Introduction Diabetic retinopathy is the leading cause of vision loss in working-aged people in developed countries due to vascular leakage in the central part of the retina (macular oedema) and/or ischaemia and subsequent retinal angiogenesis (proliferative retinopathy). Despite the emerge of new pharmacotherapies, like e.g. intravitreal injections of antibodies against vascular endothelial growth factor, photocoagulation remains the golden standard treatment for diabetic retinopathy. Focal macular oedema is treated by direct laser irradiation of the leaking microaneurysms, while diffuse oedema is treated in a grid pattern over the oedematous part of the retina. Proliferative diabetic retinopathy instead is treated with panretinal photocoagulation, i.e., the retina is exposed to laser irradiation outside the central part, from the outer border of the vessel arcades towards the equator. All three modalities of photocoagulation are pigment-dependent, so the laser light is absorbed by pigmented molecules and converted to heat. In focal photocoagulation this energy conversion takes place in the haemoglobin of the blood, leading to thrombus formation and contraction of the vessel wall. In the grid technique and in panretinal photocoagulation, the energy conversion takes place in the melanin in the retinal pigment epithelium (RPE). Even though it is easier to understand the beneficial effect of focal photocoagulation, where vascular leakage is directly targeted, the molecular mechanisms leading to decreased vascular leakage and proliferation after grid or pan-retinal laser irradiation are still ill defined. This is in part due to the lack of *in vitro* models that mimic the effects of laser irradiation and to difficulties in dissecting the contribution of different cell types in the retina to these processes. Therefore, we have established a model for *in vitro* photocoagulation of RPE cells, which due to their melanin content are the primary site of laser energy absorption *in vivo*. These cells form a monolayer between the photoreceptor outer segments and choriocapillaris, playing a critical role in the maintenance of visual function. RPE cells transport toxic waste products from the highly metabolic photoreceptors to the choriocapillaris, while supplying them with nutrients. RPE cells are also a source of angiogenic and anti-angiogenic substances and therefore important regulators of retinal vessel formation. In this study, we establish an *in vitro* model of photocoagulation which replicates the changes in cellular necrosis, apoptosis, migration and proliferation observed early after laser irradiation. We also show changes in the expression of genes involved in the regulation of cell proliferation, migration and tissue repairing, as well as the induction of cytoprotective genes. We postulate that this *in vitro* model can be used to further dissect the molecular mechanisms triggered by laser irradiation and the contribution of RPE cells to the process. # Methods ## Cell Culture The human RPE cell line ARPE-19 (the American Type Culture Collection, Manassas, VA, USA) was used for all experiments. RPE cells were cultured in DMEM (Invitrogen Ltd, Paisley, UK) containing 100 mg/dL D-Glucose, Sodium Pyruvate, without L-Glutamine and Phenol Red, supplemented with GlutaMAX-I (L-Alanyl-L- Glutamine; Invitrogen) at a concentration of 4 mM, 10% FBS, Streptomycin 100 µg/ml and Penicillin 100 U/ml (Invitrogen). Cells were incubated in humidified environment containing 5% CO₂ at 37°C and medium changed every third day, reaching a final density of approximately 3×10⁶ cells per cell culture flask within seven days. For all experiments RPE cells were washed once with PBS (pH 7.4±0.05, Invitrogen) and detached from the culture flasks by treatment with 0.05% trypsin-EDTA (Invitrogen). The detached cells were plated at a density of 3×10⁴ cells in 500 µl of medium on glass cover slips (12 mm in diameter, 0.15 mm in thickness) and placed in cell culture wells (16 mm in diameter). The cell culture reached confluency (∼1×10⁵ cells per cover slip) and formed a polarized monolayer 7 days after they were plated (referred to as time “zero”), at which time laser treatment was performed. ## *In vitro* Photocoagulation Model During the photocoagulation procedure, the cover slips with ARPE-19 cells were temporarily moved to wells without culture medium and placed on top of a black paper to facilitate absorption of the laser energy, as ARPE-19 cells in culture lack pigment. The black paper had been soaked in medium for 2 h prior photocoagulation to create a thin liquid film between the paper and the cover slips, facilitating more uniform heat conduction. Photocoagulation of the confluent RPE cells was accomplished with a frequency-doubled Nd:YAG laser (Visulas 532, Carl Zeiss, Oberkochen, Germany). Each 12 mm cover slip was subjected to 50 evenly spaced laser shots to obtain a similar distribution pattern as that of pan-retinal photocoagulation. Various laser power intensities (200–300 mW) and spot sizes (100–300 µm) were tested in order to determine the settings that yielded higher reproducibility in terms of lesion size and morphology. Laser irradiation time was 0.1 s regardless the setting. Fresh complete medium was added after photocoagulation and the cells were returned to the CO₂ incubator. ## Morphology ARPE-19 cells were washed once with PBS and fixed with HistoChoice (Amresco Inc., Solon, OH, USA) at 0h, 30 min, 2h, 6h, 12h, 24h, 48h, 72h and 168h (seven days) after laser treatment. Cells were stained with Haematoxylin (Scharlab S.L., Barcelona, Spain) and Eosin (H & E; Histolab AB, Gothenburg, Sweden) according to the manufacturer’s instructions, inspected on a Nikon Eclipse E800 microscope (Nikon, Tokyo, Japan) and imaged using a Nikon DS-5Mc camera and control unit (Nikon DS-U1) and NIS-Elements version 3.22 (Nikon) software. ## Cell Death For visualization and discrimination of live and dead ARPE-19 cells after photocoagulation, a LIVE/DEAD Viability/Cytotoxicity Assay Kit was used (Molecular Probes, Eugene, OR, USA). Briefly, cells were washed with PBS before simultaneous staining with green-fluorescent calcein-AM to indicate intracellular esterase activity and red-fluorescent ethidium homodimer-1 to indicate loss of plasma membrane integrity. Ubiquitous intracellular esterase activity and an intact plasma membrane are characteristic of live cells. Both dyes were used at 1µM and cells were incubated for 45 min. Cells were inspected on a Nikon Eclipse E800 microscope using the fluorescence mode with appropriate filter sets, and imaged at 0h, 30 min, 2h, 6h, 12h, 24h, 48h, 72h and 168h (seven days) after laser treatment. ARPE-19 cell death was also measured by quantification of lactate dehydrogenase (LDH) activity in the culture medium using a Cytotoxicity Detection Kit (Roche Applied Science) according to the manufacturer’s instructions. For visualization of apoptotic ARPE-19 cells, the In Situ Cell Death Detection Kit, POD (Roche Applied Science, Mannheim, Germany) was used according to the manufacturer’s instructions. This assay is based on labeling of DNA strand breaks (TUNEL technology) and therefore, preferentially detects apoptosis vs. necrosis. Cells were washed with PBS before fixation with HistoChoice and imaged at 2h, 6h, 12h, 24h, 48h, 72h and 168h after photocoagulation. For quantification of apoptosis, cytoplasmic histone-associated DNA fragments were measured in cell homogenates and in culture media using the Cell Death Detection ELISA^plus assay (Roche Applied Science) at 0h, 2h, 6h, 12h, 24h, 48h, 72h and 168h after photocoagulation according to the manufacturer’s instructions. Briefly, at the different time-points, 100 µl culture medium was removed and placed on ice for 30 min while the cells were lysed. Both the culture medium and the cell lysate were analyzed. An enrichment factor was calculated as the mean absorbance ratio between the photocoagulated samples and the corresponding non-irradiated controls. Six to 8 samples were analyzed for each time-point. ## Cell Proliferation and Migration ARPE-19 cells were fixed in ice cold methanol at 0h and at 2h, 6h, 24h, 48h, 72h and 168h after photocoagulation. Immunofluorescence experiments were performed using a monoclonal antibody against PCNA (Proliferating Cell Nuclear Antigen, PC10, Santa Cruz Biotechnology, Santa Cruz, CA, USA) and a secondary antibody, Cy5 donkey-anti-mouse IgG (1∶300, Jackson ImmunoResearch, Suffolk, U.K.). Nuclear regions and individual cells were identified using SYTOX Green (1∶3,000, Molecular Probes, Eugene, OR, USA). Three coverslips were examined for each time point at x10 (0.30 numerical aperture) in a Zeiss LSM 5 Pascal laser scanning confocal microscope (Carl Zeiss, Oberkochen, Germany) and nuclear PCNA intensity was quantified using the Zeiss LSM 5 analysis software. For laser-treated cells, 13–33 laser spots were imaged for each time point. Four to six boxes were positioned at defined distances from the lesion centre to quantify nuclear PCNA intensity in different regions. For non-laser treated control cells, 3–7 images were obtained from each coverslip. For discrimination between cell proliferation and migration during repair of the laser lesions, we first determined the non-toxic concentration of docetaxel (Sigma Aldrich, Stockholm, Sweden) that effectively blocks migration without affecting proliferation and the non-toxic concentration of mitomycin C (Sigma Aldrich) that effectively blocks proliferation without affecting migration. For cell proliferation, ARPE-19 cells were plated as explained above under “Cell culture” at a density of 3×10⁴ cells and cultured for 7 days until they reached confluency, at which time (0 h) they were incubated with docetaxel (0.001–10 nM) or mitomycin C (0.1–10 µM) for 72 hours. After treatment, cells were trypsinized and live cells were counted using a Bürker chamber after exclusion of Trypan Blue (Sigma Aldrich) positive cells. Data is expressed as percentage growth from treatment start (0 h). Cell viability was also assessed using a CellTiter 96 AQueous One Solution Cell Proliferation Assay Reagent (Promega, Stockholm, Sweden) according to the manufacturer’s instructions. This assay is based on the spectrophotometric detection of a colored formazan product converted from an (3-(4,5-dimethylthiazol-2-yl)-5-(3-carboxymethoxyphenyl)-2-(4- sulfophenyl)-2H-tetrazolium) (MTS) compound by NADPH or NADH via metabolically active cells. After culture, fresh culture medium was added together with CellTiter 96 Aqueous One Solution Cell Proliferation Assay Reagent and incubated for 1 h before measuring absorbance at 490 nm with a 96-well plate reader. For the effects of the drugs on cell migration, confluent ARPE-19 cells on 12 mm cover slips were scratched with a sterile pipette tip and photographed just after scratch (0 h) and 24 h later after incubation with or without docetaxel or mitomycin C at various concentrations. Cells were inspected on a Nikon TMS microscope and imaged using a Nikon DS-Fi1 camera and NIS-Elements F version 2.20 software. Scratch areas were measured using Image J and data were expressed as percentage of scratch area closed after 24 h. The differential effects of docetaxel and mitomycin C on migration and proliferation during the repair of laser lesions were then evaluated by measurement of the size of the cell-free region at the center of the lesions 12, 24, 48 and 72 h after photocoagulation. ## RNA Extraction RNA extraction was performed using Nucleospin RNA XS (Machery-Nagel, Düren, Germany) at 6h and 24h after photocoagulation, according to the manufacturer’s description. RNA quality and concentration was assessed using the NanoDrop 1000 Spectrophotometer (Thermo Fisher Scientific, Wilmington, DE, USA). RNA samples were stored at –80°C until further analysis. ## Quantitative RT-PCR (qPCR) cDNA was synthesized from RNA using the RevertAid First Strand cDNA Synthesis Kit (Fermentas, St. Leon-Rot, Germany). mRNA levels were analyzed with the real- time RT-PCR 7900HT system (Applied Biosystems, Foster City, CA, USA) using 10 µl TaqMan Gene Expression Master Mix (Applied Biosystems), 1 µl TaqMan Gene Expression Assay, 1 µl cDNA and 8 µl deionized, diethylpyrocarbonate (DEPC) treated water (Fermentas). All assays were from Applied Biosystems: *HSPA6* Heat shock 70kDa protein 6 (Hs00275682_s1); *CTGF* Connective tissue growth factor (Hs00170014_m1); *FOS* FBJ murine osteosarcoma viral oncogene homolog (Hs00170630_m1); *HMGA2* High mobility group AT-hook 2 (Hs00971724_m1); *IL33* Interleukin 33 (Hs01125943_m1); *IL1B* Interleukin 1, beta (Hs01555410_m1); *ADAMTS6* ADAM metallopeptidase with thrombospondin type 1 motif, 6 (Hs01058097_m1); *IL8* Interleukin 8 (Hs00174103_m1); *TGFBR2* Transforming growth factor, beta receptor II (70/80kDa) (Hs00234253_m1); *ANKRD1* Ankyrin repeat domain 1 (cardiac muscle) (Hs00173317_m1); *TIMP3* TIMP metallopeptidase inhibitor 3 (Hs00165949_m1). Cyclophilin B (Hs01018502_m1) was used as endogenous control. The relative quantity of the target gene was calculated using the comparative threshold method (ΔΔCt). If the standard deviation of the duplicate Ct value differed more than 0.16, the sample was rerun. ## Statistics Kruskal-Wallis and Mann-Whitney U test were used to test for significance. *P* values \<0.05, \<0.01 and \<0.001 were used after Bonferroni correction for multiple comparisons. Statistical calculations were performed on SPSS software v 20 (IBM, Armonk, NY, USA). Data is presented as mean ±SEM in all graphs. # Results ## Photocoagulation Settings The immediate clinical appearance of a laser lesion is characterized by a pale discoloration due to denaturation of proteins within the retina and RPE, as observed in. Within a couple of weeks the lesions usually become pigmented as a result of RPE cell accumulation. In order to achieve reproducible *in vitro* lesions with similar size and spacing pattern as those observed *in vivo,* we tested the impact of varying the laser intensity and the spot size. *In vitro*, the highly conductive glass cover slips will favour the lateral transfer of the thermal transients, whereas *in vivo*, these are limited due to the water content of biological samples. To compensate for this difference in conductance, laser spot sizes *in vitro* were smaller than those generally used *in vivo* for pan-retinal photocoagulation (100–300 µm vs. 500 µm). Of all tested experimental laser setting, the combination of 300 mW of laser power, 200 µm spot size and 0.1 s irradiation duration were found to yield the most reproducible lesions and therefore used throughout this study. shows images of H & E stained ARPE-19 cells 24 h after photocoagulation using all tested laser setting combinations. As also evidenced in, the laser leaves a circular impression of the same size of the beam (200 µm) as it hits the pigment source (black paper). Within one minute, numerous small gas bubbles appear between the paper and the glass cover slip, forming the pale punctuated rings observed around the lesions in. The small bubbles often merge into a bigger central bubble within the next five minutes (white arrowheads in) and gradually disappear within the following 5–10 minutes. These bubbles are not in contact with the cells but contribute to the artifactual appearance of the *in vitro* lesions in. The monolayer nature of the *in vitro* system as opposed to the complex multilayered architecture of the intact retina and the fact that cultured cells are temporarily moved to wells without culture medium during photocoagulation may also contribute to the somewhat different appearance of the *in vitro* lesions when compared to *in vivo* lesions in. ## Time Course of Repair Photocoagulation of ARPE-19 cells *in vitro* resulted in visible morphological changes already at 30 min. An increased turbidity of the H & E staining was observed at the centre of the lesions (mean diameter 456.9±6.3 µm, N = 104 lesions), likely reflecting protein denaturation and the result of the high thermal input received by these cells. Two hours after laser irradiation, cell detachment occurred at the periphery of the lesions, as evidenced by an empty rim in the H & E images. This detachment is probably due to cell damage caused by heat dissipation outwards from the centre of the lesion. The images also highlight that in this experimental model the thermal transients are at least able to travel as far as to the outer edges of the empty rims (i.e. a mean rim thickness of 38.7±4.0 µm, N = 158 lesions). At 6 h and 12 h, these empty areas were smaller due to gradual re-population with migrating cells from outside the lesion. At 24 h, the empty rim was completely covered by cells organized in a densely packed circular ring. At 48 h, the accumulated cells at the circular rim had started to cover the centre of the lesion, while at 72 h this area was almost fully covered by ARPE-19 cells. Seven days after photocoagulation, cells completely covered the lesion areas, but were still organized in a less homogeneous pattern. ## Necrosis and Apoptosis One of the consequences of the laser irradiation is the generation of heat, which per se can induce necrosis through membrane damage or apoptosis through aggregation of toxic denatured proteins or activation of the intrinsic apoptotic pathway. Using a LIVE/DEAD assay based on the uptake of red-fluorescent ethidium homodimer-1 in dead cells due to loss of membrane integrity and on the staining of intracellular esterase activity with green-fluorescent calcein-AM in living cells, we demonstrated that necrotic ARPE-19 cells were clearly restricted to the laser irradiated areas and appear early after photocoagulation, with a peak at 6 hours. No dead cells (red) and only live cells (green) were detected in areas in between laser lesions or on cover slips that have not been laser treated. As a complementary approach, we also measured the release of LDH to the culture medium at various time-points after photocoagulation and found, in agreement with the fluorescence experiments shown in 3A, that LDH was significantly higher in the medium of laser treated cells compared to untreated controls at 30 min, 2 and 6 h after photocoagulation, but not at later time- points. Apoptotic ARPE-19 cells were also evident in the laser irradiated areas but these appeared at later time-points exhibiting a maximal intensity 24 h after photocoagulation. Using a complementary approach, we measured DNA fragmentation in ARPE-19 cell homogenates and in the culture medium. DNA fragmentation reflecting apoptosis was higher in the samples from laser treated cells when compared to control samples. In agreement with the *in situ* apoptosis assay, apoptosis levels in the cell homogenates peaked 24 h after photocoagulation. The later peak at 72 h measured in the medium was likely due to accumulation of apoptotic DNA fragments released from the cells in the culture media. This is supported by the fact that DNA fragmentation levels in the culture media at 168 h were lower than those measured at 72 h, a time-point after which culture media was routinely exchanged. ## Cell Proliferation and Migration To study the effect of photocoagulation on cell proliferation, we measured the expression of the S-phase marker PCNA in ARPE-19 cells at various time points after laser irradiation. Using confocal immunofluorescence, we first examined the basal levels of nuclear PCNA in non-irradiated control cells maintained in culture up to 7 days (168 h). In this as in all other experiments, the culture medium was replaced by fresh medium after photocoagulation (0 h) and after 72 hours in culture. As shown in, proliferation reached a peak at 2–6 h, returned to baseline at 24 h and declined further thereafter. This is in agreement to what others have described, namely that ARPE-19 cells can continue to proliferate despite confluence. In contrast, photocoagulated cells displayed dramatically reduced PCNA levels after 2 h that remained reduced at 6 and 24 h (region “I”). This is in part due to lower number of cells in this region due to photocoagulative necrosis as shown in. However, the same was true for cells located further away from the more central region I, with the inhibitory effect decaying with increasing distance but still evident in regions \>400 µm away from the centre of the laser spot (regions “II”, “III” and “IV”), where no necrosis was detected. This depression of cell proliferation was transient; as evidenced by restored PCNA levels 24–48 hours after irradiation depending on the region examined. Moreover, 72 hours after irradiation, ARPE-19 cells in all regions, exhibited significantly higher PCNA levels than those in control cells. Measurements in areas \>600 µm away from the centre of the laser spot were not different from those at \>400 µm (not shown). shows original confocal images depicting the changes in PCNA expression at various times after laser irradiation summarized in. It is well recognized that the repair of the laser lesion involves not only cell proliferation, but also migration of the adjacent RPE cells. To discriminate between these two processes, we used docetaxel to block cell migration or mitomycin C to block cell proliferation and compared the size of the cell free areas of the laser lesions to untreated controls at 12, 24, 48 and 72 h after photocoagulation. In order to determine non-toxic concentrations of docetaxel that effectively block migration without affecting proliferation and of mitomycin C that effectively block proliferation without affecting migration, dose-response experiments were performed prior photocoagulation experiments. Using an *in vitro* scratch assay, we found that both 1 and 10 nM docetaxel effectively compromised cell migration as evidenced by significantly reduced scratch area covered with cells 24 h after wounding. Given that 10 nM was found to cause cytotoxicity as determined using the MTS assay (data not shown), docetaxel was used at 1 nM. Mitomycin C was found to limit cell proliferation at all doses tested, but 0.3 and 1 µM were used based on the lack of cytotoxic effects or effect of these doses on cell migration. Using these previously titrated doses, we found that both docetaxel and mitomycin C effectively delayed the repair process of the laser lesions. Inhibition of cell migration with docetaxel resulted in larger cell-free areas in the center of the lesions than non-treated controls and this difference was noticeable 24 h after photocoagulation. Inhibition of cell proliferation on the other hand, slowed down the repair process but at a later time-point (48 h), suggesting that changes in cell migration may precede cell proliferation during the repair process after laser photocoagulation. ## qPCR Several genes known to promote cell proliferation were significantly up- regulated 24 h after photocoagulation, such as the cytokines IL1β, IL8 and the oncogene and regulator of pluripotency in stem cells HMGA2. Production of both cytokines has been demonstrated in RPE cells, but the HMGA2 finding is novel. Six hours after photocoagulation FOS was significantly up-regulated, a target known to promote proliferation in RPE cells via cyclin D1 expression. We also found increased mRNA expression of the TGFβ receptor 2 (TGFBR2), of the matrix metalloprotease ADAMTS6, known to be up regulated in ARPE-19 cells upon TNFα stimulation and CTGF, as well as decreased expression of TIMP3, recently established as a signature gene of potential role in AMD pathogenesis all consistent with enhanced tissue remodeling capacity important for lesion repairing. ANKRD1, a co-activator of p53 and pro-apoptotic gene was down regulated at 24 h. The cytokine *IL33* which is proposed to be an alarmin released from necrotic cells and heat shock 70 kDa protein 6 (HSPA6), an inducible heat shock protein known to increase stress tolerance and survival were readily induced by *in vitro* photocoagulation in ARPE-19 cells. The induction was clear 6 h after the laser irradiation and still significant at 24 h. # Discussion In this study, we have established a method for *in vitro* photocoagulation of human RPE cells which results in reproducible lesions of similar size and spatial distribution to the ones observed *in vivo* short after laser irradiation. In this model we have demonstrated an initial induction of necrosis (∼6 h) followed by apoptosis (∼24 h) only in the laser irradiated RPE cells. We also show that the repair of the laser lesions involves both cell migration and proliferation, but that changes in cell migration seem to precede the changes in cell proliferation. Interestingly, the changes in cell proliferation were not limited to the lesions. Also, photocoagulation of RPE cells resulted in changes in the expression of genes involved in the regulation of cell proliferation, migration and tissue repairing, as well as the induction of the alarmin *IL33* and cytoprotective heat shock protein *HSPA6*. We believe that this is a suitable model to study the role of RPE cells in the processes that occur after laser coagulation and to investigate their potential contribution to the beneficial effects observed in clinical practice. Laser-tissue interactions are affected by the wavelength, energy delivered, spot size and duration of application. The parameters used in our experimental setup (300 mW of power, 200 µm spot size and 0.1 s duration, with 50 spots evenly distributed per 12 mm cover slip) were found to be optimal to obtain reproducible spots. Previous studies have used similar approaches to induce laser photocoagulation in cultured RPE cells, however the nature of the laser (i.e. diode lasers, less often used for retinal vascular disease since they cause pain) and experimental settings (i.e. smaller (50 µm) or larger (625 µm) spot sizes) used in these studies deviated to a larger extent from standards use in clinical practice. The strength of our model is the use of a laser system currently also applied for treatment of patients and experimental settings that comply to the standards established by the Early Treatment Diabetic Retinopathy Study, including 0.1 s exposures and spot sizes ranging from 100–500 µm. Also, we tried to obtain a similar spot spacing *in vitro* as that normally required for panretinal photocoagulation, since the pattern density of the lesions has been suggested to have an impact on clinical efficacy. Apoptosis of RPE cells is a not only an important event in the pathogenesis of age-related macular degeneration, but also the result of high-glucose induced oxidative damage in diabetic retinopathy. Given that RPE cells are essential for the support of photoreceptor function; efforts have focused on identifying mechanisms to prevent apoptosis in an attempt to maintain visual function. Our results shows that laser photocoagulation has indeed a negative impact on RPE cell integrity by virtue of increased photocoagulative necrosis and apoptosis, but this effect is transient and limited to the fraction of cells at the centre of the lesions and exposed to the highest thermal input. This is in line with the study of Barak et al who demonstrated an increase in apoptosis in RPE cells hours after diode laser irradiation. The fact that RPE cells underwent apoptosis relatively late after photocoagulation would suggest a mechanism triggered by the toxic accumulation of aggregated proteins typical of cells with high heat tolerance, as opposed to a mechanism dependent on MAPK signalling and the intrinsic apoptotic pathway characteristic of cells with low heat tolerance. In any case, the loss of apoptotic RPE cells seems to be compensated by an increased proliferative capacity after laser irradiation. The overall decrease expression of ANKRD1 mRNA at 24 hours may also contribute to limiting the apoptotic response. Short after *in vivo* photocoagulation, lesions usually become pigmented as a result of RPE cell accumulation, as shown in. This has been attributed, at least in animal models to changes in RPE cell proliferation and migration. Using our *in vitro* model, we could temporally and spatially dissect the changes in RPE migration and proliferation induced by laser coagulation. While the proliferative capacity of RPE cells was depressed during the first hours after photocoagulation, cells recovered and were back to control levels already 24 hours after laser irradiation to then exhibit significantly higher proliferating rates during the following days. Interestingly, this pattern was not exclusive to the lesion cells, but also a feature of neighbouring cells, suggesting thermal spread of the laser beyond the lesion area. Both theoretical and experimental data show that the temperature decreases rapidly with increasing distance from the lesion and that the spread of the retinal damage would be negligible 500 µm from the lesion. In our model in which a higher thermal conductance can be anticipated, the central zone of the lesions where the protein denaturation occur and the periphery, where cell detachment and empty rims were observed averaged 456 µm in diameter. This zone is clearly affected by the thermal transient, but effects on cells outside this zone, although less severe, may still be possible. The fact that the effects in proliferation decayed with increasing distance from the lesion and were significant in areas outside the lesions (\>600 µm), would support this idea. However, it is also possible that the changes outside the lesions may not be solely induced by heat, but also mediated by a diffusible factor from RPE cells upon laser irradiation. A potential candidate is pigment-epithelium-derived factor (PEDF), which is up regulated in human retinal pigment epithelial cells after photocoagulation and has been shown to control RPE proliferation. Another growth factor that has been shown to be secreted by RPE cells upon laser photocoagulation is transforming growth factor-beta (TGFβ) 2 and important regulator of cell migration. The here described increased expression of *TGFBR2* may suggest potentially enhanced signaling via TGFβ in this context. Expression results indicate that several important mediators of cell proliferation and tissue remodeling are engaged upon photocoagulation in this model, mediators that can be predicted to play a role in the lesion repairing process and in maintaining RPE integrity. Another observation was the induction of *HSPA6* mRNA, normally not expressed in untreated RPE cells. The induction of heat shock proteins is a well-described effect of the elevated temperature upon laser irradiation is and critical for orchestrating a cytoprotective response against severe cellular stress. Interestingly, *IL33*, a novel cytokine of the IL1 family involved in the polarization of T cells towards T helper 2 cell phenotype was also increased. IL33 is proposed to be released from necrotic cells as an alarmin and very recently, it was suggested to play a role in the pathogenesis of AMD. In conclusion, further research will be required in order to understand the sequence of biological events triggered by laser photocoagulation leading to reduced risk of vision loss from retinal vascular disease. Our simplified *in vitro* model lacks potential influences from other surrounding cells types also affected by the laser, but allows a better characterization of the contribution of RPE cells to the process. # Supporting Information We thank Maj-Lis Smith, Bodil Israelsson and Anna Maria Dutius-Andersson for skilful technical assistance. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: PTK LMB CDA MFG EA. Performed the experiments: PTK LMB EA. Wrote the paper: PTK LMB. Developed the experimental set-up: EA PTK. Interpretated the data: PTK LMB CDA MFG EA. Wrote the manuscript: PTK LMB CDA MFG EA.

# Introduction The participation of next of kin (NoK) is a cornerstone of high-quality palliative care, and as has been expressed in policy documents and translated into practice through advanced care planning. However, a previous study found that staff in nursing homes focus on ‘getting the work done’ rather than on the relationship or communication with NoK. NoK who not have a good relationship to staff members during palliative care have expressed feelings of powerlessness and being left out, which can lead to estrangement and a less good death for the older person. Several barriers to communication have been identified in nursing homes, such as NoK role confusion, conflicting responsibilities, understaffing, turnover and inadequate staff training. Inadequate training for staff members is a well-known barrier in the provision of palliative care; however, few initiatives for staff training in palliative care have been implemented, and this is especially true for initiatives focusing on the participation of NoK. A recent scoping review of studies aiming to improve staff members’ competence in palliative care with a focus on relationships with NoK found 22 articles describing educational initiatives but only three that focused on NoK relationships. The study concluded that there is a need for further research that explores NoK outcomes using robust methods, including robust outcome measures and conducting appropriate statistical comparisons of different groups. First, in terms of outcome measures, we have previously described the concepts included in the Next of Kin Participation in Care (NoK-PiC) questionnaire, which is designed to measure NoK’s participation in care in nursing homes. Specifically, these concepts are trusting the staff, being present, conversations and information, relationships with the staff, completing tasks, being respected for one’s knowledge and being acknowledged as part of the care team. It has been found that participation in care can be measured using the NoK-PiC scale, which includes the Communication and Trust (CaT) and Collaboration in Care (CiC) subscales. The development of these two subscales was based on Rasch model analysis (RMA). This article presents results from the first application of the NoK-PiC in an intervention study of palliative care. Second, in terms of statistical analysis, traditional group comparisons have been found to hide significant outcomes at person level. Hobart et al. stated that ‘group-based analyses should be complemented by legitimate analyses at the individual person level’ (p. 1048). The scales and analytical approaches used to evaluate the effects of interventions contribute to decisions about which interventions to implement in practice. In this decision-making process, rating scales and comparative statistics play crucial roles in choices that impact clinical practice. This study evaluated an intervention outcome by applying different analytical approaches: traditional within-group comparison of raw ordinal scores; traditional within-group comparison of transformed interval scores and modern within-individual (person-level) interval score comparisons. The same analytical approaches as recommended by Hobart et al. were used in this study to evaluate in detail the effects of the intervention. To the best of our knowledge, no previous intervention studies that had a control group and were based on rigorous outcome measures have explored the participation of NoK in care following an intervention using within-person interval score comparisons. Hobart et al. (2010) emphasised the clear need for further work using scales that fit the Rasch model requirements to explore significant outcomes at the person level. This study focuses on the NoK-PiC scale as the outcome measure. The aim of this study was to describe and evaluate the outcome of NoK’s participation in care in an intervention and a control group through different analytical approaches (traditional within-group comparison of raw ordinal scores, traditional within-group comparison of transformed interval scores and modern within-individual (person-level) comparisons) after an educational intervention focusing on palliative care for older persons in nursing homes that was directed towards nursing home staff members. A further aim was to compare the intervention group and control group based on the individual person-level of change. # Methods ## Design This study used a descriptive and comparative evaluation design in a pre–post intervention setting. ## Setting ### The research setting This study is part of the ‘Implementation of Knowledge-based Palliative Care’ project, which is abbreviated in Swedish as the KUPA project (*Implementering av KUnskapsbaserad PAlliativ vård*). The KUPA project was a non-randomised experimental two-armed crossover study. A mix of nursing homes participated in the study, including larger (\> 100 older persons), middle-sized (25–100 older persons) and smaller (\< 25 older persons) facilities located in both urban and rural areas. The project implemented an educational intervention across 30 nursing homes (intervention \[n = 10\], control \[n = 10\], and first control and then intervention \[n = 10\]). The intervention consisted of five 2-hour seminars addressing the knowledge and skills considered necessary to provide evidence-based palliative care in nursing homes. The content was based on fundamental principles of palliative care found in two Swedish documents: a national care programme produced by the Regional Co-operative Cancer Centres and a national knowledge support document produced by the National Board of Health and Welfare. Both of these Swedish documents are based on the World Health Organization definition of palliative care. The topics covered by each seminar were: 1) the palliative approach and dignified care; 2) NoK; 3) existence and dying; 4) symptom relief and 5) collaborative care. ### Sampling and the study group The participants in this study were recruited from all 30 nursing homes in the KUPA project, which are located in two counties in the south of Sweden. The inclusion criteria for individual participants were being NoK to an older person living in one of the included nursing homes in the KUPA project and being able to speak and understand Swedish. A total of 203 NoK (n = 95 in the intervention group and n = 108 in the control group) were included. There were no significant differences between the intervention group and the control group at baseline in terms of age, sex or type of relationship to the older person living in the nursing home. ### Data collection A contact person (a nurse assistant or a manager) at each of the included nursing homes informed the NoK who fulfilled the inclusion criteria about the study and asked whether they were interested in participating. If they responded positively, the contact person sent the NoK’s contact information to the researcher, who then contacted the NoK by telephone, further informed them about the study and asked them whether they consented to participate. When the NoK provided consent, the researchers sent the questionnaire along with a consent form for the baseline assessment via regular mail to the NoK. The researchers then sent a follow-up questionnaire by regular mail 3 months after the completion of the educational seminars (i.e. 9 months after the NoK responded to the first questionnaire). ## Questionnaire The self-report NoK-PiC questionnaire was developed on the basis of a review of the literature, and some items were inspired by the Family Collaboration Scale. The NoK-PiC scale consists of two subscales, the CaT and the CiC, which can either be used separately or combined to produce a total score. The scales meet rigorous measurement standards, having, for instance, no differential item functioning, high person separation indexes and no disordered thresholds. The subscales contain nine items each, and all items are scored from 0 to 4 (*do not agree at all* \[= 0\]; *agree to a low extent* \[= 1\]; *partially agree* \[= 2\]; *agree to a high extent* \[= 3\] and *totally agree* \[= 4\]). Possible scores range from 0 to 36 on each of the two subscales and from 0 to 72 on the total scale. ## Data analysis The percentages of persons with the lowest (floor) and highest (ceiling) scores were calculated. It is recommended that these not exceed 20%. Comparisons were made to explore whether there were differences in baseline characteristics between the intervention group and the control group. These independent group comparisons were made using the chi-square test, the *t*-test and the Mann–Whitney *U* test. IBM SPSS Statistics for Windows, Version 23.0 (IBM Corp., Armonk, NY, USA) was used for these analyses. In the context of this study, RMA advances the possibilities for evaluating intervention effects. First, RMA allows interval-level (linear) measurements to be estimated from ordinal-level scores. Second, RMA enables the examination of changes at the individual person-level, beyond traditional group-level comparisons. Ordinal scores were calculated by summing the score of each item and then transformed into interval-level measurements through RMA. These estimates, termed ‘person locations’, are in log-odds units (logits). For each person location, the RMA also generated a standard error (SE). The raw score transformation to logits and SEs has previously been presented for the NoK-PiC and its two subscales (the CaT and the CiC). RMA has been explained in detail elsewhere. In this study, the RMAs were conducted using RUMM2030 Professional Edition 5.4 (RUMM Laboratory Pty Ltd, Duncraig, Australia). ### Traditional within-group comparisons Within-group comparisons of total ordinal scores were conducted using the paired samples *t*-test and the Wilcoxon signed-rank test. The ordinal data were normally distributed (skewness and kurtosis were within the range of -1 to +1 \[specific range: -0.912 to 0.164\]). Comparisons of raw scores that were linearly transformed to person locations (logits) were made using the paired samples *t*-test. Responsiveness, in terms of the magnitude of change over time, was estimated using the Kazis effect size (ES = mean change score/standard deviation \[SD\] of baseline score) and the standardised response mean (SRM = mean change score/SD of change score), following the analytical approaches used by Hobart et al. (2010). Because there is a lack of uniform and widely accepted criteria to give meaning to the size of an effect, we chose to use the cut-offs suggested by Cohen, although these are based on calculations with the pooled SD. Thus, ES and SRM were interpreted as trivial when the value was \< 20, small when the value was ≥ 0.20 and \< 0.50, moderate when the value was ≥ 0.50 and \< 0.80, and large when the value was ≥ 0.80. ### Modern within-individual (person-level) comparisons Within-individual comparisons were conducted based on the approach described in detail by Hobart et al. (2010) and McCarthy et al. (2012), and the change at the individual level was assessed by computing, for each person, the significance of the individual’s own change in participation in care following the intervention (Sig Change). First, the size of the change for each person was computed (follow-up location − baseline location). Second, the size of the error associated with the change was computed (SE of the difference; SEdiff). Third, the significance of the change for each individual was computed by dividing their change score by their SEdiff. Finally, the significance of each person’s change was categorised into one of five groups according to the size and direction of the significance of their change value. The formulae are as follows: $$\text{Sig}\ \text{Change} = \left( {\text{follow-up}\ \text{location} - \text{baseline}\ \text{location}} \right)/\text{SEdiff},$$ where SEdiff = square root \[(SE baseline location)² + (SE follow-up location)²\]. The categorisation of the significance of each person’s change into five groups was made as follows: - Significant improvement: Sig Change ≥ +1.96 - Non-significant improvement: 0 \< Sig Change ≤ +1.95 - No change: Sig Change = 0 - Non-significant worsening: -1.95 ≤ Sig Change \< 0 - Significant worsening: Sig Change ≤ -1.96 ### Between-group comparisons based on individual person-level outcomes Although the main focus of this study was on within-group and within-individual (person-level) comparisons, between-group analyses were also conducted for comparisons between the intervention and control groups. Between-group comparisons were conducted by counting the number of people achieving each level of significance of change, and the distributions were compared using chi-square tests and relative risks (RR). The RR is estimated as the absolute risk in the intervention group divided by the absolute risk in the control group. A value of one indicates no difference in risk, greater than one indicates increased risk, and a value lower than one indicates decreased risk. The RRs and their 95% confidence intervals were calculated using an online resource: <https://www.medcalc.org/calc/relative_risk.php>. Effect sizes for mean differences between groups with unequal sample sizes within a pre–post control design were also calculated using an online resource: <https://www.psychometrica.de/effect_size.html#cohc>. These effect sizes were interpreted as described above. ## Ethical considerations The KUPA project, including this study, has been approved by the Regional Ethics Review Board in Lund, Sweden (no 2015/69), and the project is registered in the ClinicalTrial database for clinical research (NCT02708498). This study was based on informed consent and guided by the ethical principles for medical research in the Declaration of Helsinki. Information provided before the start of the study presented the aim and design of the study, as well as describing participants’ right to withdraw from the study at any time without suffering any consequences. Both oral and written informed consent was received from each participant before they responded to the questionnaire. The management of the data is in agreement with the General Data Protection Regulation, and the code lists identifying individual study participants are stored in locked cabinets separate from the questionnaire forms. The participants’ confidentiality was respected, and the results have therefore been reported at group level or using non-traceable case numbers. # Results In the intervention group (n = 95), 80 (84%) of the participants completed the CaT subscale, 77 (81%) completed the CiC subscale and 70 (74%) completed the total NoK-PiC scale at both baseline and follow-up. The corresponding figures in the control group (n = 108) were 95 (88%) both for the CaT subscale and for the CiC subscale, and 86 (80%) for the total NoK-PiC scale. Participants for whom a total NoK-PiC score could not be computed because of nonresponse to relevant items at baseline and/or follow-up (n = 47) did not differ significantly at baseline from those for whom these scores could be computed (n = 156) in terms of age or gender. There were no significant differences between the intervention and control groups at baseline on CaT, CiC or NoK-PiC raw ordinal scores. ## Traditional within-group comparisons When comparing ordinal raw scores and interval scores between baseline and follow-up, there were no significant changes in any of the scales within the two groups. All ESs and SRMs could be considered trivial. ## Modern within-individual (person-level) comparisons Despite traditional within-group comparisons both at the level of raw ordinal scores and of linearly transformed interval scores revealing no change in any of the groups, significant changes were seen when using individual person-level comparisons. This is illustrated in, where individual person-level changes are shown for five cases illustrative of: significant improvement (Panel A); non- significant improvement (Panel B); no change (Panel C); non-significant worsening (Panel D); and significant worsening (Panel E). ## Between-group comparisons based on individual person-level outcomes When comparing individual person-level outcomes, there was a difference in CaT between the intervention group and the control group (p \< 0.0005;). More specifically, there was a significantly greater chance for improvement in CaT score in the intervention group (25.0%) than in the control group (6.3%, p = 0.002). CiC also differed significantly between the intervention group and the control group (p \< 0.0005), in both the percentage experiencing significant improvement (31.2% vs. 10.5%, p = 0.002) and the percentage experiencing significant worsening (35.1% vs. 8.4%, p \< 0.0005). There was also a significant difference in the total NoK-PiC scale between the intervention group and the control group (p \< 0.0005). A larger percentage of persons in the intervention group (32.9%) than in the control group (11.6%) experienced significant improvement between baseline and follow-up (p = 0.002;). All ES values between the intervention group and the control group were trivial/small (for ordinal/interval scores respectively, 0.40/0.11 for CaT, -0.11/-0.06 for CiC and 0.02/0.06 for NoK-PiC). # Discussion Based on the significant results on the individual level, the intervention can be regarded as successful within the area of CaT, although the success regarding CiC is more doubtful. However, two contradicting conclusions can be reached depending on which analytical approach is chosen for the analysis of the data and interpretation of the results. First, the ordinal and interval scores showed no change in participation within any of the groups. Second, considering individual person-level change, both significant improvement as well as worsening were revealed in both the intervention group and the control group. In between-group comparisons, there was an improvement in participation with respect to both CaT and CiC in the intervention group, compared with the control group. However, there was also a higher risk of significant worsening in CiC in the intervention group than in the control group. For the total NoK-PiC scale, there was a greater individual person-level positive change in the intervention group than in the control group. This study has three major findings that need to be discussed. The first relates to whether or not the intervention can be regarded as successful. The second relates to the diverse findings resulting from using different analytical approaches for comparison. The third relates to the conflicting findings regarding the CiC subscale. First, when focusing on the individual level of change, the intervention can be regarded as successful for CaT but not as successful for CiC. This finding was somewhat expected because it has been hypothesised that it is easier for care interventions to achieve success in improving CaT than in improving CiC. Previous research has shown somewhat similar results. For example, Maas and colleagues evaluated an intervention including 8 hours of training for staff members over three sessions to promote family participation in nursing homes. The study showed significant improvements in family members’ perceptions of relationships with the staff; however, no differences were found concerning their perceptions of partnership. Another study, conducted by Beck and colleagues, used study circles about palliative care for nurse assistants to implement a palliative care approach in nursing homes. The qualitative study showed that the nurse assistants experienced a deeper understanding of NoK’s needs and stressed the importance of including NoK in the care if they wished to participate. However, the study did not evaluate whether this was actually done in practice, from the NoK’s perspectives. The second important result from the present study relates to the diverse findings when using different analytical approaches for comparisons. Depending on whether we relied on comparisons of ordinal scores, interval scores or individual person-level significant change, we reached different conclusions. This finding raises several questions: What analytical approaches should be used for evaluating the effects of interventions? What are the implications of different analytical approaches for evaluating intervention outcomes in future studies? Are current ordinal score-based standards for evaluating intervention effects, including meta-analysis, not the best path forward? It might be that we need to develop a more individual, person-based evaluation of interventions. As illustrated in this study, this approach can be achieved through Rash measurement-based criteria. Thus, the findings from this study might have implications for the assessment of previous outcome evaluation research and for the analysis conducted in future research. It is possible that many successful interventions have not been discovered because of the predominance of a group- level focus on ordinal score significance rather than a person-centred focus on individual and interval level-based significance. However, we do not have a full explanation for our findings, and there is thus a clear need for further work in line with the analytical approaches described here and in previous work conducted by Hobart et al. ‘to elaborate upon what we have uncovered here and to ultimately pin down its root cause’ (p. 1047). The findings from the present study indicate that standard group-level analyses are limited and possibly misleading. Since individual person-level analytical approaches can unpick the nuances beyond aggregated group-level data, it becomes possible to find out to what extent an intervention is effective, in other words, what works, for whom and in what circumstances. Traditional evaluation efforts that focus on aggregate effectiveness have been criticized for representing an oversimplification. Through individual person-level analytical approaches we can begin to explore why interventions worked for certain persons and not others. This is in line with the recommendation by Pawson and Tilley, developers of the realist evaluation approach, to conduct evaluations for subgroups within programmes; they advised researchers to be cautious since there might be more than one mechanism at work within each subgroup, generating mixed results, which was also revealed by the findings in this study. Realist evaluation is situated between positivism and realism, and it attempts to explore contextual circumstances in which mechanisms are triggered and lead to outcomes. Interestingly though, many realist evaluations, although being neutral on the qualitative-quantitative spectrum, tend to be small-scale, mixed-method or qualitative case studies. One criticism is that they lack generalisability beyond the case study unit of analysis, and sometimes the evaluations are unclear about context, mechanism and/or outcome. The area of outcome assessment tends to be especially problematic. Taken together, these issues can make it hard for realist evaluations to gain scientific credibility. However, this is a point for debate as realism, for some, is not necessarily congruent with generalisability, i.e. neither from a positivism nor an interpretivism point of view. It has been stated that “Prevailing statistical models which by their nature are aggregate /…/ may have limited utility in the analysis of complex systems” (p 388). We claim that modern individual person-level analytical approaches may support larger scale realist evaluations as part of a mixed-methods study design. The third important finding relates to the diverse results regarding CiC. In the intervention group, compared with the control group, there were significantly higher numbers of both persons with significant improvement and persons with significant worsening in terms of CiC. CiC measures collaboration, meaning involving more action from the NoK’s perspective, in contrast to the CaT subscale, which focuses more on the prerequisites for participation. NoK involvement in care results from both the staff members’ expectations and the NoK’s will/ability to get involved. Correspondingly, because NoK functioning as actors in care must be voluntary, it can be more difficult for interventions to achieve high scores on the CiC scale than on the CaT scale. Although many NoK want to participate in the care of their relatives, others experience pressure to take on more tasks than they wish. Thus, it is possible that staff members express a desire to involve the NoK in care as a ‘requested’ indirect side effect of the intervention. This may be welcomed by the NoK, leading to an increase in involvement, but it may also be the case that they expect the staff to involve them more than actually occurs, leading to a decrease in the rating of participation as measured by the CiC scale. Thus, NoK participation in care is a balancing act between NoK maintaining their own responsibility while also ceding responsibility to the nursing home staff. Our findings indicate a possible imbalance between staff expectations and NoK’s will/ability developing over time as a side effect of educational interventions for staff members. A first methodological aspect of the present study that needs to be discussed relates to responsiveness and ES. Scale responsiveness and treatment effectiveness are inseparably linked. ES, computed using change in scores from baseline to follow-up, is an indicator of both the ability of a scale to detect change and the magnitude of the intervention effect. There is, however, a lack of consensus on the most appropriate effect size indicator to use. Further, in matched-pair studies, the cut-offs suggested by Cohen (1977) cannot be used interchangeably for the SRM because of the correlation between scores within pairs. However, because we used ES and SRM mainly for explorative purposes and because low ES was expected, we followed the interpretation suggested by Cohen (1977); this was also necessary because no substantiated alternative exists for the interpretation of ES. SRM was calculated in addition to ES to determine whether the findings were specific to one computation, and this was not found to be the case. A second methodological aspect relates to the understanding of meaningful change from a perspective ‘outside’ the individual level and from an ‘inside’ person- level perspective. It should be noted that, although ES has been used as a proxy for clinically important change, this measure does not provide a complete understanding of the meaningfulness of the observed change. What does the change mean for the respondent? A respondent may perceive even a small change as significant, but this might not be captured by ES or other traditional analytical approaches for group comparison. This aspect of the subjective perception of important change is of course also relevant for the individual person-level of significance. Thus, there seems to be a need for approaches that take more explicit account of the person’s own perceptions of improvement or worsening. Individuals’ preferences are thus essential pieces of information. RMA offers a pathway towards individual person-level analysis of intervention outcomes, especially if it is based on person-centred outcome measures focusing on what individuals themselves perceive as important. # Conclusions Can an educational intervention for staff members focusing on palliative care be recommended for increasing NoK’s participation in care? The answer is both yes and no. If the goal is to improve CaT, the answer is yes. If the goal is also to improve NoK’s CiC, the results of this study provide no clear answer. Different conclusions regarding intervention outcomes can be drawn, depending on which analytical approaches are used for comparisons. Modern individual person- level methods for analysis of intervention outcomes uncover individual significance that cannot be detected by traditional group-level comparisons and would thus otherwise be hidden. The findings of our study have the potential to impact future outcome studies and may also necessitate a re-evaluation of previous studies of this type. We thank the researchers at Lund University who participated in the data collection: Birgitta Wallerstedt (PhD, RN), Helene Åvik Persson (PhD student, PHN) and Anne Molina Tall (research assistant, RN). We would also like to acknowledge the cooperation of Birgit Rasmussen (professor) and Tove Lindhardt (RN, PhD). We also thank Jennifer Barrett, PhD, from Edanz Group (<https://en- author-services.edanzgroup.com/>), and Catriona Chaplin, PhD, CMC Scientific English for Publication (<https://scientificenglish.se/>), for editing a draft of this manuscript. 10.1371/journal.pone.0244600.r001 Decision Letter 0 Selman Lucy Ellen Academic Editor 2021 Lucy Ellen Selman 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 Oct 2020 PONE-D-20-16638 Evaluating participation of next of kin after an educational intervention in nursing homes using methods for individual person-level comparison PLOS ONE Dear Dr. Westergren, 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 review the recommendations and suggestions from the reviewers. In particular please attend to the title and abstract to ensure they adequately reflect the main messages and methods of the paper, and revise the Methods, Results and Discussion sections to ensure consistency.  ​ ============================== Please submit your revised manuscript by Dec 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, Lucy Ellen Selman, BA, MPhil, 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\. 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. \[Note: HTML markup is below. Please do not edit.\] 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: Overall, the manuscript was concise and comprehensive. The author uses two statistical analyses on the data, group, and individual, arriving at different conclusions. This was a unique method to demonstrate significance in alternative results. Also, the author provides explanations as to why results may not have been significant (e.g., next of kin not wanting to take on additional care). However, it may have been beneficial if the author focused on the individual changes instead of comparing group to individual results. It was challenging to determine which overall findings the reader should accept. If this was a psychometric study, it needs to be clearly stated at the beginning of the manuscript. Reviewer \#2: Thank for you inviting me to review this manuscript, which makes the case for individual person-level methods for analysis in the context of an educational intervention in nursing homes. Generally, I found this manuscript to be well written and interesting. There are, however, a couple of exceptions in respect of the clarity of the manuscript. 1\) Upon first reading the title and abstract, I didn’t feel that they adequately describe the paper. For example, it is not clear from the abstract (see objective) what the three methods are, and I think they may be better described as analytical approaches. In my view, a method refers to a means of data collection (e.g. interview or survey), whereas I think the authors are talking about analytical approaches. 2\) The three analytical approaches should be described before the Results section, given this is the key focus of the paper. This only becomes clear in the results and conclusion, and even then not introducing it earlier makes the key focus of the paper unclear in the abstract. 3\) I think the authors should consider naming the three analytical approaches in the title, and then briefly mention and describe in the background and objective sections. This will then mean the manuscript has an obvious and clear focus. 4\) Under Trial registration it says that the manuscript is a psychometric study – but this is the first time it is mentioned. Something perhaps also to consider adding to the title and abstract. As an active researcher in this field I found the rest of the paper to be of significant interest, particularly the development of the new scales/outcome measures (which I may even use myself at some point). One additional point to consider including in the discussion is whether this manuscript can feed into the relatively emerging field of realist methodology and methods. Because the researchers highlight that individual person-level methods of analysis can unpick the nuances behind aggregate/sample level data, my interpretation of this is that this analytical approach is capable of going beyond finding out whether an intervention is effective (as denoted by sample level reporting) and could highlight how it might, for whom and in what circumstances – based on being able to begin to explore why interventions worked for certain people, and not others. This is what realist evaluation and realist research attempts to do (by exploring the contextual circumstances in which mechanisms fire and lead to outcomes). Interestingly though, many realist evaluations tend to be small scale mixed method or qualitative case studies, and one criticism is that they lack generalisability beyond the case study/unit of analysis (though this is a point for debate as realism, for many, is not necessarily congruent with generalisability. There are many strands of realist philosophy!). When reading lines 346-364 I am left wondering whether this manuscript may be capable of adding some interesting discussion to the emerging debates on realist research, particularly in identifying an analytical approach that may support larger scale realist evaluations (as part of a mixed methods study design). \*\*\*\*\*\*\*\*\*\* 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.0244600.r002 Author response to Decision Letter 0 26 Nov 2020 Reply to reviewers We wish to thank the academic editor and reviewers for your positive feedback and relevant suggestions for revisions. Below we respond to each point raised. Academic Editor Please review the recommendations and suggestions from the reviewers. In particular please attend to the title and abstract to ensure they adequately reflect the main messages and methods of the paper, and revise the Methods, Results and Discussion sections to ensure consistency. Response: Thank you for this advice. We have now made changes according to the reviewers’ recommendations, regarding the title, Abstract, Methods, Results and Discussion. Reviewer \#1 R1: Overall, the manuscript was concise and comprehensive. The author uses two statistical analyses on the data, group, and individual, arriving at different conclusions. This was a unique method to demonstrate significance in alternative results. Also, the author provides explanations as to why results may not have been significant (e.g., next of kin not wanting to take on additional care). Response: Thank you for these positive remarks. R1:Q1: However, it may have been beneficial if the author focused on the individual changes instead of comparing group to individual results. It was challenging to determine which overall findings the reader should accept. Response: Instead of deleting the between-group comparisons, which add valuable information when interpreting the findings, we have now clarified that the within-group and individual person-level comparisons are the primary findings. We have also clarified this by adding matching subheadings in the Methods and Results sections. For example (from the Data analysis section): Between-group comparisons based on individual person-level outcomes Although the main focus of this study was on within-group and within-individual (person-level) comparisons, between-group analyses were also conducted for comparisons between the intervention and control groups. R1:Q2: If this was a psychometric study, it needs to be clearly stated at the beginning of the manuscript. Response: Thank you for pointing this out. We have now deleted “psychometric” under the heading “Trial registration”. Reviewer \#2 R2: Thank for you inviting me to review this manuscript, which makes the case for individual person-level methods for analysis in the context of an educational intervention in nursing homes. Generally, I found this manuscript to be well written and interesting. There are, however, a couple of exceptions in respect of the clarity of the manuscript. Response: Thank you for your positive feedback. R2:Q1: Upon first reading the title and abstract, I didn’t feel that they adequately describe the paper. For example, it is not clear from the abstract (see objective) what the three methods are, and I think they may be better described as analytical approaches. In my view, a method refers to a means of data collection (e.g. interview or survey), whereas I think the authors are talking about analytical approaches. Response: Thank you for pointing out this terminology issue. We have now used “analytical approach” instead of “methods”. R2:Q2: The three analytical approaches should be described before the Results section, given this is the key focus of the paper. This only becomes clear in the results and conclusion, and even then, not introducing it earlier makes the key focus of the paper unclear in the abstract. Response: We are grateful to you for this observation. We have now clarified the analytical approaches in the Abstract, Introduction, Methods and Results, for instance by using subheadings. R2:Q3: I think the authors should consider naming the three analytical approaches in the title, and then briefly mention and describe in the background and objective sections. This will then mean the manuscript has an obvious and clear focus. Response: Thank you for this suggestion. To clarify the focus, we have now included these approaches in the title and in various places in abstract, introduction, methods and in results (see also R2:Q2). R2:Q4: Under Trial registration it says that the manuscript is a psychometric study – but this is the first time it is mentioned. Something perhaps also to consider adding to the title and abstract. Response: Thank you for pointing this out. We have now deleted “psychometric” under the heading “Trial registration”, since we do not consider this paper to represent a psychometric study. However, the paper preceding this one was a psychometric study. R2: As an active researcher in this field I found the rest of the paper to be of significant interest, particularly the development of the new scales/outcome measures (which I may even use myself at some point). Response: We are delighted that you found the paper relevant and interesting. R2:Q5: One additional point to consider including in the discussion is whether this manuscript can feed into the relatively emerging field of realist methodology and methods /…/. Response: This was an excellent idea! We have added the following paragraph about this in the Discussion: Since individual person-level analytical approaches can unpick the nuances beyond aggregated group-level data, it becomes possible to find out to what extent an intervention is effective, in other words, what works, for whom and in what circumstances. Traditional evaluation efforts that focus on aggregate effectiveness have been criticized for representing an oversimplification \[32, 33\]. Through individual person-level analytical approaches we can begin to explore why interventions worked for certain persons and not others. This is in line with the recommendation by Pawson and Tilley \[34\], developers of the realist evaluation approach, to conduct evaluations for subgroups within programmes; they advised researchers to be cautious since there might be more than one mechanism at work within each subgroup, generating mixed results \[34\], which was also revealed by the findings in this study. Realist evaluation is situated between positivism and realism \[35\], and it attempts to explore contextual circumstances in which mechanisms are triggered and lead to outcomes \[34\]. Interestingly though, many realist evaluations, although being neutral on the qualitative-quantitative spectrum, tend to be small-scale, mixed- method or qualitative case studies. One criticism is that they lack generalisability beyond the case study unit of analysis, and sometimes the evaluations are unclear about context, mechanism and/or outcome \[32, 36-38\]. The area of outcome assessment tends to be especially problematic \[32\]. Taken together, these issues can make it hard for realist evaluations to gain scientific credibility \[39\]. However, this is a point for debate as realism, for some, is not necessarily congruent with generalisability, i.e. neither from a positivism nor an interpretivism point of view \[40\]. It has been stated that “Prevailing statistical models which by their nature are aggregate /…/ may have limited utility in the analysis of complex systems” \[33\](p 388). We claim that modern individual person-level analytical approaches may support larger scale realist evaluations as part of a mixed-methods study design. 10.1371/journal.pone.0244600.r003 Decision Letter 1 Selman Lucy Ellen Academic Editor 2021 Lucy Ellen Selman This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 14 Dec 2020 Next of kin participation in the care of older persons in nursing homes: a pre–post non-randomised educational evaluation, using within-group and individual person-level comparisons PONE-D-20-16638R1 Dear Dr. Westergren, 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, Lucy Ellen Selman, BA, MPhil, PhD Academic Editor PLOS ONE Additional Editor Comments (optional): Reviewers' comments: 10.1371/journal.pone.0244600.r004 Acceptance letter Selman Lucy Ellen Academic Editor 2021 Lucy Ellen Selman 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-20-16638R1 Next of kin participation in the care of older persons in nursing homes: a pre–post non-randomised educational evaluation, using within-group and individual person-level comparisons Dear Dr. Westergren: 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. Lucy Ellen Selman Academic Editor PLOS ONE [^1]: The authors have declared that no competing interests exist.

# Introduction Duck virus hepatitis (DVH) is an acute, rapidly spreading, highly contagious disease of young ducklings that is characterized by hepatitis. Hepatic injury is the only visible pathological manifestation of this disease. The pathogen was defined as duck hepatitis virus (DHV). DHV includes three distinct serotypes: DHV-1, DHV-2 and DHV-3. DHV-1 was recently classified as a member of the *Picornaviridae* family and renamed duck hepatitis A virus (DHAV) according to sequence analysis. Among these serotypes, DHAV is considered to be the most harmful to the large-scale duck industry all over the world. Although an attenuated DHAV vaccine and a hyperimmune serum are effective, they cannot be widely used and other effective preventive and therapeutic medicines have not been developed. If clinical cases of this disease were to emerge, it would cause irreparable damage. The fact that viral infections can induce a large number of free radicals has been confirmed by many studies, and the injuries caused by free radicals are considered to be one of the pathogenic mechanisms that is widely associated with RNA viruses. Our previous research also demonstrated that a large amount of free radicals are generated during DHAV infection, and the mortality rate from this disease can be effectively improved by reducing free radical levels. Free radicals are widely produced by living organisms and are highly reactive, leading to the peroxidation of lipids, proteins and DNA. Excessive free radical production is one of the driving forces behind hepatic injury, and it plays a critical role in various liver diseases. Studies on polysaccharides and flavonoids, main ingredients in traditional Chinese medicines (TCM), have suggested that both of them possess free radical scavenging abilities. TCM has been used in multiple combinations of compounds in the form of processed natural products used to treat human and animal diseases by both Chinese and other Asian populations for thousands of years. The most striking pathological change induced by DVH is hepatic injury, such as hepatorrhagia and hepatic congestion, and the main clinical symptoms are cramps, convulsions, opisthotonus and sudden death. According to the theory of syndrome differentiation and treatment in traditional Chinese veterinary medicine, DVH is considered to be a syndrome of wind-heat in the liver channel and of body fluids being consumed due to excessive blood heat induced by an exogenous evil source. It is considered proper to clear liver heat, to extinguish liver wind, to cool the blood and to remove toxic materials. Therefore, we used *Hypericum japonicum*, *Radix Rehmanniae Recens* and *Salvia plebeian* to formulate a *Hypericum japonicum*-*Radix Rehmanniae Recens*-*Salvia plebeian* (HRS) prescription to treat DVH. *Hypericum japonicum* refers to the entire *Hypericum japonicum Thunb* herb, which is also named Diercao in China. The entire *Salvia plebeia R*. *Br*. (Lizhicao) herb is used in the TCM system. *Radix Rehmanniae Recens*, the dried root of *Rehmannia glutinosa Libosch*, is also called Shengdi. These drugs have been used singly or in compound drugs in water decoctions to fight hepatitis and other diseases for a long time in Chinese folk medicine, and modern pharmacological studies have confirmed their hepatoprotective effects. It is encouraging that Tianjihuang (*Hypericum japonicum*) injections have been applied in a clinical setting and shown provide notable hepatoprotective effects. With the deepening of this area of research, phytochemical analyses have also revealed that the major bioactive ingredients of *Hypericum japonicum* and *Salvia plebeian* are flavonoids, and that polysaccharides play a major role in the effects of *Radix Rehmanniae Recens*. Meanwhile, many studies have suggested that both the *Hypericum japonicum* flavone (HJF) and the *Salvia plebeia* flavone (SPF) have antiviral and antioxidant effects. Furthermore, it has also been shown that RRRP has an effect against skin oxidative stress in experimental mice. In this work, we compare the anti-DHAV effects of an HRS prescription to the effects of its three individual ingredients *in vitro* and *in vivo*, and we confirm that HRS is a competitive candidate drug to use against DHAV infection. Moreover, the free radical scavenging activity of HRS was also studied to explore its hepatoprotective mechanisms. # Materials and Methods ## Ethics statement All animal experiments were carried out in accordance with the guidelines set forth by the Institutional Animal Care and Use Committee (IACUC) and Nanjing Agricultural University IACUC, and the protocol was approved by the IACUC with the project number 2012GGC15003. All efforts were made to minimize the number of animals used and their suffering. During the entire experimental session in vivo, the ducklings were infected with virus by intramuscular injection only once. The intramuscular injection was one of the most commonly methods used in animal and human without obvious pain, so we did not use analgesics. A humane endpoint was used during the study *in vivo*. To ameliorate suffering, the ducklings were humanely euthanized by CO₂ gas when the typical clinical symptoms of DVH, such as opisthotonos and convulsions, appeared. CO₂ euthanasia was an effective method to ameliorate suffering of the animals which can be used alone. So we didn’t use analgesics and anaesthetics in this process, either. The health status of the ducklings was monitored 6 times per day. During the entire monitoring time, there were no ducklings died as a result of opithotonos or convulsions. During the entire experimental session, the ducklings were carefully nursed to reduce all types of stress, and each process was carried out strictly in accordance with the regulations of the animal protection committee to minimize injuries. ## Reagents and virus DMEM (GIBCO) supplemented with 100 IU/mL benzylpenicillin, 100 IU/mL streptomycin, glutamine (0.75 mg/mL) and 10% heat-inactivated fetal bovine serum was used as nutritive medium to cultivate the cells. For maintenance medium, which was used to dilute the drug and virus and to maintain the cells, the fetal bovine serum concentration was reduced to 1%. Dulbecco's Hanks balanced salt solution (D-Hank’s) was used to wash the duck embryo tissue and to block the cells. The reagent 3-(4, 5-dimethyl-2-thiazolyl)-2, 5-diphenyl tetrazolium bromide (MTT, Amresco Co.) was dissolved at 5 mg/mL in calcium and magnesium- free phosphate-buffered saline. Trypsin (Amresco Co.) was dissolved with D-Hank’s to a concentration of 0.25% and stored at -20°C. These reagents were filtered through two 0.22 μm millipore membranes, and the pH was adjusted to 7.4 with a solution of 5.6% NaHCO₃. DMEM, MM and D-Hank’s were stored at 4°C, and MTT solution was stored in a brown bottle at 4°C. Ethanol, Dimethyl sulfoxide (DMSO, Lot no. 20141013) and other chemicals used in the experiments were analytical grade and were purchased from Guoyao Co., Ltd. Superoxide dismutase (SOD) assay kit (Lot no. 20141211), catalase (CAT) assay kit (Lot no. 20141211), glutathione peroxidase (GSH-Px) assay kit (Lot no. 20141215), malondialdehyde (MDA) assay kit (Lot no. 20141211), nitric oxide synthase (NOS) assay kit (Lot no. 20141216) and total antioxidant capacity (T-AOC) assay kit (Lot no. 20141216) were purchased from Nanjing Jiancheng Bioengineering Institute. The RNAiso Plus Reagent (Lot no. 9108), PrimeScript^™ RT Master Mix Kit (Lot no. AK 3101) and SYBR^® Premix Ex Taq^™ (Tli RNaseH Plus) Kit (Lot no. AK 5303) were Takara products. DHAV (*LQ*_*2* strain) was provided by the Shandong Institute of Poultry. The TCID₅₀ was 1 × 10⁻⁴ according to a Reed- Muench assay. The analytic 50 TCID₅₀ (5 × 10⁻³) was used for antiviral assays *in vitro*. ## Drug preparation The *Radix Rehmanniae Recens* polysaccharide (RRRP) was extracted in our laboratory. The content of RRRP was 84.5% according to the phenol-sulfuric acid method. The crude polysaccharide was purified by the Sevage method and by column chromatography using Sephadex G-200 and it didn’t contain flavone. The *Hypericum japonicum* flavone (HJF, net content of 60%) and *Salvia plebeia* flavone (SPF, net content of 70%) were purchased from Nanjing Zelang Medical Technology Co., Ltd. For HJF extract, its major flavones were quercetin and isoquercetin and it didn’t contain polysaccharide. As the reference component, the level of the quercetin was 10.35% according to HPLC analysis. For SPF extract, its major flavones were homoplantaginin, nepetin-7-glucoside and luteolin-7-glucoside and it didn’t contain polysaccharide. As the reference component, the level of the homoplantaginin in the extract was 9.59% according to HPLC analysis. The proportion of each ingredient in the prescription of HRS was based on the formula theory of traditional Chinese medicine. The composition proportion of net HJF, SPF and RRRP in the prescription was 2:1:2. Based on our previous research, RRRP, HJF, SPF and HRS were diluted to their maximal safe concentrations of 1250 μg/mL, 12.500 μg/mL, 2000 μg/mL and 2500 μg/mL, respectively, in twofold serial dilutions with MM and then sterilized and stored at 4°C. For *in vivo* tests, the RRRP, HJF, SPF and HRS prescription was diluted to 5 mg/mL and stored at 4°C. ## Preparation of duck embryonic hepatocytes (DEHs) DEHs were prepared as previously described in detail by Chen et al. In brief, sterile livers were ground and digested to obtain single cells. Cells were then incubated in a humid atmosphere of 5% CO₂ at 37°C at a seeding density that was adjusted to 0.8×10⁶−1.2×10⁶/mL. The hepatocytes grew into a monolayer and were taken as ready approximately 48 hours later. ## Comparison of anti-DHAV effects of the three single ingredients versus the combined prescription *in vitro* and *in vivo* ### Comparison of antiviral activity *in vitro* RRRP, HJF, SPF and HRS solutions were diluted to five concentrations from prepared concentrations. When DEHs had grown into monolayers in 96-well plates, after being cultivated for 48 hours, the wells in each 96-well plate were separated into three groups: cell control (CC) group, virus control (VC) group and drug group. A 100 μL volume of virus solution was added to each well except the CC wells, to which maintenance medium was added at an equal volume. The supernatant was removed and the cells were washed twice with D-Hank’s after the plates were incubated for 2 hours at 37°C in a humid atmosphere of 5% CO₂. A volume of 100 μL RRRP, HJF, SPF and HRS solution in a series of concentrations was then added to the drug group wells. At the same time, 100 μL of maintenance medium was added to each well of the CC group and the VC group. All plates were then placed into a 5% CO₂ incubator at 37°C. When clear cytopathic effects appeared in the VC group (after approximately 96 hours), the cells were washed twice with PBS, and then the MTT colorimetric method was applied to measure the cytoactivity in each plate. The viral inhibitory rate was calculated based on the formula: Viral inhibitory rate (%) = ($\overset{-}{A}~$_{drug + virus}−$\overset{-}{A~}$_{virus control})/($\overset{-}{A~}$_{cell control}−$\overset{-}{A~}$_{virus control}) × 100%. Antiviral activity among RRRP, HJF, SPF and HRS was evaluated compared to the results of A₅₇₀ and the viral inhibitory rate. ### Comparison of clinical curative effect A total of 210 one-day-old cherry valley ducklings (Purchased from the Chaoyang hatchery, Anhui province, China) that were not inoculated with a DHV vaccine were housed in wire cages (100 cm × 60 cm × 40 cm) in air-conditioned rooms at 37°C with 24 hours of light for the first several days. The temperature was then gradually decreased to room temperature and the light decreased to 12 hours per day. The ducklings were fed a commercial diet provided by the Feed Factory of the Jiangsu Academy of Agricultural Science. On the fourth day, the ducklings were challenged with 0.2 mL of DHAV by intramuscular injection, except for the ducks in the blank control (BC) group, and then randomly divided into RRRP, HJF, SPF, HRS and Virus Control (VC) groups. The ducklings in the separately reared BC group were injected with an equal volume of physiological saline. Two hours later, ducklings in the RRRP, HJF, SPF and HRS groups were administrated with prepared drug solutions at a dosage of 3 mg net drug per duckling by drinking water, once a day for five days. The dosage of drugs was determined based on our previous studies. All procedures involving animals were conducted strictly in accordance with the Chinese legislation for the use and care of laboratory animals, and the experimental use of animals and procedures were approved by the Nanjing Agricultural University Animal Care Committee. To ensure consistency across tests, ducklings in the BC and VC groups were treated with the same volume of solvent-added solution. To evaluate the clinical curative effect of each drug, the mortality rate and the time of death ending were recorded. Because DHAV-induced DVH is an acute disease, the ducklings died quickly once the typical clinical symptoms appeared. Therefore, to ameliorate suffering, the ducklings were humanely euthanized by CO₂ gas when the typical clinical symptoms of DVH, such as opisthotonos and convulsions, appeared. Only when the pathological change was identified as DVH was the death counted. Dead ducklings were executed and disposed of in bio-safety containers in accordance with local standard protocols. The mortality rate in each group was calculated according to the formula: mortality rate (%) = the number of dead ducklings/the number in the sample group × 100%. ## Curative effect of HRS prescription and its hepatoprotective effect ### Animal grouping and treatment Based on results comparing the clinical effects of the HRS prescription to its single ingredients, the HRS prescription was indicated as the most effective drug against DVH. According to the feeding and management procedures described above, a total of 180 four-day-old cherry valley ducklings that were not inoculated with a DHV vaccine were divided into three groups: BC, VC and HRS. The following challenge and treatment procedures are the same as above. At the acute phase (4 hpi and 8 hpi) (hpi: hours post-injection) and recovery phase (54 hpi), blood samples were randomly taken from 5 feathers collected from birds in each group, and these ducklings (15 feathers per group) were not included in the mortality rate. ### Evaluation of hepatic injury In the interest of evaluating the severity of DVH, both visual changes in pathological anatomy and the index of hepatic injury were studied. Once the typical clinical symptoms appeared, the ducklings were humanely euthanized by CO₂ gas and the changes in pathological anatomy in their livers were graded and recorded. The degree was classified on a 0–5 point scale, with the degree increasing with increases in the aggravation of the hepatic injury. A 0 point score meant no pathological changes were found; 1 point: less than 10 hemorrhagic spots or only one ecchymosis appeared on the liver surface; 2 points: the color of the liver appeared essentially normal and the number of hemorrhagic spots was approximately 10 to 100; 3 points: the color of the liver slightly changed to red and one third of the liver surface appeared to have hemorrhagic spots; 4 points: the color of the liver slightly changed to yellowish red and half of the liver appeared to have hemorrhagic spots; and 5 points: the whole liver appeared to be experiencing diffuse hemorrhage, the color of the liver changed to yellowish red and more than 5 ecchymoses were identified on the liver surface. The average score was calculated with the following equation: (0 × the number of survival ducklings + 1 × the number of ducklings scored with 1 point + 2 × the number of ducklings scored with 2 points + …… + 5 × the number of ducklings scored with 5 points). The plasma contents of alanine aminotransferase (ALT), aspartate aminotransferase (AST), albumin (ALB), lactate dehydrogenase (LDH), alkaline phosphatase (ALP) and total protein (TP) in the blood were tested at 4 hpi, 8 hpi and 54 hpi using an Automatic Biochemistry Analyzer (7180 Automatic Biochemistry Analyzer, HITACHI, Japan). To investigate the free radical-scavenging activity of HRS *in vivo*, the plasma levels of SOD, CAT, GSH-Px, NOS, MDA and T-AOC were detected using the appropriate kit according to the kit instructions. ### Relative DHAV gene expression in each group Total RNA was extracted from blood samples with RNAiso Plus Reagent according to the kit instructions. Then, cDNA was synthetized using a PCR machine (2720 Thermal Cycler PCR instrument, Applied Biosystems, America) with a PrimeScript^™ RT Master Mix Kit. The cycling program was 37°C for 15 min, 85°C for 5 s, and 4°C for 7 min. Finally, the semi-quantitative analysis of viral replication was conducted using a RT-PCR machine (StepOnePlus^™ Real Time PCR instrument, Applied Biosystems, America) and a SYBR^® Premix Ex Taq^™ (Tli RNaseH Plus) Kit. The primers for DHAV and β-actin were designed in our previous study. The primer sequences used were as follows: DHAV forward: `5’- GCCACCCTTCCTGAGTTTGT-3’`, DHAV reverse: `5’-TACCATTCCACTTCTCCTGCTT-3’`; β-actin forward: `5’-CTTTCTTGGGTATGGAGTCCTG-3’`, β-actin reverse: `5’-TGATTTTCATCGTGCTGGGT-3’`. The reaction parameters were as follows: 95°C for 30 s, 95°C for 5 s (40 cycles) and 60°C for 30 s. ### Correlation analysis The correlation among hepatic function indexes, peroxidation damage evaluation indexes, mortality rate, viral gene expression levels and hepatic pathological severity scores at 54 hpi were determined by Pearson’s correlation coefficient using SPSS Software Package v.20.0. ## Statistical analysis All data are expressed as the mean ± S.D. Duncan’s multiple range test was used to analyze the difference among groups and a χ² -Test was used to analyze the difference in mortality rates with SPSS 20.0 software. The 2^-ΔΔCT method was used to analyze relative gene expression data. Significant differences were considered as *p* \< 0.05. # Results ## Results from the comparison of anti-DHAV activity by the three single ingredients to the combined prescription, *in vitro* The A₅₇₀ values and viral inhibitory rates on DEHs are presented in. The A₅₇₀ values of the HJF and HRS groups at different concentrations were significantly increased compared with the VC group (*p* \< 0.05). However, only at a concentration of 1000 μg/mL was the A₅₇₀ value of the SPF group significantly higher than that of the VC group (*p* \< 0.05). The A₅₇₀ value of the RRRP group was significantly higher than that of the VC group at a concentration of 78.125–312.500 μg/mL. The highest inhibitory rates of the HJF, SPF, RRRP and HRS groups were 37.0%, 20.2%, 32.6% and 103.9%, respectively. Furthermore, the viral inhibitory rate of the HRS group was generally higher than that of the HJF, SPF and RRRP groups, and the HJF group generally presented a higher viral inhibitory rate than the SPF and RRRP groups. ## Results showing anti-DVH effects in a comparison of the three single ingredients and the combined prescription The mortality rates and the time of death ending in the HJF, RRRP, SPF and HJF groups are presented in. In the VC group, the mortality rate (74.3%) and mortality (26 birds) were both the highest. In contrast, in comparison to the VC group, the mortality (15 birds) and the mortality rate (42.9%) of the HRS group were significantly decreased (*p* \< 0.05). The mortality rates in the HJF, SPF and RRRP groups were also reduced to 57.1%, 62.9% and 60.0%, respectively. The HJF group had a lower mortality rate and was not significantly different compared to the SPF and RRRP groups (*p* \> 0.05). In the overall ranking of the time of death ending, the VC group was the longest lived (up to 144 hours), followed by the RRRP group (120 hours), the SPF group (120 hours), the HRS group (108 hours) and the HJF group (54 hours). ## Results showing the curative and hepatoprotective effects of HRS ### Results showing the curative effect of HRS The curative effect of HRS on DVH is presented in. The mortality rate in the VC group was the highest among the three groups and was significantly higher than that in the BC and HRS groups (*p* \< 0.05). After treatment with HRS, the mortality rate in the VC group was decreased by 22.2%. ### Visual changes in pathological anatomy and scores of hepatic pathological changes Visual changes in pathological anatomy in the BC, VC and HRS groups are illustrated in (A, B and C). As picture C shows, the liver surface was lubricious and no change in pathological anatomy was observed in the BC group. In the VC group (picture B), a large number of hemorrhagic spots (arrow c) and ecchymoses (arrow b) were identified on the liver surface and the color of the liver changed to yellowish red. On the contrary, the pathological anatomy in the HRS group was clearly alleviated, in that they displayed fewer hemorrhagic spots (arrow a) and no ecchymosis were found on the liver surface in this group (picture A). Moreover, the color of the liver of the HRS group was much closer to that of the BC group than in the VC group. The quantitative analysis of average liver injury severity in each group is illustrated in. The liver injury severity score in the BC group was zero, indicating no pathological change. However, the VC group presented the highest average injury severity score, which was significantly higher than that of the BC and HRS groups (*p* \< 0.05). ### Results of hepatic function biochemical indexes shows the hepatic function biochemical indexes in each group. All of these indexes are shown at 4 hpi, including the TP values at the three sampling times, which showed no significant difference among the BC, VC and HRS groups (*p* \> 0.05). At 8 hpi, ALT, AST and LDH levels in the VC group were significantly higher than those in the BC group (*p* \< 0.05). The levels of ALT and AST in the HRS group both decreased compared with levels in the VC group, but they were close to the levels in the BC group. The LDH level in the HRS group was significantly lower than that in the VC group. The levels of TP and ALB were almost the same among the BC, VC and HRS groups. At 54 hpi, significant increases in the ALT, AST and LDH levels were detected in the VC group compared with levels in the BC group (*p* \< 0.05). In the HRS group, the ALT and LDH levels were significantly lower than those in the VC group (*p* \< 0.05), and the AST level decreased but was not significantly different compared to that in the VC group (*p* \> 0.05). The levels of both ALT and AST in the HRS group were the same as in the BC group (*p* \> 0.05). ### Results of evaluation indexes of peroxidation damage The evaluation indexes of peroxidation damage are shown in. The plasma levels of SOD, CAT, GSH-Px, MDA and T-AOC in the VC group were not distinct from those in the BC group; however, plasma NOS activity was significantly elevated in these animals at 8 hpi (*p* \< 0.05). Furthermore, the plasma levels of CAT, GSH-Px, MDA and T-AOC were almost the same between the VC and HRS groups. In addition, the activity of both SOD and NOS was significantly increased in the HRS group compared with those in the VC group (*p* \< 0.05). At 54 hpi, the levels of plasma SOD, CAT, GSH-Px and T-AOC among the three groups showed the same relationship: the HRS group was close to the BC group, and both were significantly higher than the VC group (*p* \< 0.05). The plasma MDA level in the HRS and BC groups was almost the same, and both were significantly lower than that of the VC group (*p* \< 0.05). The plasma NOS level presented no difference among the three groups (*p* \> 0.05). ### Results of relative expression of DHAV in the blood Relative gene expression of DHAV in the blood at 4 hpi, 8 hpi and 54 hpi is listed in. Viral gene expression in the VC group at 4 hpi was set to 1. The DHAV gene expression level in the VC and HRS groups was significantly higher than that of the BC group at all time points (*p* \< 0.05). At 8 hpi and 54 hpi, the viral gene expression level in the VC group was 1.46 times and 1.93 times higher, respectively, and was significantly higher than that in the HRS group (*p* \< 0.05). The rate of viral gene expression inhibition by HRS was as high as 59.3% at 8 hpi and 65.9% at 54 hpi. \[The viral gene expression inhibition rate was calculated using the formula: Viral gene expression inhibitory rate (%) (at the same time point) = (the relative viral gene expression in the VC group—the relative viral gene expression in the drug group)/the relative viral gene expression in the VC group × 100%\]. ### Results of the correlation analysis The Pearson correlation coefficients among hepatic function indexes, evaluation indexes of peroxidation damage, mortality rate, viral gene expression level and hepatic pathological severity score at 54 hpi are listed in. As shown in this list, ALT, AST and LDH were positively correlated with MDA, NOS and viral gene expression level and mortality rate but negatively correlated with SOD, CAT, GSH-Px and T-AOC. The correlation analysis also showed that ALB was positively correlated with T-AOC but negatively correlated with SOD, CAT, NOS, GSH-Px, MDA, viral gene expression level and mortality rate. Moreover, the hepatic pathological severity score was significantly and positively correlated with the mortality rate and viral gene expression (*p* \< 0.01). # Discussion The MTT method is one of the most convenient assays for testing cellular activity in cell cultures, *in vitro*. As an intuitive index used to evaluate the activity of cells, the higher the A₅₇₀ value, the higher the activity and the better the antiviral activity displayed by the drug. The A₅₇₀ values of the HJF, SPF, RRRP and HRS groups in *in vitro* experiments were significantly higher than those in the corresponding VC group at 0.781–12.500 μg/mL, 1000 μg/mL, 78.125–312.500 μg/mL and 156.250–2500 μg/mL (*p* \< 0.05). These results indicate that all of these drugs have significant anti-DHAV effects *in vitro*. The viral inhibition rate as an index directly reflects the antiviral intensity of the drug. In comparison to the highest viral inhibition rate of the four drug groups (HJF: 37.0%, SPF: 20.2%, RRRP: 32.6%, and HRS: 103.9%), the HRS group presented the best result, and therefore had better and stronger anti-DHAV activity *in vitro* than the three single ingredients. To further evaluate the anti-DHAV effects of the three single ingredients and the combined HRS prescription, we tested the curative effect of these drugs in artificially infected ducklings. The mortality rate is the most useful index for evaluating the clinical curative effect. Although the mortality rates in groups given one of the three single ingredients or the HRS prescription groups were lower than in the VC group, only the HRS group showed a significant decrease (*p* \< 0.05). This result was consistent with our *in vitro* results and demonstrated that the HRS prescription conferred a more superior clinical curative effect than any other three single drugs. Therefore, further research is required to determine the anti-DHAV mechanisms that function in the HRS prescription. In this verification experiment, the results from the final mortality rate experiment were consistent with those in the preliminary experiment. The visual changes in pathological anatomy were clearly alleviated after treatment with HRS. The hepatic pathological severity score provided a quantitative analysis index of the liver injury severity: the higher the score, the more serious the injury. The score in the VC group was the highest and was significantly higher than that in the HRS group (*p* \< 0.05), which was consistent with the visual changes observed in pathological anatomy. These results serve as evidence that validates the effective clinical curative effect of this HRS prescription for treating DVH. Moreover, the significantly positive correlation between hepatic pathological severity scores and mortality rates further confirms that the curative effect of HRS against DHV was closely related to its actions that alleviated hepatic injury. The biochemical markers AST, ALT, TP, ALB and LDH are widely used to diagnose many types of human hepatitis. These markers are of great significance in differential diagnoses, curative effect monitoring and evaluations of hepatitis. In this experiment, all of the biochemical markers that were tested showed no significant difference among the three groups at 4 hpi. However, in the VC group, the levels of ALT, AST and LDH were significantly higher than in the BC group at 8 hpi (*p* \< 0.05). This indicates that the hepatic injury appeared at 4 to 8 hours following infection with DHAV and that the biochemical markers AST, ALT and LDH displayed higher sensitivity than the other markers. The decrease in the three markers in the HRS group suggests that HRS effectively protects the liver from injury caused by DHAV infection during the acute phase. Similarly, the AST, ALT and LDH levels in VC group remained the highest among the three groups at 54 hpi. Meanwhile, the level of ALB was significantly lower than that in the BC group. In the HRS group, the levels of AST, ALT and ALB were the same as those in the BC group at 54 hpi. This indicates that the hepatic injury remained serious and that the prognosis of the disease was poor in the VC group ducklings. Conversely, HRS alleviated the injury and had a protective and curative effect during the recovery phase. The correlation analysis also showed that hepatic function indexes (AST, ALT, and LDH) were positively correlated with the mortality rate. These results suggest that hepatic injury was positively correlated with the mortality rate. These results therefore confirm that HRS confers hepatoprotective effect from another angle. Health status and viral gene expression levels are closely related, and a higher viral gene expression level indicates a worse health status. The replication cycle of DHAV includes adsorption, penetration, uncoating, biosynthesis, assembly and release, and requires approximately 6 to 8 hours. In the early period, before 8 hpi, many specific defense mechanisms in the body have not yet been triggered, and the antiviral effect of a drug mainly depends on its direct inactivating activity. Although DHAV gene expression levels in the blood in the VC group was higher than that in the HRS group, there was no significant difference between them at 4 hpi (*p* \> 0.05). This indicates that HRS possesses a faint inhibitory effect against viral proliferation at this time point. After being challenged with DHAV for 8 hours, the blood DHAV gene expression level increased significantly in both the VC and HRS groups, and the level in the VC group was significantly higher than that in the HRS group (*p* \< 0.05). This suggests that the virus finished at least one replication cycle and that a large amount of virus had been released into the blood. HRS presented an effective ability to inhibit the proliferation of DHAV *in vivo*, which was consistent with the experiment *in vitro*. In the recovery phase (54 hpi), the DHAV gene expression levels in the VC and HRS groups were clearly decreased, and the expression level in the HRS group was much lower. This reveals that DHAV was going through the natural process of being gradually cleaned out after peak virus replication. Meanwhile, the correlation analysis results also showed that DHAV gene expression levels were positively correlated with hepatic injury indexes and the pathological severity score. This indicates that when the viral load in the blood increases, more hepatocytes are infected, leading to exacerbation of the hepatic injury. Hence, it is possible that HRS exerts its hepatoprotective effect by inhibiting the proliferation of DHAV. Oxidative stress is one of the negative effects produced by free radicals in the body, and it is believed to be an important factor leading to aging and disease. A large amount of evidences accumulated over the past decades suggested that patients with RNA virus infections are under oxidative stress. In response to this damage, cells protect themselves by using a variety of free radical scavengers, such as SOD, CAT, GSH-Px, and vitamins E and C. T-AOC level is an important and direct index used to evaluate total radical scavenging activity. NO is a relatively short-lived inorganic free radical that is synthesized by the different isoforms of nitric oxide synthase NOS (i.e., nNOS, eNOS and iNOS) and can cause oxidative stress and fight against bacteria, parasites and viruses. MDA is one of the main products of lipid peroxidation, and its elevated level reflects the degree of lipid peroxidation injury. The NOS plasma levels in both the VC and HRS groups were significantly higher than that in the BC group at 8 hpi (*p* \< 0.05), and the level in the HRS group was the highest. Meanwhile, the SOD and T-AOC plasma levels were also significantly increased (*p* \< 0.05). This indicates that NO might play a pivotal role against viral infections, and that HRS enhances its synthesis to more effectively inhibit viral proliferation. However, all the other indexes in the VC group were at the same level as those in the BC group, which might indicate that the metabolism of free radicals was in an almost balanced state, despite the injury. In addition, the elevated levels of plasma SOD and T-AOC suggested that HRS enhanced the free radical scavenging effect. In the recovery phase (54 hpi), all of the indexes were significantly different between the VC and BC groups except NOS (*p* \< 0.05). This suggests that the balance in free radical metabolism had been destroyed, resulting in large amounts of free radicals accumulating in the body. Meanwhile, the correlation analysis results showed that the hepatic function indexes, including ALT, AST and LDH, were positively related to MDA and NOS and negatively related to SOD, CAT, GSH-Px, and T-AOC. This indicates that the large number of free radicals being generated caused the hepatic injury to become aggravated, resulting in significant increases in the levels of ALT, AST and LDH in the VC group. Therefore, it is clear that peroxidation damage is an important contributor to DHAV-induced hepatic injury. The level of NOS in the VC group seemed to bounce back to a normal level, but a large number of other types of free radicals remained, except for NO and T-AOC, which was significantly decreased, suggesting that the free radical scavenging effect was attenuated. It therefore made no sense to change the imbalance. On the contrary, in ducklings treated with HRS, all of the indexes presented as no difference than those in the BC group, suggesting that the hepatoprotective effect of HRS might be related to its antioxidant activity. However, there were dead ducklings after the balance returned and this remains unexplained. Further investigated is therefore required to further clarify these mechanisms. # Conclusion The HRS prescription had superior anti-DHAV activity compared to the three single drugs both *in vivo* and *in vitro*. Research into the curative mechanism suggests that HRS confers a hepatoprotective effect that is closely linked to its antioxidant activity. These results indicate that the HRS prescription can be expected to be developed into a new candidate of anti-DHAV drug. We are grateful to all other staff in the Institute of Traditional Chinese Veterinary Medicine of Nanjing Agricultural University for their assistances in the experiments. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: JL. Performed the experiments: HD SZ MS YXW LZ YC WX JY FY YW DW YH. Analyzed the data: SZ YXW MS. Contributed reagents/materials/analysis tools: JL. Wrote the paper: HD.

# Introduction Peripheral artery disease is a complex vascular pathology characterized by flow limiting obstruction of the peripheral conduit vessels and microvascular dysfunction. Despite limited treatment modalities designed to alleviate conduit vessel obstruction and reduce risk factors associated with endothelial dysfunction (e.g., lipid lowering, plasma glucose levels, blood pressure), the morbidity and mortality associated with PAD remains very high. Critical limb ischemia is the most severe form of PAD and often leads to limb amputation and poor quality of life. Effective treatments to improve limb perfusion are vital to prevent amputation, however these modalities heavily favor invasive surgical and percutaneous treatments which have significant morbidity. Moreover, the limited efficacy of surgical intervention in vascular beds distal to the popliteal artery and the high procedural costs should necessitate the development of non-invasive therapies. A potential therapeutic method is the application of light energy to the skin. Numerous reports have associated light treatment with improved wound healing. In animal studies, polychromatic visible light stimulates vasodilation. Pharmacological inhibition of vessel segments stimulated with polychromatic light or high intensity red light has suggested multiple mechanisms for vasodilation including nitric oxide and the small and intermediate conductance potassium channels. Interestingly, the vasodilation stimulated under these conditions has been endothelial independent. Previously, our laboratory identified energy in the visible red to near infrared spectrum (R/NIR, 670 nm wavelength) could produce nitric oxide within the blood and myocardium from sources such as iron nitrosyl-hemoglobin and nitrosyl- myoglobin. The NO produced from these moieties is significant, as it is cardioprotective in acute ischemia and stimulates angiogenesis. The mechanism of NO generation by R/NIR is independent of functional nitric oxide synthases, which increases its clinical relevance since many patients with PAD have impaired NOS function. Recently, we have shown R/NIR to stimulate vasodilation of physiologically pressurized external carotid vessels in mice through a NO and endothelial dependent manner, which differs from previously reported light stimulated vasodilation studies where vasodilation required inhibition of the small and intermediate potassium channels. Unlike these studies which rely on pharmacological inhibitors of the NO pathway, we sought to characterize the effect of R/NIR on smooth muscle cell ion channel activity. We demonstrate R/NIR increases the open state probability of the large conductance potassium channel. # Materials and methods All experimental procedures and protocols used in this investigation were reviewed and approved by the Animal Care and Use Committee of the Medical College of Wisconsin. Furthermore, all conformed to the *Guiding Principles in the Care and Use of Animals* of the American Physiologic Society and were in accordance with the *Guide for the Care and Use of Laboratory Animals*. ## R/NIR source A 670 nm light emitting diode source was utilized for all pressure myography experiments and patch clamping experiments (NIR Technologies, Waukesha, WI). The patch clamping experiments were performed with a fiber optic cable attached to the R/NIR light source. Power output was measured with an irradiance meter (X97, GigaHertz-Optik). The light sources were placed 2.5 cm from their target in all pressure myography experiments. In vitro, the tip of the fiber optic cable was positioned approximately 5 cm above the bath. ## Pressure myography Segments of facial arteries (160–260 μm ID) or femoral arteries (380–460 μm ID) from C57Bl6/J mice were transferred to a water-jacketed perfusion chamber and cannulated with two glass micropipettes (tip diameter ≈30 μm) at their in-situ length. The arteries were bathed in the PSS-equilibrated solution maintained at pH 7.4 and 37°C. The micropipettes were connected to a reservoir filled with physiological saline solution and the arteries were pressurized to 60 mmHg. The internal diameter of the arteries was measured with a video system composed of a stereomicroscope (Olympus CK 40), a charge-coupled device camera (Panasonic GP- MF 602), a video monitor (Panasonic WV-BM 1410), and a video measuring apparatus (Boeckeler VIA-100). All experiments were performed with the endothelium intact. After a 1-hour equilibration period, the arteries were pre-constricted by ∼50% of their resting diameter with a thromboxane A₂ analog, U-46619. To test the stability of U46619 we measured vessel diameters in a separate group up to 45 min after steady state. Diameters did not change during this test period. The vessels were placed in the dark and once steady-state contraction was obtained; they were treated with 10 mW/cm² of 670 nm light for 5 minutes. The vessels were then allowed to recover in the dark for 10 minutes and then the R/NIR light was reapplied for another 5 minutes. The vessels were then allowed to recover in the dark for another 5 minutes. Passive vasodilator responses to papaverine (10⁻⁴ M) were determined at the end of each experiment. Vascular responses are expressed as percent maximal relaxation of the U-46619-induced constriction, with 100% representing the passive increase in diameter with papaverine. ## Dissociation of mouse femoral arterial smooth muscle cells Femoral arteries were isolated from wild type mouse (8–10 weeks of age) anesthetized with inhalational 2% isoflurane. Isolated femoral arteries were placed in low-calcium dissociation solution composed of (in mM): NaCl 134, KCl 5.4, MgSO₄ 1.2, KH₂PO₄ 0.24, CaCl₂ 0.01, glucose 11 and HEPES 10 (pH adjusted to 7.4 with NaOH). The femoral arterial segments were transferred into 1 mg/ml bovine serum albumin (Sigma) in low-calcium dissociation solution and kept for 10 minutes at room temperature. The arterial segments were then transferred into a low-calcium dissociation solution containing 0.6 mg/ml papain (Worthington) and 1 mg/ml dithiothreitol (DTT) (Sigma) and incubated for 15 min at 37°C. The femoral arterial segments were washed twice with fresh low-calcium dissociation solution and transferred into a 1 ml vial containing 0.5 mg/ml collagenase (Sigma) and 0.5 mg trypsin inhibitor (Sigma) and incubated at 37°C. Supernatant fractions were collected every 5 minutes and diluted to 1 ml with fresh low-calcium dissociation solution and checked for the appearance of dispersed cells under a microscope. The procedure was repeated by incubating the remaining femoral arteries with 1 ml fresh low-calcium dissociation solution. The collected femoral arterial cell suspension was placed on ice and used for patch-clamp experiments within 6 hours. ## BK_ca channel current recordings Single-channel K⁺ currents were recorded at room temperature from cell-attached and excised inside-out membrane patches of freshly isolated mouse femoral arterial muscle cells using the patch-clamp technique. Channel current recording pipettes were fabricated from borosilicate glass pulled on a 2-stage micropipette puller (PC-84) and heat-polished under a microscope (Narishige MF-83 heat polisher). The recording pipettes were mounted on a three-way hydraulic micromanipulator (Narishige, Tokyo, Japan) for placement of the tips on the cell membrane. High resistance seals (greater than 2–3 giga ohm) were established by applying a slight suction between fire-polished pipette (with tip resistance values between 3-l0 megohm) and cell membranes. The offset potentials between pipette and bath solution were corrected with an offset circuit before each experiment. Pipette potential was clamped, and single-channel currents were recorded using Axopatch 200B amplifier and Digidata 1440A digitizer (Molecular Devices, Sunnyvale, CA) at a 50-kHz sampling rate and filtered online at 5 kHz using a low-pass Bessel filter. Single-channel currents were analyzed using a pClamp software package (Molecular Devices, Sunnyvale CA, USA, pClamp version 10.3) to determine event frequency and open state probability. Slope conductance was determined by fitting the unitary current-voltage relation using least- square linear regression. The effects of NIR light on single-channel BKCa single-channel activity was recorded from 1 to 2 independent replicates of cell- attached or excised inside-out membrane patches obtained from femoral arteries isolated from one mice and mean values per mice (animal) were used to obtain a data point. ## Patch-clamp solutions Pipette solutions for both cell-attached and excised inside-out patches contained (in mM): KCl 145, CaCl₂ 1.8, MgCl₂ 1.1, HEPES 5, with the final pH adjusted to 7.2 with KOH. During recording from either cell- attached or inside-out membrane patches the bath solution was composed of (in mmol/L): KCl 145, CaCl₂ 1.8, MgCl₂ 1.1, HEPES 5, ethylene- glycol-bis (β-aminoethyl ether)-N, N, N’, N’-tetraacetic acid (EGTA) 10, with pH adjusted to 7.2 with KOH. This resulted in a calculated final bath \[Ca²⁺\] of 10⁻⁷ M. The bath was contained in a volume of 1 ml that was continually exchanged with fresh solution at a rate of 2 ml/min by gravitational flow. To study the calcium sensitivity of single-channel K⁺ currents recorded from inside-out patches of mouse femoral arterial muscle cells the \[Ca²⁺\] in the recording bath solution was increased from 1 μM to 3 μM. ## BKca knockout mice A breeding pair of heterozygous Slo ^-/+ mice developed by AL Meredith were obtained from the lab of Harpreet Singh (Drexel University College of Medicine). ## PCR genotyping of Slo/Mice Tail snips were digested overnight at 55°C in 750 μl of SNET (20 mM Tris, pH 8, 1mM EDTA, 1% SDS, 0.4 M NaCl) plus 15 μl of 20 mg/ml proteinase K and extracted with an equal volume 1:1 phenol/chloroform. DNA was ethanol-precipitated and 500 ng of DNA (or 2 μl of supernatant) was used in PCR reactions (Promega, Madison WI, supplier’s reaction condition plus 2% DMSO); amplification conditions: 94°C, 2 min; (94°C, 30 s; 55–50°C, 30s; 72°C, 2 min) x 5 cycles; (94°C, 30 s; 50°C, 30 s; 72°C, 2 min) x 30 cycles; 72°C, 5’;4°C, hold). Neo 5’ (5’-`ATA GCC TGA AGA ACG AGA TCA GC`-3’) and RA 14025 3’ (5’-`CCT CAA GAA GGG GAC TCT AAA` C-3’) amplify the Slo ^-/- allele product of 800 bp. Exon 1 5’-3 (5’-`TTC ATC TTG CTC TGG CGGACG`-3’) and WT 3’-2 (5’-`CCA TAG TCA CCA ATA GCC C`-3’) amplify the wild-type product of 332 bp. ## Statistics All values are expressed as mean Standard Error of the Mean (SEM). Comparisons were made using a Student’s t-test, with the exception of a one-way ANOVA with Holm-Sidak post hoc analysis for vasodilation studies performed over time. Values for p\<0.05 were considered significant. # Results ## R/NIR stimulates vasodilation and requires functional potassium channels R/NIR stimulated vasodilation was previously tested with the mouse facial artery, a surrogate for microvascular function. Consistent with previous observations, there was a 14.2% (±0.81) increase in vessel diameter after 5 min of light exposure to pre-constricted facial arteries. Since potassium channels are key mediators of vasodilation, we tested the impact of channel inhibition on NIR vasodilation. Tetraethylammonium (TEA), a selective voltage gated potassium channel inhibitor, abolished light mediated vasodilation (1.4 ±1.31)). This inhibitory effect was also observed when paxilline, a specific inhibitor of the large conductance potassium channel (B_Kca) was administered prior to light treatment (1.6 ±0.47). ## Electromagnetic energy increases voltage dependent K⁺ channel current activity Since increased potassium channel activity is a key mediator of vasodilation, we measured the effect of electromagnetic energy on channel function. Since the facial artery could not yield sufficient dissociated smooth muscle cells, the femoral artery smooth muscle was utilized as a source of cells for patch clamping. To ensure consistency in the physiological responses and the patch clamping current analysis, the response of the femoral artery to R/NIR was tested by pressure myography. As expected, light exposure to the femoral artery elicited a 12.3% change after 5 min (±1.24). The effect plateaued when light was removed and then increased upon re-stimulation with light. There was no significant difference in the magnitude of R/NIR vasodilation in femoral vessels when compared to the facial artery in or previously published observations. Patch clamp K⁺ channel current recording from freshly dissociated mouse femoral arterial muscle cells using symmetrical K⁺ (145 mM) recording solutions revealed existence of a 273 pS single K⁺ channel current that displayed voltage-dependent increased openings and sensitivity to inhibition by the selective large conductance Ca²⁺-activated K⁺ channel (BK_Ca) antagonist paxilline (1 μM). Exposure to R/NIR induced a significant increase in the opening frequency and open state probability (NPo) of this 273 pS single–channel BK_Ca channel current indicating that R/NIR light has the capacity to activate the channel. This activation of BKCa channel by R/NIR was completely attenuated following treatment of the mouse femoral arterial muscle cells with the specific BK_Ca channel inhibitor paxilline (1μM) evidence that this channel is the putative mediator of R/NIR induced femoral arterial vasodilation. To characterize light’s actions on the regulation of BK_Ca channel activity, single channel currents were recorded in excised inside-out membrane patches of wild type femoral arterial smooth muscle cells and the effects of exposure to R/NIR light was examined. As depicted in, no significant increase in the opening frequency and the open state probability (NPo) of the 273 pS BK_Ca single-channel current was observed in response to exposure to 670 nm light. These findings reveal that the mechanism of action of light on the BK_Ca activity is not a direct modulatory effect on the opening of the channel. To further confirm the contribution of BK_Ca channel activity on light stimulated vasodilation, pressure myograph studies were performed on isolated femoral and facial arterial segments of wild type and BK_Ca Slo-1 knockout mice. Exposure to R/NIR light, resulted in a predictable increase in vessel diameter after 5 min in the control group. However, the response to light treatment was blunted in the BK_Ca Slo-1knockout (9.4% SEM 1.4, p\<0.05), when compared with the wild-type genetic control (17.6% ±2.25) suggesting the BK_Ca channel was a significant contributor to R/NIR light induced vasodilation. Patch clamp recordings of single-channel K⁺ currents in SMC isolated from BK_Ca Slo-1 knockout demonstrated the loss of light stimulated activation of the 273 pS single–channel BK_Ca channel current and supported the observed blunted R/NIR light induced vasodilation response in the arterial segments isolated from BK_Ca Slo-1knockout mice. Functional loss of the BK_Ca current was confirmed by examining the vasodilatory actions of the specific BK_Ca channel activator BMS 191011. Treatment with BMS 191011 (3 μM) evoked increased openings of BK_Ca single-channel currents recorded from wild type cell-attached patches but had no effect on those recoded from cell-attached patches of the BK_Ca Slo-1 KO mice isolated femoral arterial muscle cells. Although the effect of light was not completely abolished in the knock-out vessels, there was residual dilation in the BMS 191011 stimulated control. These results suggest genetic deletion of BK_Ca Slo-1 KO can significantly attenuate light induced vasodilation. # Discussion This study convincingly demonstrates that 1) electromagnetic energy (R/NIR light, 670 nm) can significantly dilate the facial and femoral arteries 2) this dilation can be abolished by the inhibition of the large conductance potassium channels with paxilline and low concentration of TEA, 3) Exposure to R/NIR light increases the opening frequency and open state probability (NPo) of a 273 pS BK_Ca single-channel current in wild type mice but not in BK_Ca Slo-1 KO mice isolated SMC, 4) The mechanism of action for electromagnetic energy (R/NIR light) to cause dilation requires functional BK_Ca channel as the induced vasodilation is eliminated pharmacologically using the specific BKCa channel inhibitor paxilline and by genetic knockout of BK_Ca Slo-1. 5) There is no direct action of light in modulating BK_Ca channel current activity in the murine arterial SMC. The action of light energy to modulate ion channel function remains relatively novel. In algae, channelrhodopsins 1 and 2 are established light gated cation channels which are activated at wavelengths in the blue and green spectrum (436 nm to 587nm). Photobiomodulation, the application of low energy visible and near infrared light to tissues, has been attributed to modulation of the TrpV1 and TrpV4 channels, which are non-selective cation channels responsible for modulating neural responses to pain, pressure, and heat. However, in whole cell patch recordings in trp1 expressing oocytes, the activation of the TrpV1 channel by red light was minimal when compared to green wavelength. To our knowledge there is no published evidence describing light as a regulator of BKCa channel function. TrpV4 indirectly activates BKCa channel function by increasing calcium conductance, however TrpV4 activation is limited to light in the blue spectrum. The present study, although limited to light in the far-red spectrum, demonstrates robust activation of the BK_Ca channel current. Vascular reactivity depends on the regulation of smooth muscle contractility through a complex interplay between intracellular potassium and calcium handling. Rising intracellular calcium levels trigger compensatory efflux of K⁺ ions which act as a negative feedback mechanism through repolarization. The large conductance potassium channel (BK_Ca) is the major contributor to re-establishing membrane potential and hyperpolarizing the smooth muscle cell. Its structure consists of an alpha subunit containing 7 transmembrane domains attributed to voltage sensing and ion pore formation and a regulatory beta subunit has two transmembrane domains. The BK_Ca channel is a major end effector of vascular reactivity. Factors which contribute to its modulation include intracellular calcium concentration, membrane potential, and phosphorylation. Nitric oxide is a well described regulator of BKCa channel activity through both direct and indirect mechanisms. Direct activation of the channel by light appears to be unlikely, since exposure to R/NIR light failed to modulate the open state probability (NPo) of BK_Ca single-channel current when recorded from excised inside-out patches of mice. Therefore, the R/NIR light induced modulation of BK_Ca is likely to be indirect and possibly mediated by increases in intracellular NO. The NO/cGMP/PKG regulation of BKCa mediated vasodilation is well described. NO binding to the soluble guanylate cyclase leads to activation of PKG-I and phosphorylation of the channel. Since light treatment increases the open state probability in cell attached patches, the indirect activation of the channel by endogenous production or release of NO is a candidate mechanism of action. This hypothesis will be the focus of future investigation. Although significant cellular NO production is through the enzymatic oxidation of L-arginine to L-citrulline and NO, there is increasing evidence that NO is present in other chemical forms (e.g., nitrite, iron nitrosyl globin compounds, S-nitrosothiols and di nitrosyl iron compounds). There is growing evidence that R/NIR increases NO levels through its action on these compounds. Previously we identified the action of light on S-nitrosothiol compounds, a potential NO source. We have also described the ability of light to release NO from nitrosyl- hemoglobin and nitrosyl-myoglobin in the blood and cardiac tissue. These are likely candidate pools for generating intracellular NO to activate the cGMP/PKG pathway to activate the BKCa channel. There are some limitations to the current study which require discussion. The complete cellular mechanism for R/NIR activation of the BK_Ca channel necessitates further study and our current results suggest indirect action of light. Further experiments will be required to dissect the intracellular mediators activated by R/NIR light to alter channel function. Moreover, the relative significance of light stimulated SMC activation in the larger context of vascular function needs further study. Our previous results using the same vessel type indicates R/NIR induced vasodilation is largely endothelial dependent, yet clearly the BKCa channel is activated in the absence of the endothelial cell further highlighting the need for further investigation. One potential explanation for these differences may exist in the relative abundance of the mediator within endothelial and smooth muscle cells. Light exposure to intact vessels results in exocytosis of S-nitrosothiols into the bath and vasodilation in a cGMP dependent manner. The effects of SMC channel activation by red light can be detected in patch clamp recording but appear less significant when compared to the endothelial derived NO source. It is important to note both the smooth muscle and endothelial pathways of vasodilation share the BKCa channel as an end effector and therefore can be inhibited by paxilline. We have purposefully kept light intensity and wavelength standardized to ensure proper comparisons between the endothelial cell and the smooth muscle cell responses to R/NIR light. A final limitation is the observed residual vasodilation in knock out animals, despite the absence of BK_Ca channel activity, could be regarded to reflect compensation by other ion channels to maintain survival. In summary, 670 nm light can stimulate arterial vasodilation and is markedly reduced by pharmacological inhibition of the BK_Ca channel. Exposure to R/NIR light increases the open state probability of the BK_Ca channel, and this increase is abolished with genetic knockout and by specific pharmacological inhibition of the channel. Exposure to light does not directly modulate the BK_Ca channel current activity, as demonstrated by the lack of effect during recoding from excised inside-out membrane patches. These findings are exciting as they show that R/NIR light energy stimuli can regulate BK_Ca channel activity in arterial muscle cells and assist in our understanding of applying R/NIR light to noninvasive therapies for wound healing and impaired vascular function. # Supporting information The authors thank Grant Broeckel and Dr. Agnes Keszler for editing and figure preparation. [^1]: The authors have declared that no competing interests exist.

# Introduction The tendency for species number to increase with area (i.e., the species-area relationship, hereafter ‘SAR’) is one of the oldest and well-documented patterns in ecology. The pattern holds true for a wide range of taxa and habitats and thus has been referred to as one of ecology’s few ‘laws’. Numerous causal explanations have been offered to explain the SAR, including influence of a variety of deterministic (i.e., species traits) and stochastic (i.e., extinction and colonization) factors; however, there is little consensus about the relative importance of such factors. Traditional niche theory, for example, emphasizes the joint importance of species traits (dispersal ability, niche-breadth, and fecundity) and habitat diversity in generating SARs. In contrast, island biogeography theory ignores the functional importance of such traits and argues instead that SARs arise simply from dynamic colonization and extinction processes. More recently, researchers have called for a more integrative approach that includes both stochastic and deterministic factors in modelling SARs. One integrative approach to understanding factors underlying SARs is to include species traits as an additional parameter in the model (e.g.,). This approach invokes both concepts of niche and neutral theories of island biogeography by incorporating both deterministic and stochastic factors within a single model. Furthermore, inclusion of species traits can improve the predictive power of SARs, while also identifying specific traits or combinations thereof that are important in conservation planning. Comparing slopes for species with different traits, for example, can help to identify species that are most sensitive to changes in area that might occur through habitat loss or fragmentation. While numerous studies have focused on how variation in traits between taxa may influence SARs, fewer have considered trait-specific variation at the species level. Here, we consider for the first time variation in SARs on lake islands in relation to ecologically relevant traits of carabid beetles (Coleoptera: Carabidae). Carabid beetles are well suited to studies of island ecology and have been widely employed in this context. Carabid dispersal is influenced by wing length and wing muscle development, thus effects of differences in dispersal ability on patterns of island occupancy can be studied. Likewise, both habitat use and patterns of reproductive activity vary among carabid species and may explain variation in carabid distributions on islands. Variation in body size has often been studied for island faunas because it influences many characteristics associated with immigration potential, ecological interactions, and resource requirements. Although variation in body size of vertebrates has been relatively well-studied on islands (with differential effects in relation to average body size termed the ‘island rule’, see), there have been fewer studies of variation in body size among invertebrates (but see), and we are unaware of any studies of intraspecific variation in carabid body size on true islands. In the context of habitat ‘islands’ in mainland systems, contiguous (i.e., larger) mature forests generally are characterized as having large-bodied species, whereas more disturbed and isolated (i.e., smaller) habitats are dominated by small-bodied species. It is not well understood, however, whether differences in body size are related to habitat quality within these patches or patch area *per se*. In this paper, we explored how patterns in carabid life-history traits are related to area and isolation of 30 islands in Lac la Ronge; a large (1,413 km²) freshwater lake in the boreal forest of northern Saskatchewan, Canada. We focused on beetle traits that are related to processes of colonization and extinction including body size, wing length, and breeding period. Species-area and abundance-area relationships were considered in relation to these traits and measurements of individual beetles were also analyzed to assess intraspecific variation in body size on the islands. Specifically, we tested the following predictions: 1) species traits modify the slope of species-area and abundance-area relationships; 2) species with large body size, flightlessness, and autumn-reproduction increase in abundance with island area; and 3) body size of carabid individuals increases intraspecifically with island area. # Materials and methods Permission to conduct this study was provided by the Saskatchewan Ministry of Environment. ## Site description Lac la Ronge (55°06’ N, 105°01’ W) is a large (1,413 km²) lake located in the boreal forest of Saskatchewan, Canada. Approximately 10,200–9,000 years BP this region was covered by Lake Agassiz, a huge glacial lake that formed at the margins of the retreating Laurentide Ice Sheet. The current remnant of that lake now called Lac la Ronge sits at the boundary of the Canadian Shield, and is thus surrounded by geologically distinct regions to the north and south. The southern reaches of the lake have few islands with sediments being mainly gravel, sand, and clay. In contrast, rugged igneous and metamorphic bedrock is exposed in the northern and central reaches, giving rise to a collection of \>1,300 islands. These islands are covered mainly by mixed and conifer-dominated forest, including some gaps surrounded by forest, with sand beaches, bogs, and meadows comprising small areas on some islands. Except for a few small cabins, islands are not impacted by human activity. Wildfire has been the primary post-glacial disturbance in the region. Parisien et al. estimated that the fire return interval on the mainland was 99–104 years, but islands in the region have a much longer fire interval. Although no work has specifically addressed the fire interval on the islands of Lac la Ronge, fire mapping in the area since 1980 and the presence of late-successional species, such as *Picea glauca* (Moench) and *Abies balsamea* (Linnaeus) suggest that none of the study islands have burned recently. ## Sampling protocol Carabids were sampled continuously between 2 June– 23 August, 2013 (the approximate frost-free period for La Ronge, Saskatchewan; <http://climate.weather.gc.ca/climate_normals>) on 30 islands, varying in area (0.2–980.7 ha) and distance to mainland (0.1–10.7 km;). We placed eight sleeved pitfall traps (11.2 cm diameter, see) in similar habitats along 120 m transects on each island, for a total of 240 pitfall traps. Transect position and bearing was determined at random and small adjustments to the direction of the transect were made to ensure a minimum distance of 10 m from the shoreline. Traps were positioned along the transect with the first trap located at 7.5 m, and 15 m between subsequent traps to help ensure a more representative catch. Each trap was filled with propylene glycol to a depth of 2–3 cm to kill and preserve captured beetles, and covered by a small plywood lid (15 x 15 cm) suspended from corner posts above the trap to exclude rainwater and debris. Transects were visited a total of five times at 14–17 day intervals to collect samples and replenish the preservative in each trap. Pitfall catches measure activity- density that is generally interpreted as a measure of relative abundance (hereafter called abundance; see). Because traits like body size and wing length are known to vary with the openness of habitat, forest canopy above each transect was measured using the line-intercept (0.1 m minimum increment) method for all woody species \> 1 m height and diameter breast height (DBH) ≥ 4.5 cm. Proportion canopy cover was then calculated as the sum of canopy cover overlapping the transect divided by transect length. Because of forest gaps, canopy cover varied from 28 to 100% among transects; however, this variation was not related to island area (*r* = 0.20, *P* = 0.29). ## Life-history traits Body length was measured as total length (tip of mandibles to apex of elytra) of specimens for each species and taken as a comparative estimate of body size. Preserved beetles were flattened, positioned dorsoventrally next to a ruler, and photographed using a Canon Rebel T3i digital camera and Tamron 90 mm macro lens. Beetle length was calculated using the ruler for calibration in each photo using ImageJ software. To compare abundance and species richness of carabids by body size, species were categorized as ‘small-bodied’ (\< 13.9 mm) or ‘large-bodied’ (\> 14.0 mm) based on median body length for each species. The differences between large and small-bodied species corresponded to a break that divided the smallest and largest specimen most nearly in half (i.e., 3.5–24.5 mm). The following five common species that represent the range of sizes studied were selected to assess intraspecific variation in body size across islands: *Carabus taedatus* Fabricius (median: 21.6 mm), *Carabus chamissonis* (Fisher von Waldheim) (18.0 mm), *Pterostichus punctatissimus* (Randall) (16.6 mm), *Pterostichus adstrictus* (Eschscholtz) (12.1 mm), and *C*. *ingratus* (8.4 mm). A minimum of 20 males and 20 females of each species, chosen at random from each island, were measured, except in cases where fewer specimens in good condition were available. In an effort to better understand the island effect on body size, we also measured a sample for each of the above listed carabid species collected with pitfall traps from mainland sites near Lac la Ronge. Wing length of all specimens was determined by raising the elytra to check wing development, and species were classified with respect to wing length as either (1) macropterous, i.e., hind wings fully developed in all specimens; (2) all specimens flightless, either brachypterous or apterous; or (3) dimorphic, having both macropterous and flightless individuals. Wing dimorphic species were excluded from analyses that required that flight ability was designated to a single category. Carabids of temperate and boreal regions are generally classified as either spring-breeding or autumn-breeding species (hereafter called spring-breeders and autumn-breeders,). Spring-breeders overwinter as adults and reproduce during early summer, while most autumn-breeders overwinter as larvae and complete development the following spring or summer with new adults breeding at earliest in late July. Although this oversimplifies some of the complexity in carabid life-cycles, Zalewski found that autumn-breeders were less common on small islands and suggested that they are more prone to extinction on islands because overwintering larvae are less tolerant to fluctuating environmental conditions than adults. Information about carabid breeding periods was obtained from the literature. Because the life-cycle of *Cicindela longilabris* Say cannot be easily classified, it was omitted from the analysis of seasonal breeding in relation to island area. ## Data analysis Interactions between life-history traits and island area in determining abundance and species richness were analyzed using negative binomial (generalized linear model) and linear regression, respectively. For species richness models, we included distance to mainland as a measure of isolation due to its general importance as a predictor of island diversity. We also included canopy cover (one measure of habitat quality) as a covariate, and used a model selection procedure to determine whether the form of this covariate should be linear or quadratic. We used Akaike information criterion (AIC_c) with small sample correction, choosing the candidate model with smallest AIC_c and thus the largest Akaike weight (*w*_*i*) as the most supported. We then assessed the interaction between life-history traits and island area using results from that model. We included the number of trap days as a covariate to account for minor differences in trapping effort due to lost traps (see). Residuals for models assessing species richness by island area met the assumptions of normality and equal variance. All analyses were conducted in R statistical software. Relationships between individual body size and island area were analyzed for all five species assessed using a linear mixed-effects model (*lme4* package) with island identity used as a random effect. Sex was included as a fixed effect in all models to account for larger body size of females. Statistical significance of the mixed model was evaluated using a likelihood ratio χ² test comparing models that included island area with those that did not. All comparisons of carabid body size relationships between islands and the mainland were based on 95 confidence intervals (CI) with non-overlapping confidence intervals considered to indicate populations of significantly different body size. Confidence intervals were constructed using non-parametric bootstrapping with replacement in which body size for each species was randomly sampled 10,000 times from the original data pool. # Results ## Effects of species traits In total, 10,018 adult carabids representing 37 species were collected on the 30 islands studied in Lac la Ronge. Neither species richness, nor abundance varied with island area (*P* = 0.47, R² = 0.02, and *P* = 0.25, D² = 0.08, respectively). However, when species traits and their interactions with island area were incorporated in the model, variance explained increased to 26–52% and 40–83% for a range of abundance and species richness models, respectively. Furthermore, a significant interaction between species traits and island area revealed that abundance and richness of small-bodied and macropterous species increased with island area, while the opposite pattern was observed for large-bodied and flightless species. A list of carabid species and their associated life-history traits is given in. ### Carabid abundance A linear model of canopy structure provided the most supported explanation of abundance for body size and wing length models, while the non-linear (quadratic) canopy structure model was the most supported for the breeding season model. However, after controlling for the effects of canopy cover, we found clear support for relationships between both body size and wing length and island area in local abundance of species. Models that included body size and wing length were also significant (*P* \< 0.001, D² = 0.52 and *P* \< 0.001, D² = 0.42, respectively) with interactions between body size and island area and wing length and island area both supported (*P* \< 0.001). Abundance of large-bodied, flightless species increased with island area, while abundance of small-bodied, macropterous species was inversely related to island area. The model including breeding season was also significant (*P* = 0.001 and D² = 0.26), although the interaction between breeding season and island area was not significant (*P* = 0.15). Abundance of spring-breeders was inversely related to island area (*P* = 0.004), while abundance of autumn- breeders did not vary with island area (*P* = 0.49). ### Species richness A linear model of canopy cover was the most supported explanation of richness for the body size model, while a non-linear (quadratic) term was most supported for both the wing length and breeding season models of richness. Models including body size and wing length were significant (*P* ≤ 0.001, R² = 0.83, and *P* ≤ 0.001, R² = 0.66), with an interaction supported between body size and island area and wing length and island area (*P* ≤ 0.001). Richness of large-bodied and flightless species increased with island area, whereas the opposite pattern was observed for small-bodied and macropterous species. Similarly, the breeding season model and an interaction between breeding season and island area were significant in explaining richness (*P* ≤ 0.001, R² = 0.38, see). Richness of spring-breeders was inversely related to area (*P* = 0.007), although richness of autumn-breeding species did not vary with island area (*P* = 0.360). Distance to mainland was significant in each of the models relating species richness to island area and species traits. However, contrary to predictions of island biogeography theory, richness of small-bodied (*P* = 0.046), macropterous (*P* = 0.021), and spring breeding (*P* = 0.019) species increased with distance to mainland. This pattern was not observed for large-bodied, flightless, and autumn-breeding species for which richness did not vary with distance to mainland. ## Intraspecific variation in body size Body size was measured for a total of 3,204 individuals from five species. The two smallest species, *P*. *adstrictus* and *C*. *ingratus*, were sufficiently abundant on nearly all 30 islands to provide sufficient samples for the full range of island sizes. In contrast, the three larger species, *P*. *punctatissimus*, *C*. *chamissonis*, and *C*. *taedatus*, were absent from two, six, and 19 of the islands, respectively, and tended to be absent or lower in abundance on smaller islands. For example, aside from a single individual collected on island ‘LG’, *C*. *taedatus* was absent from samples from the 14 smallest islands (≤ 7.5 ha). As is typical for carabids, females were significantly larger than males for each species studied. Body size was similar between islands and the mainland for four of the five species. However, body size varied significantly with island area for the largest species, *C*. *taedatus* (*P* = 0.003, χ² = 8.89, df = 2). However, this pattern was due entirely to variation in size of females (*P* = 0.002, χ² = 9.50, df = 1), which also showed greater range in body size ($\overline{\text{x}}$ = 22.0 ± 0.06 SE mm, 19.58–24.48 mm, *n* = 153) compared to males ($\overline{\text{x}}$ = 21.1 ± 0.07 SE mm, 19.08–22.69 mm, *n* = 122). Finally, female body size in *C*. *taedatus* was significantly larger on the islands (CI \[21.85–22.11 mm\], *n* = 153) than in samples from the mainland (CI \[21.13–21.73 mm\], *n* = 36), although as above, males sizes were similar. # Discussion Neutral theories challenge the traditional understanding that life-history traits are causally related to spatial distributions of species. Recent attempts to include both stochastic and deterministic factors in distribution models have, however, shown promise in providing better predictions of distributions. Our results, along with those from several other recent studies, clearly demonstrate that species traits can be useful additions in species-area and abundance-area modelling. Predictive power of both our species-area and abundance-area models for ground-beetles was significantly improved with inclusion of species traits. Furthermore, the strong interaction between species traits and island area revealed opposing patterns for two of the three traits examined: species richness and abundance of small-bodied, macropterous species was negatively related to island area, while the opposite pattern was observed for large-bodied, flightless species. In fact, we did not detect a positive SAR if species traits were not considered, likely because of the associated divergent response in richness and abundance. Thus, inclusion of species traits in SAR modelling can help reveal complex interactions between species traits and area, as well as reconcile both niche theory and neutral theory in a single framework. Large-bodied carabids were missing from many small islands and generally less abundant when present, with the opposite true for smaller-bodied species. Similar findings have been reported in relation to size of forest patches. Species richness and abundance of large-bodied carabids is also known to vary with successional stage and level of disturbance, with large-bodied species being more abundant in older forests and less disturbed, closed-canopy habitats. Interestingly, we found a residual relationship between body size and island area after controlling for habitat variation associated with canopy cover within mature island forests. Together, these findings suggest that both habitat quality and patch size are important for persistence of large-bodied forest specialists. Local extinction appears to be a generally important process structuring species composition of island communities. Carabid populations are generally short-lived with local extinctions occurring on decadal time scales. Thus, it is possible that even small differences in extinction risk related to body size could contribute to the body size and island area associations observed in our study. Below, we suggest three lines of evidence supporting the inference that large- bodied carabids are more prone to extinction on small islands. First, large-bodied species tend to have smaller populations, even among arthropods and are thus more prone to local extirpation on small islands. In our study, each of the three large-bodied species (*C*. *taedatus*, *C*. *chamissonis*, *P*. *punctatissimus*) collected in sufficient numbers were absent in samples from some small islands, although smaller-bodied species were more widely distributed across the island area gradient. The largest of these species, *C*. *taedatus*, was represented by a single individual among the 14 smallest islands (≤ 7.5 ha), but was collected commonly on ten of the 16 largest islands, suggesting that smalls islands rarely sustain populations of this species. Similar patterns were reported in a large-scale study of collembolans in Europe, in which the four largest-bodied species were absent from the islands, but occurred on the nearby mainland. Second, large-bodied species, such as those mentioned in the previous paragraph, are typically flightless in the boreal forest and thus may be ‘rescued’ more infrequently from local extinction on islands by dispersal events. Immigration of carabids over water indicates that macropters dominate, even in drift material, suggesting that most drifting carabids are blown into the water from flight, and thus drift is less probable for large, flightless species. Our general observations of fewer flightless species on the islands are supported by other studies of carabids in freshwater systems (but see) and may result from occasional local extirpation on small islands with more infrequent re- colonization. Third, we observed that body size in females of *C*. *taedatus*, the largest species examined, varied directly with island area. Studies of intraspecific variation in body size of carabids are limited, although two published examples suggest that beetles from less suitable habitats are smaller: 1) *Carabus nemoralis* declined significantly in body size towards the city center of an urban-rural gradient and 2) four large-bodied species (ca. 21–30 mm) declined in body size in response to forest thinning and disturbance of ground vegetation in forests, although this effect was not observed among two smaller-bodied species (11.5–14.5 mm,). In our study, the relationship between island area and body size in *C*. *taedatus* was strongest in females, possibly due to sex-specific resource requirements associated with breeding. A general relationship between diet deficiency and body size is well-established in insects and in carabids food composition influences egg number and size. In general, female carabids invest in reproduction only after basic energy demands have been met, and because full reproductive potential is rarely achieved, Lövei & Sunderland concluded that food shortages are common. Thus, the smaller body size observed for *C*. *taedatus* females on small islands likely reflects more limited food availability than on larger islands. Of course, this hypothesis assumes that these populations are long-lived enough for such selection to act. Reduced body size on small islands has also been reported for the tenebrionid beetle *Asida planipennis* in the western Mediterranean and for collembolans in island-mainland systems of southern Europe. These studies, together with ours, broaden application of the ‘island rule’ to include some invertebrates. We observed that body size of *C*. *taedatus* was largest on the largest islands and was, in turn, reduced in size on the smallest islands and adjacent mainland. These findings are consistent with Palmer who observed the largest body size of *A*. *planipennis* on islands 11.5 km² in area with a reduction in body size for islands beyond or below this size (see also for discussion on optimal body size). Future studies will help determine if there are patterns of size variation for invertebrates on islands and what factors may be involved in maintaining them. It is important to note that while we considered species traits separately here, they are undoubtedly interrelated. Indeed, much of the focus on body size in ecological studies of islands relates to the influence of body size over other important characteristics of island faunas, such as immigration potential, ecological interactions, and resource requirements. Similarly, a major challenge in studying dispersal relates to the multiple covarying traits (body size, sex, condition, and behaviour) that influence its propensity. It is therefore likely that the effects of body size and wing length act in concert to shape carabid distributions in our study. Contrary to predictions of the theory of island biogeography, we observed a positive relationship between richness and distance to mainland (isolation) for small-bodied, macropterous, and spring-breeding carabids. Although these findings seemed initially surprising, similar increases in species richness with isolation have also been reported for invertebrates in other island systems. We offer two possible explanations for these patterns. First, greater flight potential of macropterous species (which in our study also tend to be small- bodied, spring breeders) allows these species to colonize more distant islands than their flightless counterparts. Second, isolated islands may experience less top-down control by vertebrate predators. For example, Jonsson et al. suggested that predation by birds occurs less frequently on isolated islands because they are suboptimal foraging sites compared to nearby habitats. Although we have no direct evidence of predation on invertebrates, by-catch of small rodents (*Peromyscus maniculatus* and *Sorex* spp.) in our pitfall samples indicates that abundance of these insectivores decrease with distance to mainland (n = 99, r = − 0.36, *P* = 0.05). Predation by small mammals can significantly alter island invertebrate assemblages, thus, it is possible that the increase in richness with distance to mainland is related to fewer insectivorous predators on the most distant islands. Our observation of negative abundance-area and species-area relationships for small-bodied species is also at odds with predictions of the theory of island biogeography. In a related study, we showed that the majority of negative co- occurrences between carabids on these islands involved large and small-bodied carabids. It is generally though that competition does not significantly influence carabid assemblages except through intra-guild predation or at high densities as may be expected on true islands. However, the density compensation hypothesis posits that higher densities on small islands may arise when there are fewer large-bodied species, predators, competitors, or more stable environments relative to large islands. It is therefore possible that higher abundance and richness of small-bodied species on the smallest islands occurs because regulatory control over small-bodied species by their large-bodied counterparts is relaxed on these islands. Zalewski hypothesized that autumn-breeding carabid species that overwinter as larvae are more vulnerable to environmental fluctuations than are spring- breeders. Although we observed an effect of breeding season, this pattern was driven entirely by higher abundance and richness of spring-breeding species on the smallest islands. In contrast, abundance and richness of autumn-breeding species was not influenced by island area, prompting us to reject Zalewski’s hypothesis that autumn-breeding species are more sensitive to less stable environments on small islands. # Conclusions Our study, together with others, suggests that combining aspects of both niche theory and neutral theory provides better explanations for biological distributions than does either theory alone (Kadman and Allouche 2007; Lomolino and Brown 2009; Öckinger et al. 2010; Franzén et al. 2012; Dondina et al. 2016). Large-bodied, flightless carabids were less frequently captured on small islands indicating lower abundance. Although body size did not vary with island area for most species, females of *C*. *taedatus*, the largest-bodied species in our study, were larger on the largest islands. Together, these findings suggest that species with traits like large body size and flightlessness are sensitive to reductions in island area. In contrast, abundance and richness of small-bodied, macropterous, and spring-breeding carabids were inversely related to island area, possibly due to reductions in vertebrate predation and intra-guild predation by large-bodied carabid species on the smallest islands. Overall, our study supports the notion that life-history traits influence the distribution of carabid beetles on lake islands, and that such understanding can be used in accordance with classic SAR modelling to better understand island distributions. # Supporting information We would like to extend our thanks to C. Frost and two anonymous reviewers for helpful suggestions on the manuscript; E. Waite, E. Boyes, B. White, A. Van Meppelen, C. Dubnick, K. Spooner, and B. Hoemsen for assistance in the field; E. Hedlin, G. Oltean, J. Pinzon, and C. Bergeron for help with analyses; D. Larson for his helpful comments throughout this project; and J. Acorn for his encouragement and for providing the high quality photograph of *Carabus tadeatus* used in. Logistical support was provided by the Bell brothers, Mark Duffy, and C. Dallyn of the Ministry of Environment and the Water Security Agency, Saskatoon. [^1]: The authors have declared that no competing interests exist.

# Introduction The master circadian clock, located within the surprachiasmtic nuclei (SCN), orchestrates the cyclic rhythms of many behaviors, tissues, hormones, genes, and other physiological processes. Studies over the last decade have demonstrated that the molecular clock mechanism operates in a cell autonomous fashion in most, if not all, peripheral tissues. Under normal conditions, peripheral clocks remain synchronized to the 24-hour day by the SCN. However, during times of desynchronization, peripheral tissues can become out of phase with the external light:dark cycle, the SCN, and other tissues. Such desynchronization, as occurs during classic restricted feeding and desynchronized feeding (DF), could lead to cardiometabolic health consequences. In both humans and rodents, feeding during the ‘wrong’ circadian time has been implicated in body weight gain. Night-shift workers, who must be awake, active, and eating during the night are at a higher risk for obesity and cardiometabolic diseases. Nocturnal rats placed in a slowly rotating wheel for 8 hours during the day (e.g. a shift-work schedule) show a loss of glucose rhythmicity, an inverted triglyceride rhythm, and an increase in body weight compared to ‘day’ working rats. Even without forced activity, mice eating a high-fat diet only during the ‘wrong’ circadian phase (e.g. the 12-hour light phase) gain 2.5x more weight than mice fed the same diet during the mouse’s natural feeding period. Additional studies have further identified that a high-fat diet consumed during the later portion of the active period, as opposed to the early portion, results in body weight gain and signs of the metabolic syndrome. It is hypothesized that eating during the ‘wrong’ circadian time contributes to circadian desynchronization and increased weight gain. With nocturnal rodents, a desynchronized feeding (DF) protocol uses two conflicting environmental stimuli to create circadian desynchronization: one from the light:dark cycle, which works in unison with the central circadian clock to cause rest and fasting during the light phase; and the other from behavioral feeding, which is experimentally restricted only to the circadian light phase. In this way, DF acts as a metabolic and circadian challenge by restricting access to a palatable high-fat diet to the ‘wrong’ circadian phase, going against cues from the light:dark cycle and the circadian clock to fast and rest. A wild type B6 mouse challenged with DF will gain significantly more weight than a B6 mouse fed only during the normal circadian feeding phase (i.e. synchronized feeding, SF). The mechanisms that lead to excessive weight gain from DF are currently unknown. The present study tests the hypotheses that leptin plays an important role in weight gain during DF and additionally, that the circadian expression of leptin itself influences that weight gain. Since its discovery, leptin has been indicated as an important component of metabolism, reproduction, motivated behaviors, and others. Leptin is secreted by adipose tissue in proportion to body fat amount and relays fat storage information to the brain. When leptin is given peripherally, feeding is reduced, reflecting leptin’s role in feeding behavior and energy balance. Obese individuals have accompanying high leptin levels and are generally considered leptin resistant due to the ineffectiveness of leptin treatment to decrease body weight. Naturally occurring fluctuations in leptin levels occur over the period of a day and are closely associated with feeding and insulin release in both humans and rodents. Notably, when meals are shifted by 6.5 hours in humans, leptin expression also shifts by 5–7 hours in the same direction. Leptin’s tight coupling with feeding could suggest that leptin may be able to communicate meal timing information to the brain in addition to fat storage information. The leptin rhythm itself may also be influenced directly by the circadian clock. Normally, the circadian clock dictates the nocturnal nature of mouse feeding behavior, and feeding in turn influences leptin expression. However, under constant and continuous feeding conditions, a circadian leptin rhythm persists, suggesting an endogenous cycle independent of meal timing. Conversely, long periods of fasting have been noted to lead to a reduction in leptin levels and eliminate the leptin diurnal rhythm. A circadian misalignment between behavior and circadian timing leads to lower overall leptin levels suggesting that leptin responds to the endogenous circadian clock independently of some behaviors like feeding. Additionally, evidence from cultured adipocycte cells suggests the presence of a circadian rhythm of leptin secretion influenced by the endogenous circadian clock of the adipocycte. Leptin has also been found to advance the master circadian clock, the SCN, *in vitro* suggesting a bidirectional relationship between the circadian clock and leptin. Together, these data suggest that the circadian leptin rhythm may provide a bridge between feeding cues, metabolic state, and circadian timing. Despite a wealth of literature describing the natural circadian variation in leptin over the 24-hour day, surprisingly few studies have examined the functional significance of the circadian leptin rhythm for health and treatment (e.g. chronotherapy). To specifically examine the significance of a leptin rhythm, the present study uses the *ob/ob* mouse, on a C57BL/6J (B6) genetic background, which has a mutation in the leptin gene rendering it ineffective throughout the mouse. As a result of the absence of leptin, *ob/ob* mice are hyperphagic, obese, and develop characteristics associated with the metabolic syndrome in humans including elevated glucose, and insulin insensitivity. By using this mouse model with no endogenous leptin, we are able to create a leptin rhythm with peaks occurring at any circadian time by using injections, or conversely, a complete absence of rhythm by using a constant release mini- osmotic pump. In the present study, we use the *ob/ob* mouse to test the hypothesis that non-rhythmic continuous leptin will have differential effects on body weight gain compared to rhythmic leptin. We present data indicating that the presence of leptin alone is not sufficient to lead to weight gain from DF, instead, the presence of leptin and a rhythmic peaking of leptin is necessary for weight gain to occur from DF suggesting that the endogenous leptin rhythm is a possible mechanism leading to excessive weight gains from ‘wrongly’ timed feeding. # Results Here we present results indicating that a circadian rhythm of leptin is necessary for increased weight gain from desynchronized feeding (DF) in mice (e.g. light-phase feeding). As our previous study demonstrated, wild type B6 mice gain excessive weight when fed a high-fat diet during the light-phase (DF) compared to groups fed the same diet only during the circadian dark phase (synchronized feeding, SF). Contrary to wild type B6 mice, leptin deficient *ob/ob* mice do not gain excess weight during DF (experiment 1). Since the primary difference between the *ob/ob* and the B6 mouse is the presence of leptin, we next gave *ob/ob* mice leptin using a continuous release leptin pump (experiment 2). Interestingly, continuous leptin failed to cause excessive weight gain during DF. However, when treated with once daily leptin injections, to mimic a rhythmic leptin profile, *ob/ob* mice respond like wild type B6 mice and gain excessive weight during DF (experiment 3). ## Experiment 1. Leptin deficient *ob/ob* mice fail to gain excess weight during desynchronized feeding To test the hypothesis that leptin may be central to the weight gain seen during ‘wrongly’ timed feeding, leptin-deficient *ob/ob* mice were placed on DF and monitored for body weight, food intake, and activity levels. To control for the length of feeding and fasting periods, DF *ob/ob* mice are compared to synchronized feeding (SF) *ob/ob* mice that have access to food for the same length of time but only during the normal circadian feeding period (e.g. the dark phase). ### Body Weight Body weights of untreated leptin-deficient *ob/ob* were similar under both DF and SF conditions. The timing of the high-fat diet failed to affect body weight gains in leptin deficient *ob/ob* mice (F(1, 14) = 1.07, p = 0.32). Importantly, this result is contrary to what is observed in the wild type B6 where we demonstrate that caloric intake occurring during the ‘wrong’ time of day leads to increased weight gain. These data suggest that the presence of leptin may play a key role in weight gain from ‘wrongly’ timed feeding since the absence of leptin leads to equal weight gains in *ob/ob* mice under both DF and SF conditions. As expected, both DF and SF *ob/ob* mice have a significant increase in body weight over the course of the study (F(14, 196) = 1306.1, p\<0.001) indicating that all *ob/ob* mice gain weight while on the high-fat diet. ### Caloric Intake No differences were seen in overall weekly caloric intake between DF and SF *ob/ob* mice (F(1, 13) = 0.49, p = 0.50). Similarly, no differences were seen in cumulative caloric intake between DF (980±13 kcal) and SF (1002±30 kcal) *ob/ob* mice over the six week period (*t*-Test, *t*(8) = −0.67, p = 0.52). ### Activity Over the course of the six week experiment, no differences were seen in cumulative activity counts (-Test, *t*(14) = 0.51, p = 0.62). However, there was significant difference in the amount of activity occurring during the circadian dark phase: the DF group had significantly less activity during the dark phase than the SF group (*t*-Test, *t*(14) = −2.50, p = 0.03). This decrease in dark activity was not accompanied by a statistically significant increase in light phase activity within the DF group most likely due to the large within group variation (*t*-Test, *t*(14) = 0.85, p = 0.41). ## Experiment 2. Leptin deficient *ob/ob* mice fail to gain excess weight during desynchronized feeding when provided with non-rhythmic, continuous leptin treatment To test the hypothesis that the presence of leptin alone could lead to excessive weight gains from DF in the *ob/ob* mouse, leptin was provided non-rhythmically by a continuous release osmotic pump (described in methods) for two weeks. This length of time was chosen based on previous data indicating that B6 mice on DF gain significantly more weight than SF mice within two weeks. ### Body Weight When treated with non-rhythmic continuous leptin, DF *ob/ob* mice weighed the same as SF *ob/ob* mice suggesting that the presence of leptin alone was not sufficient to lead to weight gain from DF as seen in wild type mice. The timing of feeding did not affect body weight in *ob/ob* mice provided non-rhythmic leptin (F(1, 12) = 2.314, p = 0.15). As expected, a significant increase in body weight was seen in both groups over the course of the study (F(13, 156) = 27.23, p\<0.001) indicating that *ob/ob* mice with non-rhythmic leptin still gain weight while on the high-fat diet. ### Caloric Intake No differences were seen in overall weekly caloric intake between DF and SF *ob/ob* mice (F(1, 12) = 0.42, p = 0.53). Similarly, no differences were seen in the cumulative caloric intake between DF (251±11 kcal) and SF (263±14 kcal) *ob/ob* non-rhythmic leptin treated mice over the two week period (*t*-Test, *t*(12) = −0.65, p = 0.53). ### Activity Over the course of the two week experiment, no differences were seen in cumulative activity counts (-Test, *t*(12) = -0.64, p = 0.53). Similarly, no statistical differences were observed in the amount of activity occurring during the circadian light phase (*t*-Test, *t*(12) = 1.23, p = 0.24) or the circadian dark phase (*t*-Test, *t*(12) = −1.40, p = 0.19). However, as in the previous experiment, the SF group appears to favor being active during the circadian dark phase more than the DF group. These data indicate that the presence of leptin alone is not sufficient to cause excessive weight gain from DF. ## Experiment 3. Excessive weight gain during desynchronized feeding occurs when leptin deficient *ob/ob* mice are given rhythmic leptin treatment To test the hypothesis that rhythmic leptin is necessary to lead to excessive weight gain from DF, we gave a once daily leptin injection to *ob/ob* mice (see methods). In addition to examining the effects of feeding time (e.g. the DF and SF groups), we also examined the timing of the leptin injection. Half of each feeding group were given a leptin injection during the feeding phase to produce a naturally occurring feeding-associated leptin peak (synchronized leptin  =  SL) while the other half of the feeding group were given a leptin injection during the fasting period to produce an abnormally occurring leptin peak desynchronized from feeding (desynchronized leptin  =  DL). Therefore, the four experimental groups (each N  =  8) were as follows: 1. SF-SL, having normally timed feeding and a normal feeding-associated leptin peak, 2. SF-DL, having normally timed feeding and a feeding-dissociated leptin peak, 3. DF-SL, having ‘wrongly’ timed feeding and a normal feeding-associated leptin peak, and 4. DF-DL, having ‘wrongly’ timed feeding and a feeding-dissociated leptin peak. ### Body Weight Only when leptin was given by daily injection did DF *ob/ob* mice gain significantly more weight than SF *ob/ob* mice (F(1, 30) = 9.09, p = 0.01). This suggests that a rhythm of leptin is needed in order to gain excessive weight from ‘wrongly’ timed feeding. Further analysis of the four groups (F(3, 28) = 3.21, p = 0.04) demonstrates that *ob/ob* mice within the most circadian desynchronized DF-DL group gained the most weight and were significantly different from the most synchronized SF-SL group (p = 0.01) and the moderately synchronized SF-DL group (p = 0.04). Additionally, DF-SL mice gained more than SF-SL mice (p\<0.05). When focusing on the weight gains at the end of the two week study (one-way ANOVA, F(3, 28) = 3.55, p = 0.03), DF-DL gained significantly more than SF-SL (p = 0.01) and SF-DL mice (p = 0.01). Additionally, DF-SL mice showed a trend to gain more than SF-DL (p  =  0.08) and SF-SL mice (p = 0.07), suggesting that a desynchronized feeding rhythm leads to more metabolic impairments than a desynchronized leptin rhythm. Interestingly, leptin injection time by feeding-association (DL vs. SL) alone had no effect on body weight gains in *ob/ob* mice (F(1, 30) = 0.71, p = 0.41) suggesting that a desynchronized leptin rhythm by itself would not lead to significant weight change without the additional component of ‘wrongly’ timed feeding. However, the timing of the leptin peak may influence body weight to some degree as represented by the beginning of a separation of body weight gains between the DF-DL and DF-SL groups. As expected, all groups gained weight over the duration of the study (F(14, 392) = 601.42, p\<0.001). There was a significant interaction effect (F(14, 392) = 5.4722, p\<0.001) between feeding phase and study duration suggesting that the DF *ob/ob* mice gain weight differently over the two week period than SF *ob/ob* mice regardless of leptin injection time. ### Caloric Intake Of the four groups, no differences were seen in overall weekly caloric intake (F(3, 28) = 1.50, p = 0.24). Similarly, an one-way ANOVA of the cumulative caloric intake over the two week period failed to show any differences in caloric intake (F(3, 28) = 1.50, p = 0.24) among the four groups (SF-SL 203±9 kcal; SF-DL 214±10 kcal; DF-SL 206±10 kcal; DF-DL 188±4 kcal). Neither feeding time (F(1, 28) = 1.78, p = 0.19), leptin injection time (F(1, 28) = 0.13, p = 0.72), or their interaction (F(1, 28) = 2.60, p = 0.12) had an effect on caloric intake. ### Activity Over the course of the experiment, no differences were seen in cumulative activity counts (F(3, 28) = 0.02, p\>0.99). Similarly, no statistical differences we observed in the amount of activity occurring during the circadian light phase (F(3, 28) = 0.50, p = 0.68) or the circadian dark phase (F(3, 28) = 1.45, p = 0.25) among the four groups. However, when comparing DF and SF groups, a trend was noted for an increase in circadian dark phase activity (F(1, 30) = 4.01, p = 0.05) among the SF groups. Together, these data indicate that a leptin rhythm is necessary for the excessive weight gain from DF. # Discussion ‘Wrongly’ timed feeding, as from DF, has been previously described to lead to excessive weight gains in rodents. The mechanisms leading to this feeding-time dependent weight gain are unknown but we hypothesize that the endogenous leptin rhythm itself may influence weight gain due to its interconnections with body weight regulation, , feeding behavior, and circadian rhythms. In addition to the feeding rhythm (DF vs. SF) in the present study, we also integrated differences in feeding time with the rhythmic or non-rhythmic behavior of the metabolic hormone leptin. Notably, we examine two circadian aspects of leptin expression: 1) rhythmic vs. non-rhythmic expression and 2) synchrony vs. desynchrony with feeding behavior. Within normal, healthy humans and rodents, leptin is naturally rhythmic and associated with feeding time (equivalent to the SF-SL group in experiment 3). In obese individuals, leptin levels are elevated but the rhythm of leptin remains relatively unchanged with the exception of a reduction in amplitude. Wild type mice on DF (equivalent to the DF-SL injection group in experiment 3) however have a significant change in peak phase (12 hours) and a lower leptin trough compared to SF mice. This scenario may be similar to a human shift-worker who must be active and eating during the ‘wrong’ circadian phase. Periods of desynchronized leptin (DL) are much less common in the mouse and human population, but may occur during small windows of entrainment time, such as when adjusting to a new feeding schedule or when experiencing jet-lag. In experiment 1 and 2, we focus on the rhythmic vs. non-rhythmic expression pattern of leptin. Our data indicate that DF does not lead to excessive weight gains in animals with non-rhythmic leptin profiles, including untreated, leptin- deficient *ob/ob* mice and continuously leptin-treated *ob/ob* mice. In a follow-up study, giving continuous leptin after six weeks of DF, replicated these results, indicating that continuous leptin treatment also does not lead to differential weight loss during ‘wrongly’ timed feeding. Treating *ob/ob* mice with continuous leptin essentially returns the *ob/ob* mouse to the wild type state with one major exception – continuous leptin treatment does not mimic the natural circadian peaks and troughs within the leptin expression profile. Indeed, B6 mice on DF have leptin peaks associated with feeding behavior as opposed to the light:dark cycle. This association with meal timing has been previously shown in both rodent and human studies. These data from experiments 1 and 2 suggest that the presence of leptin alone is not sufficient to lead to excessive weight gains from ‘wrongly’ timed feeding. Given that non-rhythmic leptin failed to increase weight gain during DF, we next examined if the presence of a leptin rhythm could lead to increased weight gains during DF. To create a leptin rhythm in the *ob/ob* mice, DF was used in combination with once daily leptin injections (experiment 3). When rhythmic leptin was present, *ob/ob* mice gained excessive weight during DF. As the endogenous leptin profile of wild type mice is also rhythmic, it is possible that the circadian expression of leptin provides additional cues leading to excessive body weights from DF. The data from experiment 3 indicate that a rhythmic profile of leptin is needed to induce weight gains from DF. As an alternative explanation, these data also suggest that rhythmic leptin administration is less effective at preventing weight gain than continuous leptin administration. Indeed, *ob/ob* provided continuous leptin gain less weight over time than *ob/ob* mice given the same amount of leptin in a single daily dose. Since leptin naturally peaks in association with feeding time, we additionally examined synchronous vs. desynchronous leptin rhythms in relation to feeding time, to determine if this level of desynchrony could also affect body weights during DF. We hypothesized that if ‘wrongly’ timed feeding could lead to weight gain, perhaps ‘wrongly’ timed leptin peaks would lead to exaggerated weight gain. *Ob/ob* mice fed in the dark and receiving a leptin injection during the dark (SF-SL) would closely mirror a normal wild type mouse which has its feeding behavior synchronized to the light:dark cycle and its leptin rhythm synchronized to the feeding behavior. At the opposite extreme is the DF-DL group that was fed exclusively during the ‘wrong’ phase (e.g. light phase) and received leptin injections dissociated from the feeding time (e.g. during the dark phase). Intriguingly, these two groups showed the most profound differences in body weight with the DF-DL mice gaining the most weight and the SF-SL mice gaining the least amount of weight over time. These data suggest that minimal circadian disorganization can lead to optimal body weight regulation whereas maximal circadian disorganization can lead to the greatest impairments in body weight regulation (summarized). While leptin in known to acutely inhibit food intake in rodents, the data from experiment 3 indicate that leptin’s inhibitory effects are not leading to the observed body weight differences between the groups. Indeed, if the acute effects of leptin were primarily affecting body weight more so than circadian timing effects, one would expect similar body weights between the groups receiving leptin during their feeding phase (e.g. SF-SL and DF-SL groups). However, we observe increased body weight (p\<0.05) in mice eating during the light phase (and receiving leptin at that time) than in mice eating during the dark phase (and receiving leptin at that time). Statistical analyses of the two variables “feeding synchrony” and “leptin synchrony,” indicate that only feeding synchrony will lead to differential weight gain during DF. This suggests that circadian mistimed feeding will lead to a greater impairment of body weight regulation than a mistimed leptin peak. This in conjunction with experiments 1 and 2 suggests that while a rhythm of leptin is necessary to induce increased weight gains from ‘wrongly’ timed feeding, the actual time of the leptin peak is not essential. This is intriguing and mirrors other rhythmic hormones such as GnRH, which is known to have little signaling effect when given continuously but signals robustly when given rhythmically, regardless of the circadian time of day that rhythm is given. It is interesting to note that the mice receiving the once daily leptin injection consumed less calories but gained more weight than the mice receiving the constant release leptin pump. This relative decrease in caloric intake may be due to the presence of the injection (of either leptin or saline) during the feeding phase which leads to increased arousal, exploratory activity, and other stereotypical behaviors including an acute affect on feeding. The relative increase in body weight within the leptin injection mice suggests a different utilization of calories within the leptin injection mice, one that intensely favors weight gain and possibly fat storage. The rising and falling of leptin levels over the 24-hour day trigger fluctuations in other feeding hormones such as ghrelin as well as fluctuations in leptin receptor (e.g. the long-form of the leptin receptor, Ob-Rb) expression which communicate circulating leptin levels to the brain and in turn influence feeding, circadian rhythms, and also feedback on leptin itself. During a short fast, leptin levels decrease and leptin receptor number and expression are up- regulated helping to promote feeding. With increasing leptin levels, as occurs during re-feeding, leptin receptor number and expression are then down- regulated. We hypothesize that by using a subcutaneous low-dose constant release leptin pump, leptin receptors fail to receive the necessary fluctuations required for leptin receptor number and expression regulation and therefore remain in a down-regulated state which favors body weight maintenance and negative energy balance. We hypothesize that the fluctuations in leptin receptor number and expression are necessary for the increased weight gains observed in mice feeding during the ‘wrong’ circadian phase. Indeed, *ob/ob* mice having no leptin or receiving non-rhythmic leptin do not gain increased body weight from ‘wrongly’ timed feeding. Within a wild type mouse, for example, fluctuations in leptin and leptin receptor number and expression naturally occur coincident with feeding, however when leptin fluctuations occur during ‘wrongly’ timed feeding increased weight gain follows, perhaps through interactions with food-driven mechanisms and/or other circadian-driven mechanisms. Indeed, other research has hypothesized that a major role the leptin circadian rhythm may be to maintain a set threshold of leptin to order to prevent excessive caloric intake. Our results also indicate that despite both DF and SF mice receiving the same length of fasting (12 hours), fasting during the dark phase (and feeding during the light) leads to a greater drop in leptin than fasting during the light (and feeding during the dark;). This suggests that the circadian timing of feeding/fasting has an effect on leptin rhythms and moreover that mistimed feeding rhythms, in relation to the light:dark cycle, will result in an increased drop in fasting leptin levels which could lead to an increased up- regulation in leptin receptors, thus favoring positive energy balance and weight gain. In summary, feeding during the wrong circadian phase leads to a greater drop in leptin during the fasting period, which in turn may lead to greater fluctuation and up-regulation of leptin receptors and promote a positive energy balance. The circadian expression of leptin appears important for the cycling of hormonal, behavioral rhythms, and health and this may be due in part to daily fluctuations in leptin receptor number and expression. It is important to highlight that the *ob/ob* mouse used in this study is, by itself, a metabolic model that develops without leptin and becomes grossly obese. However, given the importance of carefully and accurately controlling the leptin rhythm, we chose to use the *ob/ob* mouse in order to experimentally control the leptin rhythm completely. To alleviate some of the expected obesity in this model, *ob/ob* mice began the experiment at 6-weeks of age when their body weights were relatively reduced (∼30g). However, it is possible that confounding metabolic factors present in the *ob/ob* mouse may lead to misinterpretation of the data. Therefore, additional experiments using wild type mice or another non-metabolic model would make ideal follow-up studies. To our knowledge, these data present the first evidence that leptin has differential effects depending on a peak in leptin levels. One previous study did examine the treatment effects of a leptin injection given in the morning, evening, or both in the morning and evening in obese humans. However, no differences were seen in weight loss among the groups. Importantly, this study used obese individuals, a population which is now mostly considered to be leptin ‘resistant’. To our knowledge, no study has examined the role of the circadian leptin rhythm in non-obese individuals’ health and our study is the first to present data which suggests that leptin’s once daily peak in expression may be an important signal for body weight regulation particularly involved in ‘wrongly’ timed feeding. It is interesting to integrate our results from an evolutionary prospective. It would be maladaptive, for instance, for a nocturnal animal to be awake and seeking food during the daytime when predators are readily available. But if a nocturnal animal critically requires calories, it may override the circadian clock (therefore being awake and eating during the ‘wrong’ circadian time) and it would then be adaptive to utilize the forged calories in such a way so to limit future food seeking behavior during the ‘wrong’ time. This can be accomplished by storing the calories and increasing body weight. An increase in body weight is precisely what happens during desynchronized feeding and it could be the endogenous leptin rhythm coupled to the feeding time that promotes such a response. Perhaps once a beneficial response promoting fitness, body weight gain from ‘wrongly’ timed feeding now has detrimental implications for many humans consuming a significant portion of their calories during non-ideal circadian times, including non-breakfast eaters and night-shift workers both of which an increased risk for weight gain and obesity. # Materials and Methods All experimental mice were housed and handled according to the Federal Animal Welfare guidelines. All studies were conducted at Northwestern University using procedures approved in advance by the Institutional Animal Care and Use Committee, university assurance number A3283-01. ## Animals Male *ob/ob* mice were ordered from Jackson Laboratory (Bar Harbor, ME) at 4–5 weeks of age and immediately singly-housed within temperature, light, and humidity controlled light boxes on a 12 hour light, 12 hour dark schedule. Mice begin experimentation at 6-weeks of age after having one week of environmental acclimation. On average, *ob/ob* mice were ∼30g when beginning the experiment. Male C57BL/6J (B6) mice were bred at Northwestern University and began experimentation at 10–11 weeks old within temperature, light, and humidity controlled light boxes on a 12 hour light, 12 hour dark schedule. On average, B6 mice were ∼24.5 g when beginning the experiment. ## Diets All mice were initially fed a standard chow (LabDiet \#5K52, JL Rat and Mouse/Auto 6F, Arlington Heights, IL distributor) of which 16% of the calories are from fat until the start of the desynchronized feeding protocol. After the start of desynchronized feeding, mice were fed a high-fat diet in which 60% of the calories are from fat (Research Diets \#D12492, New Brunswick, NJ). A subset of *ob/ob* mice were allowed continuous free-access to food and were maintained on the standard chow diet in order to limit weight gain during ad libitum feeding. *Desynchronized Feeding*. Mice were placed onto a restricted feeding protocol as previously described. Desynchronized Feeding (DF) mice had unlimited access to the high-fat diet during the 12-hour circadian light phase only, during the 12-hour dark phase no food was provided. Synchronized Feeding (SF) mice were given the opposite feeding schedule, receiving unlimited high-fat diet during the 12-hour dark phase but no food was provided during the 12-hour light phase. At all times, all mice had free access to water. Feeding conditions were maintained by manually switching each mouse every 12 hours between the individual’s habituated light cage and the individual’s habituated dark cage. ## Leptin pumps 14-day constant release leptin pumps (Alzet mini-osmotic pump \#1002, Cupertino, CA) were filled with a 100ug/kg/day leptin solution containing 0.9% saline solution. Pumps were activated in a warm saline bath for 12 hours prior to implantation. Pumps were implanted subcutaneously into isofluane-anesthetized mice in the dorsal interscapular region just prior to their designated feeding phase. Previous studies have indicated that an osmotic pump delivers leptin stably and consistently. Leptin pumps were surgically removed 14 days after implantation. ## Leptin injections Subcutaneous leptin injections to the dorsal interscapular region were given once daily at either ZT4 (i.e. four hours after light onset) or ZT16 (i.e. four hours after dark onset) at the same dose as the leptin pumps, 100 ug/kg/day. Both the dose and the timing of the leptin injections were chosen to mirror the physiological leptin levels of wild type mice fed the same SF or DF protocol. The single 100 ug/kg/day leptin dose was determined sufficient to slow weight gain while on a high-fat diet, to be absent from circulating plasma 4–6 hours post injection, and to elevate leptin to wild type physiological levels. Since DF wild type mice experience the highest physiological levels of leptin from ZT4-ZT10 and SF wild type mice experience the highest physiological levels of leptin from ZT16-ZT22, ZT4 and ZT16 were chosen as the injection times to elevate leptin in *ob/ob* mice to wild type levels during the same 4-6 hr period similarly fed wild type would experience a physiological leptin peak. A saline injection was given at the counter ZT hour to control for the stress of injection at both ZT4 and ZT16. It was previously determined that the timing of the leptin injection alone while on *ad libitum* standard chow feeding did not lead to differences in body weight. *Activity*. Overall activity was recorded continuously using a one-dimensional, three-beam infrared (IR) monitoring board. Beam breaks were recorded and data analyzed using an online recording program (*Clocklab*, Actimetrics, Wilmette, IL). ## Statistical Analysis Unless otherwise stated, data were analyzed using a repeated measured ANOVA with the aid of a statistical program (StatsSoft Statistica, Tulsa, OK). ## Experiment 1 *Ob/ob* mice were randomly separated into two groups: DF (N  =  8) and SF (N  =  8). Each group continued on the designated feeding schedule for six weeks during which time body weights were taken regularly at ZT0 and ZT12 corresponding to food onset and offset. Caloric intake during the light and dark period was measured weekly by manual food removal. ## Experiment 2 *Ob/ob* mice were randomly separated into two groups: DF (N  =  7) and SF (N  =  7). Before the onset of restricted feeding, mice were implanted with a leptin pump 24 hours prior to the first high-fat feeding period. Each group continued on the designated feeding schedule for two weeks during which time body weights were taken regularly at ZT0 and ZT12 corresponding to food onset and offset. Caloric intake during the light and dark period was measured weekly by manual food removal. ## Experiment 3 *Ob/ob* mice were randomly separated into four groups: SF-SL (N  =  8), SF-DL (N  =  8), DF-SL (N  =  8) and DF-DL (N  =  8). Mice begin leptin injections 24 hours prior to the onset of restricted feeding. All groups continued on the designated feeding schedule for two weeks during which time body weights were taken regularly at ZT0 and ZT12 corresponding to food onset and offset. Caloric intake during the light and dark period was measured every 48 hours by manual food removal. ## Supplemental Experiments To determine endogenous leptin levels, wild type mice were randomly separated into two groups: DF (N  =  6) and SF (N  =  6). After two weeks of the designated feeding schedule, plasma blood leptin was measured at six points equally spaced over the 24-hour day (ZT 2, ZT 6, ZT 10, ZT 14, ZT 18, and ZT 22). To determine if constant, non-rhythmic leptin leads to differential weight loss during DF, *ob/ob* mice from Experiment 1 were then implanted with a leptin pump after six weeks of DF and then maintained on the DF feeding schedule for an additional two weeks. Body weights were measured daily for the first week and then every 2–3 days. The leptin pump was surgerically removed after two weeks and body weight gain was measured for an additional week after leptin removal. To determine if the low-dose leptin injection (100 ug/kg/day) was sufficient to slow body weight gain in *ob/ob* mice, *ob/ob* mice were administered the low- dose of leptin injection (100 ug/kg/day; N  =  5), a high-dose of leptin injection (1 mg/kg/day; N  =  4), or saline (N  =  4) over the course of one week. Body weights were measured once a day just prior to the leptin injection at ZT 11. To determine the rate of clearance of the low-dose leptin injection in *ob/ob* mice and if the leptin injection time lead to body weight differences independently of temporal feeding restriction, *ad libitum* fed, ob/ob mice were injected with the low-dose leptin injection at either ZT 5 (N  =  7) or ZT 17 (N  =  6) for two weeks. Body weights were taken daily, 12-hours after the leptin injection. Plasma leptin was measured on the 11^th day from 1-5 hours after the leptin injection. The authors would like to thank Ms. Carolyn Goldschmidt for her help with data collection and animal care. [^1]: Conceived and designed the experiments: DMA MHV FWT. Performed the experiments: DMA. Analyzed the data: DMA. Contributed reagents/materials/analysis tools: FWT. Wrote the paper: DMA. [^2]: The authors have declared that no competing interests exist.

# Introduction Accurate information on overall and cause-specific mortality is essential to prioritize the activities of health systems and to efficiently invest scarce public health and medical care resources. The availability of such information is also important for epidemiologic studies. The standard method to determine the cause of death is certification by an attending physician, based on valid medical documents, but this approach may yield unreliable results in many low- and middle-income countries, especially in rural and suburban areas. This is mainly due to the lack of infrastructure and the high cost of collecting the data, which limit access to information from diagnostic tests and post-mortem pathology services. Mortality data from these countries are therefore limited and potentially biased. One relatively simple and low-cost alternative for determining a person's cause of death which is available in most low-resource countries is the so-called verbal autopsy (VA). The VA methodology was first developed for investigating epidemics and was later used for evaluations of outcomes of specific interventions, and national mortality surveillance systems, principally in low-income countries such as India. Several studies have shown that VA gives more valid causes of death than routine death certificate data in many developing countries. In VA, a trained interviewer ascertains the symptoms, signs and events during the period leading up to death from family members or primary care givers of the deceased. This information is analyzed to derive a probable cause of death. The most commonly used method for analysis of the collected information is a “physician's review”, generally performed by more than one physician. Other methods, such as algorithms that can be applied by computer, have been tried but found to lack validity. During 2004–2008, the Golestan Cohort Study (GCS) enrolled more than 50,000 adults in Golestan Province, in northeastern Iran, following a pilot study. Golestan is a low-resource area of the country, and consequently, reliable clinical data are not available to determine the causes of death of the residents. Thus we have applied the VA method as a tool to identify the causes of death in the GCS. It is estimated that about 60% of the GCS participants will die at home, and some of them will not have any medical records accurately documenting their cause of death. VA represents an appealing approach to determine the cause of death in this group of subjects. However, it is necessary to validate the VA questionnaire in this adult population. The majority of VA validation studies have focused on neonatal and childhood mortality,. Only a few studies have investigated the validity of VA in adults. Although VA is prone to erroneous estimates of cause-specific mortality rates due to misclassification, several studies have demonstrated its ability in valid identification of the most common causes of death in many settings. And even those who think VA is an imprecise tool for detecting the leading causes of death suggest that in the absence of other more reliable methods, VA may be useful as a secondary tool to determine causes of death in rural areas. Our study is the first attempt to validate an adult VA questionnaire to be used in a longitudinal study in a medium income country. # Materials and Methods ## The Golestan Cohort Study The methods of the Golestan Cohort Study (GCS) have been previously described in detail. In brief, 50,045 adult middle-aged individuals were enrolled in eastern Golestan Province, Iran between January 2004 and June 2008. Participants are actively followed through annual telephone contact to ascertain their vital and health status. If a participant cannot be reached, family members, friends, or local health workers are contacted. Moreover, local health workers in rural areas, called “Behvarz”, are contacted monthly to inquire about any possible outcomes, including death. In the event of death, the follow-up team performs two main tasks in parallel. First, a trained general practitioner goes to the homes of the family members or primary care givers of the deceased and conducts a VA interview. Second, the team determines which physicians or hospitals were visited by the decedent and obtains all medical documents (charts, X-rays, pathology reports, etc) that could be used to identify the cause of death. These documents may be available in Golestan or in neighboring provinces. The GCS follow-up team uses the adult VA questionnaire originally developed by World Health Organization (WHO) and the International Network of field sites with continuous Demographic Evaluation of Populations and Their Health in developing countries (INDEPTH), with some modifications to adapt to the local situation in Golestan. We tailored the standard VA questionnaire based on cultural background and education of study population. We made special attention to the most common disease and causes of death in the study area. We added some disease-oriented questions for specific diseases (cardiovascular, stroke, cancer (esophageal and gastric), diabetes, hypertension, tuberculosis and asthma) to collect more information by VA. Since we have already collected the data of life style and personal habits of the study participants at the enrollment phase of GCS, we excluded this part of VA questionnaire to save time. Local terms for some signs/symptoms such as “dysphagia” were applied when we translated the VA questionnaire to Farsi. After the VA interview and medical document search are completed, the results are given to two internists to ascertain the cause of death. The two internists who review the VA and other documents are unaware of each other's diagnosis. When they disagree on the cause of death, a third senior internist reviews the VA, the available documents, and also the diagnoses of the first two internists and makes the final decision. All causes of death are coded according to the core three digit codes of the International Classification of Diseases, Tenth Revision (ICD-10). The cause of death obtained by this method was considered as the gold standard for the current validation study. Seventy cases (32%) had no medical documents, so in these cases the VA-based diagnoses confirmed by the above method were used as the gold standard. ## Validation study This validation study was conducted on all 219 deaths reported in GCS participants by the end of January 2005. Copies of all 219 completed VA questionnaires were given to two trained internists, henceforth referred to as the “reviewers”, who were different individuals from the internists who made the first GCS cause of death determinations. The reviewers studied the completed VA questionnaires independently, and made their decisions on the cause of death based on the VA questionnaire alone, without having any other medical documents. In order to get an estimate of within-reviewer reliability of the VA diagnoses, the same two reviewers were asked to review the VA's a second time one month later, without the knowledge that this was a repeat review. For the purpose of this study the causes of death were categorized into one of nine major categories. To estimate the reliability, kappa statistics were calculated for the agreement between the reviewers' diagnoses. Both within- reviewer reliability (comparing the first and second diagnoses of the same reviewer) and between-reviewer reliability (comparing the diagnoses made by the two reviewers) were calculated. Multi-rater agreement was calculated and its confidence interval was calculated using bootstrap technique. To estimate validity, the VA diagnoses made by the two reviewers were compared to the gold standard diagnoses and the sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV) and kappa statistics were calculated for each reviewer-gold standard comparison. As a sensitivity analysis, the VA validity was once calculated in the subgroup of 149 cases (68%) who had both VA and medical documents available for the gold standard cause of death determinations, and then in all 219 cases. All study participants had signed written informed consent at enrollment phase of GCS and ethical approval for the present study was obtained from the ethics committee of Digestive Disease Research Center of Tehran University of Medical Sciences. # Results Of the 219 deceased participants, 133 (60.7%) were male and 86 (39.3%) were female. The mean age (±standard deviation) at death was 64.4±10.7 years. Among the deceased, 91 (41.6%) were urban and 128 (58.4%) were rural dwellers. In most cases (85%), the respondent lived with the deceased at the time of death. Of the 219 deaths in the validation study, 70 (32%) had no medical record other than the completed VA. presents the major causes of deaths according to the gold standard diagnoses, among the total study population and the subgroup of 149 subjects (68%) who had both VA and medical documents available. Ischemic heart disease, cancers, cerebrovascular events, and transportation accidents were the most common causes of death, respectively, and were responsible for, approximately 80% of deaths. shows the results of kappa statistics for the within and between reviewer diagnoses and the comparison of the VA diagnoses with the gold standard, based on the 149 deaths with documentation available. The overall multi-rater agreement across four reviews was 0.84 (95%CI: 0.78–0.89). Most pairwise kappas were higher than 0.80, indicating good within- reviewer and between-reviewer reliability, the within-reviewer reliability being somewhat better than between-reviewer reliability. Agreement between each reviewer and the gold standard was also good (kappa\>0.75). Sensitivity, specificity, and predictive values for the four most common causes of death are presented in. To analyze sensitivity, these were calculated for the A1 review which had the lowest agreement with the gold standard and then for the one with the highest agreement (A2). All estimates were higher than 0.81 which indicate good validity. As expected transportation accidents had the highest validity. Since the main goal for Golestan Cohort Study is to study the causes of upper GI cancers in particular and other cancers in general, we also tested the validity of VA for different types of cancer. Of 41 cancer deaths (in 149 deaths), 13 were due to esophageal cancer, the others being due to gastric cancer (n = 5) liver cancer (4), lymphoma (4), lung cancer (3) leukemia (3), breast cancer (2) and other cancers (7). In the comparison between A1, A2, B1 and B2 review results versus GS, the kappas were 0.82, 0.85, 0.78, and 0.85, respectively for all types. The multi-rater agreement for four reviews was 0.93 (95%CI: 0.85–0.99). In addition, the validity of VA in detecting upper GI cancers was 0.95 for all reviews. To check the differences between documented and non-documented causes of death, we did the same analysis on 219 VA. The numbers are comparable to those in. # Discussion Verbal autopsy seems to be a reliable and valid supplemental method to assess causes of death in the Golestan Cohort Study with comparable results in men and women and for patients from both rural and urban areas. One major reason for the usefulness of VA in the GCS may be the appropriate modifications made in the adult questionnaire prepared by WHO and the INDEPTH, to adapt it to the local setting. Our results are consistent with those of most previous studies, showing that the VA is a reasonably valid tool to ascertain causes of death. Some other studies are less supportive of the VA, and some even suggest that VA is not a very precise tool for detecting the leading cause of death among adults. The reason for inconsistency in results of VA validation studies may be that VA is a developing method itself. Thus, there are several variations of VA methodology and questionnaires, and some studies have not use the ICD coding system for their cause of death diagnoses. WHO has recently published instructions to improve the quality and standards for use of this method. We used the VA method in the GCS, which is the first large-scale prospective population-based cohort study of cancer in the Middle East, to improve the accuracy of diagnosing the causes of death of cohort members. The majority of families in the rural area of Golestan Province prefer their family members to die at home after a diagnosis of end-stage cancer. About 60% of the decedents in the current study died at home, and only half of these had a prior hospital- based diagnosis; for the other half, the VA seems to be a promising approach to identify at least a general cause of death. The kappa statistics obtained in the current study show that VA generates highly reliable results, at least among the 9 major categories of causes of death used in our study. Our results showed both high within–reviewer and between-reviewer reliability. We also found this method to be valid, with high sensitivity and specificity when compared to the gold standard diagnoses. Our results for making the diagnosis of different cancer subtypes also seem promising. This is especially true for upper gastrointestinal (UGI) cancers, the main focus of the GCS. Dysphagia, the main symptom of esophageal cancer, is very characteristic of this disease, and the availability of at least 10 UGI endoscopy clinics in the region, three providing free-of-charge endoscopy services to the GCS subjects, has made it possible to have accurate histologic diagnoses for almost all UGI cancers. There are of course several caveats and methodological considerations related to this method. The gold standard was set by a combination of diagnoses made by two internists, or a third internist when the first two internists did not concur. These physicians used both VA and other clinical documents to adjudicate the results, but in 32% of the cases both the original internists and the reviewers in our study had only the VA answers to review. This lack of additional clinical documents in a third of the cases might raise concern that our validity estimates were falsely elevated, but this does not seem to have been the case, since these estimates were essentially identical in the full group of cases and in the subgroup which had additional clinical documents. On the other hand, our results may underestimate the potential sensitivity and specificity of the VA method because they tested the diagnosis made by each reviewer separately. In the actual GCS, at least two internists and perhaps a third one decides on the final diagnosis, and therefore the results may be more accurate than the judgment of just one physician. In conclusion, our results suggest good reliability and good validity of verbal autopsy in determining the causes of death in a large-scale adult cohort study in a predominantly rural area in a developing country. These results add to the current literature on the use of VA for cohort studies in adult populations. The authors wish to thank all the study participants and Behavarz for their cooperation. We also would like to show our appreciation to all follow-up team. We received special support from the Social Security Organization of Iran, Golestan branch. [^1]: Conceived and designed the experiments: HK AE MN BA AP PB SMD RM. Performed the experiments: HK BA AP MB AH RM. Analyzed the data: HK AE FK FI CCA PDP PB PB SMD. Contributed reagents/materials/analysis tools: HK AE FK MN RS BA MB AH RM. Wrote the paper: HK AE FK MN RS AP MB AH FI CCA PDP PB PB SMD RM. [^2]: The authors have declared that no competing interests exist.

# Introduction Over past decades, although breakthroughs have been achieved in the development of cancer therapies, resistance and nonspecific toxicity of conventional drugs are still bottle-neck issues for potential clinical practices. Hence, it is urgently required to develop novel drugs with different modes of action which can overcome the shortcomings of many available drugs. Currently, the potential applications of anticancer peptides (ACPs) as therapeutic agents for the treatment of cancer progression attract more attention than conventional chemotherapy mainly because of the following properties: (1) high specificity. The positively charged peptides selectively target cancer cells that carry negative charges and have different membrane components from normal cells;(2) novel mode of action. It could avoid established multidrug-resistance mechanisms; (3) synergistic anticancer effect with chemotherapeutics. It has been reported that certain ACPs can produce synergistic anticancer activity when combined use with different conventional anticancer drugs. The general mechanism of peptide-induced cell death is cytoplasmic membrane disruption via micellization or pore formation, although some of ACPs are reported to trigger apoptosis by death receptor pathway and/or mitochondrial pathway. Moreover, the pore formation on the cell membrane and the change of the membrane permeability may provide a better channel for the entry of other anticancer drugs into cells and enhance their anticancer activities. Our previous study has proven that HPRP-A2 can induce the cell death and simultaneously enhanced DOX/epirubicin (EPI)-induced apoptosis in HeLa and HepG2 cell lines. In addition, due to the D-amino acid composition, HPRP-A2 is resistant to proteolytic cleavage and retains equivalent anticancer activities to its L enantiomers. Based on the previous studies, we aim to accomplish two objectives in this study: to delineate the underlying anticancer mechanism of HPRP-A2 and to investigate the synergistic anticancer effect on BGC-823 and SGC-7901 cells when combined HPRP-A2 with DOX. # Materials and Methods ## Cell lines and cell culture Human gastric cancer cell lines BGC-823 and SGC-7901 were obtained from the American Type Culture Collection which authenticates the cell lines by short- tandem repeat DNA testing, were used within 6 months of resuscitation and grown in DMEM with fetal bovine serum (FBS; 10% v/v), penicillin (100 U/ml), and streptomycin (100 U/ml) in a humid atmosphere at 37°C with 5% CO₂. ## Peptide synthesis and purification The peptide was synthesized by the solid-phase peptide synthesis using Fmoc (9-fluorenyl-methoxycar-bonyl) chemistry as described previously. Further characterization was detected by mass spectrometry and amino acid analysis. DOX·HCl was purchased from Meilun Biology Technology Co., Ltd. (Dalian, China). ## Cell viability assays BGC-823 and SGC-7901 cells (5×10³) were plated in triplicates in 96-well microtiter plates. Complete medium was replaced after 24 h with 100 μl of fresh medium containing various concentrations of drugs. After a further 24 h, cells were incubated with 3-(4,5-dimethyl-2-thiazolyl)-2,5-diphenyl-2H-tetrazolium bromide(MTT)at 37°C for 4 h. Thereafter dimethyl sulfoxide (DMSO) was added to dissolve the formazan crystals and the absorbance at 492 nm was measured with a microplate reader (GF-M3000; Gaomi Caihong Analytical Instruments Co., Shandong, China). Jin’s formula was used to further quantify the synergistic effect of the combination treatment of HPRP-A2 and DOX. The formula is: Q = Ea+b / (Ea + Eb—Ea × Eb), where Q is the combination index; Ea+b represents the cell proliferative inhibition rate of the combined drug; Ea and Eb are signs of the cell proliferative inhibition rate of each drug. After calculation: Q\<0.85, Q\>1.15 and 0.85\<Q\<1.15 indicate antagonism, synergy, and additive effect, respectively. ## Hemolysis activity assay Hemolysis activity analysis was performed as described previously. To obtain red blood cells, fresh human blood stabilized with EDTAK was centrifuged at 1,000 rpm for 5 min, washed twice with PBS and diluted to a final concentration of 2% in PBS. 70 μL of 2% human erythrocytes were added to a round-bottomed 96-well plate, followed by 70 μL of different concentrations of HPRP-A2. After incubation at 37°C for 1 h, the plate was then centrifuged at 3,000 rpm for 10 min and 90 μL of supernatant was transferred to a flat-bottomed 96-well plate. The release of hemoglobin was determined by measuring the absorbance of the supernatant at 540 nm. Erythrocytes in PBS and distilled water (dH₂O) were used as negative and positive controls, respectively. The hemolytic activity was calculated as the percentage of experimental group and positive control, after subtraction of negative control respectively. Data are the mean ± SD of three independent experiments. ## PI assay BGC-823 cells (1×106 cells/well) were seeded in six-well plates. After incubation with HPRP-A2 (5, 10, 15 μM) for 1 h, the cells were collected and then treated with 5 μg/ml PI at 4°C for 10 minutes in the dark. Cells were washed with PBS for three times and then detect the fluorescence intensity of PI using flow cytometry (FACSCalibur, Becton-Dickinson, San Jose, CA, USA). ## LDH release LDH release activity was measured by LDH assay kit (Jiancheng Bioengineering, Ltd., China) according to the manufacturers’ instructions. BGC-823 cells were seeded at 5 ×10³ cells/well in a 96 well plate. After incubation with HPRP-A2 (5, 10, 15 μM) for 1 h, the release of LDH in the supernatant was measured with a microplate reader (GF-M3000; Gaomi Caihong Analytical Instruments Co., Ltd. Shandong, China) at 450 nm. Cells without treatment or lysed with triton X-100 was used as negative and positive controls, respectively. All experiments were carried out in triplicates. LDH activity was calculated as the percentage of experimental group and positive control, after subtraction of negative control respectively. ## ROS assay ROS assay kit (BestBio, Co. Shanghai, China) was used to detect the generation of ROS. Cells (1×106) were treated by HPRP-A2 (5, 10, 15 μM) for 1 hour. The subsequent procedures were performed according to the manufacturer’s protocol. In brief, cells digested by trypsin were centrifuged at 1,500 rpm for 3 min, washed three times with PBS and then suspended in 500 μL PBS. After incubation with 2’,7’-dichlorofluorescin diacetate (DCFH-DA) fluorescent probe for 20 min at 37°C,cells were washed with PBS for three times and detected the green fluorescence intensity (in Geomean) by flow cytometry. The green fluorescence intensity was positively correlated with the level of ROS. ## Measurement of MMP MMP was determined by the mitochondrial membrane potential assay kit with 5,5′,6,6′-tetrachloro-1,1′,3,3′-tetraethylbenzimidazolcarbocyanine iodide (JC-1), which is a probe of mitochondrial activity (BestBio, Co., Shanghai, China). JC-1 is always used to detect mitochondrial depolarization occurring in the early stages of apoptosis. When cells with high MMP, JC-1 gathered in the mitochondrial matrix, forming J-aggregates and produce red fluorescence. Conversely, cells containing JC-1 monomer monomer monomermonomer have low MMP and show green fluorescence. Mitochondrial depolarization was determined by a decrease in the red (590 nm, FL-2 channel)/green (530 nm, FL-1 channel) fluorescence intensity ratio. Briefly, after treatment with HPRP-A2 (5, 10, 15 μM) for 1 h, BGC-823 cells were collected and incubated with 0.5 ml JC-1 working solution for 20 min at 37°C, then washed twice with PBS, suspended in PBS, and analyzed by flow cytometry (FACSCalibur, Becton-Dickinson, San Jose, CA, USA). ## Caspase activity assay Cells were treated with HPRP-A2 (10, 15 μM) for 24 h and then levels of caspase activities were measured using the corresponding caspase activity detection kits (BestBio, Co., Shanghai, China), according to the manufacturers’ instructions. Average data are presented as the mean ± SD of at least three independent experiments. ## Cell cycle analysis BGC-823 cells (1×10⁶) were seeded in 6-well plates for 24 h. After induction with HPRP-A2 (5, 10, 15 μM) for 24 h, cell cycle distribution was determined by a FACScan cytometer and Cell Quest software (FACS Calibur, Becton- Dickinson, San Jose, CA, USA). All experiments were performed in triplicates. ## Statistical analysis The data are presented as means ± SD of three independent determinations. Statistical significance of differences between groups were analyzed by *t*-test, with significance accepted at *P* \< 0.005 (\*) and *P* \< 0.001 (\*\*). # Results ## Peptide and cytotoxicity As shown in, peptide HPRP-A2 is a 15-residue α-helical amphipathic membrane- active peptide composed of all D-amino acids. Comparing the selective toxicity of HPRP-A2 towards gastric cancer cells and normal cells (human red blood cells), we can easily find that the IC₅₀ (the concentration of drug at which cell viability was reduced by 50% compared with untreated cells) values are far less than the minimal hemolytic concentration (the concentration of drug that resulted in 20% cell hemolysis) of the HPRP-A2. These results indicated that HPRP-A2 can selectively kill the gastric cancer cells and spare the normal cells (Figs). Similar anticancer activities of the two cell lines (BGC-823 and SGC-7901) indicated that there was a broad-spectrum effect in the anticancer action of HPRP-A2. Owing to its membrane-active characteristic, HPRP-A2 shows the anticancer therapeutic potential since it is more selectively toxic towards tumor cells than normal cells. ## HPRP-A2 induced the enhancement of membrane permeability In order to verify the change of membrane permeability after incubation with HPRP-A2, the cellular uptake of PI and extracellular release of LDH were investigated with flow cytometry and microplate reader toward BGC-823 cells. As shown in, the flow cytometric graphs of the PI move gradually to the direction of high fluorescence intensity in a concentration-dependent manner, and the increased release of LDH was also observed in the cells incubated with HPRP-A2. That is to say, HPRP-A2 could cause the damage of cell membrane and result in the enhancement of cell membrane permeability. ## HPRP-A2 caused the damages of mitochondrial function The intracellular reactive oxygen species (ROS) release and mitochondrial membrane potential (MMP) were detected with FACS to reflect the mitochondria function of BGC-823 cells *in vitro*. As shown in, the flow cytometric histogram of the cells incubated with higher concentration of HPRP-A2 revealed higher fluorescence intensity after incubation for 1 h. The corresponding flow cytometric quantitative comparison of fluorescence intensity in Geomean at these different concentrations showed a similar trend, namely that the increased release of ROS in BGC-823 cells in the presence of HPRP-A2 was concentration- dependent. Similar concentration-dependent changing trend was also observed in, the ratio of FL2-H/FL1-H markedly decreased, indicating a depolarization of MMP. ## Caspase activation and cell cycle analysis Activation of caspase-3, -8 and -9 in HPRP-A2-induced cells was measured using a with a microplate reader at 405 nm. As shown in, caspase-3, -8 and -9 were all increased after treatment with HPRP-A2 for 24 h in BGC-823 cells,which suggests that the induction of apoptosis by the HPRP-A2 may be caspase-dependent. As shown in, BGC-823 cells treated with the different concentrations of HPRP-A2 (5, 10, 15 μM) for 24 h resulted in the increases in sub-G1 arrest and in G1 arrest. These findings also strengthened the increased caspase-3 activity after the treatment of HPRP-A2. ## HPRP-A2-induced increase in DOX cytotoxicity BGC-823 and SGC-7901 cells were selected to study the synergistic anticancer effect of HPRP-A2 and chemotherapeutic drug DOX. Cells were treated with HPRP-A2 (6 μM) and/or Dox (1.6 μg/ml) for 4, 24 and 48 h. MTT assays were used to evaluate the combinational anticancer effects on cells. The drug concentrations selected in this study were based on the IC₅₀ values of each drug alone. There was no obvious cytotoxicity or growth reduction when each drug was used alone. In contrast, when used in the peptide/drug combinations (HPRP-A2/DOX) at the same doses of being used alone, the combination exhibited significant cytotoxicity. It is also clear that anticancer activity of HPRP-A2 was not much affected by the incubation time; in contrast, with the increase of incubation time to 48 h, DOX shows much greater anticancer activity than that of 4 h, indicating the dramatically different mechanisms of action between ACP and chemotherapeutic drug. According to the Jin’s formula,all Q (combination index) values were greater than 1.15, which indicates that there were significant synergistic effects between the α-helical peptide HPRP-A2 and the conventional anticancer drug DOX in BGC-823 and SGC-7901 cells. # Discussion Anticancer peptides (ACPs) recently have received great attentions as promising chemotherapeutic agents that avoid the drawbacks of current drugs. Many studies have verified that some synthetic and natural cationic peptides possess a rapid and broad spectrum of anticancer activity towards tumor cells rather than normal cells such as human red blood cells. Moreover, ACPs were also verified to have ability to overcome the multidrug-resistance mechanism, and synergistic effects in combination treatment. HPRP-A2 possesses a rapid and broad spectrum of anticancer activity, however, some differences also occur in the sensitivity of the HPRP-A2 to different cell lines. Based on the different IC₅₀ values for HPRP-A2 to BGC-823 (8.65 ±0.38 μM), SGC-7901 (10.42 ±0.30 μM), PC3 (21.38±0.56 μM), and B16 (19.16±0.38 μM), we chose BGC-823 and SGC-7901 cells as research targets. Besides, the anticancer activities of HPRP-A2 to other cancer cell lines such as HeLa and HepG2 have been published in our previous paper. Both of BGC-823 and SGC-7901 cells belong to gastric cell lines, thus, we selected BGC-823 as an example to investigate the anticancer mechanism of HPRP-A2 *in vitro*. In this study, we have shown that HPRP-A2 is an amphipathic α-helical peptide with significant anticancer activity to BGC-823 and SGC-7901 cell lines. Our studies have indicated that HPRP-A2 exhibited a cancer-selective toxicity, mainly because that cancer cells are composed of more anionic phospholipids and contain O-glycosylated mucin, which increases the negative charge on the cancer cell surface. Moreover, more microvilli on cancer cells can increase the concentration of binding peptide by expanding the membrane surface and thereby show stronger cytotoxicity against cancer cell membranes. ACPs are capable of disrupting cell membrane, which may cause cell membrane permeability changes. In this study, the change of BGC-823 cell membrane was observed by detecting PI-permeabilization and LDH release. The gradual increases of PI permeabilization and LDH release are in concentration-dependent manners after the treatment with HPRP-A2. Moreover,HPRP-A2-induced cell death is associated with the generation of reactive oxygen species, the depolarization of MMP, the activation of the caspase activities and the block in G1 phase of cell cycle. In addition to the cytotoxicity of HPRP-A2, we proved that it could increase the efficacy of DOX against BGC-823 and SGC-7901 cells, which is consistent with our previous study. In our previous study, we have explored combinational anticancer therapy using α-helical peptides HPRP-A1 and HPRP-A2 with the chemical drugs DOX and EPI in HeLa and HepG2 cell lines. The synergy shown permits the uses of relatively low concentrations of peptides and drugs to achieve significant anticancer effects *in vitro* and *in vivo*. This dose reduction minimizes drug side-effects on normal cells and enables an effective apoptosis-mediated anticancer effect. Our present study has implications in that HPRP-A2 may become a promising anticancer therapeutic agent with high anticancer selectivity and strong synergistic effect in combination therapy. Our studies mainly illustrate the mechanism of HPRP-A2-induced cell death and may be helpful in design of chemotherapeutics against gastric cell lines. # Conclusions HPRP-A2 shows strong anticancer activity to BGC-823 and SGC-7901 cell lines and low toxicity against human red blood cells. HPRP-A2 induced cancer cell death through both direct membrane-destructive effect and intracellular mechanisms, including a dramatic increase in caspase-3, -8 and -9 activation, a reduction of mitochondrial membrane potential (MMP), and the generation of ROS and cell cycle arrest in G1. Besides, HPRP-A2 synergized strongly with DOX to enhance the efficacy of killing gastric tumor cells *in vitro*. Our results underscore the broad anticancer potential of HPRP-A2 and elucidate its mechanism of action. We believe that endowing ACPs with more effective and tumor-targeting properties will open up new ways to combat cancer successfully. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: JZ YBH YXC. Performed the experiments: JZ XYH DL. Analyzed the data: JZ YBH YXC. Contributed reagents/materials/analysis tools: YBH YXC. Wrote the paper: JZ YBH YXC.

# Introduction Patients with cancer commonly develop anemia arising from the effect of the tumor itself or from treatments such as chemotherapy. Since anemia is an independent risk factor for mortality, treating anemia with consideration of the associated risks remains important for the care of patients with cancer. Erythropoiesis-stimulating agents (ESAs) are recombinant glycoproteins that stimulate red-blood-cell production using the same molecular mechanism as endogenous erythropoietin. Randomized, controlled trials have shown that ESA use in patients with cancer receiving chemotherapy reduces red-blood-cell transfusion requirements. However, safety concerns have arisen around ESA use with various trials reporting that ESA administration promoted tumor progression and adversely impacted overall survival rates (trials reviewed in). The mechanism(s) by which ESAs might affect survival or stimulate tumor-cell growth is not clear. Indirect mechanisms have been proposed including (i) ESA- mediated promotion of thrombovascular events leading to increased mortality and (ii) ESA-mediated stimulation of angiogenesis leading to increased tumor growth (10) although recent results do not support this mechanism. An alternative hypothesis is that ESAs may directly increase tumor cell proliferation and survival by activating EpoR on tumor cells. Studies examining the hypothesis that ESAs directly stimulate tumor growth by activating canonical EpoR signaling pathways (i.e. PI3K/AKT, RAS/RAF/ERK and JAK/STAT) on tumor cells have reported conflicting results. Several studies reported that primary tumor tissue and tumor-derived cell-lines express *EpoR* mRNA transcript and contain EpoR protein as shown by Western blot analysis or immunohistochemistry (IHC). A possible confounding variable in these studies is that mRNA analysis of bulk tumor tissue includes representation of stromal cells and other infiltrating cell types from blood. Also, the quantities of *EpoR* mRNA detected in some tumor and normal cells outside the erythroid compartment are relatively low (at levels10- to 1000-fold lower than in positive controls) and calls into question whether these mRNA levels are adequate to produce relevant amounts of functional EpoR protein. Many studies employing Western blotting and IHC often used commercially available polyclonal EpoR antibodies that have been shown to lack EpoR specificity. Importantly, these studies could not address erythropoietin- dependent EpoR function in tumor tissue. More recent, detailed studies have reported that tumor cell-lines, tumor biopsies, and endothelial cells did not contain increased levels of *EpoR* mRNA, or protein compared with normal tissues and that there was no amplification of the *EpoR* gene in tumor cells. Additionally, xenograft models were conducted using a limited number of breast cell-lines that suggested co-administration of rHuEpo resulted in diminished efficacy of Her2 directed agents in Her2+ cell-lines. In contrast, several studies using tumor cell-lines showed an improved tumor response with administration of ESAs. These results demonstrate the continuing challenges in the field that confound clear conclusions to be drawn regarding the ESA tumor stimulation hypothesis. However, as methods to study signaling in freshly- derived human tumor cells have only recently become available, the literature has mostly relied on cell-lines whose relevance is uncertain. It has been hypothesized that ESAs are able to directly stimulate tumor cells. This study was performed to specifically address this hypothesis. EpoR pathway activation was analyzed on viable human tumor cells obtained directly from human tumor tissues representing a range of different primary tumor types. This included evaluation of receptor function and protein expression in freshly- derived human breast tumor tissues, including both Her2+ and Her2- tumors to address specific questions related to the biological relevance EpoR in breast cancer. We examined if ESA exposure could activate signaling pathways by treating viable primary human tumor cell isolates with recombinant human erythropoietin (rHuEpo) and analyzing the effect on the activation state of multiple signaling proteins downstream of cell-surface receptors. Cell-surface expression of EpoR, as well as total EpoR (assessed in breast cancer sample cohort) was also analyzed using specific EpoR monoclonal antibodies. # Materials and Methods ## Cell Culture The megakaryoblastic leukemia cell-line, UT-7/Epo was a gift from Dr. Norio Komatsu, Jichi Medical School, Minamikawachi, Japan. The colorectal adenocarcinoma cell-line HT29 was purchased from ATCC (Rockville, MD). Although not formally authenticated, control cell-line performance was consistent over the duration of the study, with no aberrant changes observed with regards to receptor level expression and response to cytokines as measured in flow cytometry experiments. Prior to growth factor stimulation, HT29 and UT-7/Epo were starved overnight in media containing 0.1% (w/v) bovine serum albumin (BSA). UT7/Epo and HT29 cells were harvested and washed by centrifugation twice with Ca²⁺/Mg²⁺ free phosphate-buffered saline (PBS; Invitrogen). Aqua Viability Reagent (Invitrogen) was added per the manufacturer’s protocol for exclusion of dead cells. Cell densities were adjusted to 10⁶ viable cells/mL prior to growth factor stimulation. For additional cell culture conditions, see. ## Erythroid Progenitor-Cell (EPC) Assay Human CD34+ progenitor cells (AllCells Inc., Emeryville, CA) were isolated from bone marrow using CD34 immunomagnetic purification (Miltenyi Biotec, Auburn, CA), per the manufacturer’s protocol. Differentiation of EPCs was induced with rHuEpo (0.1 U/mL), IL-3, IL-6, and stem cell factor (SCF) (R&D Systems, Minneapolis, MN). The use of CD36+/CD34- expression as markers for erythroid lineage development has been previously described. EpoR function was analyzed by exposing the culture at various time points to a range of rHuEpo concentrations from 0 U/mL (rHuEpo formulation buffer “vehicle” control: 100 mM NaCl, 20 mM NaCitrate, 0.25% human serum albumin \[HAS\], pH 6.9) to 300 U/mL for 5 or 30 minutes, which includes the physiological range of endogenous Epo (5–30 mU/mL) and exceeds the pharmacological levels observed in patients treated with ESAs (reported to be a mean of 1 U/mL in plasma). Thus, the upper range of the titration was at least 300-fold higher than the theoretical exposure of tumor cells in patients administered ESAs. EpoR cell-surface expression was determined by flow cytometry using the EpoR specific antibody MAb307 (R&D Systems) incorporating 7-amino-actinomycin-D (7-AAD; Invitrogen) staining to select live cells with intact plasma membranes. ## Tissue Processing Samples were obtained from Asterand USA, BIO OPTIONS, Inc, and the MT Group. Tumor content was characterized by: 1) A qualified, independent pathologist (all samples included in this study were confirmed to be tumors). 2) H&E staining (tumor content ranged from 6% to 100%;.). 3) DNA content (median tumor aneuploidy percentage, interquartile range, was 56%;). 4) pERK and pAKT induction of 19 matched normal colon and tumor samples stimulated by growth factors (tumor samples had higher pathway induction than normal colon samples;.). In addition, flow cytometry experiments were designed to isolate an equivalent number of viable tumor cells from each sample (see details in the next section). For disaggregation, human tumor tissues were digested with 0.34 U/mL dispase (Roche Diagnostics, Indianapolis, IN) and 1mg/mL DNAse (Worthington Corp., Lakewood, NJ) for 30 minutes at 37°C. Cell suspensions were washed, cell viability and density determined, and cell densities adjusted to 10⁶ viable cells/mL in medium. Aqua Viability Reagent was added to cell suspension per manufacturer’s protocol to label dead cells. ## Effect of Growth Factor and rHuEpo Addition Aqua Viability Dye-labeled cell-lines and tumors were stimulated for 5 or 30 minutes with vehicle, a serial dilution of rHuEpo (vehicle to 300U/mL), or a cocktail of epidermal growth factor (EGF)/hepatocyte growth factor (HGF)/insulin-like growth factor-1 (IGF-1) (EGF: 100 ng/ml, Roche, Basel, Switzerland; HGF: 200 ng/ml, IGF-1: 100 ng/ml, R&D Systems) for 5 and 30 minutes in aliquots of 10⁶ cells. Treated cells were fixed by adding an equal volume of pre-warmed (37°C) fix buffer I (BD Biosciences) and incubated for 10 minutes at 37°C. Fixed cells were then permeabilized in ice-cold 90% (v/v) methanol and stored at -20°C prior to flow cytometry analysis. ## Analysis of Intracellular Signaling by Flow Cytometry Fixed and permeabilized cells were washed twice with ice-cold Fluorescence- Activated Cell Sorter (FACS) Stain Buffer (2% FBS \[v/v\], 0.09% \[w/v\] NaN₃; BD Biosciences). Samples were stained for 1 hour at room temperature with fluorochrome-conjugated antibodies specific for phosphorylated forms of AKT (AF-647; Cell Signaling Technology, Danvers, MA), ERK1/2 (AF-647; Cell Signaling Technology), and STAT5 (AF-488; BD Biosciences). EpCAM (PerCP- Cy5.5; BD Biosciences), pan-cytokeratin (PE; BD Biosciences), and active Caspase-3 (AF-405; Cell Signaling Technology) were multiplexed with phospho- specific antibodies listed above to allow for gating of viable, non-apoptotic epithelial cells. Stained samples were run on an LSR II (Becton Dickinson, Franklin Lakes, NJ) with 10⁴ viable epithelial events acquired. The analysis used a gating strategy for viable EpCAM or cytokeratin positive events excluding active caspase-3 (apoptotic) or Aqua Viability dye-positive events. Results are reported as mean fluorescence intensity (MFI) fold change in treated samples compared to vehicle. ## Analysis of Cell-surface Receptors in Live Cells by Flow Cytometry Tumor cells and control cell-lines were stained with antibody cocktails (see for the antibodies used). Prior to flow cytometry, samples were stained with 5 μg/mL 7-AAD to exclude dead and apoptotic cells. Flow cytometry was performed using an LSR II (BD Biosciences). Receptor levels were reported as a ratio of MFI values relative to the appropriate isotype control. The analysis employed a cell-gating strategy that selected viable EpCAM positive cells and excluded CD45-positive and dead/apoptotic cells. For each sample, 10⁴ viable EpCAM positive events were acquired. ## Analysis of EpoR Protein Expression by Western Blot Tumor tissues were homogenized by sonication on ice in a lysis buffer containing 50 mM Tris-HCl, pH 7.2, 150 mM NaCl, 1% Triton X (v/v), 0.1% Na deoxycholate (w/v) and a protease inhibitor cocktail (0.1 mg/mL 4-(2-aminoethyl) benzenesulfonyl fluoride hydrochloride \[Pefabloc-SC\], and 10 mg/mL pepstatin). Lysates from cell-line controls were prepared in the same manner. Lysates were subjected to SDS-PAGE (NuPAGE; Invitrogen) and transferred to Invitrolon polyvinylidene fluoride membranes (Invitrogen) and processed as described in the. # Results ## EpoR Pathway Activation in an EPC Assay To define assay sensitivity and specificity, EpoR function and cell-surface expression was evaluated during 8-day cultures of a normal, physiologically relevant Epo-responsive tissue, differentiating primary human EPCs. CD34+ cells were isolated from human bone marrow and exposed to SCF, IL-3, IL-6, and a low concentration of rHuEpo (0.1 U/mL). During the time course, EPC numbers increased as shown by the accumulation of the CD36+/CD34- cells from the CD34+ enriched population. Cell-surface EpoR expression was analyzed by flow cytometry using the MAb307 antibody. Levels of EpoR expression increased from undetectable levels on day 0 to maximum levels by day 8. At each time point, a range of rHuEpo (0 U/mL \[vehicle\] up to 300 U/mL) was added to EPCs. Levels of phospho-proteins (pSTAT5, pAKT, and pERK) known to be induced by activated EpoR were measured by flow cytometry. EpoR function was determined by comparing levels of each phospho-protein in the rHuEpo-treated cells relative to the vehicle treatment. Similar approaches have been reported in the literature. As shown in, no rHuEpo-dependent signaling was observed on day 0. Dose-dependent increases in pSTAT5 were observed on day 1 when only 6.9% of cultured cells were CD36+ and low levels of cell-surface EpoR were observed. At each of these time-points, concentrations of rHuEpo above 10 U/mL stimulated pSTAT5 to maximal levels and no further induction was observed at higher concentrations of rHuEpo. By day 8, when \> 90% of the culture had an erythroid phenotype and were EpoR positive, significant levels of pSTAT5 were observed at rHuEpo concentrations as low as 0.02 U/mL. Representative histograms demonstrating rHuEpo driven pSTAT5 induction are shown in. Similar data were observed for pAKT and pERK (data not shown). The sensitivity and specificity of the EPC assay ensured that EpoR function could be examined in primary cells exposed to rHuEpo levels that are physiologically relevant (range 0.005 U/mL to 0.03 U/mL) and approximately 30-fold lower than maximal concentrations of serum ESA observed in patients treated with ESAs. rHuEpo dose response was also evaluated in the positive control cell-line UT-7/Epo and the negative control cell-line HT29. UT-7/Epo is a known Epo- responsive cell-line and demonstrates pathway responsiveness along PI3K/AKT, RAS/RAF/ERK and the JAK/STAT signaling pathways. HT29 cells do not express detectable levels of EpoR and are not responsive to rHuEpo and thus serve as a negative control for rHuEpo addition. In UT-7/Epo cells, maximum pSTAT5, pAKT, and pERK induction levels were reached by 10 U/mL rHuEpo and no further induction was seen in concentrations up to 300 U/mL rHuEpo. In HT29 cells, no induction was observed across all rHuEpo concentrations. For every human tumor that was analyzed for response to rHuEpo and also for EpoR expression, both UT7/Epo and HT29 cell-lines were processed in parallel as experimental controls. Representative histograms demonstrating pSTAT5, pAKT, and pERK induction in the UT7/Epo cell-line are shown in. A representative tumor sample from head-and-neck cancer (BIOH002) is shown in. In contrast to the Epo responsive UT-7/Epo cell-line, no pSTAT5, pAKT, and pERK induction was observed at any rHuEpo concentration. A separate pool of cells isolated from the head-and-neck cancer biopsy were stimulated with a growth factor cocktail of EGF/ HGF/IGF1 and demonstrated response along pAKT, pERK and pSTAT5 pathways. The growth factor cocktail response served as a positive control and clearly demonstrated that tumor cells isolated from this patient sample responded to known tumor growth factors in a robust and measurable fashion. ## EpoR Pathway Activation in Tumor Cells Isolated From Human Tumor Tissues To evaluate EpoR function in tumors, tissues were obtained from surgical resections (using IRB-approved protocols) from various tumor types, stages, and grades. All tumors, except where indicated, were collected from patients not previously treated with chemotherapy. A number of metastatic tissues were also included to determine whether response to rHuEpo was altered during disease progression. Disaggregation of tumor tissue to obtain single-cell populations of both tumor and stromal compartments requires proteolytic enzymes which may compromise functional analysis of cell-surface receptors. To address this concern, dispase was used because it has a narrow range of specificities. In the EpoR dependent UT-7/Epo cell-line, no effect of dispase on EpoR function or expression was observed below 0.5 U/mL dispase compared with undigested cultures (.). Similarly, in the colorectal cancer cell-line HT29, no effect on EGFR, c-MET or IGF-1R function or expression was observed below 0.5U/mL dispase (data not shown). A concentration of 0.34 U/mL dispase was selected for tissue disaggregation as this amount had no observable effect on EpoR function in the UT7/Epo cell-line control model and also no effect on EGFR, c-MET or IGF-1R function in HT29 cells. Single-cell suspensions from the disaggregated tumors were stimulated with a range of rHuEpo (0 to 300 U/mL) for 5 and 30 minutes. While maximal effects of rHuEpo are observed at \~ 1 U/ml, concentrations up to 300 U/mL were used to account for the possibility that tumor cells are less responsive to rHuEpo than EPCs. Early and late time points accounted for differing kinetics of PI3K/AKT, RAS/RAF/ERK, and JAK/STAT pathway induction observed in control cell-lines. Epo driven STAT5 phosphorylation was specifically targeted in this study because unlike PI3K/AKT, RAS/RAF/ERK, activation of STAT5 is more specific to the EpoR pathway and therefore provides a good assessment of specific EpoR activation. As a positive control to demonstrate that tumor cells were viable and responsive, they were stimulated with a cocktail of known tumor growth factors consisting of EGF, HGF, and IGF-1. Following addition of rHuEpo or the growth factor cocktail, cells were fixed, permeabilized, and analyzed by flow cytometry using a panel of fluorescently conjugated antibodies specific for pAKT and pERK, which are induced when EpoR, EGFR, c-Met, and IGF-1R are activated. pSTAT5 was included as it is tightly regulated by EpoR in EPCs. To ensure that a lack of response was not simply due to inactive preparations of rHuEpo and to also verify potential false-positive or false negative events, signaling in HT29 and UT-7/Epo was analyzed in parallel with every tumor cell preparation. HT29 was selected as a positive control because it is known to express EGFR, c-Met, and IGF-1R, and is responsive to EGF, HGF, and IGF-1, activating the PI3K/AKT and RAS/RAF/ERK signaling pathways. These cells also served as a negative control for the rHuEpo addition. UT-7/Epo is a known Epo- responsive cell-line and served as a positive control. Both control cell-lines performed as expected and the observed effects were highly reproducible. Representative histograms showing pathway response to 30 minute stimulation with vehicle, 300 U/mL rHuEpo, or growth factor cocktail are depicted in. Viable, non-apoptotic tumor cells were analyzed through the use of antibodies specific for epithelial tumor cell markers (EpCAM and pan-cytokeratin), a viability dye (debris exclusion), and an activated-caspase-3 antibody. These analyses provided a sensitive and specific method to characterize Epo-dependent effects on signaling pathways that are known to be critical for the growth and survival of tumor cells. The use of the growth-factor cocktail positive control also allowed comparisons with known tumor growth factors. Stimulation of pAKT or pERK was observed in response to growth factors in tumor cells from colorectal (n = 48 patients), breast (n = 38 patients), NSCLC (n = 45 patients), and ovarian (n = 37 patients). This pathway-activation was the positive control and activation of at least one pathway was evident in each cell sample (although activation of all pathways was not always seen). In contrast, no rHuEpo-mediated activation of pERK, pAKT, or the canonical EpoR pathway constituent, pSTAT5, was observed in any of the tumor samples that were analyzed at any concentration of rHuEpo (the highest concentration employed, 300 U/mL, is represented). Similarly, no rHuEpo-mediated activation of pSTAT3 was observed in any of the tumor samples (.). In each experiment robust phosphorylation of STAT5 was observed with UT-7/Epo cells indicating that EpoR was specifically activated under the conditions used. Identical observations were made from head and neck (n = 5), kidney (n = 4), pancreatic (n = 2), cervical (n = 1), and gastric tumors (n = 3) with no response to rHuEpo detected across a broad range of epithelial tumors types. As shown in, there was no response in cells from metastatic lesions. Similarly, in samples derived from patients who had undergone various chemotherapeutic regimes there was no response to rHuEpo although all displayed pathway-activation in response to the growth factors. Together these findings suggest that while tumor cells did respond to known epithelial-growth factors, they did not respond to rHuEpo in the original tumors, in metastatic lesions, nor following anti-cancer treatments. In interpreting these data, it is critical to understand what represents a significant induction of phosphorylation as represented by MFI. To evaluate the variability of repeated measures of replicate vehicle-treated tumor cells, disaggregated cells from 10 tumor tissues were treated with vehicle in replicate (n = 10), and the ratio of the upper 95% CI of each MFI relative to the mean calculated. The ratios determined had a mean of 1.14 (range 1.07–1.15) for pAKT and a mean of 1.16 (range 1.08 to 1.22) for pERK. Therefore, a significant increase was defined as a response above a threshold ratio of 1.14 for pAKT and 1.16 for pERK. In principle, small numbers of tumor cells in the population could respond to rHuEpo and not be represented in MFI. To determine the sensitivity of the assays in terms of percentage of tumor cells at the selected thresholds for pAKT and pERK, mixing experiments using growth factor-stimulated and vehicle-treated populations were performed. The data demonstrate that a population MFI ratio relative to vehicle of 1.14 for pAKT and 1.16 for pERK would correspond to approximately 1% to 2% of responding tumor cells, suggesting that the method employed would be capable of detecting even a small sub- population of responsive cells in the rHuEpo-treated samples (data not shown). To further validate this, visual inspection of flow cytometry data was carried out for all rHuEpo-treated cells compared to the corresponding vehicle control. This rigorous exercise revealed no instance of activation of any sub-population of cells in rHuEpo-treated tumors, either in the tumor cell compartment or among the viable, non-apoptotic, non-epithelial stromal compartment of the tumor tissues. ## EpoR Protein Expression in Tumor Cells Isolated From Human Tumor Tissues To support the lack of Epo driven pathway activation observed in disaggregated tumor samples, cell-surface EpoR was also analyzed in disaggregated, live tumor cells from the cohort of 186 patients by flow cytometry using the specific anti- EpoR antibody MAb307. This antibody was previously validated as specific by immunoprecipitation followed by Western blot using the A82 antibody and by flow cytometry with live cells. However, in other experiments it was observed that MAb307 cannot be used directly for Western Blot or IHC because of lack of detection of unfolded EpoR. The specificity of MAb307 when used for flow cytometry was demonstrated using the EpoR-expressing positive control cell-line UT-7/Epo and the EpoR-negative control cell-line HT29. It is important to note that in this study, EpoR expression was evaluated by flow cytometry on the extracellular membrane of viable cells. Apoptotic cells and debris generated during the disaggregation process were excluded from analysis to avoid false positive EpoR detection. Under these conditions MAb307 was shown to be specific for cell surface EpoR in positive control cell-lines. In EpoR negative cell- lines, MAb307 MFI signal intensity was equivalent to a matched isotype control. MAb307 had similar sensitivity and specificity for EpoR as the A82 antibody (.). The membrane-impermanent DNA stain 7-AAD was used to select intact, viable cells thereby excluding contributions from functionally irrelevant intracellular EpoR. Additionally, an anti-EpCAM antibody was used to identify epithelial tumor cells in each disaggregated tissue. Levels of EGFR, c-Met, and IGF-1R were also measured using specific antibodies. A wide range of cell-surface expression of EGFR, c-Met, and IGF-1R was observed among the tumor tissues that were analyzed, whereas in no case were significant levels of EpoR detectable on the cell-surface of the tumor cells from each of these tissues that were above the negative control cell-line or isotype control (Figs.). Significant levels of EpoR were not detected in the small cohorts of tumor tissues from metastatic tissues or from patients that had been previously treated with chemotherapy (Figs.). The lack of EpoR expression suggested that EpoR was not induced in tumor cells during disease progression and was also not induced in response to treatment. As previously mentioned, cell-lines (HT29 and UT-7/Epo) were processed in parallel as controls for cell-surface receptor expression with the analysis of every tumor tissue. In the analyses of all tumors described above, IGF-1R, EGFR, c-Met (using HT29 cells), and EpoR (UT-7/Epo cells) cell-surface expression were consistently demonstrated in the appropriate control cell-line. Analysis of EpoR protein expression was also evaluated by Western blot analysis of 30 tumor tissues lysates from the breast cohort, including 6 Her2 positive tissues. Her2 status was assessed by a qualified pathologist for all breast tumor tissues. Consistent with reported Her2 prevalence, 6 of the 34 breast tumor tissues (same tissue samples used) were Her2 positive as determined by IHC using the Dako Herceptest scoring system (25% tumor cells stain 2+ or 3+) using formalin fixed tissue collected from the tumor biopsy material (See for additional details on Her2 status). Western blot analysis was conducted with the EpoR-specific A82 antibody (Amgen, Inc) that was previously shown to be suitable for Western Blot analysis. This method also differed in that total EpoR (intracellular and surface) was examined. No EpoR protein was detectable using this method in any of the breast tumor tissues analyzed. Lysates were also prepared from EPCs that had been differentiated from human bone marrow as a biologically relevant positive control. Expression of the mature EpoR protein was readily detectable (additional bands correspond to intracellular proteolytic fragments of EpoR as demonstrated by Mass Spectrometry protein sequencing). Analysis of *EpoR* mRNA expression levels was also conducted. *EpoR* mRNA was largely undetectable in the breast tumor tissues analyzed but readily detectable in the control UT7/Epo cell-line and in human bone marrow cells. Taken together, these data support the conclusions derived from flow cytometry-based analysis of EpoR in that no expression or function was detectable in tumor tissues (Figs.) but EpoR expression and function was observed in EPCs at this stage of differentiation. # Discussion Although early literature was inconsistent with functional EpoR expression in non-hematopoietic cells, including tumor cells and cell-lines, safety signals reported in some recent clinical studies re-ignited interest in this question. Eight of sixty controlled ESA trials reported increased mortality and/or disease progression with ESA use in the oncology setting and are included in the ESA product labeling information. Many controlled ESA oncology trials have not reported safety signals, and differences in study design exist between these and studies that have reported mortality and progression safety signals. In addition, a meta-analysis of 26 studies indicated that ESA use did not significantly affect disease progression. Publications have since emerged suggesting functional EpoR is expressed in human tumors and human tumor cell-lines. They have formed the basis of an “EpoR tumor stimulation hypothesis”. Several potential issues have been identified for those reports that may confound their conclusions. Many used preparations of polyclonal anti-peptide antibodies in IHC and flow cytometry assays, assays for which those antibodies had not been validated. Other studies reported *EpoR* mRNA expression but without examining EpoR protein and function. Functional studies with ESAs on tumor cell-lines were also conflicting and difficult to interpret, in part because they lacked appropriate positive and negative controls to detect false-negative or false-positive effects. *In vivo* Epo antagonism studies have reported that the blockade of Epo:EpoR inhibited tumor growth. However, these results are inconsistent with *in vitro* findings that showed the same cell-lines had little/no EpoR expressed and had no detectable *in vitro* response when treated with ESAs, although it is possible that Epo in combination with other local and systemic growth factors may have an effect on tumor growth. In 31 different *in vivo* xenograft studies, no effect of Epo on tumor cell growth was observed. Cytoprotection studies have also been conducted to assess whether ESAs have non- hematopoietic effects. In a number of animal studies, ESAs were reported to enhance angiogenesis after injury. However, the results from these studies may be related to RBC increases, such as enhanced oxygen delivery or changes in ferrokinetics. In other *in vivo* studies, ESAs did not provide non- hematopoietic protective effects and the reported cytoprotective effects have generally not correlated with a clinical benefit in humans. Most recently, more sensitive and specific reagents, controls and targeted protocols have been generated and the question of EpoR protein expression and function on tumor cell-lines and other cell types has been re-examined. Consistent with the earliest literature, those analyses have demonstrated that functional EpoR expression is essentially restricted to erythroid cells. This study was designed to answer two questions using disaggregated tumor cells isolated directly from patient samples and employing rigorous protocols and controls: (i) do freshly isolated human tumor cells demonstrate Epo-induced signaling and (ii) do they express detectable EpoR (surface or intracellular, independent of the first question). No functional response to rHuEpo was detected in freshly-derived primary tumor cell populations. In addition, according to two different criteria, flow cytometry and Western blotting, no EpoR was detected in any tumor sample. It is possible that EpoR is expressed at levels below the threshold of detection for these assays or that an as yet unidentified receptor mediates “EpoR-like” pathway activation. However, the lack of a measurable pathway response to rHuEpo stimulation suggests that if EpoR was expressed at low levels, it was insufficient to drive a meaningful biologic response as defined by MAPK, PI3K, and pSTAT5 signaling. Controls using validated phospho-specific and anti-EpoR antibodies, demonstrated assay sensitivity using Epo driven pathway induction of pSTAT5 and detectable expression of EpoR in differentiating erythroid cells. These studies demonstrated rHuEpo-dependent, functional EpoR early in differentiation of the culture when levels of cell-surface EpoR are known to be low. The late-stage erythroid cells have relatively higher levels of EpoR (approximately 800 to 1000 receptors on the cell-surface) and these levels were readily detected with the reagents used. Additionally, viable epithelial tumor cells were specifically examined, and other cell-surface tumor growth factor receptors (EGFR, IGF-1R, and c-Met) were shown to remain functional under the conditions employed. Addition of a cocktail of known tumor growth factors to tumor cells uniformly demonstrated the activation of at least one of the key pathways that were interrogated. This further demonstrates that the methodology was sensitive and could detect meaningful responses in tumor cells. These pathways included the RAS/RAF/ERK and PI3K/AKT pathways that are essential for proliferation, survival, and metastatic spread of tumor cells. Although we have attempted to perform these studies in a rigorous fashion, several possible caveats exist: (i) a very rare tumor cell subpopulation (ie, less than the assay sensitivity of \~2%) may express functional EpoR, albeit with questionable clinical relevance, (ii) the incidence of tumors with EpoR function may be too infrequent to have been detected in this study, and (iii) the actual receptor responding to rHuEpo is unique and does not utilize PI3K/AKT, MAPK, or STAT5 for signaling. These scenarios would represent a substantial departure from current knowledge and would require a fundamentally new understanding of Epo biology in tumors, and thus be a substantial departure from the assumptions underlying the hypothesis that Epo is a “tumor growth factor”, and that EpoR is widely expressed and functional on epithelial tumors. With several possible explanations for the lack of a positive result, we conclude that evidence of a rHuEpo-responsive phenotype in patient derived tumor cells does not exist. Furthermore, EpoR was not detectable on human tumor cells isolated directly from multiple epithelial tumor types. This demonstrated that rHuEpo did not act directly on these tumor cells to promote growth and survival pathway signaling. Finally, the approach presented here may have considerable utility to address other important questions in tumor-cell signaling. # Supporting Information We thank Yanyan Tudor for technical assistance; Jenn Hawkins and Rachel Hooks for tissue management assistance; Robert Loberg, Brian Kotzin, Angus Sinclair, Angela Coxon, David Reese, Chet Bohac, and Richard Markus for critically reviewing the manuscript; Linda Rice and Shawn Lee at Amgen for editorial and writing assistance. [^1]: All authors are or were employees of Amgen Inc. and stockholders in Amgen, Inc., which manufactures and markets erythropoiesis-stimulating agents. CGB is currently employed by TetraLogic and is a stockholder. IM is currently employed by Genentech Inc. and is a stockholder in Roche (parent company of Genentech Inc.). This does not alter the authors’ adherence to PLOS ONE policies on sharing data and materials. [^2]: Conceived and designed the experiments: IM SDP CGB VDF SE. Performed the experiments: IM KP LB JMR. Analyzed the data: IM SDP CGB VDF JMR SE. Wrote the paper: IM SDP CGB VDF JMR KP LB SE. [^3]: Current address: Gilead Sciences Inc., Foster City, California, United States of America [^4]: Current address: Kite Pharma Inc., Santa Monica, California, United States of America [^5]: Current address: Research and Development, TetraLogic, Malvern, Pennsylvania, United States of America [^6]: Current address: Oncology Biomarker Development, Genentech Inc., South San Francisco, California, United States of America

# Introduction Heart disease is the leading cause of death for patients with type I or type II diabetes. This is in part because diabetes directly impacts cardiac function independently of other comorbidities. This is termed diabetic cardiomyopathy and it is a multi-factorial condition resulting from the metabolic stresses of disrupted insulin signaling, hyperglycemia and hyperlipidemia, and mitochondrial dysfunction. In addition, there are also disruptions in protein kinase A (PKA) signaling, the molecular pathway that mediates the metabolic and contractile responses to sympathetic stimulation. While the molecular mechanisms contributing to diabetic cardiomyopathy are highly interrelated, the relationship between metabolic perturbances and changes in PKA signaling are not fully understood. In the healthy heart the sympathetic nervous system functions through β-adrenergic signaling to increase cardiac contractility. Catecholamines bind to Gα_s-coupled β-adrenergic receptors, stimulate adenylate cyclase, and subsequently increase cAMP to activate PKA. PKA then phosphorylates proteins involved in calcium cycling (troponin, SERCA, and phospholamban) and proteins that affect metabolic substrate selection (phosphofructokinase-2 (PFK-2) and acetyl-CoA carboxylase-2). Glucose uptake and oxidation are the primary means of meeting the rapid increase in energy demands in response to sympathetic stimulation. In this way the increase in contractility is orchestrated with activation of metabolic pathways to ensure energy demands are met. As an insulin sensitive tissue, the heart is affected by either decreases in circulating insulin or by the loss of insulin signaling that occur with type 1 or type 2 diabetes. The primary role of insulin is to increase glucose uptake and metabolism. Thus, the decrease in insulin signaling contributes to the metabolic inflexibility whereby the heart increases reliance on fatty acid oxidation, at the expense of decreased glucose usage, to meet energetic demands. Over the long term, this metabolic inflexibility promotes lipotoxicity, mitochondrial dysfunction, and oxidative stress. Increasing evidence suggests there are interactions between insulin and β-adrenergic signaling. For example, hyperinsulinemia can blunt PKA signaling via an increase in phosphodiesterase 4 which increases cAMP hydrolysis. In our own work, we found PKA signaling is affected in a type 1 diabetic mouse model via changes in PKA activity that are downstream of receptor activation and adenylate cyclase activity. Furthermore, we identified that a loss of insulin signaling, in both type 1 and type 2 diabetic conditions, decreases the content of the PKA substrate, PFK-2. In the healthy heart, phosphorylation of PFK-2 increases the production of fructose-2,6-bisphosphate, an allosteric activator of PFK-1 which is a committed and rate-limiting step of glycolysis. Thus, the loss of insulin signaling disrupts a mechanism whereby β-adrenergic signaling increases glycolysis to meet energetic demands. The goal of the present work was to define how the loss of insulin signaling impacts β-adrenergic signaling in cardiomyocytes. Adult mouse cardiomyocytes (ACMs) were isolated from control and Akita diabetic mice and then cultured in the presence or absence of insulin. ACMs were subsequently stimulated with β-adrenergic agonists and PKA signaling was determined by immunofluorescence microscopy. We have identified a striking decrease in PKA signaling in wild type ACMs cultured in the absence of insulin. This effect was mirrored in ACMs isolated from Akita type 1 diabetic mice, regardless of the presence of added insulin. Using phospho-specific antibodies, we found that the phosphorylation of proteins involved in calcium regulation were unaffected by the absence of insulin. In contrast, the metabolic target, PFK-2, was highly sensitive. Our results demonstrate the effects of PKA on cardiomyocyte function is dependent upon the actions of insulin. # Materials and methods ## Adult mouse cardiomyocyte isolation Adult cardiomyocytes from 5-month C57BL/6J or C57BL/6J-*Ins2*^*Akita*/J male mice (Akita, The Jackson Laboratory 003548) were isolated and cultured as previously described. Akita mice are a well-established model of hypoinsulinemia and hyperglycemia. Blood glucose was measured by a glucose test strip (Contour) at the time of sacrifice to confirm hyperglycemia. All Akita mice had blood glucose levels of at least 400 mg/dl. Briefly, after isoflurane administration the heart was excised, the aorta was cannulated, and it was then perfused with type II collagenase (Worthington \#LS004176). Calcium was reintroduced to the subsequent single cell suspension and cells were plated on laminin (Corning 354232) coated plates. Media was switched to serum-free culture media (minimal essential medium with Hanks’ balanced salt solution, Gibco (11575–032) supplemented with 0.2mg/mL sodium bicarbonate, penicillin-G, 0.1%BSA, glutamine, 10mM butanedione monoxime, and 5μg/mL insulin as indicated. Cells were cultured 18h at 37°C and 5%CO₂ with indicated drugs as described in figure legends. All procedures were approved by the Oklahoma Medical Research Foundation Animal Care and Use Committee. ## Antibodies and drugs Rabbit polyclonal antibodies to phospho-PKA substrate (9621S), PKA C-α (4782S), phospho-PFK2 (13064S), phospho-Ser16/Thr17-phospholamban (8496S), and phospho- Troponin I were purchased from Cell Signaling Technology. Rabbit polyclonal anti-PDE4D (ab14613) was purchased from Abcam. Alexa Fluor 488 goat anti-rabbit IgG (A11034) and Alexa Fluor 546 phalloidin (A22283) were purchase from Invitrogen. Insulin solution human (19278),-Isoproterenol hydrochloride (16504), 3-Isobutyl-1-methylxanthine (15879) and 8-Bromoadenosine 3’,5’-cyclic monophosphate sodium salt (B7880) were purchased from Sigma. Phosphodiesterase inhibitor Tocriset containing Milrinone, Cilostamide, Zardaverine, (*R*)--Rolipram, and Ro 20–1724 (Cat. No. 1881) along with MMPX (Cat. No. 0552) and EHNA hydrochloride (Cat. No. 1261) were purchased from Tocris. ## Microscopy Methods for immunofluorescent staining have been previously described and adapted for primary mouse cardiomyocytes (ACMs). Briefly, ACMs were plated on laminin coated coverslips (Fisherbrand Microscope Cover Glass, 12-545-80) (1 coverslip per well, 24-well plate) for 1h post isolation. Cells were cultured overnight and treated with drugs as described in the figure legends. Following incubation, cells were washed 1X with PBS (Gibco 14190–144) and fixed for 20min in 4% paraformaldehyde (Electron Microscopy Sciences 15710). Cells were washed 2X with PBS and blocked for 1 hr in 2% Blocker BSA (Thermo 37525). Coverslips were inverted onto 50μL of block solution containing 0.1% Triton X-100 (Sigma T9284) and 1–250 dilution of primary antibodies as indicated on parafilm covered 150mm gridded tissue culture dish (Falcon 353025) and incubated overnight at 4°C. Coverslips were returned to tissue culture dish and washed 3X with block solution and then inverted on 50μL of block solution containing.1% Triton X-100 (Sigma T9284) and 1–250 dilution of secondary antibodies/phalloidin for 1hr at room temperature. Coverslips were then washed 2X with block solution and 1X with PBS and then inverted onto 4μL Vectashield mounting media with DAPI (Vector Laboratories H-1200) and sealed with nail polish (Electron Microscopy Sciences 72180). Cells were imaged on a Zeiss LSM-710 confocal (Carl Zeiss). The microscopy settings were kept the same between each experiment to facilitate unbiased comparisons of fluorescence intensities. Micrographs are maximum intensity projections of 12 picture z-stacks average step size of 1.5μm. Projection images were quantitated in Zen Black (Carl Zeiss version 2012 SP5 FP3) from maximum intensity projection by drawing a free polygon outline around the cell and measuring mean fluorescence intensity. Each experiment constitutes of at least three biological replicates, indicated in Figure Legends, with at least six individual cells per experiment for a total of at least eighteen total cells quantitated for each data point. ## Western blot analysis Cardiomyocytes were cultured in 12 well plates, with or without insulin, and drugs were added as indicated in the figure legends. Media was then removed, cells were washed with 0.5 mL PBS, and 75 uL of 1X sample buffer containing 25 mM DTT and 1X Halt Protease/Phosphatase Inhibitor Cocktail (ThermoFisher \#78442) was added per well. Samples were heated at 95°C for 5 min, resolved by SDS-PAGE (4–12% NuPAGE Bis-Tris gel, Thermo Fisher), transferred to nitrocellulose membranes, and blocked for 30 min with Odyssey TBS blocking buffer (LI-COR). Antibodies were diluted 1:2000 in block buffer and added to blots overnight at 4°C, subsequently washed the following day, and the secondary antibody (IRDye 800CW, LI-COR; 1:10,000 dilution) was incubated for 1h. Following additional washing, blots were imaged on an Odyssey CLx system and analyzed using the Image Studio software (LI-COR). ## PKA activity PKA activity was assayed using the PepTag Non-Radioactive cAMP-Dependent Protein Kinase Assay Kit (Promega) as previously described. Briefly, ACMs were isolated and cultured for 18hr in the presence or absence of insulin. ACMs were then treated with 0.25μM isoproterenol for 30min. Media was removed and ACMs were lysed on ice with 100μL RIPA buffer (100mM KCL, 100mM NaPO₄, 0.1% NP40, PH 7.4) containing HALT phosphatase and protease inhibitor cocktail (ThermoFisher) and 10mM IBMX (Sigma). Following protocol guidelines, reaction buffer, PepTag A1 peptide, and water were premixed. Next, 20μL of freshly prepared lysate (at approximately 0.1mg/ml of total protein) was added and incubated for 15 min at RT. Conditions were optimized to ensure the reaction rate was linear at this time point. The reaction was stopped by the addition of 1mM cAMP Dependent Protein Kinase Inhibitor peptide (PKI-tide; Sigma) in a 50% glycerol solution to a final concentration of 45uM. Phosphorylated peptide was separated from unphosphorylated peptide by electrophoresis on a.8% agarose gel (100 V for 30 min). The gel was imaged with a D-Digit scanner (LI-COR) and the intensities of phosphorylated peptides were analyzed using ImageJ software (National Institutes of Health, Bethesda, MD). The protein concentration of each lysate was determined by Bradford assay and was used to standardize activities. ## Statistical analysis GraphPad Prism 7.02 was used for statistical analysis and mean fluorescence intensity values were evaluated using one-way ANOVA with multiple comparisons using Tukey’s test. Statistical analysis was performed on the total number of cells analyzed, comprised of at least 18 cells from 3 unique cardiomyocyte preparations and is indicated in the Figure Legends. Similar statistical significance and the same conclusions are reached if instead the data from a given cell preparation are averaged and then analyzed. Statistical significance is noted in the figure legends. # Results ## β-adrenergic signaling is decreased under diabetic conditions in cardiomyocytes Initial experiments were performed to validate methodology for evaluating PKA signaling by immunofluorescence in adult mouse cardiomyocytes (ACMs). ACMs were isolated from control mice and cultured for 18h in the presence of insulin and then stimulated with isoproterenol (ISO, 0.25μM) for 30min. Cells were then fixed and PKA activity was visualized by immunofluorescence microscopy using an antibody that specifically recognizes the protein consensus phosphorylation sequence (RRXS/T, where S or T is phosphorylated) that is specific for PKA substrates. This antibody is widely used in the literature (123 citations per [CiteAb.com](http://CiteAb.com)) and has been previously used as a means to identify changes in PKA activity by immunohistochemistry. We observed low levels of PKA substrate phosphorylation basally and this was increased approximately 3-fold by ISO treatment (top panels). Detection of PKA activity by this manner revealed largely diffuse staining but with increased intensity proximal to the sarcolemma and intermittently in areas consistent with Z-bands. We next examined how the lack of insulin affects PKA signaling. Freshly isolated ACMs were cultured in insulin-free media for 18h and then stimulated with ISO. PKA-substrate phosphorylation was significantly blunted both basally and following ISO treatment. This decrease in PKA-substrate phosphorylation was not due to altered kinetics. A time course study, with increasing durations of ISO stimulation, revealed phosphorylation reached a maximal threshold within 10min regardless of whether insulin was present. In contrast to the immunofluorescence data, no significant differences were observed when PKA substrate phosphorylation was examined by Western blot. It is possible that less abundantly expressed proteins may be differentially phosphorylated by the presence or absence of insulin but not detected by Western blot. We also measured PKA activity directly and found that culturing ACMs without insulin decreased basal PKA activity. However, the activation of PKA by ISO was not significantly affected by the insulin status. This supports that the loss of immunofluorescence intensity with the PKA substrate antibody is not due to overt loss of PKA activity. Rather, unique epitopes may be detected by the antibody under conditions that maintain native protein confirmations, as with immunofluorescence detection, as compared to the denaturing conditions of SDS- PAGE. We next examined whether the chronic hypoinsulinemia that occurs with type 1 diabetes is also associated with changes in PKA signaling. Akita mice develop type I diabetes in the absence of obesity and insulitis and this is mediated by a mutation in the Ins2 gene that results in its improper release in response to glucose. Akita ACMs were isolated and cultured overnight in the presence or absence of insulin. As shown in, Akita ACMs had substantially reduced PKA substrate phosphorylation upon stimulation with ISO as compared to WT. Furthermore, insulin did not enhance ISO-stimulated PKA substrate phosphorylation in Akita ACMs. This supports that decreased insulin signaling affects PKA signaling and that overnight insulin treatment of Akita ACMs is insufficient for rescue. ## Insulin signaling is necessary downstream of cAMP production The decrease in PKA substrate phosphorylation in cardiomyocytes cultured without insulin may be due to changes in β-adrenergic receptors or their response to ligand binding, thereby leading to a dampened response to ISO stimulation. We therefore examined direct activation of PKA using the cell permeable cAMP analog, 8Br-cAMP. Like ISO, 8Br-cAMP induced a robust increase in PKA-substrate phosphorylation in ACMs isolated from control mice and cultured with insulin. However, substrate phosphorylation stimulated by 8Br-cAMP was significantly blunted in ACMs cultured overnight without insulin. Likewise, ACMs isolated from adult Akita mice had significantly blunted response to 8Br-cAMP as compared to WT. Furthermore, insulin did not enhance 8Br-cAMP-stimulated PKA substrate phosphorylation in Akita ACMs. While we cannot completely rule out changes in β-adrenergic receptor content or activity, these results support that the defect in PKA signaling induced by the absence of insulin is downstream of receptors and cAMP production. Alterations in PKA signaling could be attributed to fluctuations in enzyme content. PKA catalytic subunit levels were therefore examined in ACMs from control and Akita mice cultured with or without insulin. As shown in, there were no significant differences in PKA content under the experimental conditions as determined by immunofluorescence or Western blot (not shown). This indicates the loss of PKA substrate phosphorylation is not mediated to decreased PKA catalytic subunit content. ## Phosphodiesterase inhibition increase PKA signaling but do not restore deficits induced by loss of insulin Phosphodiesterases (PDEs) hydrolyze cAMP and are an essential component in modulating proper PKA signaling. Thus, insulin mediated changes in PDE activity could contribute to the observed effects on PKA signaling shown in. We therefore examined the effect of 3-isobuty-1-methylxanthine (IBMX), a nonspecific phosphodiesterase inhibitor, on ACMs from control and Akita mice to determine whether blocking PDE activity is sufficient to recover PKA signaling. Addition of IBMX, in the absence of other PKA agonists, was sufficient to stimulate PKA signaling by 2.5-fold. ACMs isolated from adult Akita mice had significantly blunted response to IBMX as compared to WT. Furthermore, insulin did not enhance IBMX-stimulated PKA substrate phosphorylation in Akita ACMs. This demonstrates PDE inhibition is sufficient to stimulate PKA substrate phosphorylation similarly to a PKA agonist when insulin is present. However, this effect is blunted when WT ACMs are cultured in the absence of insulin. The stimulatory effects of IBMX are blunted in Akita ACMs and is unaffected by the insulin status. We next sought to determine the combined effects of ISO and IBMX on PKA signaling. ACMs from control or Akita diabetic mice were cultured overnight in the presence or absence of insulin and then treated with combinations of IBMX and ISO. IBMX enhanced ISO stimulation of PKA substrate phosphorylation under all conditions. This demonstrates the importance of PDE activity in attenuating catecholamine mediated PKA signaling. However, the maximum PKA substrate phosphorylation was nevertheless blunted in control ACMs cultured in the absence of insulin. Akita ACMs exhibited a similar additive increase in PKA substrate phosphorylation with the combination of ISO and IBMX. However, the maximum intensity was decreased as compared to WT ACMs and unaffected by the insulin status. Recent work has identified a relationship between cardiac insulin signaling and the content and activity of PDE4. Specifically, the hyperinsulinemia that occurs with type 2 diabetes is associated with increased PDE4B content which thereby decreases cAMP and attenuates β-adrenergic signaling. We therefore tested to see if reciprocally the lack of insulin affects specific phosphodiesterases. Control ACMs were treated with a panel of phosphodiesterase inhibitors including IBMX (nonspecific PDE inhibitor), EHNA (PDE2 specific), MMPX (Calmodulin sensitive cyclic GMP specific), milrinone (PDE3 specific), clostramide (PDE3 specific), RO-20-1724 (PDE4 specific), rolipram (PDE4 specific), and zardavsine (PDE3/4 specific). All of the inhibitors, except EHNA, stimulated phosphorylation of PKA substrates. However, the most pronounced effect was observed with inhibitors of PDE4. Next, we tested if PDE4 inhibition could increase PKA signaling in the absence of insulin. PKA signaling in ACMs from control mice treated with RO-20-1724 closely approximated the effects of IBMX. The drug enhanced PKA signaling on its own or in combination with ISO, but nevertheless signaling was decreased in ACMs cultured in the absence of insulin. We also examined PDE4 by immunofluorescence and determined that the presence or absence of insulin had no effect on its content. Consistent with previous reports, this supports that PDE4 is the primary regulator of cAMP degradation in mouse cardiomyocytes, that its content is not affected by acute changes in insulin, and that its inhibition is not sufficient to fully recover PKA activity when insulin signaling is absent. ## Phosphorylation of PFK2 is decreased in diabetic conditions Our results demonstrate that the absence of insulin affects PKA substrate phosphorylation downstream of β-adrenergic receptors and that this is not mediated by changes in adenylate cyclase or PDE activities. We next evaluated PKA signaling using phospho-specific antibodies to identify substrates that may be differentially phosphorylated in the absence of insulin. The bifunctional enzyme 6-phosphofructo-2-kinase/fructose-2,6-bisphosphatase (PFK-2) is a glycolytic regulator and substrate of PKA. When the cardiac isoform (gene produce of *pfkfb2*) is phosphorylated, PFK-2 increases the production of fructose-2,6-bisphosphate, a potent allosteric activator of the glycolytic enzyme phosphofructokinase-1 (PFK-1). Furthermore, we have previously reported that the content of PFK-2, is regulated by insulin signaling. Control ACMs were cultured for 18h in the presence or absence of insulin and then stimulated with either ISO or IBMX. As shown in, PFK-2 phosphorylation was significantly increased in wildtype ACMs stimulated by either ISO or IBMX regardless of whether cells were cultured with insulin. However, for each condition the magnitude of PFK-2 phosphorylation was significantly less in ACMs cultured in the absence of insulin as compared to those cultured with insulin. In Akita ACMs, PFK-2 phosphorylation was significantly repressed basally and was largely unresponsive to stimulation by either ISO or IBMX. We next examined whether the phosphorylation of other well-described PKA substrates exhibit insulin sensitivity. Phospholamban (PBN) is an inhibitor of the sarcoplasmic reticulum calcium dependent ATPase (SERCA2) channel. PBN mediated inhibition of SERCA2 is relieved upon phosphorylation by PKA, thereby increasing calcium cycling in response to β-adrenergic signaling. Troponin I (TnI) is part of the calcium-sensitive troponin complex that decreases myosin- actin crossbridges. Phosphorylation of TnI by PKA decreases the sensitivity of the complex to calcium and is important for increasing the inotropic response. As shown in, PBN and TnI were robustly phosphorylated in response to ISO or IBMX and this was unaffected by 18h of insulin starvation. These results demonstrate that there is a differential sensitivity to insulin among β-adrenergic targets, with the metabolic target, PFK-2, being significantly repressed while those involved in contractility are sustained. ## PKA substrate phosphorylation remains depressed upon acute insulin administration Lastly, we sought to determine whether the loss of insulin has a sustained effect on PKA signaling. ACMs from wildtype mice were cultured for 18h in the absence of insulin and then stimulated acutely with combinations of ISO and insulin. As shown in, the acute addition of insulin alone failed to increase PKA substrate phosphorylation. Furthermore, the acute addition of insulin had no additive effect on ISO mediated PKA substrate phosphorylation. This lack of effect was not from deficiencies in the insulin signaling pathway. A time course of Akt phosphorylation was examined by Western blot in ACMs cultured in the presence or absence of insulin and then acutely treated with combinations of insulin and ISO. ACMs that had been cultured overnight with insulin had low levels of Akt phosphorylation. Acute additions of ISO and insulin minimally increased Akt phosphorylation. This suggests that desensitization of the pathway occurs when cells are cultured continuously with insulin. In contrast, ACMs cultured overnight in the absence of insulin showed a robust increase in Akt phosphorylation upon acute insulin treatment regardless of the presence of ISO. The signal was maximal at 10min and then decreased over time. Thus, the lack of acute effect by insulin on PKA signaling was not due to unresponsiveness of the insulin signaling pathway. # Discussion β-Adrenergic and insulin signaling pathways are the primary means of modulating moment-to-moment changes in cardiac function and metabolic substrate selection. Nevertheless, interactions between these two pathways are not fully understood. This is important to understand in regard to diabetes where insulin signaling is disrupted and PKA signaling is dysfunctional. In an effort to further understand the interrelationship of these two pathways, we used adult mouse cardiomyocytes as a model system. ACMs have the advantage that they more closely represent in vivo signaling and metabolic conditions as compared to immortalized cardiomyoblasts, such as H9c2 and HL-2 cells. ACMs can also be isolated from different genetic and disease models, as in the Akita mice used here, to interrogate alterations in function at the cellular level. A disadvantage is that mouse ACMs are not amenable to genetic manipulation and have a limited lifespan. For these reasons, we chose to optimize conditions to monitor PKA signaling using an immunofluorescence technique. This approach, to the best of our knowledge, has not been taken before in adult ACMs. The antibody that we used to measure PKA signaling is validated by the observations that the immunofluorescence intensity is responsive to a β-agonist (ISO), a PKA agonist (8Br-cAMP), and PDE inhibitors. Thus, by this methodology we were able to visualize how insulin affects PKA signaling at the cellular level, specifically in a homogenous cardiomyocyte population, and independently of other systemic factors. Little difference was seen in the subcellular distribution and pattern of PKA substrate phosphorylation in wildtype cells stimulated with either ISO or 8Br-cAMP when cells were culture in the presence of insulin. This suggests that similar pools of PKA were activated by both agonists, resulting in phosphorylation of downstream substrates throughout the cell. When wildtype ACMs were cultured in the absence of insulin, though, PKA phosphorylation was substantially blunted in response to all three agonists examined. In Akita ACMs PKA substrate phosphorylation was decreased as compared to wildtypes. Furthermore, insulin had no beneficial effects on PKA substrate phosphorylation in Akita ACMs. Interestingly, though, when PKA phosphorylation was examined by Western blot analysis using the same antibody as in the immunofluorescence experiments, no significant differences were seen in the pattern of proteins phosphorylated. This was further corroborated by measuring total PKA activity. We saw that the absence of insulin reduced basal PKA activity. However, ISO stimulated PKA activity similarly in ACMs cultured with or without insulin. The cause of this reduced basal PKA activity is not due to reduced PKA catalytic subunit content and needs further investigation. One possibility is that the absence of insulin induces changes in ACM morphology and impairs PKA substrate detection by immunofluorescence. This is countered, though, by the finding that the phosphorylation of PBN and TnI were robustly induced by agonists even in the absence of insulin. Rather, we suggest that there is a differential recognition of PKA substrate epitopes depending upon the protein status. The immunofluorescence protocol maintains proteins in a native conformation, as compared to Western blot where proteins are denatured. The effects of insulin on global PKA substrate phosphorylation followed a similar pattern to that observed with phospho-PFK-2 staining. The cardiac isoform of PFK-2 is phosphorylated in response to insulin or β-agonists to increase the levels of fructose-2,6-bisphosphate, an allosteric activator of PFK-1. This serves to increase glycolytic flux. We have previously reported that PFK-2 content is decreased in the absence of insulin. In the streptozotocin toxin-induced type 1 diabetic model this results in a constitutive decrease in its content. In addition, the PFK-2 that remains is not phosphorylated in response to PKA activation. Our immunofluorescence data follows a similar pattern. The absence of insulin decreases basal PFK-2 phosphorylation and dampens its phosphorylation in response to PKA agonists. In Akita ACMs, we observed a constitutive decrease in phospho-PFK-2. Interestingly, though, culturing Akita ACMs overnight with insulin failed to rescue PFK-2 phosphorylation. This suggests that the mechanisms that normally regulate the expression and activation of PFK-2 are not affected by administration of insulin for 18h. While the pattern of PKA substrate phosphorylation approximates that of phospho-PFK-2, we cannot rule out that other PKA substrates are also affected by the insulin status. Furthermore, insulin also activates the mitogen-activated protein kinase/extracellular-regulated kinase pathway and effects related to the stimulation of these other signaling cascades must be further evaluated. We demonstrate here that insulin affects PKA signaling in ACMs. Reciprocally, previous studies have shown that PKA can affect insulin signaling. In neonatal rat cardiomyocytes and mice with chronic β-adrenergic stimulation there is a PKA-dependent decrease in insulin signaling. The mechanism involves insulin receptor desensitization, mediated by Akt, and is manifested by a decrease in GLUT4 content and translocation upon insulin stimulation. This PKA-mediated effect contributes to the insulin resistance that is manifested in failing hearts. These role of PKA in modulating insulin signaling, though, are dependent upon the duration of PKA activation. With short-term activation of PKA there is synergistic enhancement of Akt phosphorylation, GLUT4 translocation, and glucose uptake. This mediates the increase in glucose uptake and oxidation in response to acute β-adrenergic stimulation. Our results provide new insights into how diabetes may impact the heart. We show a decrease in PKA signaling in adult cardiomyocytes when insulin is absent. A novelty of this study was the use of immunofluorescence microscopy as a means of monitoring PKA signaling. As we show, other methodologies, such as Western blot, would have missed these apparent changes in substrate phosphorylation. Another novel aspect of this study was using primary adult cardiomyocytes isolated from Akita mice. Immortalized cells cannot recapitulate the unique morphology and metabolic aspects of primary cells. We demonstrate that the lack of insulin affects the phosphorylation of PFK-2, while the phosphorylation of contractile proteins was similar in control and Akita ACMs. Future studies must be performed to more exhaustively identify what other substrates may be affected. # Supporting information 10.1371/journal.pone.0231806.r001 Decision Letter 0 Kanzaki Makoto Academic Editor 2020 Makoto Kanzaki 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. 8 May 2020 PONE-D-20-09116 Diabetes induced decreases in PKA signaling in cardiomyocytes: the role of insulin PLOS ONE Dear Dr. Humphries, 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. Your manuscript was reviewed by two knowledgeable referees in this area and their comments are appended. As you will see they had a numerous major concerns that will need to be properly addressed by the authors before I can proceed further. In particular, they both raised technical issues including the fluorescent-based evaluation methods. In addition, they felt that conclusion drawn is not fully supported by the results presented. The authors need to address/respond all their comments to fully satisfy both reviewers. We would appreciate receiving your revised manuscript by Jun 22 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, Makoto Kanzaki, 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\. PLOS ONE now requires that authors provide the original uncropped and unadjusted images underlying all blot or gel results reported in a submission’s figures or Supporting Information files. This policy and the journal’s other requirements for blot/gel reporting and figure preparation are described in detail at <https://journals.plos.org/plosone/s/figures#loc-blot-and-gel- reporting-requirements> and <https://journals.plos.org/plosone/s/figures#loc- preparing-figures-from-image-files>. When you submit your revised manuscript, please ensure that your figures adhere fully to these guidelines and provide the original underlying images for all blot or gel data reported in your submission. See the following link for instructions on providing the original image data: <https://journals.plos.org/plosone/s/figures#loc-original-images-for-blots-and- gels>. In your cover letter, please note whether your blot/gel image data are in Supporting Information or posted at a public data repository, provide the repository URL if relevant, and provide specific details as to which raw blot/gel images, if any, are not available. Email us at <plosone@plos.org> if you have any questions. \[Note: HTML markup is below. Please do not edit.\] 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: Partly \*\*\*\*\*\*\*\*\*\* 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: 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: This study shows that PFK-2, a regulator of glucose metabolism, of wild type cardiac myocytes is downregulated in insulin-dependent manner. However, PFK-2 of Akita diabetic model mice was decreased and was not recovered by the addition of insulin. It was suggested that insulin-dependent decrease of PFK-2 expression is a mechanism of metabolic regulation by insulin in diabetes. However, there are several concerns as described below. However, in vitro assay system does not mimic in vivo phenomena as described in following comments. It is difficult to conclude that PFK-2 is an insulin- dependent target molecule in diabetic heart. The followings are comments. 1\. When the effects of insulin are due to the decrease of PFK-2 expression level, overexpression of PFK-2 should restore the effects of cAMP signal-induced responses. Authors described at lines 398 o 399 that ‘a disadvantage is that ACMs are not amenable to genetic manipulation and have a limited lifespan’. However, there is the report that adenovirus is successfully used for expression of FRET biosensors in adult mouse cardiomyocytes (Methods Mol Biol. 2015; 1294:103-15). Therefore, expression of genes of interest is possible. The experiment of PFK-2 overexpression should be done. 2\. In Fig. 1B, PKA substrates detected by anti-PKA consensus sequence-antibody represents all the proteins phosphorylated by PKA. However, minor but important PKA-phosphorylating proteins are not apparent from this detection system. 3\. Immunofluorescence of PKA substrates is not quantitative. The mRNA expression levels should be measured. Then, it will provide information that insulin regulates transcription, translation, stability of protein. 4\. In many cases, signaling intensity by Western blot is shown as a ratio of phosphorylated (active) form to total protein content. When the phosphorylated form decreases in diseased state and the total protein content also decreases, the ratio of phosphorylated form to total protein content is almost the same as that of control group. In this case, fold activation of diseased mice by β-adrenergic receptor stimulation will be about the same as control mice. When fold stimulation is same between control and diseased states, it cannot be concluded that decreased PKA signaling is due to the decreased expression of total protein content. 5\. PKA signaling is mediated by signaling complex being comprised of PKA, A-kinase anchoring protein (AKAP), target protein, and other signaling proteins. Intensity, efficiency and specificity of PKA signaling depend on AKAPs in many cases. Therefore, the effects of insulin on decreased PKA signaling may be explained by the decreased expression of AKAPs. Western blot and immunofluorescence data of AKAPs expressing in the myocytes are necessary to substantiate the conclusion. 6\. When phosphorylation states is really wanted to be examined, phospho- proteomics analysis may be better than Western blot instead of using anti-PKA consensus sequence-antibody. 7\. Insulin has growth factor-like effects. Therefore, overnight culture without insulin may affect cell growth or maintenance. Thus, the effects of insulin on the isolated cells cultured overnight without insulin may be growth factor-like actions but not metabolic actions as suggested. 8\. Insulin treatment may modify epigenetic modification during development of diabetes. It is interesting to examine insulin-induced epigenetic modification in myocytes of Akita diabetic model mice. Minor comment 1\. At lines 456 to 458, authors described that ‘our results support that the deficiency of PKA signaling is not mediated by loss of β-adrenergic receptors, PKA protein, or cAMP production/degradation’. However, authors do not measure number of β-adrenergic receptors, and amounts of PKA protein and PDEs/adenylyl cyclases. It is too much to say that. 2\. There are many proteins that their activities re regulated by insulin in myocytes. The decrease of PFK-2 level may not be enough to explain insulin- dependent metabolic changes in the heart. Reviewer \#2: In the manuscript titled “Diabetes induced decreases in PKA signaling in cardiomyocytes: the role of insulin”, the authors test the hypothesis that decreased insulin signaling contributes to dysfunctional PKA response in adult rat ventricular myocytes. The authors use a novel technique of immunofluorescence microscopy to monitor PKA signaling in fixed myocytes cultured from WT or Akita type 1 diabetic mice. The results demonstrate that a lack of insulin in culture conditions is associated with reduced PKA signaling when measured by immunofluorescence. Further, the authors explore the possibility that phosphorylation of the glycolytic regulator PFK2 and not contractile proteins were responsible for the overall reduction in PKA signaling. The authors conclude that deficient insulin signaling decreases PKA signaling in adult myocytes, which may provide insights into how diabetes impacts the heart. Major issues 1\. There are several inconsistencies between the written text and what is shown in the figures. For example, line 191 onward describes WB data in Figure 1B, Figure 1B however, shows something different, not WB data. Continuing from that point, the authors state in line 199 "culturing myocytes in the presence of insulin for 18hrs had no enhancing effect on PKA substrate staining upon ISO stimulation". The authors need to be clear if they are comparing this to WT or baseline, as compared to baseline there is an effect with ISO stimulation in all groups. Far too much of my time was spent trying to interpret the data and I would urge the authors to proofread and ensure the manuscript is concise and to the point. 2\. No in vivo mouse data is provided in the manuscript. While the Akita model has previously been reported to be associated with diabetes, there is no data provided demonstrating that this model serves as a mechanism of low insulin as a result of diabetes in this study. It would be interesting to determine the overall insulin levels of Akita mice and how does this compare to WT without insulin. 3\. The use of immunofluorescence microscopy to monitor PKA signalling, as has been demonstrated in this study, has not been done before in adult myocytes. I am, therefore, surprised the authors have not validated this technique with other well-know measurements of PKA signaling such as a PKA activity kit to confirm the results. Immunofluorescence can often lead to false results; therefore the use of an antibody control is strongly advised. The manuscript would also benefit from clarification of how the mean fluorescence intensity images were normalized and the control measures taken to ensure there are no false positive signals. 4\. Extended time in culture leads to great changes in myocyte morphology. For example, the structural integrity of the cells, such as t-tubule organization becomes disrupted after prolonged culture. Both insulin receptors and B-receptors are located on the t-tubule membrane, thus it cannot be ruled out that signaling alterations could be a consequence of culture. Moreover, just the presence of insulin in the culture media may have metabolic benefits that prove more advantageous for long culture conditions. To address this, a time-course study would be useful, starting with the effects of insulin on PKA signaling in freshly isolated cells. 5\. There are concerns regarding the authors’ interpretation of the pPFK2 data. Especially, with the conclusion that PFK2 phosphorylation was decreased in an insulin-dependent manner. Although blunted compared with WT insulin hearts, WT hearts in the absence of insulin showed increased pPFK2 in response to ISO and IBMX. In the Akita hearts, however, ISO and IBMX failed to have any effect. This suggests that it is not just the absence of insulin, but something specific to the Akita mice leading to decreased pPFK2 activity. Minor issues 1\. Were the stats performed on cells or animals, have the authors considered performing nested statistics to account for cells and animals? 2\. It would be useful if the graphs displayed individual data points. 3\. The authors state PKA staining is found on the z-bands, but there is no mention as to which software was used to determine the position on the PKA activity staining? \*\*\*\*\*\*\*\*\*\* 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.0231806.r002 Author response to Decision Letter 0 30 Jun 2020 We’d like to thank the reviewers for their careful evaluation of our manuscript. We have addressed their concerns to the best of our ability. This includes extensive revision of the manuscript and an additional experiment. Reviewer 1: Reviewer 1’s general comment is that “It is difficult to conclude that PFK-2 is an insulin-dependent target molecule in diabetic heart.” However, this current study is supported by, and builds upon, our previous report that PFK-2 content is regulated by insulin signaling (see Bockus et al. 2017, Journal of American Heart Association; <https://www.ahajournals.org/doi/full/10.1161/jaha.117.007159>). We concur that other targets are likely to be affected by the presence or absence of insulin, but this and our previous studies support phosphorylation of PFK-2 is abnormal in the diabetic heart. 1\. “When the effects of insulin are due to the decrease of PFK-2 expression level, overexpression of PFK-2 should restore the effects of cAMP signal-induced responses. Authors described at lines 398 o 399 that ‘a disadvantage is that ACMs are not amenable to genetic manipulation and have a limited lifespan’. However, there is the report that adenovirus is successfully used for expression of FRET biosensors in adult mouse cardiomyocytes (Methods Mol Biol. 2015; 1294:103-15). Therefore, expression of genes of interest is possible. The experiment of PFK-2 overexpression should be done.” Thank you for providing the suggestion and the reference. As noted in this elegant paper by Zaccolo et al., there are limitations with adenovirus expression in adult mouse cardiomyocytes. First, it takes a high MOI for protein expression. We have also found this to be true, and as Zaccolo noted, this has toxicity issues. Secondly, the referenced paper reports that the timing of an adenoviral transduction experiment is dependent upon the protein that is being expressed. In the case of the referenced work, they are expressing a FRET reporter with favorable fluorescent properties that allows detection by 24h. In our experience it takes approximately 48-72h, if at all, to detect the expression of unlabeled proteins in adult mouse cardiomyocytes. Unfortunately, at this point there is also significant loss of viable cells which may be further confounded by the insulin status. Thus, while such a rescue experiment would help to support our conclusions it is technically unfeasible. We have edited the manuscript to reflect some of the limitations. Furthermore, we focus on the phosphorylation status of PFK-2 and not on its total content. 2\. In Fig. 1B, PKA substrates detected by anti-PKA consensus sequence-antibody represents all the proteins phosphorylated by PKA. However, minor but important PKA-phosphorylating proteins are not apparent from this detection system. We agree that there might be additional substrates we can’t detect by Western. We clarified in the text that other substrates may be phosphorylated but not detected (lines 220-222). 3\. “Immunofluorescence of PKA substrates is not quantitative. The mRNA expression levels should be measured. Then, it will provide information that insulin regulates transcription, translation, stability of protein.” We have attempted to make the immunofluorescence data as quantitative as possible by keeping the experimental conditions constant. This includes how the cardiomyocytes were prepared and the microscopy conditions. We have previously shown that in the diabetic heart PFK-2 protein levels decrease without a change in transcript levels (Bockus et al., 2017). We concluded that PFK-2 stability is greatly decreased in the absence of insulin signaling. While we agree the transcript levels of PKA substrates may be affected by insulin, such an effect may be very selective. It is also worth noting that the phosphorylation of other substrates, such as phospholamban and troponin, were unaffected by insulin. 4\. In many cases, signaling intensity by Western blot is shown as a ratio of phosphorylated (active) form to total protein content. When the phosphorylated form decreases in diseased state and the total protein content also decreases, the ratio of phosphorylated form to total protein content is almost the same as that of control group. In this case, fold activation of diseased mice by β-adrenergic receptor stimulation will be about the same as control mice. When fold stimulation is same between control and diseased states, it cannot be concluded that decreased PKA signaling is due to the decreased expression of total protein content. This is true and a limitation of the immunofluorescence technique. Nevertheless, we can tell that the response to a PKA agonist is different depending upon whether cardiomyocytes were in the presence or absence of insulin. I made it clear that we are referring to the phosphorylation of proteins and not total proteins in our immunofluorescence experiments. Furthermore, the limitations of the experiments are now more clearly stated. 5\. PKA signaling is mediated by signaling complex being comprised of PKA, A-kinase anchoring protein (AKAP), target protein, and other signaling proteins. Intensity, efficiency and specificity of PKA signaling depend on AKAPs in many cases. Therefore, the effects of insulin on decreased PKA signaling may be explained by the decreased expression of AKAPs. Western blot and immunofluorescence data of AKAPs expressing in the myocytes are necessary to substantiate the conclusion. This is a good point. Changes in PKA may be due to alterations in its subcellular localization and mediated by AKAPs. Over 30 different AKAPs have been identified (<https://doi.org/10.1016/j.cellsig.2017.05.012>). It is unfortunately beyond the scope of this paper to determine if changes in AKAPs contribute to the insulin-dependent effects on PKA activity. We have previously reported though that there is no apparent change in AKAPs in the hearts of STZ model of type 1 diabetic mice using a binding assay (Bockus and Humphries, J. Biol. Chem., 2015, 290(49): 29250). In addition, we have now directly measured PKA activity (new Fig 1E). We show that the lack of insulin decreases basal PKA activity but not its activation by ISO. While we cannot rule out alterations in subcellular PKA, this supports that the primary defect is in discrete substrate phosphorylation. 6\. When phosphorylation states is really wanted to be examined, phospho- proteomics analysis may be better than Western blot instead of using anti-PKA consensus sequence-antibody. Phospho-proteomics is a powerful technique that will be used in future experiments, but it is beyond the scope of this work. 7\. Insulin has growth factor-like effects. Therefore, overnight culture without insulin may affect cell growth or maintenance. Thus, the effects of insulin on the isolated cells cultured overnight without insulin may be growth factor-like actions but not metabolic actions as suggested. We agree that insulin has many effects on cellular function that extend beyond metabolism. We do not see any change in cell number or cell size with overnight +/- insulin treatment. However, we cannot rule out other effects and have added this caveat to the Discussion (lines 482-484). 8\. Insulin treatment may modify epigenetic modification during development of diabetes. It is interesting to examine insulin-induced epigenetic modification in myocytes of Akita diabetic model mice. That is an interesting suggestion, but the epigenetic effects of insulin are beyond the scope of this work. Minor comments: 1\. At lines 456 to 458, authors described that ‘our results support that the deficiency of PKA signaling is not mediated by loss of β-adrenergic receptors, PKA protein, or cAMP production/degradation’. However, authors do not measure number of β-adrenergic receptors, and amounts of PKA protein and PDEs/adenylyl cyclases. It is too much to say that. 2\. There are many proteins that their activities re regulated by insulin in myocytes. The decrease of PFK-2 level may not be enough to explain insulin- dependent metabolic changes in the heart. 1)The last paragraph of the Discussion was edited and this statement was removed. 2) We agree that insulin has many effects on cardiac metabolism such as increasing glucose uptake, activating glycolysis, and increasing glucose oxidation via the activation of pyruvate dehydrogenase. The novel aspect of this study is identifying that the lack of insulin affects PKA signaling. We have updated the Discussion to reflect other possibilities. Reviewer 2 1\. There are several inconsistencies between the written text and what is shown in the figures. For example, line 191 onward describes WB data in Figure 1B, Figure 1B however, shows something different, not WB data. Continuing from that point, the authors state in line 199 "culturing myocytes in the presence of insulin for 18hrs had no enhancing effect on PKA substrate staining upon ISO stimulation". The authors need to be clear if they are comparing this to WT or baseline, as compared to baseline there is an effect with ISO stimulation in all groups. Far too much of my time was spent trying to interpret the data and I would urge the authors to proofread and ensure the manuscript is concise and to the point. I apologize for the lack of clarity. The references to Fig 1 in the text now correspond to the figures accurately. In addition, we have changed the text extensively to improve clarity. Please note the highlighted regions throughout the Results section where it is made clear what groups are being compared. 2\. No in vivo mouse data is provided in the manuscript. While the Akita model has previously been reported to be associated with diabetes, there is no data provided demonstrating that this model serves as a mechanism of low insulin as a result of diabetes in this study. It would be interesting to determine the overall insulin levels of Akita mice and how does this compare to WT without insulin. The Akita mice are a well-established model of diabetes. The first study that characterized this model showed that mice are hypoinsulinemic and hyperglycemic. In vitro perfusion studies showed pancreatic beta cells do not release insulin in response to high glucose (Yoshioka et al. Diabetes, (1997) 46: 887). I agree it would be interesting to see how well we are approximating insulin levels in Akita mice, in vivo, to WT ACMs without insulin. We unfortunately didn’t measure circulating insulin levels in these experiments. However, we did check blood glucose levels in the mice at the time of sacrifice. Blood glucose levels of all Akita mice were \> 400 mg/dl. This information has been added to the Methods section. 3\. The use of immunofluorescence microscopy to monitor PKA signalling, as has been demonstrated in this study, has not been done before in adult myocytes. I am, therefore, surprised the authors have not validated this technique with other well-know measurements of PKA signaling such as a PKA activity kit to confirm the results. Immunofluorescence can often lead to false results; therefore the use of an antibody control is strongly advised. The manuscript would also benefit from clarification of how the mean fluorescence intensity images were normalized and the control measures taken to ensure there are no false positive signals. Based on the reviewer’s suggestion, we have performed additional experiments to measure PKA activity in ACMs cultured in the presence or absence of insulin. This is now shown in Fig. 1E. Graduate student Maria Newhardt completed these experiments and has been added as an author. We report that basal PKA activity is decreased in wildtype ACMs cultured in the absence of insulin. Nevertheless, PKA was activated to a similar extent in the -/+ insulin conditions upon the addition of ISO. This supports that the robust insulin-dependent change in PKA substrate phosphorylation observed by immunofluorescence is likely due to changes in specific substrates and not a global defect in PKA activity. We agree that immunofluorescence data can be problematic. This is alleviated, in part, by using commercially sourced and validated antibodies. For the PKA substrate antibody, we have confidence in the results and specificity because of the responsiveness of the fluorescence signal to 3 different PKA activators: ISO, 8Br-cAMP, and PDE inhibitors. We also note that our experiments use the same secondary antibodies throughout the study. Staining was essentially undetectable in secondary antibody alone control experiments. This gives us confidence that the signal we are seeing is from the specificity of the primary. The confocal microscopy images were collected using the same settings between all experiments. This ensured reproducibility and allowed for unbiased comparisons between experimental conditions. This information has been added in the Materials and Methods section. MFI was calculated on maximum intensity projection images using the Zeiss Zen Black software. 4\. Extended time in culture leads to great changes in myocyte morphology. For example, the structural integrity of the cells, such as t-tubule organization becomes disrupted after prolonged culture. Both insulin receptors and B-receptors are located on the t-tubule membrane, thus it cannot be ruled out that signaling alterations could be a consequence of culture. Moreover, just the presence of insulin in the culture media may have metabolic benefits that prove more advantageous for long culture conditions. To address this, a time-course study would be useful, starting with the effects of insulin on PKA signaling in freshly isolated cells. It is a limitation of ACMs that cells begin to dedifferentiate in culture. Our experiments our completed within 18h of isolation and we see no overt changes in morphology or viability when insulin is excluded from the media. Regarding the receptors, our experiments in Fig 7 show that reintroduction of insulin after 18h of culture results in robust phosphorylation of Akt. This would argue that our immunofluorescence results are not from decreased insulin receptors or their accessibility. Likewise, we see PKA signaling dysfunction when ACMs are cultured in the absence of insulin and are then stimulated with 8Br-cAMP or PDE inhibitors. 8Br-cAMP activates PKA directly and argues against a B-receptor defect. Changes in receptors cannot be ruled out, though. Regarding the time course, we have previously attempted such an experiment but it was subject to high variability. The reason, we believe, is that the cardiomyocytes are required to be in plating media containing horse serum when they are first isolated. This is then changed to serum-free culture media. We surmised that the variability arose from the persisting effects of the sera. For this reason, we chose 18h because cells sustained viability (regardless of -/+ insulin), morphology was not overtly different, it approximates a physiological fasted state, and the results were very consistent. 5\. There are concerns regarding the authors’ interpretation of the pPFK2 data. Especially, with the conclusion that PFK2 phosphorylation was decreased in an insulin-dependent manner. Although blunted compared with WT insulin hearts, WT hearts in the absence of insulin showed increased pPFK2 in response to ISO and IBMX. In the Akita hearts, however, ISO and IBMX failed to have any effect. This suggests that it is not just the absence of insulin, but something specific to the Akita mice leading to decreased pPFK2 activity. I agree with the reviewer’s concern and I have modified the text to reflect other potential defects affecting PFK2 phosphorylation. I have also removed the sentence: “Thus, the immunofluorescence staining of phospho-PFK-2 follows closely to that of PKA substrate antibody.” Minor issues 1\. Were the stats performed on cells or animals, have the authors considered performing nested statistics to account for cells and animals? We used Akita and WT mice that were, whenever possible, littermates. In addition, we attempted to minimize variability with our cell preparations by starting the experiment at the same time of day, using aliquots of reagents, and having the same personnel perform the experiments. I have consulted with a statistics expert, Albert Batushansky (see Batushanksy, Humphries, et al. Metabolomics 2019), and he confirmed our statistics are appropriate for the experimental design. 2\. It would be useful if the graphs displayed individual data points. We believed the whisker plots were appropriate because it makes it easier for the reader to visualize differences between experimental conditions. Here is a representative plot with all the data points presented for the reviewer’s consideration (please see the response to reviewers in the PDF). We are willing to modify the figures or include all the data points in supplemental figures. 3\. The authors state PKA staining is found on the z-bands, but there is no mention as to which software was used to determine the position on the PKA activity staining? The staining was largely diffuse but with apparent increased intensity along the membranes. There was also staining of the filaments that was qualitatively consistent with Z-bands. We have modified this statement. 10.1371/journal.pone.0231806.r003 Decision Letter 1 Kanzaki Makoto Academic Editor 2020 Makoto Kanzaki This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 15 Jul 2020 PONE-D-20-09116R1 Diabetes induced decreases in PKA signaling in cardiomyocytes: the role of insulin PLOS ONE Dear Dr. Humphries, 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. Your revised manuscript was reviewed by the original referees, and their comments are appended. As you will see they both recognize that the revised manuscript has adequately improved, while reviewer \#2 pointed out "n" issue. The authors need to properly address this issue. Please submit your revised manuscript by Aug 29 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, Makoto Kanzaki, 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 \#2: (No Response) \*\*\*\*\*\*\*\*\*\* 2\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Yes Reviewer \#2: Partly \*\*\*\*\*\*\*\*\*\* 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: Thank you for the opportunity to review this revised manuscript. The authors have thoroughly revised the paper to address my primary concerns. The authors have modified the text to provide clarity and clearly state the limitations of the study; making the study more accessible. Furthermore, adding additional experiments to measure PKA activity in ACMs make the results more robust and support the immunofluorescence data on insulin dependent change in PKA substrate phosphorylation. At this stage, I have one minor concern, it is still not clear what exactly is N in this study. Are the statistics are performed on biological repeats (n=3) or cells (n= approx. 30) as is shown on the example data point plot provided? If it is biological repeats were the cell data averaged for each animal? On the contrary, if the stats were performed on cells, the authors need to be wary of pseudoreplication. It is recommended that a line is added to the methods section to clarify this. \*\*\*\*\*\*\*\*\*\* 7\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No Reviewer \#2: No \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <figures@plos.org>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0231806.r004 Author response to Decision Letter 1 28 Jul 2020 We’d like to thank the reviewers for their evaluation and constructive critiques of our original and revised manuscripts. Reviewer 2: At this stage, I have one minor concern, it is still not clear what exactly is N in this study. Are the statistics are performed on biological repeats (n=3) or cells (n= approx. 30) as is shown on the example data point plot provided? If it is biological repeats were the cell data averaged for each animal? On the contrary, if the stats were performed on cells, the authors need to be wary of pseudoreplication. It is recommended that a line is added to the methods section to clarify this. I apologize that this wasn’t clear. The experiments under a given condition were done on 3 cell preparations from 3 hearts. The statistics were performed on the sum of all the cells that were imaged. I have added additional information in the Methods section and in each of the Figure Legends to make this clear. I agree that n= appox. 30 can be considered as pseudoreplication, or as we considered them repeated measurements, and this could be a problem. However, when the cells from a given preparation are averaged and analyzed as n=3 similar conclusions are still reached. This information has been added in the “Statistical analysis” section in the Methods. For example, here is Figure 1 plotted with all the cell data shown or with the data plotted as averages of the 3 cell preparations. (These example Figures can be seen in our "Response to reviewers" in the PDF). We believe that showing the whisker plots encompassing all the cells more clearly demonstrates the data distribution and variability intrinsic to the method. 10.1371/journal.pone.0231806.r005 Decision Letter 2 Kanzaki Makoto Academic Editor 2020 Makoto Kanzaki 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 Aug 2020 Diabetes induced decreases in PKA signaling in cardiomyocytes: the role of insulin PONE-D-20-09116R2 Dear Dr. Humphries, 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, Makoto Kanzaki, 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: Thank you for the clarity and the additional information to address my concern regarding the statistics. All my comments have now been fully addressed. \*\*\*\*\*\*\*\*\*\* 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.0231806.r006 Acceptance letter Kanzaki Makoto Academic Editor 2020 Makoto Kanzaki 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 Aug 2020 PONE-D-20-09116R2 Diabetes induced decreases in PKA signaling in cardiomyocytes: the role of insulin Dear Dr. Humphries: 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. Makoto Kanzaki Academic Editor PLOS ONE [^1]: The authors have declared that no competing interests exist.

# Introduction Huntington’s disease (Huntington’s chorea, HD) is a severe autosomal dominant disease caused by an increase in the number of CAG (cytosine-adenine-guanine) trinucleotide repeats in the first exon of the huntingtin (*HTT)* gene. The mutant HTT protein that is expressed from the gene with more than 35 repeats leads to death of brain cells, which causes impairment of motor and cognitive functions. Even though a mutation in the *HTT* gene was described more than 20 years ago, the molecular and cellular mechanisms of HD are still largely unclear. The pathogenesis of HD has been shown to involve impairment of mitochondrial function, Ca²⁺ homeostasis, and autophagy. Many factors contributing to HD have not yet been determined. Adverse changes in the functions and in interactions of neuronal organelles in HD have also been observed. Medium spiny neurons of the striatum undergo pathological processes at the first stage of disease development, and these processes then spread to other parts of the brain. Studies on mutant neurons have revealed significant disturbances in the structure and dynamics of mitochondria and in their contacts with endoplasmic reticulum (ER) membranes; these problems lead to impairment in calcium ion homeostasis as well as in autophagy and particularly mitophagy. Elucidation of the influence of *HTT* mutation on the fine organization of cells and intracellular organelles, such as mitochondria, ER cisternae, and components of the autophagic system, remains one of the essential issues in the HD pathology research. To understand the successive stages of development of neurodegenerative diseases under the influence of mutant proteins and to search for possible drug targets, both model animals reproducing the pathological phenotype of the disease and neuronal cell models based on patient-specific induced pluripotent stem cells (iPSCs) are currently used. Nonetheless, the results obtained via the patient- specific cell-based approach are significantly influenced by the genetic background of a cell line under study. More promising is the creation of cellular models based on isogenic lines of human cells carrying relevant mutant alleles of the *HTT* gene. Advances in genome-editing technologies based on the CRISPR/Cas9 system give investigators an opportunity to create isogenic cell clones differing only in allelic variants of a target gene. In the present study, we investigated the ultrastructure of human cells of three isogenic mutant clones with deletions or insertions in the *HTT* gene. The mutant cell clones were obtained for the first time via introduction of an HD- causing mutation by the CRISPR/Cas9 technology. A comprehensive analysis by electron microscopy showed that deletion of three CAG repeats or a functional *HTT* knockout by means of a reading frame shift had practically no effect on morphology of the cells, whereas an increased number of CAG repeats caused significant disturbances in the organization of the envelope, cristae, and matrix of mitochondria; stimulated their contacts with ER membranes; and increased the number of autolysosomes and their anomalous variants in the cytoplasm. Morphometric analysis confirmed the structural alterations observed in the mutant cells. # Materials and methods ## Generation of the HTT-mutant isogenic cell clones The isogenic cell clones were created from the HEK293 Phoenix cell line by editing the genome with the CRISPR/Cas9 system. The HEK293 Phoenix cell line was kindly provided by Dr. Maria Lagarkova. HEK293 Phoenix cells were cultivated in a medium consisting of Dulbecco’s modified Eagle’s medium (DMEM)/F12 (1:1), 10% of fetal bovine serum, 200 mM GlutaMAX (Life Technologies, USA), and 1% of a penicillin/streptomycin solution (Life Technologies, USA). We designed a 20 nt single guide RNA (sgRNA) sequence (`5′-CGTGAGGGAGCGGGGCTGAA-3′`) to cut the genomic DNA 19 bases upstream of the CAG repeat tract in *HTT* exon 1 and inserted it into the pSpCas9(BB)-2A-GFP (pX458) plasmid (Addgene, USA). HEK293 Phoenix cells were cotransfected with both the CRISPR/Cas9-encoding plasmid and donor construct. The presence of the *EGFP* marker gene in the pX458 plasmid allowed for sorting of transfected cells by green fluorescence on Cell Sorter S3e (Bio-Rad, USA). Thus, single-cell clones of EGFP-positive cells were obtained. Genomic DNA was isolated from cell clones growing in 96-well plates using the Quick-DNA 96 Kit (ZymoResearch, USA). Length of the mutant *HTT* allele was analyzed by PCR (65 mM Tris-HCl pH 8.9, 16 mM (NH₄)₂SO₄, 1.5 mM MgCl₂, 0.05% Tween 20, 3% glycerol, 6% DMSO, 0.5 U Taq polymerase, primers 0.2 μM each, 50–200 ng genomic DNA)on a C1000 Touch Thermal Cycler (Bio-Rad, USA) (96°C 5 min; 40 cycles: 96°C 20 s, 72°C 3 s; additional synthesis 72°C 15 min) using primers HTT-F `5′-CCCAAGGCCACCTCGGCTCAGAGTC-3′` and HTT-R `5′-CGCAGGCTGCAGGGTTACCGCCATC-3′`. PCR products synthesized from total genomic DNA were sequenced (BigDye Terminator v3.1 Cycle Sequencing Kit, Thermo Fisher Scientific, USA) and analyzed in sequence trace decomposition software TIDE (<https://tide- calculator.nki.nl/>,) to reveal open reading frame (ORF)-shifting mutations. Immunofluorescent staining with an anti-huntingtin (N-terminal) antibody (Sigma, H7540, Germany) was performed as described earlier. HTT expression was confirmed by RT-PCR using primers `5′-GGATGGCGGTAACCCTGCAG-3′` and `5′-GATCAGACTCTCCCTTCTCCCGAC-3′`. Cell lysates for western blotting were prepared in RIPA buffer (Thermo Fisher Scientific, USA). Cell lysate protein concentrations were determined by the BCA protein assay (Thermo Fisher Scientific, USA). Lysates (40 μg) were prepared in 4×SDS- PAGE loading buffer (200 mM Tris-HCl, pH 6.8, 400 mM DTT, 8% SDS, 0.4% bromophenol blue, 40% glycerol). The samples were boiled for 5 min at 98°C and separated by one-dimensional SDS-PAGE on a 7.5% acrylamide gel at 100 V for 4 h in running buffer (25 mM Trisbase, 3.5 mM SDS, 200 mM glycine) on ice. Overnight transfer was performed at constant voltage 30 V for 16 h onto an Immun-Blot PVDF membrane (Bio-Rad Laboratories, USA) in transfer buffer (25 mM Trisbase, 192 mM glycine, 10% ethanol) at 4°C. After blockage in 5% milk in TBST, a primary antibody (Sigma Anti-Huntingtin N terminus (3–16), H7540, 1:4000; Bethyl, anti- SMC1 antibody, A300-055A, 1:4000) was incubated with the membranes overnight at 4°C. The membranes were incubated with a secondary antibody (HRP-conjugated anti-rabbit IgG antibody; 1:10000, Jackson ImmunoResearch) at room temperature for 2 h in the blocking solution. Protein bands were detected by chemiluminescence with the Pierce ECL Western Blotting Substrate (Thermo Fisher Scientific, USA). ## Proliferation and viability assay The proliferation rates of the mutant clones were estimated using real-time cell analyzer iCELLigence (ACEA Biosciences, USA). The cells were seeded in electronic microtiter plates (E-Plates, ACEA Biosciences, USA) at 5 × 10⁴cells per well and analyzed on the RTCA iCELLigence (ACEA Biosciences, USA). The viability of mutant cell clones cultured under standard conditions was evaluated using the ApoDETECT Annexin V-FITC Kit (ThermoFisher Scientific, USA). The assay was performed in triplicate and analyzed by Wilcoxon’s test. ## Electron microscopy For electron microscopy, the cells were seeded at 5×10⁵/well in a 12-well plate. Cells growing in culture on the Melinex polyester film (175 μm thickness, Agar Scientific, UK) were fixed in 2.5% glutaraldehyde in the culture medium for 15 min, followed by fixing with 2.5% glutaraldehyde in 0.1 M sodium cacodylate buffer (pH 7.3) for 1 h at room temperature. The samples were washed three times in the same buffer and additionally fixed in 1%osmium tetroxide for 1 h, washed twice in double-distilled (dd) H₂O, and incubated in 1% uranyl acetate for 12 h at 4°C. Next, the samples were dehydrated in a graded series of ethanol solutions (from 30% to 100%, 10 min in each) and then acetone (twice, 10 min), were embedded in epoxy resin Epon 812 (Sigma, USA), and polymerized for 2 days at 60°C. The cells remained attached to the Melinex film for all treatments and embedding. The most cell-enriched areas were marked on the polymerized plates; then, a block with a diameter of approximately 2 mm was cut out. Ultrathin sections for transmission electron microscopy analysis (50 nm thickness) were cut off in parallel to the plane of the substrate by means of a diamond knife on a Leica EM UC7 ultramicrotome (Leica, Austria).The sections were examined and photographed under a transmission electron microscope, JEOL-1400 (JEOL, Japan), with a Veleta camera (Olympus, USA) and iTEM 5.1 software (Olympus, USA). ## Morphometric analysis Quantitative parameters of the organelles (autolysosomes and mitochondria) of the cells were evaluated on randomly selected sections obtained from three randomly selected areas of embedded samples. The operator was blinded to the assignment of sections to experimental groups. The proportions of mitochondria with various impairments as well as those in tight contact with ER membranes and other mitochondria were calculated as the percentage of the corresponding types of organelles in cell cytoplasm on the sections. In total, 60 randomly selected cells and approximately 600 mitochondria were examined for each cell clone. The relative number and volume density of autolysosomes were determined as the number of organelles or their area per 1 μm² area of the cytoplasm in a cell section. Autolysosomes were conventionally subdivided into two groups: small organelles having a maximum diameter of 0.1 to 0.6 μm and large ones 0.7 to 2 μm or larger. The measurements were carried out in the ImageJ freeware (ImageJ, USA, <https://imagej.nih.gov/ij/>). Significance of the differences in the compared mean values was verified by one-way ANOVA followed by Fisher’s multiple-comparison test in the SPSS Version 11.0 statistical software package (IBM Corp., Armonk, NY, USA). # Results ## Characterization of HEK293 cell lines harboring mutations in the *HTT* gene ### Creation of mutant HEK293 cell clones using the CRISPR/Cas9 genome-editing tool Using the genome-editing method based on the CRISPR/Cas9 system, we obtained a panel of isogenic cell clones with an identical genetic background differing in the number of CAG repeats in the *HTT* gene. An expanded CAG repeat tract was introduced into the first exon of *HTT* by homology-directed repair. First, we generated a donor construct bearing 215 CAG trinucleotide repeats flanked with long homology arms. Oligonucleotides containing three CAG repeats and recognition sites for type SII restriction enzymes, BsmBI and BbsI, were synthesized. The long trinucleotide repeat tract was obtained by repetitious rounds of digestion/ligation by the golden gate cloning method as described elsewhere. We used the pSMART cloning vector (Lucigen, USA) and the recA13Stbl3 *E*. *coli* strain, which are designed specifically for cloning direct repeats to reduce the frequency of recombination of long repetitive DNA tracts. Oligonucleotides containing three CAG repeats and recognition sites for type SII restriction enzymes, BsmBI and BbsI, were synthesized and introduced into the pSMART cloning vector. The plasmid was digested with restriction enzyme pairs PvuI + BbsI and PvuI + BsmBI. The plasmid fragments containing CAG repeats were then isolated and ligated to each other. The trinucleotide repeat sequence was elongated by means of repeated rounds of digestion–ligation according to the formula X = 2N–2, where N is the number of the digestion–ligation rounds. X is the resulting number of the CAG repeats. The 5′ and 3′homology arms were amplified from genomic DNA using the Phusion High-Fidelity PCR Master Mix with GC Buffer and introduced along with the expanded CAG tract into a donor plasmid (donor 215rep-HTT;). The donor construct served as a template for homologous recombination. HEK293 Phoenix cells were cotransfected with both the CRISPR/Cas9 encoding plasmid (pX458-HTT) and the donor construct. The presence of the *EGFP* marker gene in the pX458 plasmid allowed for sorting of transfected cells. Over 100 single-cell clones of EGFP-positive cells harboring both the expanded trinucleotide tract and deletions in *HTT* exon 1 were obtained. The length of the mutant *HTT* allele was analyzed by PCR. It was shown that only a truncated tract of 215 CAG repeats was introduced into the genome of cells during the homology-directed repair. It should be noted that some of the clones had more than two *HTT* alleles of different length because the HEK293 cells have a hypotriploid karyotype as confirmed by fluorescence *in situ* hybridization (FISH) analysis with a human chromosome 4 painting probe. It was revealed that the HEK293 Phoenix cells have two full-length copies of chromosome 4 and show a translocation of an additional small fragment of the chromosome 4 short arm. Cell clones with insertion into the *HTT* gene (6H cells) and two clones with deletions (8D and 8H cells) were subjected to sequencing, and the results were analyzed in sequence trace decomposition software TIDE (<https://tide- calculator.nki.nl/>,) to confirm the ORF shifts. It was found that the 8D cell clone has a deletion of three CAG triplets in the first exon of the *HTT* gene, the 8H clone bears a deletion causing an ORF shift, clone 6H has a normal allele and insertions of 100 and 150 CAG repeats into the gene. The ORF-shifting mutations are caused by the CRISPR/Cas9 retrial digest during the genome-editing procedure. ### The HTT expression assay The *HTT* gene is known to be widely expressed in different types of human tissues; therefore, we examined the mutant-HTT expression in the mutant clones. The immunofluorescent assay revealed cytoplasmic localization of the HTT protein, with a slightly increased HTT concentration in the vicinity of the nuclear envelope. No aggregates of mutant HTT were detected. ### The viability and proliferation assay Given that HEK293 cells express some neural markers that correlate with certain processes occurring in neural tissue, we decided to estimate viability and proliferation rates of the three mutant clones (8D, 8H, and 6H) compared with an isogenic control cell line. We measured proliferation rates of the three mutant clones compared to the original HEK293 Phoenix cell line by real-time cell analysis. The 6H clone has the slowest growth rate as compared to the other two clones and the original HEK293 Phoenix cell line. Increased apoptosis in mutant clones 6H and 8H compared with HEK293 Phoenix cells was revealed by an annexin V/propidium iodide cell viability assay. ## Ultrastructure of cultured HEK293 cells (control) The cells had an oval shape, large nuclei with electron-dense nucleoli, and some invaginations of the nuclear envelope. The cytoplasm contained a large number of polysomes, mitochondria, and short rough and smooth ER cisternae. Mitochondria in cells were either dispersed or grouped and could be classified into two types by morphology. The first type contained a light matrix and narrow parallelly oriented cristae, whereas the second type had a denser matrix and swollen cristae. It was observed that certain mitochondria came into contact with ER cisternae. Cells in the population could contain either both types of mitochondria or one of them predominantly. Some mitochondria had irregular arrangement of cristae together with low density of the matrix. The cytoplasm contained a well-developed Golgi complex consisting of cisternae and vesicles of various sizes. Autolysosomes of various diameters surrounded by one limiting membrane and some disintegrating mitochondria with destroyed cristae were also identified. ## Ultrastructure of mutant cell clones harboring a deletion in the *HTT* gene (clone 8D) or an *HTT* gene with a shift of the reading frame (clone 8H) The cells of clones 8D and 8Hhad similar shapes and structures of the nuclei as compared with the original control cells. Both types of mutant cells are characterized by the presence of mitochondria with a light or dense matrix, respectively, autolysosomes or autophagosomes , and a well-developed Golgi apparatus. Short and elongated rough ER cisternae contacting certain mitochondria in the cells of both clones were also observed. In some cells of both clones, mitochondria with a sparse (rarefied) matrix and dislocated cristae were identified, and accumulation of lipid droplets was detected in the cells of clone 8H. ## Ultrastructure of the mutant cells harboring the *HTT* gene with an insertion of 100–150 CAG repeats (clone 6H) The cells often had irregular shapes, contained large round or ellipsoidal nuclei, and were characterized by higher density of organelles in the cytoplasm as compared to the HEK293 cell line. In some cells, invaginations of the nuclear envelope and annular membrane inclusions in the nucleoplasm were observed. Many mitochondria had different morphological disturbances: protrusions of the outer membrane , changes in the shape of these organelles, the absence of cristae in large parts of the matrix, and the appearance of low density in the matrix. Significant deformation and degradation of mitochondrial cristae were also observed. A distinctive feature of this cell clone was the close contact between defective mitochondria and the presence of elongated or “doubled” mitochondria. Partly fragmented mitochondria or those completely lacking the envelopes were often located in the perinuclear region. Furthermore, defective mitochondria were also detected in clusters of lipid droplets associated with the membranes of the smooth ER. The cells were also characterized by the presence of both long and short cisternae of the rough ER that were closely in contact with normal and defective mitochondria. The distances between the membranes of the organelles in contact (gap) varied from 6 to 30 nm. Typical and unusual shapes of the annulate lamellae were often observed in the cytoplasm. A distinctive feature of the 6H cell clone was accumulation of early/small (less than 0.6 μm) and late/large (0.7–2.0 μm) autolysosomes, forming clusters in the cytoplasm of certain cells. Many large autolysosomes contained dense content adjacent to an empty electron- lucid vacuole and often showed disruption of the integrity of the membrane. Additionally, light inclusions of irregular shapes, without a limiting membrane and similar in size to these autolysosomes, were found in the cells. Furthermore, the cytoplasm was enriched in vesicles with a diameter of approximately 40–50 nm, observed presumably in the Golgi complex as well as near ER membranes and in the vicinity of autolysosomes. Notably, vesicles of similar size were frequently in contact with the outer membrane of defective mitochondria. Autophagosomes were randomly present in cells. ## Morphometric analysis of autolysosomes and mitochondrial parameters in HEK293 cells and in cells of different mutant clones According to the morphometric analysis, the relative number of small (early) autolysosomes, up to 0.6 μm in diameter, is significantly higher in comparison with that of these organelles having a diameter of 0.7–2.0 μm (late autolysosomes) in all the cell clones studied. Notably, the numbers of small and large autolysosomes, as well as their total number in the cells of the 6H clone doubled as compared to the HEK293 line. The number of small autolysosomes per 1 μm² of the cytoplasm was estimated as 0.023±0.003, 0.022±0.003, 0.021±0.003, and 0.04±0.006 for HEK293 cells and in clones 8D, 8H, and 6H, respectively. The number of large autolysosomes per 1 μm² of the cytoplasm was estimated at 0.006±0.002, 0.004±0.001, 0.002±0.001, and 0.015±0.003 in HEK293 cells and in clones 8D, 8H, and 6H, respectively. It was found that the proportion of small autolysosomes exceeded the relative number of large organelles 3.2-, 4.4-, 10.7-, and 3.1-fold in the HEK293 line and cell clones 8D, 8H, and 6H, respectively. The most noticeable differences were found in the cells of the 6H clone, where the numbers of small and large autolysosomes as well as their total number significantly increased (doubled) as compared to the HEK293 cell line. At the same time, the proportion of cells containing large autolysosomes increased in clone 6H in comparison with all the others (7% in clone 6H and 3% in clones 8D and 8H and in the control line). Nevertheless, the volume density of large autolysosomes did not significantly differ between cells of clone 6H and cells of the HEK293 line (0.09±0.03 and 0.08±0.03 μm²/μm², respectively). This finding suggests that the mean size of large autolysosomes in cells of clone 6H is less than that in cells of the HEK293 line. Analyses of the mitochondrial fraction with various morphological defects, such as the change in the shape of cristae, low density of the matrix, and disruption of the envelope structure, showed the presence of similar organelles in the cytoplasm of cells in all the clones. Nonetheless, their relative numbers were different. We found that matrix alterations were always accompanied by the cristae disturbances. Therefore, only the numbers of mitochondria with these two types of defects were estimated. The relative numbers of such organelles in the cells of the HEK293 line and in clones 8D, 8H were similar and amounted to 16.1%, 15.9%, and 16.5%, respectively, whereas the relative numbers of mitochondria with the envelope defects constituted respectively only 1.8%, 1.6%, and 2.31% of all mitochondria on cell sections. Cells of the 6H clone showed 5.5% of mitochondria with envelope defects and 37.8% of defective mitochondria; these mutant cells differed significantly (a 2.1-fold increase) from HEK293 cells in the proportion of mitochondria with all the above defects. A comparative analysis of the relative numbers of mitochondria with a contact uncovered a significant increase in the number of such organelles (1.4-, 1.6-, and 2.0-fold) relative to the control in cell clones 8D, 8H, and 6H, respectively. Contacts of mitochondria with ER membranes could be seen among organelles with normal or defective structure. The proportion of normal mitochondria was higher than that of defective ones 3.7-, 3.5-, and 3.2-fold in cell clones HEK293, 8D, and 8H, respectively. Whereas a significant increase (3.8-fold) in the proportion of ER-contacting defective mitochondria was recorded in 6H cells, there was a 1.8-fold increase in the total relative number of mitochondria in contact with the ER as compared with the control cells. In addition, the ratios of relative volumes of large autolysosomes in 6H cells and HEK293 line were similar (0.09±0.03 and 0.08±0.04, respectively), whereas the mean diameter of large autolysosomes in this mutant was lower than that in control cells (1.6±0.12 and 2.36±0.39 μm, respectively). # Discussion In this article, we presented a detailed description of the ultrastructural phenotype of three *HTT* mutant cell clones obtained by genome editing: a deletion mutant (clone 8D), an ORF shift mutant (clone 8H), and an insertion mutant with 100 and 150 CAG repeats (6H clone). The isogenic cell model of HD offers an opportunity to study molecular pathogenesis and to test drugs in the presence of an adequate healthy isogenic control that allows us to exclude the influence of genomic variation. In spite of the abnormal karyotype of HEK293 cells, we assumed that mutant-HTT expression should affect the cells and that mutant cell clones should recapitulate some molecular mechanisms of mHTT action because the HD-causing mutation is known to be dominant. Expression of the *HTT* gene was detected in both mutant and original cell lines, even though no mHTT protein aggregates were found in the 6H cell clone harboring mutant *HTT* alleles with expanded CAG repeat tracts. The evidence of intracellular protein aggregates with the expanded polyglutamine tract has been previously obtained in knock-in PC12 cells with the *HTT*exon1–polyQ transgene that expressed a low- molecular-weight HTT fragment. The absence of protein aggregates is supported by recent findings that formation of misfolded protein plaques is an age-dependent process that occurs only in postmitotic neurons. A similar cellular HTT localization pattern was described in COS cells as a model of HD. Nevertheless, our results suggest that mutant clones manifested enhanced apoptosis and a decreased proliferation rate as compared to an isogenic control. According to our study, when the number of *HTT* repeats in the cells was expanded, invaginations of the nuclear envelope and disturbances of the mitochondrial structure became more frequent. At the same time, increased contacts of mitochondria with membranes of the rough and smooth ER, fusion of mitochondria, and arise in the number of autolysosomes, vacuolation, and disturbances in the integrity of their membranes, were observed. A knockout or a shift of the reading frame of the *HTT* gene did not cause such changes in the morphology of mutant cells. ## Defects of mitochondria in cells harboring the *HTT* gene with an increased number of CAG repeats Beyond energy production, mitochondria are involved in the regulation of many processes in the cell: nuclear gene expression, Ca²⁺ homeostasis, cell stress, apoptosis, and necrosis; therefore, their normal structural organization, dynamics, and biogenesis are especially important for the effective functioning of brain nerve cells. According to our morphometric analysis, the total number of mitochondria containing various morphological defects varied from 16.9% to 18.9% of the entire population of such organelles in cells of the original HEK293 line with normal expression of the *HTT* gene and in both mutant clones with the deletion (clone 8D) or an ORF shift (clone 8H). We can speculate that the presence of such mitochondria in cultured cells is related to the process of mitochondria biogenesis, probably owing to their subsequent removal by autophagy.The proportion of defective mitochondria grew up to 38% in mutant 6H cells harboring an expanded CAG tract, as compared to the original cell line and the other mutant lines. This alteration apparently causes substantial dysfunction of these organelles. We believe that the accumulation of complex pathological changes in mitochondria included anomalous localization and changes in the shape of cristae as well as their partial absence, low density of the matrix of these organelles, and appearance of deformation in the envelope, probably reflecting the initial stages of their disintegration. Similar mitochondrial defects have been observed previously, in a study on the morphology of spiny neurons derived from iPSCs created from fibroblasts of HD patients with a short (42–47) tract of CAG repeats in the *HTT* gene. Besides, it has been shown that N-terminal mutant HTT fragments that are associated with mitochondria in the brain of Hdh (CAG) 150 knock-in mice affect the organelle trafficking, and this problem progresses with age and disease development. Pathological changes in the structure of mitochondria have also been described in a study on neurons obtained from brain biopsies, in patient-specific fibroblasts and myoblasts, in postmortem sections of brain cells of HD patients, and brain cells of transgenic mice modeling HD (150 CAG repeats in the gene). We suppose that the emergence of defects in the fine organization of mitochondria correlates with a decline in the glucose metabolism in brain cells and a decrease in respiratory-chain activity in mitochondria isolated from brain cells of patients with HD. Moreover, mutant huntingtin can cause mitochondrial dysfunction, which affects the activity of transcription factors regulating the expression of mitochondrial proteins in the nucleus of the cell, whereas aggregates of the mutant huntingtin protein are capable of destroying the nuclear envelope. Our experiments showed that in 6H mutant cells, strongly disrupted mitochondria were often present in close proximity or in contact with the nuclear envelope, and this arrangement probably influences the function of this membrane compartment. Similarly defective nuclei and swollen mitochondria have been identified in neurons of the striatum of transgenic mice. It has also been found that aggregates of the mutant huntingtin protein destroy the nuclear envelope of the cell. Therefore, it is likely that the observed close association of disintegrated mitochondria and nuclei in the mutant cells with an expanded length of the CAG repeat tract are similar to those found in brain neurons in HD and are probably related to the disruption of the energy metabolism of cells. Besides, our findings are consistent with the results of evaluation of the proliferation rate and the level of apoptosis of the mutant clones. These data suggest that mutations in the first exon of the *HTT* gene in HEK293 cells affect cell viability but do not cause cell death. Overall, our quantitative electron microscopy analysis showed relative contribution of different defects of mitochondrial substructures to the response of HEK293 line cells to expansion of CAG repeats in the *HTT* gene. ## Possible significance of mitochondrial contacts with the ER in mutant cells harboring the *HTT* gene with an increased number of CAG repeats In recent years, a lot of attention was given to the functional role of mitochondrial contacts with ER membranes called MAMs (mitochondria-associated membranes) related to the transfer of ions, lipids, and Ca²⁺ to mitochondria; these factors may substantially influence organelle properties. It has been demonstrated that dysfunction of such contacts leads to the development of neurodegenerative diseases. We revealed a 3.8-fold increase in the proportion of defective mitochondria that are in contact with the membranes of rough and smooth ER in the cells of clone 6H, which carries the insertion of 100 and 150 CAG repeats into *HTT* in comparison with the control HEK293 cell line. According to literature data, we supposed active participation of the ER in the increased proportion of defective mitochondria in the 6H cell clone, possibly owing to dysfunction of MAMs. It has been reported that neuronal death in HD is caused by an increased calcium ion concentration in the cytosol and a disturbance of the calcium concentration in mitochondria. We believe that the increase in the number of mitochondria in contact with ER cisternae in the mutant cells could facilitate the transport of calcium ions to the organelles and disrupt their balance between the organelles and cytosol. This notion is consistent with the data of other investigators who analyzed mitochondrial organization in the cells harboring mutant huntingtin. Besides the important role in the biogenesis of mitochondria, the ER participates in the processes of mitochondrial fission and fusion. Our experiments revealed an increase in the relative number of interacting mitochondria often associated with a low density of matrix in the cells of mutant clones 8D and 8H, and especially in cells of the 6H clone. Given that short ER membranes were observed in the vicinity of a region of contact with mitochondria, we propose that they are in some way involved in this process. We detected the amplification of annulate lamellae as well as their modification called “TUHMAs” (tubulohelical membrane arrays) with an aberrant organization. Therefore, we believe that these alterations may reflect the increase of the lipid synthesis in mutant cells. This finding is consistent with data showing that lipid accumulation is a defining feature of mutant HdhQ111 striatal progenitor cells culture. In our study, we observed contacts of defective mitochondria with 40–50 nm vesicles derived from the ER. We can speculate that this phenomenon may reflect the process of transfer of metabolites to the mitochondria not only via their direct interaction with ER cisternae but also via similar vesicles. Nevertheless, we cannot rule out that these vesicles represent early lysosomes. ## Alterations of the lysosomal system in mutant cells harboring *HTT* with an increased number of CAG repeats The maintenance of a balance between biogenesis and degradation of mitochondria and other organelles in nerve cells is provided by components of the lytic system including lysosomes, autolysosomes, and autophagosomes and it is one of the important conditions for their functional activity and survival. It has been shown that HTT is involved in the regulation of various autophagic processes such as autophagosome transport, fusion to lysosomes, abnormal protease cleavage, and an altered post-translational modification. The imbalance in the complex interactions of lytic components accompanies the development of HD and other neurodegenerative diseases. According to our data, cells of the mutant 6H clone are characterized by an increase in the total number of autolysosomes \[and especially the early (small) type\] per μm² of the cytoplasm on a section. This parameter was twice that of the parental HEK293 cell line. In addition, the lytic organelles of different sizes often formed clusters occupying a large part of the cell. Overall, these data allow us to theorize that the increase in the number of CAG repeats in the *HTT* gene activated the process of autolysosome formation. This finding exactly matches the results of other studies suggesting that the activity of early steps of autophagy during the development of HD in cell models and in mutant neurons obtained from iPSCs of HD patients is higher in comparison with normal cells. In addition, an increase in the concentration of autophagosome- and lysosome-like structures in the cytoplasm was demonstrated in the analysis of postmortem samples from patients with HD. According to our morphometric analyses, the ratio of numerical density of early autolysosomes with diameter less than 0.6 μm to the numerical density of larger autolysosomes (0.7–2.0 μm) in the control and all mutant cells was similar; this result could reflect the dynamics of the lytic components at different stages of autophagy and formation and functioning of these organelles. Whereas relative volumes of large autolysosomes in cells of clone 6H and of the HEK293 line were similar (0.09 ± 0.03 and 0.08 ± 0.04 respectively), their mean diameter in mutant cells was significantly (1.48-fold) smaller than that in control cells. One of possible explanations of this phenomenon is that the process of maturation of large autolysosomes in mutant cells was probably disrupted under the influence of the mHTT protein. It has been demonstrated that the mutant HTT protein inhibits autophagosome–lysosome fusion. Our analysis revealed a large number of defectively vacuolized large autolysosomes as well as the autolysosomes with significant disruption of the integrity of their membrane in cells of the 6H clone. Simultaneously, vacuole-like inclusions without membranes were detected in the cytoplasm, allowing us to assume that they can be derivatives of large autolysosomes, judging by their similar sizes. It has been reported that huntingtin-labeled vacuoles manifest the ultrastructural features of early and late autophagosomes (autolysosomes) in clonal striatal cells culture of the mouse brain. Additionally, huntingtin-positive vacuoles were identified in these cells. On the basis of these and our data, we suggest that an increased number of CAG repeats in the *HTT* gene initiates disturbances of the later stages of autophagy in mutant 6H cells. # Conclusions In this study, we generated a novel isogenic HD model based on the HEK293 cell line by means of the CRISPR/Cas9 system. The resultant mutant cells with 150 CAG repeats in *HTT* were found to undergo a wide spectrum of pathological changes that were confirmed by morphometric analysis. This analysis revealed structural defects of mitochondria (cristae, matrix, and envelope) together with proliferation of their contacts with the ER. We can hypothesize that these alterations should significantly disturb the energy-dependent metabolism of cells and the balance of Ca²⁺ between mitochondria and the cytoplasm as demonstrated in other studies on HD. The uncovered accumulation of early and late autolysosomes in mutant cells together with artificial vacuolization of large organelles and disruption of integrity of their limiting membrane in 6H mutant cells strongly correlate with the point of involvement of the HTT protein in stimulation of autophagy in the early period and dysregulation of this process at later stages. Our findings are consistent with studies on the morphology of neurons derived from iPSCs of patients with HD. Meanwhile, the use of isogenic cell lines allowed us to exclude the influence of the genetic background and ensured more accurate comparison of mutant cells’ structures. Thus, we for the first time characterized the HEK293 cell culture model with an expanded CAG tract (in *HTT*) obtained via the CRISPR/Cas9 technology and demonstrated that this model can be useful for research into the molecular mechanisms underlying the pathological function of mutant HTT in the development of HD. # Supporting information The study was supported by the Russian Scientific Foundation (project № 16-15-10128). We thank the Microscopic Center of Biological Objects of the Siberian Branch of the Russian Academy of Sciences for granting access to microscopic equipment and Genomics Core Facility of the Siberian Branch of the Russian Academy of Sciences for sample sequencing. We are grateful to Dr. Maria Lagarkova (Instutite of General Genetics, RAS) for providing HEK293 Phoenix cell line. We are grateful to Dr. A. Strunov (ICG SB RAS) for critical discussion. The English language was corrected and certified by [shevchuk-editing.com](http://shevchuk-editing.com). [^1]: The authors have declared that no competing interests exist.

# Introduction Ecological restoration efforts to conserve both biodiversity and ecosystem services are increasingly common. The success of restoration remains poorly known, however, because of tendencies to monitor only a few ecosystem components and for only a few years after restoration activity. The majority of projects are evaluated on progress towards restoring physical structure (flora or physical features) or vertebrate species of concern, while rarely measuring effects on other taxa or ecosystem processes. Among more systematically monitored projects, many fail to achieve the targeted ecosystem’s species composition, structure, or function. Arthropods play important roles in ecosystems as pollinators, predators, parasites, and prey. Their small size, short life cycles, and large numbers facilitate use as indicators of overall biodiversity and ecosystem stability. The use of arthropods to monitor restoration progress, however, also has a notable disadvantage: the scarcity of life history and ecological data on most arthropod species make some ecological analyses difficult. Among studies of arthropod response to restoration of native plants, some report successful arthropod community restoration compared to the reference ecosystem in the long term (30 years), and even in the short term (\<6 years). The definition of “success” varies. One of these studies, for example, found common arthropods in similar densities on planted and naturally occurring shrubs in California scrubland, but that planted shrubs were less likely to support rare species. Other studies have found markedly dissimilar communities between restored and reference sites in both the short and long term, ; one documented greater butterfly richness and abundance in restored areas than in control areas. Driven by mixed results from past analyses of arthropods in restoration, we explore the effect of efforts to restore pasturelands to montane Hawaiian forest on the parasitoid wasp family Ichneumonidae. This family, with ca. 60,000 species, is one of the most biologically diverse insect families in the world. We chose to focus on ichneumonids for a number of reasons. First, the native Hawaiian ichneumonid fauna includes at least three subfamilies and 38 species. Second, Ichneumonids are relatively host-specific parasitoids whose diversity is thought to reflect that of their hosts and sometimes other arthropod groups ; relatedly, their complex trophic roles can lead to substantial impacts on other species when icheumonids are introduced into new ecosystems. Third, they appear to disperse readily to suitable habitat and to indicate environmental change. Another important reason for selecting this taxon is that many native and non- native ichneumonids in Hawai’i parasitize Lepidoptera. Biologists working in Hawai’i have suggested that non-native dominance of ichneumonid communities may impact threatened native birds by depleting their prey through parasitism. The fairly well-known dominance of Hawaii’s ichneumonid fauna by non-natives, does not detract from the taxon’s relevance to restoration due to the extraordinary role played in Hawai’i by non-native species, including flora, arthropods, and birds. Hawai’i provides a rich example of a system for which restoration cannot proceed without accounting for non-native species. We thus seek to understand the impacts of restoration on a biologically informative taxon despite the high proportion of non-native species in its Hawaiian populations. In this study we explore whether planting of a dominant native tree species (*Acacia koa*), along with a number of native understory trees and shrubs, yields an ichneumonid wasp fauna resembling communities found in nearby native Hawaiian forest. We investigated an ongoing, large-scale restoration effort in the Hakalau Forest National Wildlife Refuge in Hawai’i, focusing on distance to native forest, planting configuration, and time since planting. We frame our habitat restoration study around both native and non-native species to address the following questions: 1) Do Ichneumonidae use reforested *A. koa* stands?; 2) How do ichneumonid communities in restored habitats compare with those in native forest, the target ecosystem?; and 3) Does ichneumonid community composition, particularly with respect to native species, vary as a function of a) age of planting, b) planting configuration (patches versus corridors), c) distance from native forest, d) surrounding tree cover, or e) components of the understory plant community? # Materials and Methods ## Ethics Statement We received permission to conduct this study and to collect insect samples in the form of Special Use Permits issued by the U.S. Fish and Wildlife Service in 2007 (SUP# 12516-07018) and 2008 (SUP#12516-08022). No protected species were sampled. ## Study Site We sampled ichneumonids in the Hakalau Forest National Wildlife Refuge (HFNWR), a United States government-owned 10,500 ha reserve on the eastern slope of Hawai’i Island. Native, mature forest covers ca. 7,000 ha of the reserve (the portion below ca. 1,750–1,900 m elevation). Dominant canopy species in this forest are *Acacia koa* (koa) and *Metrosideros polymorpha* (‘ōhi’a); the understory is comprised mostly of native trees and ferns, with some non-native trees, forbs, and lianas. The remaining 3,500 ha at higher elevations, although originally forested, were cleared for cattle pasture in the early to mid-1800’s and are now dominated by non-native grasses. The U.S. Fish and Wildlife Service acquired this land in 1986, primarily to protect and restore prime habitat for Hawaii’s highly endangered forest bird species. The HFNWR initiated reforestation in former pastureland in 1987, 20 years before we conducted this study, and reforestation efforts continue today. We sampled within stands that were planted from the late 1980s through the early 2000s and were 5–20 years old at the time of this study. ## Habitat Types HFNWR’s reforestation program involved planting primarily koa seedlings. Koa was selected for its rapid growth and ability to survive in high-light, non-forest conditions at high elevations. Planting was conducted in two designs: linear corridors stretching uphill from within the mature forest, and roughly rectangular patches of trees in the middle of pasture. Planting locations were selected without reference to soils or geomorphology. Since 2003 there has been regular planting of native understory trees and shrubs within these koa corridors and patches. At the time of this study, natural regeneration under planted koa was limited to several fern species (personal observations) and the shrub *Vaccinium reticulatum*. Corridors are approximately 40 m wide and range from 0.5 to 2.5 km in length. Patches are roughly rectangular, with sides of 70–125 m. The landscape now comprises a variety of different habitat types, referenced here as: 1. *Native forest* – contiguous native forest; the target system of the restoration effort. 2. *Remnant corridors* – corridors of mature (unplanted), native trees persisting in former pasturelands in steep gulches, somewhat protected from former cattle disturbance. 3. *Old planted corridors* – corridors planted with koa 11–20 years before this study; these corridors were planted on relatively flat land. 4. *Young planted corridors* – corridors planted with koa 5–10 years before this study; these corridors were planted on relatively flat land. 5. *Patches* – stands of koa planted 5–20 years before this study. 6. *Grassland* – former pastureland, dominated by non-native grasses. ## Ichneumonid and Vegetation Sampling We sampled plants and arthropods at HFNWR from June–August 2007 and June–August 2008. We surveyed 12 corridors (four remnant corridors, four young planted corridors, and four old planted corridors), five patches of restored koa, and four grassland sites. We identified suitable replicate corridors and stands using aerial images, historical records, and expert field-based knowledge. The entire study site extends about 6 km north-south, and 3 km east-west. All sampling points are between 1,650 and 2,000 m elevation. For each corridor, we surveyed fixed points starting 300 m within the native forest adjacent to the corridor and continuing at 150 m intervals along each corridor’s length. The first points, at 300 m and 150 m inside the native forest, constitute our native forest sampling. At each patch and grassland site, our goal was to collect a representative sample of wasps from the area. In patches, we determined center points using aerial photographs in a Geographic Information System; we measured the length and width of patches, identified the center point, and subsequently found these points in the field using UTM coordinates. We selected sites for grassland sampling by using these same tools to identify the points on the Refuge furthest from the nearest tree cover. We established survey locations at the patch center or grassland site ‘center’ and 50 m from those points at 0°, 120°, and 240°. We combined data from all four sampling locations in each patch and grassland; we situated survey points in three different directions 50 m away simply as a non-biased method for increasing sampling coverage. ### Ichneumonid sampling We surveyed arthropods using six pan traps at each point. Pan traps were 20-cm- diameter plastic bowls, placed on the ground 1–2 m apart in the vicinity of each sampling point. We used three colors (two blue, two yellow, and two white at each point) to attract a diversity of invertebrates, since species may respond differently to the color spectrum. We filled traps approximately 2/3 full of water; to decrease surface tension we added biodegradable soap (1 ml per liter of water). We left the traps out 24 hours (12 hours of daylight) unless weather conditions during part or all of that time were such that wasps were unlikely to be active (e.g., high humidity or high wind conditions). In those cases we left the traps out longer in order to ensure 12 hours of “effective” trapping time. We combined samples from all six traps at each point. We repeated this survey twice in Summer 2007 and three times in Summer 2008, for a total of five summer sampling days at each point. We built a reference collection of Ichneumonidae and identified them to subfamily using Goulet and Huber (1993) and then to genus or species using the “Key to the Ichneumonids of Hawaii”. For one species not in the key, we consulted an expert. We followed the native/non-native descriptions of Hawaii’s Bishop Museum. Our data are publicly available in the Dryad database at [www.datadryad.org](http://www.datadryad.org), <http://dx.doi.org/10.5061/dryad.dr6s2>. ### Vegetation sampling In all but remnant corridors, we recorded the abundance and size class of all vegetation in a 10×10 m² square centered on the sampling point. In the excessively steep terrain of remnant corridors, which typically had woody vegetation that extended only 5 m away from a deep gulch, we sampled vegetation in a 5×10 m² rectangle centered on the sampling point. ## Data Analysis We tested the effects of habitat type, surrounding tree cover, and distance to forest on ichneumonids using a number of analyses. First, we compared ichneumonid richness and abundance across habitat types (section 2.4.1). Second, we analyzed ichneumonid community composition variation across habitat types (section 2.4.1). Third, we examined whether three response variables \[(a) ichneumonid species richness, (b) native ichneumonid abundance, and (c) the similarity of individual sites to the pooled community of ichneumonids in forest habitats\] varied with respect to three explanatory variables \[(a) distance to forest, (b) tree cover, and (c) for abundance of native ichneumonids only, the non-koa plant community\] (section 2.4.2). We do not consider samples within the same corridor, patch, or grassland independent from one another; nor did we consider the temporal replicates from each point independent. We minimize the risk of pseudoreplication using the techniques described below and detailed in the Supporting Information (S1. Elaboration of Methods). ### Comparison of wasp communities across the landscape To assess whether ichneumonids visit restoration plantings differently than they do adjacent unrestored areas, we calculated sampled species richness and individual abundance. To determine statistical significance of observed differences, we used a Generalized Linear Mixed Model, a method described in more detail below and in the Supporting Information (S1). To compare wasp communities across habitats, we used permutational, nonparametric multivariate analysis of variance tests (“PerMANOVA”). We conducted one global analysis, and then pairwise analyses for habitat pairs of interest. We conducted PerMANOVA analyses using the “adonis” function in R. To prevent pseudoreplication in the PerMANOVA tests, we pooled temporal replicates and used only one sampling point from each corridor, patch, and grassland. The Supporting Information (S1) details the points used in each habitat type and two alternative verifications of the results. We conducted non-metric multidimensional scaling to aid in visualization of relationships presented below. We based community similarity tests on Chao similarity coefficients to account for our many rarely encountered species. We also developed a similarity-to- forest index based on Chao similarity coefficients; the index quantifies community composition similarity of each sampling point in the corridors with a set of sampling locations in the target ecosystem (12 points, each 300 m inside the forest edge; see). We calculated this index for each point by calculating Chao similarity coefficients (denoted as *C*) for each planted point (denoted as *k*) compared with each forest point (denoted as *f_i*) (see Eq. 1). We then calculated the arithmetic mean of those 12 Chao similarity coefficients for each planted point (denoted as *Mp*): To convert the results into a more easily interpretable 0–1 format, we standardized values by dividing each *Mp* value by the maximum *Mp* value in the dataset (*Mp_max*): Because the similarity-to-forest index was created from proportional data, we transformed the data using the logit transform so that they closely fit normality assumptions. ### Effects of distance to forest, tree cover, and plant community We used Generalized Linear Mixed-Effects Models (GLMM) to quantify the impact of three explanatory variables \[distance from forest, tree cover, and presence/absence of non-koa plants\] on three response variables \[ichneumonid species richness, similarity-to-forest index, and abundance of native Ichneumonids\]. All response variables exhibited spatial autocorrelation as measured by Moran’s I (Appendix A). The GLMM addressed this autocorrelation by grouping spatially proximate observations (through the specification of corridor as a random effect); sampling points within corridors were not considered independent from one another (spatially or otherwise). Similarly, when all temporal replicates were used in analyses, the GLMM addressed the non- independence of data from the same sampling point (through the specification of sampling point as a random effect). That is, the GLMM approach avoided pseudoreplication by accounting for the non-independence of sampling points while using all spatial and temporal replicates. We specified a Poisson error structure for count data, a binomial error structure for presence/absence data, and a Gaussian error structure for proportional data. All count and presence/absence data were neither overdispersed nor underdispersed, and proportional data fit normality assumptions after logit transformation. We conducted GLMM analyses in R, using the “lmer” function in the “lme4” package. details all GLMM analyses conducted. We calculated local tree cover using ArcMap10.0 and a 2010 WorldView II satellite image with 0.5 m-resolution. We performed a supervised classification to define a tree cover layer, and validated the tree cover classification through GPS-based ground-truthing. We used this tree cover layer to calculate tree cover within a 120 m radius about each survey location (this distance avoids overlap of radii between points; see Appendix A). # Results We collected 7,724 ichneumonid individuals, and identified 96% (7,399) to species and the rest (4%) to genus. Our reference collection of 21 morphospecies comprised 17 species-level identifications, and four morphospecies identified to genus. Of all individuals sampled, 3.5% by abundance (273 individuals) were identified as native to Hawai’i and 97% (265) of these natives were in the genus *Spolas* (we identified these as a single morphospecies, *Spolas* sp.1). The other natives sampled were *Pristomerus hawaiiensis* (six individuals) and *Enicospilus sp.1* (two individuals). Unless otherwise indicated, we pool abundances of the three native species in analyses of native ichneumonids below. We found 18 non-native morphospecies, three of which were numerically dominant – *Agasthenes swezyzii* (3,127 individuals, 41% of total catch), *Vulgichneumon diminutus (*1,381, 18%) and *Woldstedtius flavolineatus* (1,380, 18%). The next most abundant species were an order of magnitude less common. ## Comparison of the Abundance and Species Richness of Ichneumonidae We found that the richness and abundance of ichneumonid communities were both higher in reforested areas (both patches and corridors) than in grassland (GLMM: z = 5.50; *p*\<0.0001 (richness) and z = 6.08; *p*\<0.001 (abundance)) and forest (GLMM: z = −3.54; *p*\<0.0001 (richness) and z = −4.54; *p*\<0.001 (abundance)). Richness values were about twice as great in reforested areas, and abundance values were 5–10 times higher. ## Comparison of Wasp Communities Most of the Ichneumonidae collected in all habitat types were non-native to Hawai’i; non-native wasps comprised 100% of wasps sampled in grassland (74 individuals), 97% of wasps sampled in reforested corridors and patches (5,423, of a total of 5,584), and 92% (971 of 1,059) of wasps sampled in forest. Differences in the ichneumonid communities existed among all six habitat types (F = 4.17, *p = *0.001, df = 4,24). We focus our pairwise comparisons on reforested areas, grassland, and forest, since these habitat types are most relevant for restoration. Ichneumonid community composition differed between the grassland and reforested areas (corridors and patches) (F = 8.6, *p = *0.001, df = 1,19). Ichneumonid communities in the native forest were significantly different from those in all planted sites (patches and planted corridors) (F = 13.49, *p = *0.001, df = 1,23). This difference held for comparisons between native forest and planted corridors alone (F = 16.64, *p = *0.001, df = 1,18) and between native forest and patches alone (F = 5.77, *p = *0.003, df = 1,15). We found higher abundances of native ichneumonids in native forest than in all corridors (z = −2.00, *p = *0.045), in forest than in remnant corridors (z = −2.832, *p* = 0.004), and in planted than in remnant corridors (z = 2.48, *p = *0.013). We found no difference in native ichneumonid abundance between native forest and planted corridors (z = −0.913, *p = *0.361), nor between young and old planted corridors (z = −0.890, *p = *0.373). Ichneumonid communities did not differ by planting configuration (F = 2.33, *p = *0.130, df = 1,11). In addition, communities found in remnant corridors were not significantly different from those in planted corridors (F = 2.29, *p = *0.078, df = 1,10). Nor did ichneumonid communities differ by planting age (F = 1.67, *p = *0.313, df = 1,6). ## Effects of Tree Cover The community similarity-to-forest index was significantly correlated with tree cover within a 120 m radius of each sampling point (t = 10.61; *p*\<0.001;). We found no relationship between native ichneumonid abundance and tree cover (z = 1.33; *p = *0.183). Despite a lack of statistical significance, we observed three general patterns: (1) when surrounding tree cover within 120 m was higher than 90%, native ichneumonids were present about 90% of the time; (2) native ichneumonids were encountered about 60% of the time when surrounding tree cover was between 38 and 90%; and (3) native ichneumonids were not found when surrounding tree cover was less than 38%. ## Effects of Distance to Native Forest We found a positive relationship between native ichneumonid abundance and proximity to forest (z = −4.33, *p*\<0.001). We found no significant relationship between distance to forest and ichneumonid communities, however, presumably because native ichneumonid individuals are so few; the community similarity-to-forest index was not correlated with distance to forest (t = 0.855, *p = *0.393). ## Effects of Non-koa Plant Community Composition Among understory plant species surveyed (Appendix B), two were significantly correlated with the presence of *Spolas* sp.1 (we used only *Spolas* sp.1 in understory analyses because no other native (morpho)species was sufficiently abundant for analysis). Presence of the native understory tree *Myrsine lessertiana* (kōlea) was positively correlated with presence of *Spolas* sp.1 (z = 2.77; *p = *0.006); conversely, presence of the native fern *Pteridium aquilinum var. decompositum* (bracken fern) was negatively correlated with *Spolas* sp.1 presence (z = −2.31; *p = *0.021). # Discussion The Hakalau Forest National Wildlife Refuge restoration effort aims to re-create in former pasturelands a forest that supports the same species that occurred prior to cattle ranching and still occur in adjacent mature forest. HFNWR is particularly concerned with aiding in the recovery of endangered flora and fauna. Although restoring ichneumonid populations is not an explicit management objective for HFNWR, the refuge’s goal of providing habitat for Hawaii’s native species encompasses native ichneumonids. Refuge scientists are also particularly interested in ichneumonid populations because non-native parasitoid wasps may contribute to the decline of bird populations through reduction of important insect prey. Non-native species are an important consideration for most Hawaiian restoration efforts, ; in HFNWR, particular non-natives species of concern include feral pigs (*Sus scrofa*), several highly invasive plant species, and the approximately 25% of the avian population comprised of non-native birds. ## Non-native Species and ‘Novel Ecosystems’ Given that 96.5% of our ichneumonid samples comprised non-native species, at least the arthropod component of our study area could be considered a novel ecosystem. In such novel ecosystems, management approaches that recognize and account for novel species assemblages are needed because complete eradication of non-native species is likely infeasible. While most restoration approaches in Hawai’i are attuned to the roles of non-native species, they often lack important knowledge (for example, of population levels and species interactions) that could guide management decisions. Ecological restoration could be considered a sub-field of “intervention ecology,” a moniker alluding to both the human agency in restoration and the unlikelihood, in many circumstances, of fully re-creating a pre-existing system. The high proportion of non-native species found in our study begets many questions about restoration trajectories and how human “interventions” impact ecological interactions. In the context of a restoration intervention such as this one, for instance, it is possible that the dominance of non-native species may decline as the ecosystem responds to restoration efforts. If and how interactions between native and non-native species change over time in restoration projects, however, are important questions that warrant further inquiry, because detailed information on ecological interactions is often lacking, as it is in our study system. Our analysis did not include lepidopteran larval rearing studies, an important step in determining ichneumonid impact on ecological communities. Given that HFNWR was created to protect increasingly rare native birds whose diets comprise lepidopteran larvae, and given the overwhelming dominance of non-native Ichneumonidae in this system, such studies are especially important. To add to the complexity, we do not know if the overall abundance of ichneumonids is now higher due to the influx of non- natives. Higher populations would likely negatively impact prey species populations, at least in the short term. Conversely, it is possible that non- native ichneumonids are filling niches left vacant by extirpated or rare native ichneumonid species. Further study of this novel ecosystem would help to illuminate these interactions and impacts. ## Patterns Important for Restoration Efforts We found significantly higher abundances and greater species diversity of ichneumonids in planted sites than in adjacent grasslands and in the target ecosystem. We found that ichneumonid communities in all planted sites differed from the target system, which is consistent with many past studies of arthropod communities and restoration, for instance in Wyoming, Arizona, and California. Other past research in agricultural landscapes (as opposed to lands undergoing native habitat restoration) has found greater species richness of bees and wasps in agricultural systems than in protected forest. The consistency of our results with these previous findings leads to questions about the mechanisms underlying higher diversity of arthropod communities in agricultural and restored habitats compared with arthropod communities in native forest. As in other studies investigation of the relationship between different aspects of ichneumonid community composition and site characteristics yields information pertinent to restoration and management. Although plantings had greater overall (native and non-native) ichneumonid abundance and species richness than native forest, the abundance of native ichneumonids did not differ significantly between plantings and native forest. This is an important distinction that is encouraging for native-focused restoration efforts. We found that restoration of native forest may help to increase the abundance of native Ichneumonidae with respect to non-natives. Specifically, we found native ichneumonids only in native habitat and restored areas (not grasslands), and these individuals were more abundant in areas with higher surrounding tree cover. Further studies are needed to explore the effect of restoration on interactions between native and non-native members of this ecological community. In corridors, we found a positive relationship between native ichneumonid abundance and proximity to forest. This result, combined with the lack of a difference in native ichneumonid abundance between forests and planted corridors, confirms that the majority of native ichneumonids in corridors are found closer to the forest. The implication of this finding (that native ichneumonids are venturing from the forest into corridors) is an important consideration for restoration, especially in non-native-dominated systems. This result is consistent with past work, which has found that connectivity to target habitat can increase restoration success. We also found that, in corridors, overall tree cover at a relatively fine scale (120 m) is related to the similarity between a given point’s ichneumonid community and the community in the target native forest ecosystem. Research on other highly mobile taxa, such as birds, flying beetles, butterflies, and moths, similarly found that total area of native vegetation is a strong determinant of target species presence in restoration sites. These two findings, that ichneumonid communities are related to both proximity to native forest and the proportion of tree cover in the surrounding landscape, are consistent with prior work on the distribution of wasps in heterogeneous human-modified landscapes and suggest that landscape context is a critical concern for restoration planning. Our results related to tree cover suggest the importance of trees in particular to ichneumonids. Although we found ichneumonids throughout restoration corridors (up to 1.5 km from native forest), we do not know how far they will fly over non-preferred habitat (in this case, former pastureland) to reach preferred habitat (in this case, trees). Restoration in general should consider that the distance between a restoration site and ‘source’ habitat may be a critical factor for colonization by many taxa, and particularly arthropods. Other factors, however, may outweigh distance, especially for highly mobile arthropods such as butterflies and moths. Each species’ colonization ability will interact with distance between habitats, and for arthropods, as with many other taxa, that interaction will be critical to restoration success. Our results do not address these issues directly, but raise important questions for future research. That neither native wasp abundance nor overall community composition differ by age of planting is not surprising given ichneumonids’ mobility and the relative maturity of even our youngest sites. This result nevertheless demonstrates that ichneumonids visit restoration plantings within 10 years (this study does not reveal whether the ichneumonids were feeding, reproducing, or passing through the restoration sites). Although a past study found arthropod communities more similar to the target ecosystem in older plantings (6–17 years as opposed to 2–4 years) , rapid colonization (in \<5 years) is possible even for taxa with relatively low mobility, such as wingless arthropods. It is possible that plantings in our study are in a slower-growth stage of a non-linear transformation process; that is, major changes may have occurred within the first five years (as in), but subsequent changes (e.g., in years 5–20 after restoration) occur more slowly. Just as age of planting did not seem to impact ichneumonid communities, our results do not suggest that the configuration of planted trees (i.e. patches versus corridors) impacts ichneumonid community composition. This finding is consistent with the scarce past work on arthropods and restoration configurations. Ingham and Samways found that Hymenopterans (wasps, bees, and ants) in South Africa were not restricted to forest fragments and patches; their distributions had no relationship to obvious landscape boundaries. Native *Spolas* sp.1 were widespread at HFNWR, and their occurrence was positively and negatively correlated with a native sub-canopy tree species (*Myrsine lessertiana*) and native bracken fern (*Pteridium aquilinum var. decompositum*), respectively. Future research could focus on understanding the potential for certain plant species in this system to facilitate the return of particular invertebrate species, for example through provision of native host insects. Previous work has demonstrated this role for plants in other systems. In one site in southwest Australia, for instance, most restored plant species supported distinct assemblages of hemiptera (‘true bugs’). In a site in the United Kingdom, insect assemblages in native tree plantations with understory restoration plantings were more similar to target habitat than assemblages in native tree plantations lacking understory plantings. In a third site in California, not only plant species composition but also genotype composition affected arthropod community composition. These past findings suggest important considerations for restoration efforts and, combined with our results, inspire questions about plant-arthropod interactions in our study system. For example, does *Spolas* sp.1 parasitize a species hosted by *M. lessertiana*? Conversely, are other plants (in this case *P. aquilinum*, which spontaneously colonized our study area) associated with factors that make an area less hospitable to native Ichneumonidae? ## Limitations We sampled only in summer months; variations in multiple domains (e.g., floral resource availability; weather) suggest that results could differ with year- round sampling. Inter-annual variation (e.g., droughts or major storm events) could also impact ichneumonid communities. Further, findings might well change with a longer temporal scale than our 16-month period. Additional limitations arise from the spatial arrangement of habitat types. Remnant corridors follow gulches, and thus have different geomorphological characteristics than planted corridors. Young planted corridors are all on the northern edge of the study area and grasslands are nearer the western edge of the study area, while the other habitat types are more evenly distributed. We do not perceive these limitations as great because overall biotic and abiotic conditions in the study area are quite similar. An additional limitation may stem from the likely impact of edge effects (the variation in species populations at the nexus of two habitat types separated by an abrupt edge). Because planted corridors in our study were approximately 40 m wide, entire corridors could be considered edge habitat. While a number of studies have found edge effects for invertebrates between forest and grassland, others have not. Our forest sampling sites that were farthest from an edge were 300 m into the forest; it is possible, and perhaps likely, that edge effects impact these areas as well. One study, for instance, found that edge effects for arthropods (beetles) extended as far as 1 km. Because all corridors were roughly the same width, the likely effect of habitat edges does not jeopardize our results. As restoration proceeds and the corridors widen and eventually meet, these edge effects will be greatly reduced. ## Conclusions The results of this study can inform restoration action and future restoration research. Although most land stewards do not manage explicitly for ichneumonids, these insects may serve as indicators of various ecological processes, and may play critical roles in inter-specific interactions such as parasitism. A better understanding of the effects of time, design (corridor or patch), and landscape context (surrounding tree cover and distance from native, mature forest) on these ecologically important organisms will help landowners, public agencies, and conservation organizations to understand more about the benefits of restoration projects to a broad array of species and will aid efforts to maximize the conservation value of planted forests. Our study suggests that restoration plantings attracted native ichneumonids in numbers similar to native forest, and they also attracted a diversity of non-native wasps. Future restoration interventions may consider implementing restoration plantings in areas with higher surrounding tree cover and biotic connections (e.g., corridors) to native forest, and/or *creating* surrounding tree cover and connections to native forest through restoration plantings, to increase the chance that ichneumonid communities will more closely resemble those of existing native ecosystems. Our findings also reinforce the need for future restoration efforts, especially those in novel ecosystems, to consider the role of non- native species in recently restored ecosystems. # Supporting Information We appreciate the many individuals who helped with this project: Pete Oboyski and Paul Ehrlich for providing guidance on project design and specimen identification; Robert Peck for frequent consultation on Hawaiian Ichneumonidae; Susan Culliney, Peter Olson, Ahn Nguyet Vu, and Annie Wattles for collecting the data in the field; Daniel Karp and Benjamin Woodcock for helpful comments on the manuscript; and Akhila Bettadapur, Japna Sethi, Varshini Cherukupalli, Eugene Lee, and Lindley Mease for their hours of microscope work sorting Ichneumonidae from other arthropods. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: LP GCD. Performed the experiments: LP. Analyzed the data: RKG SGB BB SW CDM. Contributed reagents/materials/analysis tools: RKG GCD SGB. Wrote the paper: RKG GCD LP.

# Introduction Although access to anti-retroviral therapy (ART) has expanded considerably in sub-Saharan Africa (SSA), the number of new HIV infections remains unacceptably high. In SSA, there were an estimated 1.3 million new infections among adults aged ≥15 years in 2015. Men constitute 44% of these, and 36% of the new infections among young people aged 15–24 years. Although HIV incidence is substantially higher in young women than young men in SSA, young men are twice as likely to die of HIV-related causes than their female counterparts. There are strong socioeconomic drivers of the HIV epidemic, and gender inequalities and poverty combine to make adolescent girls and young women particularly vulnerable to HIV infection. However, unequal gender norms also adversely affect men, and in particular, young men. Young men are less likely than women to engage with HIV services across all stages of the HIV prevention and care cascade. Studies in SSA have consistently shown that HIV testing coverage is lower among young men than among young women. In a recently- completed Treatment as Prevention (TasP) trial in rural KwaZulu-Natal, South Africa, which included intensive 6-monthly home-based and mobile testing with evening and weekend provision, less than half of eligible men tested. Young men are also less likely to link to care, and to start and stay on ART. Estimated ART coverage in SSA among HIV positive women aged ≥15 years was 62% in 2016, compared with only 44% among men. In addition to resulting in poorer clinical outcomes, low rates of HIV testing and ART initiation among young men increases the risk of onward transmission to their HIV negative partners. Stigma, gender norms, perceived negative attitudes of health care providers, and structural barriers to access–such as time and cost of visiting a facility, distance, waiting times and opening hours–remain serious challenges to the delivery of HIV services to young people\[–\]. Young men often have other, more immediate, priorities, such as income generation, developing sexual partnerships, family support, and maintaining social networks. Health facilities are seen as ‘for women’, and health services often reinforce those beliefs, assuming that young men do not need or will not access services. This gender gap in access to health services is a persistent issue that has long been recognised, in SSA and elsewhere. However, studies have shown that well-designed interventions that are integrated into the community can engage men to change gender-related attitudes and practices around health care\[–\]. Voluntary medical male circumcision (VMMC) has been shown to reduce the risk of female-to-male HIV transmission by up to 60%, and is recognised as a highly cost-effective HIV prevention intervention. VMMC is only one component in combination HIV prevention, but has advantages in being a single event that does not require ongoing adherence, offers men lifelong benefits, and is a valuable entry point for providing a broader range of health services to men including HIV testing. Since 2007, extensive efforts have been made to scale up VMMC in settings with high HIV prevalence. In 2016, UNAIDS launched a new framework for accelerating HIV prevention and improving the health of adolescent boys and men, with VMMC at its core. The framework has set a target of 90% VMMC coverage for males aged 10–29 years in priority settings, including South Africa. Reaching young men aged 20–29 with VMMC is particularly important, as it will provide some of the greatest efficiency and HIV prevention impact. However, achieving the target VMMC coverage will require new service delivery models, and innovative approaches to overcome current barriers that discourage young men from accessing health care. The DREAMS partnership is an ambitious programme that aims to halt the high incidence of HIV infection among adolescent girls and young women in SSA through rapid scale-up of a combination HIV prevention package to target biological, behavioural, social and structural sources of risk. Whilst the main target population is adolescent girls and young women, the programme also aims to reduce the risk of HIV transmission from male partners. This hinges on improved uptake of HIV testing for male partners and programmes to link those who test positive to HIV treatment, whilst providing VMMC, condom promotion and sexual risk reduction interventions for those who are HIV negative. This study had two objectives: 1) to describe HIV incidence in young men aged 20–29 years in a hyper-endemic rural area of KwaZulu-Natal, South Africa, where the DREAMS partnership is being implemented; and 2) to describe their uptake of HIV prevention services. HIV prevalence among young men in this area, although lower than in women of the same age, is substantially higher than in other areas of South Africa, and has increased over time. When examining uptake of services, we focus on VMMC and HIV testing. These are two components of HIV prevention in male partners of adolescent girls and young women that are being targeted in the DREAMS partnership, and on which we have collected annual data since 2010, allowing us to examine trends over time. We use data from population-based surveys before the roll out of DREAMS, which provides a baseline against which change can be measured after the implementation of DREAMS. # Methods ## Setting The African Health Research Institution (AHRI) (formerly the Africa Centre for Health and Population Studies) is a Wellcome-Trust funded research institute in South Africa. Since 2000, household-based surveys have been used to collect demographic data from a population of approximately 100,000 individuals living in an area of 438 km² in rural uMkanyakude District, northern KwaZulu-Natal. Households are surveyed 2–3 times per year, to collect information on birth, deaths and migration patterns for all household members, including non-residents. In addition, since 2003, resident household members who are aged ≥15 years are invited to participate in an annual individual-level survey, which includes an interview on general health and sexual behaviour, and collection of a dried blood spot (DBS) for anonymised HIV testing. DBS results are for research purposes and are not routinely returned to participants. Eligibility lists for the individual-level survey are drawn up based on age and residency status in December of the preceding year. The cohort is one of the world’s largest population-based longitudinal HIV surveillance studies, situated at the epicentre of the dual HIV and TB epidemics, with information on a wide range of health-related and social factors measured over 17 years of follow-up. ## Population The analysis of HIV incidence uses household demographic and annual HIV serosurvey data from 2006 to 2015, as the period in which ART was widely available in the area. Individuals were included in the analysis if they were male, aged 20–29 years, resident in the surveillance area, participated in the HIV serosurvey at least twice during the period 2006–2015, and their first test was HIV negative. For the analysis of uptake of HIV prevention and care services (‘service uptake’), all men aged 20–29 years who were resident in the surveillance area and participated in the general health survey in either 2011 or 2015 were included. These time points were chosen to include the most recent survey before the introduction of DREAMS in early 2016, and an earlier survey to examine any secular trends in service uptake before the introduction of DREAMS. ## Statistical methods Data were entered and verified in an SQL database, and were analysed using Stata 14. HIV incidence rates were calculated as the number of seroconversions per person-year of observation. Person-time was defined from the date the individual first tested negative in the serosurvey until the latest of date of last negative test or seroconversion date. Periods of non-residency (if individuals moved away from the surveillance area and later returned) were excluded from the calculation of person-time. Seroconversion dates were multiply-imputed (using 250 imputations) as a fraction of the interval between the last negative test date and the first positive test date, assuming a uniform distribution. Imputed seroconversion dates which did not fall within a period when the individual was resident in the surveillance area (e.g. if the individual had temporarily out- migrated) were censored at the latest exit date (the end of the period when the individual was last resident) before the imputed seroconversion date, i.e. the seroconversion event did not contribute to the numerator, and person-time after the residency period did not contribute to the denominator. HIV incidence rates (with 95% confidence intervals (CI)) were estimated separately for each age group (aged 20–24 years and 25–29 years) and calendar period (2006–2010 and 2011–2015). To assess trends over time, Poisson regression was used to estimate rate ratios (RR) and 95% CIs for the effect of calendar period on incidence. As a sensitivity analysis, periods of non-residency were included in the calculation of incidence (i.e. all seroconversions contributed to the numerator, and person-time during periods of non-residency contributed to the denominator). Uptake of services was measured through two self-reported outcomes: 1) having had an HIV test in the past 12 months; and 2) having undergone voluntary medical male circumcision (VMMC). HIV testing was assessed based on the responses to the following questions: a) do you know your HIV status; and if yes, b) when did you last test for HIV. Since DBS results from the annual serosurvey are anonymised and not returned to individuals, participation in the serosurvey was not considered as having tested for HIV. VMMC was assessed based on the responses to the following questions: a) are you circumcised; and if yes, b) where was the procedure conducted; c) was the procedure done for cultural reasons, health reasons, or both. Participants who reported having been circumcised in a medical centre or for medical/health reasons were considered to have undergone VMMC; all others were considered not to have undergone VMMC. Participant characteristics were tabulated for each survey year, stratified by age group. Logistic regression was used to estimate odds ratios (OR) and 95% CIs for factors associated with uptake of each service; separate models were developed for each outcome. Robust standard errors were used to account for repeat participation by some participants between survey years. Analyses were not weighted for survey non-participation or non-response. Potential determinants of HIV testing, and of VMMC uptake, were examined using a conceptual framework with three levels: sociodemographic factors, general behavioural factors, and sexual behaviour. For each outcome, age and survey year were considered a priori confounders and were included in all models. First, sociodemographic factors whose age- and survey year-adjusted association with the outcome was significant at p\<0.10 were included in a multivariable model; those remaining associated at p\<0.10 were retained in a core model. Behavioural factors were then added to this core model one by one. Those that were associated with the outcome at p\<0.10, after adjusting for sociodemographic factors were included in a multivariable model and retained if they remained significant at p\<0.10. Associations with sexual behaviour factors were subsequently determined in a similar way. Since most of the sexual behaviour questions are only asked of individuals who report having had sex in the past 12 months, the analysis of sexual behaviour was restricted to this sub-group. ## Laboratory methods Laboratory testing was performed according to manufacturer’s instructions and standard operating procedures in the AHRI laboratories, Durban. DBS were tested for HIV antibody using an enzyme linked immunosorbent assay (ELISA). From 2006‒2008, Vironostika HIV Uni-Form II Ag/Ab (Biomerieux) was used, with the GAC ELISA (Biorad) as a confirmatory test. From 2008–2016, the SD HIV 1/2 ELISA (V3) was used. ## Ethics Ethical approval for the demographic surveillance study and analyses of these data was granted by the Biomedical Research Ethics Committee of the University of KwaZulu-Natal, South Africa. Separate written informed consent was obtained from all participants for the main household survey, the individual general health and sexual behaviour questionnaires, and the HIV serosurvey. For participants aged \<18 years, written informed consent was obtained from the parent/guardian, and written informed assent was obtained from the participant. # Results ## Survey response rates ### Participation in the annual HIV serosurvey for HIV incidence and prevalence estimates During 2006–2015, there were 16,050 young men aged 20–29 who were resident and therefore eligible for at least one of the annual HIV serosurveys; 84% were eligible in more than one survey round. The majority (\>90%) were successfully contacted; at the time of contact, 30–45% of individuals were no longer eligible, mostly owing to out-migration. In the survey years from 2006–2011, men aged 25–29 years were less likely than those aged 20–24 to still be eligible at the time of contact; however, from 2012–2015, there was no difference between the age groups in the proportion who remained eligible. Among those who were contacted and still eligible for the survey, annual participations rates ranged from 25%–35% among men aged 20–24 years, and 21%–32% among men aged 25–29. Participation rates in 2011 and 2015 (for HIV prevalence estimates) were 33% and 34% among men aged 20–24 years, respectively, and 29% and 26% among men aged 25–29 years. Overall, 5223 (45% of 11,605 contacted and still eligible at time of contact) young men ever participated in at least one of the HIV serosurveys. Of those, 90% were HIV negative at the first HIV test, and 93% were eligible in a subsequent survey round. 3311 contributed data to the HIV incidence analysis (i.e. tested at least twice and were HIV negative at the time of the first test). ### Participation in general health survey for analysis of service uptake 4491 and 4615 young men aged 20–29 were on the eligibility lists for the individual-level surveys in 2011 and 2015, respectively, of whom \>90% were contacted. At the time of contact, 1541 (38%) and 1602 (35%) individuals, respectively, were no longer eligible, mostly owing to out-migration. Among those who were contacted and still eligible for the survey, 1311 (52%) young men in 2011, and 1221 (41%) in 2015, consented to participation in the general health survey; 2218 men participated in only one of the survey years, and 157 participated in both years. In both survey years, men aged 20–24 years were more likely to participate than those aged 25–29 (54% vs 47% in 2011; and 45% vs 35% in 2015; p\<0.001). In general, participation rates were lower in urban areas than in rural or peri-urban areas. ## Characteristics of study participants The majority of survey participants were never married (94% in 2011, and 99% in 2015;). Overall, 37% of those aged 20–24 and 55% of those aged 25–29 had completed secondary education, with a significant increase between the two survey years. Unemployment was high, with only 17% of those aged 20–24 and 39% of those aged 25–29 having either full- or part-time current employment, and no evidence of a difference between the survey years. The majority of participants had been members of a household in the surveillance area since the start of the surveillance cohort in 2000 (83% in 2011, and 80% in 2015, with no difference between the age groups). In 2011, 69% of participants had out-migrated from the surveillance area at least once previously and returned; that proportion had reduced to 52% in 2015 (p\<0.001). Overall, men aged 25–29 were more likely to have out-migrated than those aged 20–24 (66% vs 58%). Overall, 67% of young men aged 20–24 and 92% of those aged 25–29 reported having had sex in the past year, with a small decline between the survey years in both age groups. Reported condom use declined significantly between the two survey years, in both age groups, with only 52% of sexually active participants reporting condom use at last sex in 2015. Furthermore, condom use was similarly low among men who had undergone VMMC and those who had not (52% vs 53%, respectively). Over a third (37%) of young men aged 20–24 years reported having a female partner aged 15–19 in the past 12 months, and 62% reported a partner aged 20–24; median partner age difference was –2 years (interquartile range, IQR, –3 to 0), with no evidence of a difference between the two survey years. Only 5% of men aged 25–29 years reported having a female partner aged 15–19, and 55% reported a partner aged 20–24; median partner age difference decreased from –3 years (IQR –4 to –2) in 2011 to –2 years (IQR –4 to –0) in 2015 (p\<0.001). ## HIV prevalence and incidence HIV prevalence among young men aged 20–29 years participating in the HIV serosurvey was 12.7% (95%CI = 10.4%‒15.3%) in 2011 and 14.6% (95%CI = 12.4%‒17.0%) in 2015, and significantly higher among the older age group in both survey years (27.6% vs 7.2% in 2015). HIV incidence during 2011–2015 was 2.6 per 100 person years (95%CI = 2.0–3.4) among young men 20–24 years and 4.2 (95%CI = 3.1–5.6;) among those aged 25–29 years. Although incidence in each age group was slightly lower than during 2006–2010, there was no evidence of a significant difference in the rate of new infections between the two calendar periods. However, there was some evidence of a decreasing trend in annual incidence among men aged 20–24 years from 2011 to 2015 (RR for linear trend from one year to the next = 0.86, 95%CI = 0.75–0.97, p = 0.02;). In the sensitivity analysis including periods of non-residency, the estimates of incidence were very similar and the conclusions regarding change over time were the same. ## Uptake of services In 2011, less than half (47%) of young men aged 20–29 years reported having tested for HIV in the past 12 months, with no difference between the age groups. In 2015, there was some evidence that men aged 25–29 were more likely than those in the younger age group to report having tested (50% vs 45%, p = 0.06). In 2011, only 5% of young men reported having undergone VMMC, with no difference between the two age groups. By 2015, that proportion had increased significantly, with 40% of those aged 20–24 years and 20% of those 25–29 reporting VMMC. The majority of participants (89% in 2011 and 94% in 2015) had heard about ART and knew of places where people could get ART if needed. Under a third of participants reported having visited a clinic of any type in the past 12 months, with no difference between the survey years; the proportion was higher among the older than the younger age group (35% vs 28%). ## Factors associated with HIV testing In the unadjusted analysis, older age, having above primary education, higher socioeconomic status (SES), history of having out-migrated from the survey area, knowledge of where to get ART and being sexually active were positively associated with having tested for HIV in the past 12 months. After adjusting for age group and survey year, those associations still remained. In the final adjusted analysis, education level was the only sociodemographic factor that remained associated (at p\<0.10) with HIV testing (adjusted (a)OR = 1.63, 95%CI = 1.15–2.30, comparing those who had completed secondary with those who had no secondary education). Behavioural factors that remained association with HIV testing were being sexually active (aOR = 1.98, 95%CI = 1.54–2.55), and a history of out-migration (aOR = 1.38, 95%CI = 1.03–1.84, comparing those had out-migrated two or more times with those who had never out-migrated). Among participants who were sexually active, those who reported \>1 partner in the past 12 months, or using a condom at last sex, were more likely to have tested for HIV. Those who reported having had a casual partner in the past 12 months were less likely to have tested (aOR = 0.53, 95%CI = 0.37–0.75) ## Factors associated with VMMC uptake In the unadjusted analysis, younger age, 2015 survey year, having above primary education, higher SES and knowledge of where to get ART were positively associated with VMMC uptake. VMMC uptake was lower among participants who were employed, and those who had a history of having out-migrated from the survey area. After adjusting for age group and survey year, the association between education and VMMC uptake still remained, and there was some evidence that those with higher SES were more likely to report VMMC uptake. There was strong evidence that participants who were employed were less likely to report VMMC. There was no longer any evidence of an association with migration history or with knowledge of where to get ART. Although there was no evidence of an association in the crude analysis, after adjusting for age and survey year, participants who were sexually active were more likely to report VMMC uptake. In the final adjusted analysis, uptake was positively associated with survey year (aOR = 8.04, CI = 5.95–10.88 comparing 2015 with 2011), education above primary (aOR = 2.52, 95%CI = 1.32–4.82 comparing those who completed secondary education or higher with those who had no secondary education), and being sexually active (aOR = 1.64, CI = 1.19–2.26). Older males (aOR = 0.47, CI = 0.36–0.62) and those who were employed (aOR = 0.66, 95%CI = 0.49–0.89) were less likely to report VMMC. Among participants who were sexually active, those who reported a female partner aged 15–24 years were significantly more likely to report VMMC uptake (aOR = 2.08, 95%CI = 1.19–3.67). There was weak evidence that those who used a condom at last sex were less likely to report VMMC (aOR = 0.76, 95%CI = 0.55–1.05, p = 0.10). After adjusting for all factors in the final model, there was some evidence that males who reported VMMC uptake were less likely to be HIV positive (based on serosurvey results in the same year) than those who did not (aOR = 0.62, 95% CI = 0.35–1.10, p = 0.10). The association was stronger in an analysis restricted to those who were sexually active (aOR = 0.51, 95% CI = 0.26–1.01, p = 0.05). # Discussion Overall, our findings indicate that rates of new HIV infections among young men aged 20–29 years in this rural area of KwaZulu-Natal are very high, and although somewhat lower in more recent years, do not appear to be declining significantly. Uptake of available services in general is low, with less than half of men having tested for HIV in the past year, and under a third having visited a clinic. There are encouraging signs in the substantial increase in VMMC uptake between the two survey years, with an 8-fold increase in VMMC uptake among the younger age group. However, despite these increases, VMMC coverage is still well below the UNAIDS target of 90% coverage in males aged 10–29 years. The proportion reporting recent HIV testing is also well below the first ‘90’ of the UNAIDS targets (90% of HIV positive individuals know their status, 90% of those diagnosed are on ART, and 90% of those on ART are virally supressed), and does not appear to have changed over the period 2011 to 2015. Our results suggest that the focus of the DREAMS partnership on young men primarily as a way to reduce the risk of HIV transmission to adolescent girls and young women may not be sufficient. Instead, the high HIV incidence in young men, and their lack of engagement with sexual and reproductive health services, indicates that the sizeable gaps in uptake of key HIV prevention services among young men should be a priority focus in its own right. There is a need for more innovative prevention and testing methods, and pre-exposure prophylaxis (PrEP) should be considered a priority for men as well as for women. This will ultimately serve to decrease the stigma surrounding the use of PrEP, and increase its preventive potential. Although overall HIV incidence in men aged 20–29 years during the 5 years from 2011–2015 was not significantly different from that during the preceding 5 years, there was some evidence of a decline in annual incidence during the more recent years in the sub-group of men aged 20–24 years. This may be, in part, a reflection of the increased uptake of VMMC in this age group. A decline in HIV incidence in men aged 15–54 years from 2012 to 2015 in this area has been reported by other authors; they hypothesise that the decline may reflect increased ART uptake and viral suppression in women through the successful roll out of Prevention of Mother to Child Testing option B+, thus protecting men from HIV acquisition. We found that there was some evidence that young men who perceived they were at risk to be more likely to test for HIV, although testing was lower among those who reported a causal partner in the past 12 months. HIV testing was higher among those with a history of out-migration, although it not known whether this reflects increased access to testing in other areas, or increased perception of risk. Less than a third of young men reported having visited any type of clinic in the past year, indicating generally poor engagement with health services. These findings suggest that young men are aware of risk, thus the main bottleneck in service uptake may be the model of service delivery. There is a need for more appropriate local services for young men, and given the high prevalence of HIV in the area, interventions to encourage all sexually active men to test regularly. Further studies may be needed to explore this in more depth. We also found evidence that some risk behaviours (e.g. condomless sex) appear to be increasing over the period of the two surveys, independent of VMMC. Similar trends have been reported in a population-based South African national survey New approaches may be needed to engage communities in condom provision and enhance the perception of condoms among young men. Although condom distribution in South Africa is higher than in other countries in SSA, with an estimated 6% surplus relative to condom need in 2015, barriers to usage still remain, especially among young people. Condom programmes need to be integrated into comprehensive sexual and reproductive health education, and must be responsive to the needs of young people. In the adjusted analysis, we found that young men who were employed, aged 25–29 years, or less educated were less likely to have undergone VMMC. However, unlike HIV testing, there was no evidence of a difference in VMMC uptake between those who had previously out-migrated and those who had remained in the area. Tremendous effort has been put into scaling up of VMMC in 14 priority countries in SSA, including South Africa. Although the coverage of VMMC has expanded rapidly in South Africa, the number of annual circumcisions has remained stable over the last few years, and achieving the target 90% coverage in males aged 10–29 will require innovative measures to expand access and overcome current barriers to services. The DREAMS partnership includes VMMC as a key element. Male-friendly service delivery approaches–including school-based campaigns, mobile services, extending clinic hours and community delivery of some services–may improve access and uptake. However, delivery models that are convenient for one age group may not be successful with others. Engaging young men in the design and implementation of these programmes will be essential. Both HIV testing and VMMC uptake were associated with education level, with uptake of each service being highest among young men who had completed secondary education. There has long been an emphasis on increasing educational attainment in girls as an integral part of achieving gender equality, increasing young women’s economic empowerment and reducing their risk of HIV. However, secondary education has a strong impact on the attitudes of men too, and our findings suggest that educational attainment may be equally important for HIV risk reduction in young men. Strengths of our study include a large population-based survey, with detailed interviews conducted annually and an annual HIV serosurvey. This allowed us to examine trends over time in HIV incidence and in uptake of available HIV services. Our method of calculating HIV incidence censors imputed seroconversion dates that do not occur during a residency episode, and excludes periods of non- residency from the calculation of person-time at risk; however, our findings and conclusions were similar when the non-resident intervals were included, indicating that our results are robust to these assumptions. Limitations of the study include that HIV testing and VMMC, as well as sexual behaviours, were self-reported in a face-to-face interview, thus may be subject to social desirability bias. We cannot discount the possibility that some of the apparent increase in VMMC uptake between survey years, or difference between the age groups, may be a result of reporting bias. Such bias may also differentially effect younger men, as a result of campaigns aimed at boys and young men to encourage VMMC as the ‘right’ thing to do. Participation rates in any one annual survey were relatively low, although are consistent with reported participation rates in young males in this area. Among those contacted, around a third were no longer eligible at the time of the visit, mostly as a result of having out- migrated. As a result, our findings may not be generalisable to the more mobile groups of the population. As of 2017, eligibility lists will be drawn up throughout the year, so should provide a more accurate picture of current residents. Lastly, the surveys are cross-sectional, so it is difficult to discern the direction of causation. In summary, we found a very high incidence of HIV among young men aged 20–29 years, with little evidence of decline over the past 10 years. Uptake of HIV testing and other health services was low. Although overall VMMC coverage was less than a third, uptake had increased dramatically over the period 2011 to 2015, especially among men aged 20–24, suggesting a demand for this service. Whilst this shows the potential for DREAMS to have an impact on the health of young men, there is good evidence that designing interventions with young men for young men should be considered, to protect this generation of men at very high risk of HIV acquisition and HIV-related morbidity. The focus of the DREAMS partnership on young men primarily as a way to reduce the risk of HIV transmission to adolescent girls and young women may be a missed opportunity to engage men with health services. # Supporting information We thank the communities of the Hlabisa sub-district for their continued support and participation in AHRI demographic and health surveillance, and to the AHRI field staff who collected the data, the data management team, and the DREAMS IE Publications & Presentations Group who reviewed the paper. [^1]: The authors have declared that no competing interests exist.